Canonical covariate columns

WT (canonical for body weight (baseline or time-varying))

• Description: Body weight (baseline or time-varying).
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – used with allometric scaling (WT / ref_wt)^exponent. Reference weights observed: 70 kg (adults), 75 kg, 84.8 kg, 56 kg (Kloprogge 2014 quinine cohort typical), 5 kg (infants), 25 kg (Wang 2012 Chinese pediatric epilepsy cohort median), 37.1 kg (Damnjanovic 2023 Serbian paediatric epilepsy cohort mean).
• Source aliases:
  • WEIG – weight column abbreviation used by Wang 2012 (Acta Pharmacol Sin 33:845-851); same biological quantity in kg, no value transformation. Used in Wang_2012_levetiracetam.R.
• Example models: Clegg_2024_nirsevimab.R, Hu_2026_clesrovimab.R, Zhu_2017_lebrikizumab.R, Kovalenko_2020_dupilumab.R, CarlssonPetri_2021_liraglutide.R, Cirincione_2017_exenatide.R, Grimm_2023_gantenerumab.R, Grimm_2023_trontinemab.R, Kyhl_2016_nalmefene.R, Soehoel_2022_tralokinumab.R, Xie_2019_agomelatine.R, PK_2cmt_mAb_Davda_2014.R, phenylalanine_charbonneau_2021.R, Chua_2025_mirikizumab.R, Jackson_2022_ixekizumab.R, Kotani_2022_astegolimab.R, Ma_2020_sarilumab_anc.R, Ma_2020_sarilumab_das28crp.R, Moein_2022_etrolizumab.R, Tiraboschi_2025_amlitelimab.R, Robbie_2012_palivizumab.R, Bajaj_2017_nivolumab.R, Quartino_2019_trastuzumab.R, Wang_2020_ontamalimab.R, Fau_2020_isatuximab.R, Okada_2025_rocatinlimab.R, Kunisawa_2014_olprinone.R, Xu_2020_daratumumab.R (reference 78.6 kg; power exponents 0.451 on linear CL and 0.375 on V1), Dirks_2008_cetuximab.R (reference 60 kg; additive linear-deviation effect 0.0083 per kg on V1 of a two-compartment Michaelis-Menten model for cetuximab in SCCHN; source column WGT), Struemper_2017_belimumab.R (reference 67 kg; fixed allometric exponents 0.75 on CL and Q, 1.00 on Vc, 0.8 on Vp; baseline-only, source column BWT), MedellinGaribay_2015_gentamicin.R (linear (not allometric) weight scaling on both CL and Vc: CL = theta1 * BW + theta5 * (CRCL/75), Vc = theta2 * BW; no reference weight used because the scaling is linear, not divisive; source column BW; cohort mean 6.4 +/- 2.2 kg, infants 1-24 months), Wahlby_2004_pefloxacin.R (time-varying; reference 65 kg; linear-exp form on V with coefficient 0.014), Decker_2024_baricitinib.R (baseline-only, source column WTE = weight at entry; reference 74 kg carried from the upstream adult rheumatoid-arthritis model rather than taken from the pediatric cohort; exponents fixed at 0.75 on all clearance arms and 1 on both volumes), Mouksassi_2015_thrombomodulinAlfa.R (baseline; reference 68 kg, the Table 1 combined median of the pooled 392-subject cohort; power exponents ESTIMATED rather than fixed to allometric values – 0.56 (14.15% SE) on CL and 0.50 (21.14% SE) on Vc – and body weight was screened both continuously and as a categorical < / >= 100 kg split, the continuous power form being retained), FernandezTeruel_2024_capivasertib.R (baseline, source column BBW, reference 67 kg; the SAME column enters in TWO different functional forms in one model, which is why the reference weight cannot be read off a single equation: a LINEAR-deviation effect on initial apparent clearance, CL0/F * (1 + (WT - 67) * 0.00585), and a POWER effect on the logit-scale first-order absorbed fraction, logitffo * (WT / 67)^(-1.11); Fernandez-Teruel 2024 equations (a) and (d)), Ekobena_2025_bictegravir.R (baseline, source column BW; reference 70 kg is a ROUNDED value named in the text, NOT the cohort median, which Table 1 gives as 74 kg – a reminder that the reference weight must be read off the covariate equation’s own prose rather than inferred from the demographics table; the exponent 0.35 is ESTIMATED, not fixed to an allometric 0.75, and the paper writes the term as exp(beta * log(BW / 70)) rather than as the algebraically identical power form), Isbister_2015_snake_antivenom.R (baseline, source column wt; power effect on the CENTRAL volume only, with an ESTIMATED and very small exponent of 0.132 whose 84% RSE means it is barely distinguishable from zero. The centring value is the one thing the paper does not print: the Methods say the covariate was “centred to the average weight” but no average appears anywhere, so the model uses the Table 1 MEDIAN of 57 kg as the stand-in. That substitution is nearly free here and the reason is worth recording, because it is the general test for whether an unprinted centring constant matters: the exponent is small and the cohort weight range narrow (40-70 kg), so any plausible mean in that range rescales V by (57/wt_av)^0.132, at most about 3%. Contrast the Ekobena entry above, where the reference weight is named in the text and differs from the cohort median – when the exponent is large, that distinction is load-bearing; here it is not), Francis_2019_rifapentine.R (reference 56 kg, the Table 1 pooled-cohort median; theory-fixed exponent 1 on the apparent central volume of a one-compartment rifapentine model. Worth noting for the head-to-head it records: total body weight beat fat-free mass as the size descriptor for VOLUME by 20 OFV points in the same analysis where fat-free mass beat total body weight for CLEARANCE by 23, so the model scales the two parameters on two different columns. Observed weight range 38-94 kg), Bi_2018_testosteroneCypionate.R (time-varying, entering ONLY through the within-subject deviation (WT - WT_BASE) under the Wahlby 2004 decomposition, in an exponential clearance term centred at +2.90 kg; the size effect proper is carried by the paired WT_BASE column, so a consumer supplying a constant weight column contributes nothing through this term by design).
• Notes: Universal. Verify time-varying vs. baseline-only against the source paper.

AGE (canonical for subject age)

• Description: Subject age in years.
• Units: years
• Type: continuous
• Scope: general
• Reference category: n/a
• Source aliases: none.
• Example models: Archary_2019_lamivudine.R, Budha_2023_tislelizumab.R, Chakraborty_2012_canakinumab.R, Chen_2020_luspatercept.R, Conrado_2014_alzheimer.R, Diepstraten_2013_propofol.R, Gandhi_2021_abatacept.R, Goel_2016_Sonidegib.R, Hennig_2013_tobra.R, Hong_2025_datopotamab.R, Ide_2020_elotuzumab.R, Koopman_2023_factorix.R, Kuchimanchi_2024_dostarlimab.R, Kunarajah_2017_doxorubicin.R, Kyhl_2016_nalmefene.R, Lahu_2010_roflumilast.R, Li_2006_meropenem.R, Li_2017_cediranib.R, Li_2019_abatacept.R, Lin_2024_casirivimab.R, Martinez_2019_alirocumab.R, Melhem_2022_dostarlimab.R, Mulyukov_2018_ranibizumab.R, NA_NA_tte_gompertz.R, NA_NA_tte_lognormal.R, Retlich_2015_linagliptin.R, Rosario_2015_vedolizumab.R, Svensson_2016_bedaquiline.R, Thakre_2022_risankizumab.R, Wu_2024_inotuzumab.R, Yassen_2025_asundexian.R, Yu_2022_ofatumumab.R, Zhong_2026_abatacept.R, Zhou_2021_belimumab.R, Zhu_2017_lebrikizumab.R, Ekobena_2025_bictegravir.R (median-centred FRACTIONAL-deviation form on apparent clearance), GarciaHervalejo_2026_teicoplanin.R (median-centred RAW-deviation linear multiplier (1 + 0.012 * (AGE - 62)) on the clearance of a one-compartment teicoplanin model in adults with haematological malignancies; centring value 62 years, observed range 18-85. Contrast with the Ekobena form noted above: this one is linear in the raw deviation in years and is NOT divided by the median, so the coefficient carries units of per year – which is exactly the distinction that entry warns about, seen from the other side. Two further points. The sign is POSITIVE, i.e. clearance rises with age, and the paper argues in its Discussion that this is not renal ageing but a surrogate for the hypoalbuminaemia and altered body composition of older cachectic patients: teicoplanin is 90-95% protein bound, albumin itself failed screening in this cohort, and the paper attributes that failure partly to collinearity with age. And because the multiplier is linear rather than exponential it turns negative below 62 - 1/0.012 = 8.7 years, so the term is adults-only by construction).
• Notes: Zhu 2017 normalizes as AGE/40. Ekobena 2025 uses a form worth recognising because it is easy to mis-transcribe: the paper calls it a “linear model”, but it is linear in the fractional deviation from the cohort median and sits inside an exponential – exp(beta_Age * (AGE - AgeM) / AgeM) with AgeM = 51 years. The division by AgeM is load-bearing. Dropping it and applying the coefficient to a raw deviation in years turns the paper’s stated “11% decreased CL” at age 80 into a 99.7% decrease. When a paper describes a continuous covariate effect as “linear” and centres it on a median, check whether the printed equation also normalises by that median before encoding it.

AGE_GE70 (canonical for age >= 70 years indicator)

• Description: 1 = subject is aged 70 years or older at baseline, 0 = subject is younger than 70 years. Time-fixed at study entry per subject. Used when the source paper reports an age effect as a threshold indicator (>= 70 y vs < 70 y) rather than as a continuous covariate on AGE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (< 70 years).
• Source aliases:
  • AGE >= 70 – derived from the continuous AGE column via AGE_GE70 = as.integer(AGE >= 70); used in Overgaard_2016_liraglutide.R (Overgaard 2016 Fig. 2 / Table S1 covariate label “Age >= 70 years”).
• Example models: Overgaard_2016_liraglutide.R (multiplicative log-scale effect on CL/F, coefficient -0.10 per Table S1).
• Notes: Sibling to the ECOG_GE1 / ECOG_GE2 binary threshold family; captures a paper-defined age band as a decomposed binary indicator rather than a continuous AGE term. When the source paper describes the age band as “>= 70 y” (as in Overgaard 2016), use AGE_GE70; for other thresholds (e.g., >= 65 y, >= 75 y) register a sibling canonical (AGE_GE65, AGE_GE75). Use the AGE_GT<n> spelling instead when the paper’s threshold is strictly greater than the cut point (see AGE_GT65). The underlying continuous AGE may still be recorded alongside for downstream re-derivation.

AGE_GT65 (canonical for age > 65 years indicator)

• Description: 1 = subject is older than 65 years at baseline, 0 = subject is 65 years old or younger. Time-fixed at study entry per subject. Used when the source paper reports an age effect as a threshold indicator (> 65 y vs <= 65 y) rather than as a continuous covariate on AGE. This is the conventional geriatric cut point in anti-infective and oncology popPK.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (<= 65 years).
• Source aliases:
  • AGE_GROUP (coded 1 = age <= 65 y, 2 = age > 65 y) – derived from the continuous AGE column via AGE_GT65 = as.integer(AGE > 65), equivalently AGE_GT65 = AGE_GROUP - 1; used in Zakria_2026_voriconazole.R (Zakria 2026 Table 2 rows “CL-AGE_GROUP 1” / “CL-AGE_GROUP 2”).
• Example models: Zakria_2026_voriconazole.R (additive-fractional effect on CL, coefficient -0.519 per Table 2 and Eq 2), Lalovic_2020_lemborexant.R (multiplicative factor 0.739 on CL/F; the Methods and Table 3 define elderly as > 65 years although Table 1 prints ‘>= 65 years’).
• Notes: Sibling to AGE_GE70 and to the ECOG_GE1 / ECOG_GE2 binary threshold family. The GT (strictly greater) rather than GE (greater or equal) spelling is load-bearing: a subject aged exactly 65 years belongs to the reference group under AGE_GT65 but to the indicator group under a hypothetical AGE_GE65. Match the spelling to the inequality the source paper prints – Zakria 2026 dichotomises as “<= 65 years” versus “> 65 years”, so AGE_GT65 is correct and AGE_GE65 would misclassify the 65-year-olds. Register AGE_GE65 separately if a paper genuinely uses the “>= 65 y” band. The underlying continuous AGE may still be recorded alongside for downstream re-derivation.

AGE_GT60 (canonical for age > 60 years indicator)

• Description: 1 = subject is older than 60 years at baseline, 0 = subject is 60 years old or younger. Time-fixed at study entry per subject. Used when the source paper reports an age effect as a threshold indicator (> 60 y vs <= 60 y) rather than as a continuous covariate on AGE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (<= 60 years).
• Source aliases:
  • Age>60 – derived from the continuous AGE column via AGE_GT60 = as.integer(AGE > 60); used in Rini_2013_axitinib.R (Rini 2013 Table 2 row “Age >60-yr effect on CL” and the typical-value clearance equation printed in Results “PK model”, in which Age>60 “is 1 if applicable and 0 otherwise”).
• Example models: Rini_2013_axitinib.R (linear-proportional effect on CL: (1 - 0.213 * AGE_GT60), i.e. 21.3% lower clearance above the cut point, per Rini 2013 Table 2 and the Results “PK model” equation).
• Notes: Sibling to AGE_GT65 and AGE_GE70, and to the ECOG_GE1 / ECOG_GE2 binary threshold family; registered under the instruction in the AGE_GE70 entry to register a sibling canonical for each new threshold rather than reusing a neighbouring cut point. The GT (strictly greater) rather than GE spelling matches the inequality Rini 2013 prints (“age > 60 years”, “>60-yr”), so a subject aged exactly 60 belongs to the reference group; register AGE_GE60 separately if a paper genuinely uses the “>= 60 y” band. The threshold is paper-chosen rather than physiological: Rini 2013 first modelled age continuously on CL, then dichotomised it to avoid overestimating clearance at younger ages given a non-uniform age distribution across its 17 pooled studies, and selected 60 years as the most significant of several tested cut points (Supplementary Figures S2-A and S2-B). Do not treat the effect size as transportable to a different cut point. The underlying continuous AGE may still be recorded alongside for downstream re-derivation.

AGE_GE60 (canonical for age >= 60 years indicator)

• Description: 1 = subject is aged 60 years or older at baseline, 0 = subject is younger than 60 years. Time-fixed at study entry per subject. Used when the source paper reports an age effect as a threshold indicator (>= 60 y vs < 60 y) rather than as a continuous covariate on AGE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (< 60 years).
• Source aliases:
  • OLD – derived from the continuous AGE column via AGE_GE60 = as.integer(AGE >= 60); used in Jayanti_2026_pyrazinamide.R (Jayanti 2026 Methods “geriatric (>= 60 years) with DM” and the deposited NONMEM dataset S2 File, whose OLD column spans ages 17-59 where it is 0 and 60-80 where it is 1).
• Example models: Jayanti_2026_pyrazinamide.R (enters only as the product AGE_GE60 * DIS_DIAB, giving Korean patients who are both diabetic and >= 60 years old a 26% higher CL/F; never a main effect).
• Notes: Sibling to AGE_GT60, AGE_GT65 and AGE_GE70, registered under the standing instruction in the AGE_GT60 and AGE_GE70 entries to register a separate canonical for each threshold and each inequality rather than reusing a neighbouring one. The GE (greater or equal) rather than GT spelling is load-bearing and distinguishes this entry from AGE_GT60: a subject aged exactly 60 belongs to the indicator group here but to the reference group under AGE_GT60. Match the spelling to the inequality the source actually uses, and be aware that a paper may print both forms inconsistently – Jayanti 2026 prints “Age > 60 years old” in its Table 1 row header and its S3 / S4 Table footnotes but “>= 60 years” in Methods, Results and the Fig 3 caption, and only the deposited dataset resolves it. When a paper is ambiguous and no dataset is available, prefer the form used in the Methods / Results text that describes the covariate model over the form used in a demographics-table row header. As with the sibling entries, the threshold is paper-chosen rather than physiological, so an effect size estimated at this cut point is not transportable to another one. The underlying continuous AGE may still be recorded alongside for downstream re-derivation.

LBM (canonical for lean body mass)

• Description: Lean body mass.
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a
• Source aliases:
  • LBW (lean body weight) – synonym; same biological quantity (total body weight minus body fat). Hemophilia popPK literature typically uses LBW (Hume or James formula) where mAb / general literature uses LBM. Used in Garmann_2017_BAY81_8973.R (reference 51.1 kg) and Li_2023_TQB3203.R (reference 49.53 kg).
• Example models: Kyhl_2016_nalmefene.R (reference 56.28 kg, exponent 0.626 on CL), Garmann_2017_BAY81_8973.R (alias LBW; reference 51.1 kg, exponents 0.610 on CL and 0.950 on Vc), Schoemaker_2017_brivaracetam.R (alias LBW; paediatric cohort, reference 50 kg adult typical value, fixed theoretical allometric exponents 0.750 on CL/F and 1.00 on V/F), Lee_2025_levofloxacin.R (James formula from sex, total body weight, and height; reference 47.91 kg cohort median, range 37.7-60.3; power exponent 1.75 on the PERIPHERAL volume V2 rather than on CL or Vc, which is unusual for this column – the source screened total body weight, FFM, and NFM as alternatives and retained none), Li_2023_TQB3203.R (alias LBW; reference 49.53 kg cohort median, range 32.09-59.40; screened because of the drug’s high lipophilicity and retained on THREE parameters with exponents of opposing sign – +1.18 on the central volume V1, -1.41 on the shallow peripheral volume V2, and -2.55 on the inter-compartmental clearance CL2 – but NOT on CL, which the authors attribute to insufficient sample size; the underlying body-composition formula is not stated in the paper), Yu_2020_inotersen.R (reference 51.6 kg cohort median, range 31.3-80.3; power exponents fixed at 1 on all four disposition parameters CL/F, Q/F, Vc/F and Vp/F; formula not stated in the paper, which tabulates lean body mass separately from lean body weight).
• Notes: The body-composition formula varies by source paper (James, Boer, Hume) and materially changes the value at a given weight and height; record the formula the source used in covariateData[[LBM]]$notes. The James formulae are LBM (female) = 1.07 * TBW - 148 * (TBW / height)^2 and LBM (male) = 1.1 * TBW - 128 * (TBW / height)^2, with TBW in kg and height in cm. When a paper reports LBM/LBW without naming the formula (as in Li_2023_TQB3203.R), say so explicitly in the per-model notes and record the cohort median and range so a downstream user can check that whatever formula they apply reproduces the source distribution – the formulae differ by several kg at a given height and weight, which directly rescales any median-normalized power effect.

FFM (canonical for fat-free mass)

• Description: Fat-free mass derived from body weight, height, and sex via the Janmahasatian et al. formula (Clin Pharmacokinet 2005;44:1051-1065).
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (FFM / ref)^exponent. Reference values observed: 33 kg (Rolsma 2025 cefepime, paediatric + adult cystic fibrosis), 40.69 kg (Zhou 2021 belimumab pooled adult+pediatric SLE), 45 kg (Aguiar 2021, Crohn’s disease cohort median), 50 kg (Sinha 2026 oxcarbazepine, the “reference adult” fat-free mass), 58.9 kg (Abdul-Aziz 2025 caspofungin, critically ill adult ECMO cohort median).
• Source aliases:
  • LBW (“lean body weight”) – used in Rolsma_2025_cefepime.R and AbdulAziz_2025_caspofungin.R. Same biological quantity in kg with no value transformation. Registered here rather than under LBM because the register discriminates fat-free mass from lean body mass by the estimating formula, not by the source paper’s label, and both papers cite Janmahasatian et al. – the fat-free-mass equation. Abdul-Aziz 2025 makes the identification explicit in its Discussion: “Although the Janmahasatian et al. equation was developed to estimate fat-free mass, the estimate is considered to be a representation of lean body weight and is commonly used interchangeably.” Expect this alias frequently in the critical-care and antimicrobial-dosing literature, where “LBW” is the conventional label for the Janmahasatian quantity.
  • PKWT (“pharmacokinetic weight”) – used in Sinha_2026_oxcarbazepine.R. Same biological quantity in kg with no value transformation, but PKWT is a composite fat-free-mass metric: it equals FFM for subjects aged >= 3 years and total body weight (WT) for children younger than 3 years. The WT fallback exists because the Al-Sallami et al. paediatric FFM equation was developed in children >= 3 years, so Sinha 2026 Methods 2.3.2 assumes a WT-equivalent FFM below that age. A downstream user computing the column for a cohort that includes children under 3 years must apply the fallback; above 3 years PKWT and FFM are the same quantity.
• Example models: Zhou_2021_belimumab.R (reference 40.69 kg; exponents 0.673 on CL and 0.891 on V1), Aguiar_2021_ustekinumab.R (reference 45 kg; power exponents 0.598 on CL, 0.590 on Vc, 0.586 on Vp), Sinha_2026_oxcarbazepine.R (alias PKWT; reference 50 kg; exponent 1 fixed on oxcarbazepine clearance, 0.671 on MHD-metabolite clearance, and a shared 0.752 on both central volumes; paediatric cohort spanning 44 days to 20.9 years including 52% with obesity, where the FFM-based descriptor outperformed total body weight by ~7 AIC points), Rolsma_2025_cefepime.R (alias LBW; reference 33 kg; linear normalisation (FFM / 33) on the central volume with no exponent reported, in children and adults with cystic fibrosis), AbdulAziz_2025_caspofungin.R (alias LBW; reference 58.9 kg, the cohort median; estimated power exponent 1.24 on the volume of a one-compartment model, and the ONLY covariate retained – no covariate was kept on clearance – in 8 critically ill adults on ECMO. Note the exponent’s bootstrap 95% CI is 0.08-1.91, so it is barely distinguishable from the theory-based value of 1 for a volume; a useful precedent that a supra-linear FFM exponent on volume can be reported without being well identified), Tsirizani_2025_ritonavir.R (reference 21.0 kg, the cohort median, corresponding to a child of 26 kg – the anchor Table 3 footnote b names – in 170 African children aged 3.16-15.6 years on boosting-dose ritonavir; theory-fixed exponents 0.75 on CL/F and Q/F and 1 on Vc/F and Vp/F. FFM beat total body weight by 5 OFV points and also beat fat mass. Notable as a case where the reference value is recoverable ONLY from the supplemental NONMEM control stream (TVFFM = 21.0 ;MEDIAN): the main text gives the 26 kg weight anchor but never the corresponding FFM, and the paper never reports which equation produced the FFM data column, citing only Holford & Anderson 2017 for the allometric-size theory. A downstream user must therefore supply FFM directly or pick an equation; the paper’s own 21.0 kg / 26 kg pair is the only in-paper calibration available), Nyangwa_2026_pretomanid.R (reference 45.5 kg, the cohort median, in 94 adults with rifampicin-resistant TB; theory-fixed exponents 0.75 on CL/F and 1 on V/F, and the ONLY covariate retained – FFM beat both total body weight and BMI on base-model fit. A second case, alongside Tsirizani 2025, where the reference value is not printed anywhere: the authors published their nlmixr2 code as supplementary Appendix 1, but it scales on a pre-computed logFFM data column whose centring constant is never stated. The reference is recoverable by arithmetic rather than assumption – exp(lcl) and exp(lvc) equal the Table 2 typical values, which forces logFFM to be zero at the reference, and only a cohort-median reference reproduces the Table 3 median AUC(0-24) of 64,000 ug*h/L; 70 kg would sit 39% high and an uncentred log(FFM) 94% low. Like Tsirizani 2025 the paper never reports which equation produced the FFM column, leaving the cohort’s median weight of 56.8 kg against a median FFM of 45.5 kg as the only in-paper calibration point), Francis_2019_rifapentine.R (reference 46 kg; theory-fixed exponent 0.75 on CL/F in 326 southern African adults with pulmonary tuberculosis. The first registered instance where FFM and total body weight are used as size descriptors for DIFFERENT parameters in the SAME model rather than competing for both: FFM won for clearance by 23 OFV points over total weight, and total weight won for volume, so the model carries (FFM/46)^0.75 on CL and (WT/56)^1 on V. Note also that the reference value is reported inconsistently – Table 2 footnote b and the Results body both say 46 kg, while the Abstract says 45 kg, which is also the Table 1 cohort median. 46 kg is used because it appears in the parameter table’s own footnote; the choice moves typical clearance by only (46/45)^0.75 = 1.7 percent, which is the same “does the unprinted or contested reference actually matter” test the Isbister 2015 entry above works through for weight. Like Tsirizani 2025 and Nyangwa 2026, the paper does not report which equation produced the FFM column), Goyal_2022_vancomycin.R (reference 45 kg, the cohort median, in 34 hospitalized pregnant women; exponent 0.75 on CL and Q, 1 on Vc and Vp. Janmahasatian et al. for patients >= 18 years and Al-Sallami et al. for patients < 18 years, so one column spans two equations; FFM beat total body weight by 12 OFV points. Note that the Janmahasatian formula uses total body weight, which in pregnancy includes the fetus, placenta and amniotic fluid – the source does not adjust for this).
• Notes: Distinct from LBM (lean body mass) which is sometimes computed by the Boer or Hume formulae. When the source paper reports the body-composition formula it used (e.g., Janmahasatian for FFM), record it in covariateData[[FFM]]$notes. FFM is preferred over total body weight when scaling monoclonal-antibody PK because mAb distribution is largely confined to extracellular fluid; muscle / lean tissue tracks extracellular volume better than total weight in heavier patients. Paediatric extractions should record which FFM equation applies over which age band: Janmahasatian et al. (Clin Pharmacokinet 2005;44:1051-1065) is an adult formula, and Al-Sallami et al. (Clin Pharmacokinet 2015;54:1169-1178) supplies the paediatric age-correction multiplier a + (1 - a)/(1 + (PNA/b)^-c) applied to the adult form, with a = 0.88, b = 13.4, c = 12.7 for males and a = 1.11, b = 7.1, c = 1.1 for females. Note that the female numerator (1 - 1.11) is NEGATIVE; Sinha 2026 ESM1 Eq. S4 prints it as +0.11, which is a transcription error – the multiplier must converge to 1 at adolescence so the paediatric form reduces to the adult one.

IBW (canonical for ideal body weight)

• Description: Ideal body weight in kg, typically derived from height and sex using the Devine formula or its variants. Time-fixed at baseline unless the source paper states otherwise. Used in size-normalisation of clearance / dose-rate in adult popPK models where the source paper reports IBW as the preferred size descriptor over total body weight (e.g., when overweight subjects pull clearance scaling away from the typical pattern).
• Units: kg
• Type: continuous
• Scope: specific
• Reference category: n/a – used with a linear ratio (IBW / ref) for clearance / dose-rate normalisation. Reference values observed: 60 kg (Holford 1992 tacrine adult Alzheimer’s population mean IBW).
• Source aliases:
  • IBW – standard abbreviation used directly in Holford 1992.
• Example models: Holford_1992_tacrine.R (reference 60 kg; tacrine “clearance” relative to IBW = 60 enters as the dose-rate normalisation factor (60 / IBW) driving the tacrine effect-compartment input; the Holford-Peace Devine variant is documented in covariateData[[IBW]]$notes), VanWart_2004_garenoxacin.R (reference 64.2 kg; additive linear-deviation effect 0.764 mL/min per (IBW - 64.2) kg on CL/F, plus an in-model obesity flag derived as WT > 1.3 * IBW per Van Wart 2004’s >130% IBW obesity definition), Shoji_2011_pregabalin.R (reference 62 kg; power-of-ratio scaling on CL/F (exponent 0.354) and V/F (exponent 0.819) per Shoji 2011 Table 3 final model; the IBW formula is the Devine-family variant cited as reference [12] of the paper), Dirks_2008_cetuximab.R (reference 64 kg; additive linear-deviation effect 0.0108 per kg on Vmax of a two-compartment Michaelis-Menten model for cetuximab in SCCHN; the source dataset supplied IBW pre-computed, so the Devine-family variant used is not identified in the paper), Stoschus_2025_phenobarbital.R (reference 68.8 kg, the cohort median; classical allometric scaling of a one-compartment model, exponent 1 on V and 0.75 on CL, in critically ill adults with refractory / superrefractory status epilepticus. IBW was the only covariate retained of the demographic, clinical, laboratory, and comedication set screened. The source computed IBW with the formula of Brower et al. (Stoschus 2025 reference 25) but does not reproduce the formula, so the variant is not identifiable from the paper alone), GarciaHervalejo_2026_teicoplanin.R (reference 61 kg, the rounded development-cohort median of 61.4 kg; power-of-ratio scaling (IBW / 61)^3.2 on the clearance of a one-compartment teicoplanin model in adults with haematological malignancies. The steepest IBW effect registered against this canonical by a wide margin – the exponent of 3.2 is over four times the classical allometric 0.75, and it is well identified (RSE 21%, bootstrap 95% CI 2.13-4.78, so the interval excludes any allometric reading). It makes clearance span 0.33-3.05 L/h across IBW 40-80 kg and dominates the model: both of the paper’s headline dosing findings – that IBW <= 40 kg patients are predicted to exceed the 60 mg/L safety threshold at steady state, and that IBW >= 80 kg patients cannot hold 15 mg/L on once-daily dosing – fall straight out of this term. Two consequences for reuse. First, the steepness makes descriptor substitution unusually costly: total body weight exceeds IBW by only 11% at the cohort medians (68 vs 61.4 kg) yet supplying it inflates clearance by 1.11^3.2 = 39%. Second, the source prints the Devine formula with the final operator mis-set as a subtraction, 50.0 + 2.3 * ((height_cm - 152.4) - 2.54), which gives a 170 cm man an IBW of 84.6 kg; the intended division (height_cm - 152.4) / 2.54 gives 65.9 kg and is the only reading compatible with the cohort’s own median IBW of 61.4 kg at a median total body weight of 68 kg. The paper additionally chose IBW over total body weight, adjusted body weight, BMI and BSA in a head-to-head anthropometric screen, attributing the result to the cachexia, fluid overload and malnutrition that make total body weight overstate metabolically active mass in this population), Garreau_2021_vancomycin.R (reference 64.1 kg, Devine formula; enters the central volume of a continuous-infusion ICU vancomycin model as the exponential of a centred power term, V1 = V1pop * exp((IBW / 64.1)^1.88), exactly as printed in the paper’s Table 2 final-model equations, and IBW also enters clearance indirectly through the IBW-based Cockcroft-Gault CRCL. The paper retained IBW over total body weight, arguing that daily fluid shifts make TBW more variable within a patient than vancomycin V1 itself), Periclou_2021_cariprazine_initial.R (reference 64.5 kg, the cohort median; additive-linear (IBW - 64.5) effects on cariprazine CL/F and Vc/F, DCAR CL/F and DDCAR Vc/F, and power-of-ratio (IBW / 64.5)^e effects on DCAR Vc/F (exponent 3.16) and DDCAR CL/F (1.12). The paper’s updated model replaced IBW with total body weight ‘for practical use’ because both performed similarly, so the updated Periclou 2021 model carries WT instead), Duong_2022_gentamicin_hodiamont2017ijaa.R (reference 70 kg; allometric (IBW / 70)^0.75 on CL and linear (IBW / 70) on V1 of a two-compartment gentamicin model in critically ill adults – the Hodiamont 2017 IJAA structure refit by Duong 2022 to a Quebec ICU cohort. The IBW formula is not stated, and the refit dataset recorded no height, so the refit probably held IBW at the 70 kg typical value; the re-estimated IBW coefficients are therefore not informed by Quebec data).
• Notes: Specific scope until a second adult-popPK model ratifies the name; at that point promote to general. Per-model covariateData[[IBW]]$notes should record the formula the source paper used. The Holford-Peace 1992 variant is: men IBW (kg) = 52 + 0.75 * (height_cm - 152); women IBW (kg) = 49 + 0.67 * (height_cm - 152). The classic Devine 1974 formula is: men 50 + 2.3 * (height_in - 60); women 45.5 + 2.3 * (height_in - 60). Other variants (Robinson 1983, Miller 1983, Hamwi 1964) exist; the per-paper choice should be recorded so a user simulating against IBW can match the source’s derivation. When the source dataset supplies IBW pre-computed, the column name is typically IBW directly. When only HT + SEXF are provided, the user must compute IBW externally using the source-paper formula before passing it to the model.

ABW (canonical for adjusted body weight)

• Description: Adjusted (dosing) body weight in kg – the obesity-dosing size descriptor that interpolates between ideal and total body weight, conventionally ABW = IBW + 0.4 * (TBW - IBW). Time-fixed at baseline unless the source paper states otherwise. Distinct from WT (total/actual body weight), from IBW (ideal body weight alone), and from FFM (fat-free mass, a physiologically derived composition estimate rather than a dosing convention): ABW is the weight a prescriber substitutes for total body weight when dosing a hydrophilic drug in an overweight patient, and popPK papers screen it head-to-head against TBW and IBW as competing size descriptors.
• Units: kg
• Type: continuous
• Scope: specific
• Reference category: n/a – used with a linear-deviation form (intercept + slope * (ABW - ref)) or with a ratio/power form (ABW / ref)^exponent. Reference values observed: 60 kg (Jaruratanasirikul 2021, critically ill Thai adults with and without ECMO).
• Source aliases:
  • ABW – the standard abbreviation, glossed as “adjusted body weight” in the Zhang 2025 imipenem review’s Table 3 footnote.
• Example models: Jaruratanasirikul_2021_imipenem.R (reference 60 kg; linear-deviation effect on the central volume, V1 = 13.8 - 0.348 * (ABW - 60) L, retained over actual body weight, ideal body weight and BMI, which were screened alongside it. The effect is linear with a negative slope and therefore has a bounded domain: V1 reaches zero at ABW = 99.7 kg and is negative above it, so a virtual cohort driving this model must be restricted to the study’s body-weight range – see covariateData[[ABW]]$notes in that file).
• Notes: The interpolation factor 0.4 is the usual convention but is not universal (0.3 and 0.5 appear in the literature), and the IBW variant feeding it (Devine, Robinson, Hamwi, …) also varies; record the formula the source paper used in covariateData[[ABW]]$notes, or state explicitly that it is unreported. A source that reports ABW alongside WT and IBW in its screened-covariate list is using ABW in this dosing-convention sense; do not alias it to either. Specific scope until a second model ratifies the name, at which point promote to general. Note that a linear-in-ABW effect on a volume or clearance is only interpretable inside the fitted body-weight range – check the sign of the slope and the implied zero-crossing before simulating outside it.

HT (canonical for body height at baseline)

• Description: Subject body height at baseline. Time-fixed unless the source paper states otherwise.
• Units: cm
• Type: continuous
• Scope: general
• Reference category: n/a – used with a linear-deviation form (HT - ref) or with a power-style scaling. Reference values observed: 167 cm (Naik 2016, vortioxetine adult MDD/GAD population median); 165 cm (Zhang 2018 flurbiprofen and Angeli 2016 healthy non-menopausal women).
• Source aliases:
  • HGT – height (cm) abbreviation appearing in some NONMEM control streams.
  • HEIGHT – spelled-out form (used by Angeli 2016).
• Example models: Naik_2016_vortioxetine.R (reference 167 cm; linear-additive effect 0.40 L/hr per (HT - 167) cm on CL/F, retained over weight and BMI in stepwise selection because it produced the larger reduction in CL IIV), Zhang_2018_flurbiprofen.R (reference 165 cm; linear-multiplicative effect 1 + theta_height * (HT - 165) on the effect-compartment equilibration rate Ke alongside a paired linear-multiplicative WT effect), Angeli_2016_iron_hepcidin.R (reference 165 cm; power-law multiplier (HT / 165)^32.70 on the hepcidin post-menses rebound parameter krel_hep; the very large exponent reflects the narrow height range across the cohort, 158-173 cm at the 10th-90th percentile), Wu_2014_FEV1_asthma.R (per-observation HT in cm enters directly into the exponential FEV1 disease-progression model: FEV1 = exp(theta_age*AGE + theta_ht*HT - theta_int) + ... with theta_ht = 0.0169 1/cm; paediatric population, ages 5-13 y).
• Notes: Height is sometimes retained as a size covariate when allometric scaling on weight performs less well; it is also an input to BSA, BMI, FFM, and LBM derivations, so a model that retains HT alongside one of those derived covariates should document the dependency in covariateData[[HT]]$notes. Promoted from specific to general on 2026-06-03 with the Angeli 2016 iron / hepcidin extraction (the third adult-popPK model to register HT, satisfying the original promotion condition documented when the canonical was introduced).

BSA (canonical for body surface area)

• Description: Body surface area (typically computed by DuBois, Mosteller, or Haycock from height and weight).
• Units: m^2
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (BSA / ref)^exponent or with linear centering (1 + slope * (BSA - ref)). Reference values observed: 1.70 m^2 (Yamada 2025, adult oncology), 1.11 m^2 (Park 2023, paediatric HSCT cohort median).
• Source aliases: none.
• Example models: Yamada_2025_zolbetuximab.R (reference 1.70 m^2; power scaling, exponents 1.06 on clearances and 0.968 on volumes), Roepcke_2023_rezafungin.R (reference 1.90 m^2 = the pooled-cohort median; separately estimated exponents 0.882 on CL, 1.56 on the central volume V1, and 1.17 on the shared peripheral volume V23; BSA replaced body weight and BMI, which the paper notes were highly correlated with it and would have fit comparably; the BSA formula used is unspecified in the source), Park_2023_mycophenolic_acid.R (reference 1.11 m^2; centered-linear scaling Vd/F = 89.83 * (1 + 0.854 * (BSA - 1.11)) on the apparent volume of distribution in paediatric HSCT recipients; founding centered-linear example), Zhang_2015_methotrexate.R (reference 1.62 m^2; centered-linear scaling on BOTH the intercompartmental clearance, (1 + 0.880 * (BSA - 1.62)), and the peripheral volume, (1 + 0.874 * (BSA - 1.62)), in Chinese osteosarcoma patients spanning 0.62-2.21 m^2. Three points worth carrying forward. First, the centring constant 1.62 is the printed-equation value and is deliberately NOT the Table 1 median of 1.63 – the paper’s Methods template centres on the MEAN – another instance of the equation-over-demographics-table rule recorded throughout the CRCL entry. Second, the signs above are CORRECTED: the paper prints a minus in both brackets, which would make the peripheral volume shrink four-fold across the observed BSA range, while the paper’s own Methods covariate template carries a plus and its Discussion states the correlation is positive. Third, with a linear rather than power form and a cohort reaching down to 0.62 m^2, the multiplier falls to 0.12 at the smallest BSA – still positive, but a reminder that a centred-linear size term has a zero crossing that a power term does not, so check it before extrapolating below the fitted range), Yao_2025_flurbiprofen_r.R (reference 2.6 m^2 = the cohort median; power exponent 1.37 on the apparent central volume Vc of R(-)-flurbiprofen; BSA formula unstated and the tabulated values are NOT reproducible from the source’s own height and weight – see Notes), Xu_2026_caspofungin.R (reference 0.79 m^2 in a paediatric intensive-care cohort whose median BSA is 0.660 m^2 – the standardisation constant is NOT the cohort median, and is confirmed only by cross-checking the paper’s own BSA-normalised CL, V1 and ke against the Table 2 thetas; Mosteller formula, stated and reproducible from the source’s own height and weight; FIXED exponents 0.66 shared by CL and Q and 1 shared by V1 and V2, selected over weight, lean body weight and fat-free mass on OFV and AIC).
• Notes: Oncology mAbs dosed by BSA (mg/m^2) often use BSA in place of body weight for allometric-style scaling. Document the BSA computation formula (DuBois / Mosteller / Haycock) the source paper used; if unstated, record “unspecified.” Sanity-check the tabulated BSA against the source’s own height and weight before using it as a normalisation constant. Yao 2025 reports a median BSA of 2.6 m^2 (range 2.0-3.2) for a cohort whose median height and weight are 1.61 m and 70 kg, which give about 1.76 m^2 by DuBois or 1.79 m^2 by Mosteller – the tabulated values run roughly 1.45x high and no published BSA formula reproduces them. Because a covariate exponent is estimated against whatever BSA scale the source used, a model carrying such an effect must be driven with BSA on that same scale (and its virtual cohort must sample BSA from the reported distribution rather than recomputing it from height and weight); silently substituting a correctly computed BSA would shift every predicted volume.

BMI (canonical for body mass index)

• Description: Body mass index at baseline.
• Units: kg/m^2
• Type: continuous
• Scope: general
• Reference category: n/a – used with a linear-deviation form (1 + e * (BMI - ref)) or a power form ((BMI / ref)^e). Document the reference value in covariateData[[BMI]]$notes.
• Source aliases: none known.
• Example models: Chua_2025_mirikizumab.R (reference 24.75 kg/m^2; linear-deviation effect on logit of bioavailability), NA_NA_lidocaine.R (DDMODEL00000281; binary stratification at threshold 27.93 kg/m^2 adding +0.939 to the GX rate constant K30 in the BMI > 27.93 cohort), Struemper_2017_belimumab.R (kg/m^2, reference 24.7; power exponent -0.610 on Vc; baseline-only, source column BBMI), Lalovic_2020_lemborexant.R (power exponent -0.428 on CL/F; reference 25 kg/m^2, unprinted and back-solved from the paper’s quoted BMI effect sizes).
• Notes: Universal clinical-trial demographic. Derived as WT / (height_m)^2; assume time-fixed at baseline unless the source paper states otherwise.

BLOOD_VOLUME (canonical for estimated total blood volume)

• Description: Subject’s estimated total circulating blood volume (plasma plus red-cell volume), in litres, derived from body size and sex rather than measured. A body-size descriptor in the same family as BSA and LBM; baseline unless a model’s notes say otherwise.
• Units: L
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (BLOOD_VOLUME / ref)^exponent. Reference values observed: 3.89 L (Savic 2022, close to the 3.9 L cohort median of adults and adolescents with sickle cell disease).
• Source aliases:
  • BLV – Savic 2022 Table 2 row label ‘Blood volume on Vc/F, (BLV/3.89)^TH’ (Savic_2022_voxelotor.R).
• Example models: Savic_2022_voxelotor.R (baseline blood volume, reference 3.89 L, power exponent 0.74 on voxelotor apparent central volume; Vc/F 269 L at the 10th-percentile 2.9 L and 413 L at the 90th-percentile 5.2 L).
• Notes: Several formulas are in use (Nadler from height, weight and sex; weight-and-sex-only mL/kg rules; Lemmens for obesity) and they differ by several hundred mL at a given body size, so record the formula in covariateData[[BLOOD_VOLUME]]$notes. When the source does not print it – Savic 2022 says only ‘calculated based on body weight and sex’ – say so, and record the cohort median and range so a user can check that the formula they apply reproduces the source distribution. Distinct from organ or tissue blood-volume fractions inside a PBPK model (frb_<organ> in parameter-names.md), which are model parameters, not subject covariates.

BMIZ (canonical for body-mass-index z-score (age- and sex-standardised))

• Description: Age- and sex-standardised body-mass-index z-score (number of standard deviations above or below the reference-population mean BMI for the subject’s age and sex). Distinct from raw BMI (kg/m^2): BMIZ is unitless and centred at 0 in the reference population, so the reference value used in linear-deviation effects is 0 (not a population BMI in kg/m^2). Time-varying when the source paper carries a per-visit z-score; document baseline-vs-time-varying status in covariateData[[BMIZ]]$notes.
• Units: unitless (z-score; standard-deviation units)
• Type: continuous
• Scope: general
• Reference category: n/a – used with a linear-deviation form (1 + e * (BMIZ - 0)) so the reference is 0 (population mean for the subject’s age/sex). Effect coefficients are interpreted as fractional change per 1 z-score-unit deviation from the reference.
• Source aliases:
  • BMI – when a paper uses the column name BMI for what is actually a z-score (e.g., the Harun 2019 NMTRAN control stream column BMI is documented in the dataset header as “body-mass index z-score”). The canonical column is BMIZ; the source-paper column name is recorded in covariateData[[BMIZ]]$source_name.
• Example models: Harun_2019_cysticFibrosis.R (time-varying per-visit BMI z-score; linear-deviation effect on baseline FEV1% predicted with reference 0 and coefficient +0.0382 per z-score unit), Muhamad_2025_cholecalciferol_pbpk.R (baseline WHO 2007 BMI-for-age z-score driving the Monasor-Ortola 2021 cubic that splits body weight into the fat-mass and lean-mass 25(OH)D3 distribution compartments; source column ZBMI).
• Notes: Distinct from BMI (raw kg/m^2 used in adult populations). Paediatric and adolescent studies routinely report BMI as a z-score relative to a growth reference (WHO 2007 Growth Reference for school-aged children, CDC 2000, etc.); document the reference standard the source paper used in covariateData[[BMIZ]]$notes. Promoted to general scope when Muhamad_2025_cholecalciferol_pbpk.R became the second paediatric model to use it (WHO 2007 reference; Harun 2019 is the other).

HAZ (canonical for height-for-age z-score (age- and sex-standardised))

• Description: Height-for-age (or length-for-age) z-score; the standard anthropometric marker of chronic malnutrition / stunting. Unitless and centred at 0 in the reference population. Same conventions as WAZ.
• Units: unitless (z-score; standard-deviation units)
• Type: continuous
• Scope: general
• Reference category: n/a – linear-deviation form, reference 0 unless the source centres on a cohort median.
• Source aliases:
  • LAZ – length-for-age z-score, the same quantity for children measured recumbent rather than standing.
• Example models: Wallender_2021_piperaquine.R (screened on relative oral bioavailability, 4.7% per z-score, delta-OFV -5.93; NOT retained in the final model – carried in covariatesDataExcluded), Chotsiri_2019_lumefantrine.R (screened in the stepwise covariate search; not significant and not retained).
• Notes: Stunting axis. See WAZ Notes for the WT / MAL_NOURISH distinctions, which apply identically.

WHZ (canonical for weight-for-height z-score (sex-standardised))

• Description: Weight-for-height (or weight-for-length) z-score; the standard anthropometric marker of acute malnutrition / wasting, and a WHO severe-acute-malnutrition criterion at WHZ < -3. Unitless and centred at 0 in the reference population.
• Units: unitless (z-score; standard-deviation units)
• Type: continuous
• Scope: general
• Reference category: n/a – linear-deviation form, reference 0 unless the source centres on a cohort median.
• Source aliases:
  • WLZ – weight-for-length z-score, the same quantity for children measured recumbent rather than standing.
• Example models: Wallender_2021_piperaquine.R (screened on relative oral bioavailability, 4.4% per z-score, delta-OFV -6.47; NOT retained – carried in covariatesDataExcluded), Chotsiri_2019_lumefantrine.R (screened alongside MUAC and WAZ; MUAC gave the larger drop in objective function and was retained instead).
• Notes: Wasting axis; distinct from MUAC, the other continuous acute-malnutrition measure (raw mm), and from the binary MAL_NOURISH. See WAZ Notes.

SEXF (canonical for sex)

• Description: Biological sex indicator, 1 = female, 0 = male.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (male).
• Source aliases:
  • SEXM (values inverted: SEXF = 1 - SEXM; effect coefficient sign and reference category both invert) – used in CarlssonPetri_2021_liraglutide.R.
  • SEX with "M"/"F" strings – derive SEXF = as.integer(SEX == "F").
  • SEX with 1=male / 2=female numeric coding – derive SEXF = as.integer(SEX == 2). Used in Netterberg_2017_docetaxel.R and NA_NA_miridesap.R (DDMODEL00000262 source bundle; Sahota 2015 NONMEM convention).
  • FEM (1 = female, 0 = male; same orientation as canonical, no transformation) – used in Guiastrennec_2016_gastric_emptying.R.
• Example models: Zhu_2017_lebrikizumab.R (canonical), CarlssonPetri_2021_liraglutide.R (alias SEXM), Bajaj_2017_nivolumab.R (male-indicator source; effect applied as exp(coef * (1 - SEXF)) to preserve the paper’s female-reference CL_REF / VC_REF), Fau_2020_isatuximab.R (exponential effect on Vc; reference category 0 = male), Netterberg_2017_docetaxel.R (multiplicative effect on baseline ANC: BACOV *= (1 + theta * SEXF); source column SEX with 1 = male, 2 = female encoding, decomposed via SEXF = as.integer(SEX == 2)), NA_NA_miridesap.R (DDMODEL00000262 / Sahota 2015; multiplicative effect on baseline SAP via SAP_BASE_ref * (1 + e_sexf_sap0 * SEXF) with e_sexf_sap0 = -0.30; female baseline is ~30% lower than male), Xu_2020_daratumumab.R (additive shift on V1 (1 + e_sexf_vc * SEXF) with e_sexf_vc = -0.205: female V1 is 20.5% lower than male, reference category 0 = male), Guiastrennec_2016_gastric_emptying.R (multiplicative +40.7% strengthening of the caloric-feedback slope SLPCAL on gastric emptying in females; SLPCAL_eff = SLPCAL * (1 + 0.407 * SEXF)), Wada_2023_sparsentan.R (male-indicator source with a FEMALE reference subject, so the published +0.139 log-scale male effect on CL/F is applied as exp(e_sexf_cl * (1 - SEXF)) to preserve the verbatim coefficient while keeping the canonical 1 = female orientation – the same construction as Bajaj_2017_nivolumab.R; men had ~15% higher CL/F), Wang_2019_oral1cmt_covariate.R (male-indicator source with a FEMALE reference subject, in the FRACTIONAL rather than exponential form: the authors’ TVCL=THETA(1)*CLWT*(1 + CLSEX*SEX) with CLSEX = -0.3 is applied as (1 + e_sexf_cl * (1 - SEXF)), so lcl = log(15) is the clearance of a 70 kg woman and men clear 30% more slowly; verified numerically in the vignette, which asserts CL = 15 L/h for a 70 kg woman and 10.5 L/h for a 70 kg man).
• Notes: When translating a model that used SEXM, document the sign/reference-category inversion in the model’s covariateData notes.

PREG (canonical for pregnancy status indicator)

• Description: 1 = pregnant, 0 = non-pregnant. Usually time-fixed per subject, in trial cohorts that enrol pregnant and non-pregnant women in parallel. It is genuinely time-varying in designs that re-sample the same women after delivery: Sawe_2025_levofloxacin.R pools matched antepartum and postpartum profiles, so a subject carries PREG = 1 on the third-trimester visit and PREG = 0 on the postpartum visit.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (non-pregnant).
• Source aliases: none known; source NONMEM control streams typically use PREG directly.
• Example models: Sawe_2025_levofloxacin.R (time-varying across matched antepartum / postpartum visits; fractional multiplicative effect 1 + e_preg_cl * PREG raising levofloxacin clearance 38.1% in the third trimester, on top of the separately-modelled FFM and serum-creatinine effects. Pregnancy was screened as a three-level categorical covariate – pregnant / postpartum / never pregnant – but only the pregnant level was retained, so postpartum records take the reference category and their intermediate exposures are reproduced by the size and renal-function covariates alone), Birgersson_2019_artesunate.R (multiplicative effect on dihydroartemisinin clearance; the published structural CLM = 190 L/h is reported with the source-paper reference category PREG = 1, so the model file applies the effect via (1 + e_preg_cl_dha * (1 - PREG)) with e_preg_cl_dha = -0.214 to preserve verbatim source values; non-pregnant women have ~21% lower CLM relative to pregnant women).
• Notes: Use this canonical for adult clinical-trial models that test a pregnancy-vs-non-pregnancy contrast (typical settings: malaria-in-pregnancy PK, antiviral-in-pregnancy PK). Trimester or gestational-age stratification within the pregnant cohort should use a separate canonical (e.g., gestational-age weeks via GA or a trimester indicator, ratified separately when needed). The canonical convention is reference category 0 (non-pregnant) following the broader pharmacology default; source papers that use the pregnant cohort as the reference (Birgersson 2019) preserve their published structural values via a (1 - PREG) form on the effect coefficient.

NEUTERED (canonical for surgical neutering / gonadectomy status)

• Description: 1 = the animal has been surgically neutered (a spayed female, i.e. ovariectomy or ovariohysterectomy, or a castrated male, i.e. orchiectomy); 0 = sexually intact. The routine companion-animal demographic descriptor, reported in veterinary papers alongside sex as “spayed female” / “castrated male” / “intact”. Time-fixed per animal in practice, since the procedure precedes enrolment in every study that has used it so far.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (sexually intact).
• Source aliases:
  • sterilization status / sterilized – Beguin 2024 Table 1 and Results wording for the same 0/1 indicator. Same orientation, no value transformation. Prefer the canonical NEUTERED: “sterilized” collides with the microbiological sense of the word, whereas “neutered” is unambiguous in the companion-animal literature.
  • neutered status – Beguin 2024 Methods wording; the source column name recorded in that model’s source_name.
• Example models: Beguin_2024_carboplatin_dog.R, Beguin_2024_carboplatin_thrombocytopenia_dog.R (carboplatin clearance in client-owned dogs; multiplicative log-scale effect exp(e_neutered_cl * NEUTERED) with e_neutered_cl = -0.22, so neutered dogs clear carboplatin about 20% more slowly, reference category 0 = intact).
• Notes: Orthogonal to SEXF, and both may be retained in the same model: a cohort routinely holds intact females, spayed females, intact males and castrated males, and Beguin 2024 screened the two separately, retaining NEUTERED and discarding SEXF. Distinct from the PRIOR_<procedure> surgical-history family (PRIOR_GAST, PRIOR_SPLEN): those record a disease-driven resection that forms part of the treatment history, whereas neutering is an elective demographic attribute recorded at intake. Ratified as a bare-word demographic canonical alongside SEXF and PREG (PMC11664934, 2026-09-11). Interpret a retained effect cautiously: Beguin 2024 found neutering significant on clearance but argue it is most likely confounded with age and plasma creatinine rather than causal, since neutering is not expected to alter GFR and a 2021 iohexol study in dogs found no neutering effect.

CHILD (canonical for child age-cohort indicator)

• Description: 1 = subject is a child, 0 = not a child. In non-human models this is the general juvenile-versus-adult age-cohort indicator (1 = juvenile animal, 0 = adult animal); the concept is the same binary immature-versus-mature contrast, only the species-specific age cutoff changes.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (not child, i.e., adult baseline).
• Source aliases:
  • PED – used in the Schoemaker 2018 LEV / BRV pediatric extrapolation (DDMODEL00000239) as the pediatric-vs-adult indicator that gates the Markov-amplitude term, the overdispersion IIV, and the four pediatric offsets on log baseline rate / mixture / placebo / Emax / EC50.
  • adult_calve – used in Winter_2024_oxytetracycline_cattle.R (and in the raw data set distributed with Winter 2024 as code_adult_calve) as the calf-vs-adult-cattle indicator on all three volumes and all three clearances. Same coding as the canonical: 0 = adult, 1 = calf.
  • AGEgr – used in CarlssonPetri_2015_liraglutide.R as the paper’s pediatric-vs-adult age-category contrast (ESM Table 1 row Cov CL-AGEgr). Same coding as the canonical: 0 = adult, 1 = pediatric.
  • adult – used in Papathanasiou_2021_somatropin.R (Methods section 2.6, ‘adult is a discrete value taking 1 or 0’), switching the typical Emax and its weight relationship and the GH residual error between adults and children. INVERTED coding relative to the canonical: CHILD = 1 - adult.
• Example models: CarlssonPetri_2015_liraglutide.R, CarlssonPetri_2021_liraglutide.R, Papathanasiou_2021_somatropin.R, Schoemaker_2018_levetiracetam.R (DDMODEL00000239), Winter_2024_oxytetracycline_cattle.R.
• Notes: Paper’s age cutoffs must be captured in covariateData[[CHILD]]$notes. That requirement is load-bearing for veterinary and preclinical models, where the cutoff is species-specific and bears no relation to the human paediatric ranges (Winter 2024 defines a calf as under 6 months of age, or any animal the original study authors declared to be a calf). CHILD is often paired with ADOLESCENT when a paper splits the immature population into two bands (CarlssonPetri_2021_liraglutide.R: children 7-11 y, adolescents 12-17 y), but it is equally the right column when a paper models a single pediatric-vs-adult contrast and ADOLESCENT would have no referent – CarlssonPetri_2015_liraglutide.R uses CHILD alone for its 10-17 y pediatric cohort against a 33-73 y adult cohort, and Schoemaker_2018_levetiracetam.R likewise. Record which band the indicator spans in the per-model notes so the two usages are never confused.

ADOLESCENT (canonical for adolescent age-cohort indicator)

• Description: 1 = subject is an adolescent, 0 = not.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0.
• Example models: CarlssonPetri_2021_liraglutide.R.
• Notes: Paired with CHILD. Document age cutoffs.

FM (canonical for fat mass)

• Description: Fat mass in kg derived as the complement of fat-free mass, FM = TBW - FFM, where FFM is computed by the Janmahasatian et al. (2005) formula (or, equivalently, by direct measurement via bioelectrical-impedance analysis / DXA when the source paper specifies it). Time-fixed at baseline.
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (FM / ref)^exponent. Reference values observed: 19 kg (Robarge 2017 efavirenz; cohort median calculated FM).
• Source aliases: FAT, FATMASS.
• Example models: Robarge_2017_efavirenz.R (reference 19 kg; exponent 1 fixed on V_p/F).
• Notes: Natural complement of FFM. Use both in the same model when the paper scales different disposition parameters by FFM vs FM (e.g., Robarge 2017 efavirenz: CL/F by FFM, V_p/F by FM, motivated by lipophilic partitioning into adipose tissue). When the source paper reports the body-composition formula it used (Janmahasatian for FFM is the most common; direct DXA / BIA also acceptable), record it in covariateData[[FM]]$notes.

VISCERAL_ABDOMINAL_FAT (canonical for visceral abdominal fat area from imaging)

• Description: Visceral abdominal fat area measured from a single computed-tomographic (CT) slice at the L2 / L3 vertebral level (standard IVGTT / metabolic-imaging protocol; Jensen et al. 1995). Captures the metabolically active visceral adipose-tissue depot that drives insulin resistance independently of subcutaneous fat. Time-fixed at baseline.
• Units: cm^2 (single CT slice). Document per-model via covariateData[[VISCERAL_ABDOMINAL_FAT]]$units when a paper uses a different anatomical level (L3 / L4, L4 / L5) or a different summary form (volumetric L1-L5 cm^3, MRI-derived).
• Type: continuous
• Scope: specific
• Reference category: n/a – used with mean-centred linear-in-log form log(SI) = log(SI_ref) + theta * (VISCERAL_ABDOMINAL_FAT - ref). Reference values observed: 141.8 cm^2 / CT-slice (Denti 2010 Table 1 pooled-population mean across 204 healthy adults aged 18-87).
• Source aliases:
  • VAF – universal abbreviation in the IVGTT / glucose-metabolism literature (Denti 2010, Basu 2003, Basu 2006). The canonical column is VISCERAL_ABDOMINAL_FAT; the source paper’s column name is recorded in covariateData[[VISCERAL_ABDOMINAL_FAT]]$source_name.
• Example models: Denti_2010_glucoseMinimal.R (introduces canonical; reference 141.8 cm^2 / CT-slice; linear-in-log effect coefficient -0.00208 per cm^2 deviation on insulin sensitivity lsi).
• Notes: Specific scope because the value reported is tied to the imaging modality (single-slice CT at L2 / L3 in the founding example). Volumetric CT, multi-slice averages, and MRI-derived volumetric adiposity from future papers can ratify under the same canonical with the assay method and units recorded per-model in covariateData[[VISCERAL_ABDOMINAL_FAT]]$notes. Distinct from BODYFAT_PCT (whole-body adiposity fraction by DEXA) and from FFM / LBM (fat-free mass / lean body mass). The H3-name uses underscores to preserve readability; future authors may abbreviate the source column as VAF in their own datasets and declare source_name = "VAF".

BODYFAT_PCT (canonical for percent total body fat (whole-body adiposity fraction))

• Description: Percent total body fat, measured as the fraction of total body mass classified as fat tissue by an imaging assay (typically dual-energy X-ray absorptiometry, DEXA). Expressed as a percentage (0-100). Time-fixed at baseline unless the source paper states otherwise. Captures whole-body adiposity, distinct from regional adiposity (visceral, abdominal, subcutaneous) and from fat-free mass.
• Units: % (percent; 0 = no body fat, 100 = all body fat). Document per-model via covariateData[[BODYFAT_PCT]]$units when a paper uses a fractional 0-1 scale.
• Type: continuous
• Scope: general
• Reference category: n/a – used with mean-centred linear-in-log form log(VD) = log(VD_ref) + theta * (BODYFAT_PCT - ref), or as the direct determinant of an adipose-compartment volume in whole-body PBPK. Reference values observed: 32.39 % (Denti 2010 Table 1 pooled-population mean across 204 healthy adults); 25 % (Levitt PKQuest standard human, i.e. 17.5 kg adipose of a 70 kg total, used by Pei_2023_tacrolimus_pbpk.R).
• Source aliases:
  • %TBF – paper symbol in the IVGTT / glucose-metabolism literature (Denti 2010, Basu 2003). The leading % character is dataset-incompatible (NMTRAN / R column-name rules); the canonical column is BODYFAT_PCT and the paper’s symbol is recorded in covariateData[[BODYFAT_PCT]]$source_name.
  • PTBF – numeric-safe rendering of the paper symbol used in some derived datasets.
• Example models: Denti_2010_glucoseMinimal.R (introduces canonical; reference 32.39 %; linear-in-log effect coefficient -0.0101 per percentage-point deviation on glucose distribution volume lvd), Pei_2023_tacrolimus_pbpk.R (whole-body PBPK organ-volume scaling: the adipose compartment weight is set directly as WT * BODYFAT_PCT / 100 and the remaining lean organ weights and blood flows scale by WT * (1 - BODYFAT_PCT / 100) / 52.5 kg; Pei 2023 Methods 2.5 names “the proportion of adipose tissue” as one of the two determinants of organ volume and Table 5 reports its local sensitivity).
• Notes: Promoted from Scope: specific to Scope: general on 2026-08-05 alongside the Pei 2023 tacrolimus PBPK extraction. The original specific scope reflected the founding example’s DEXA assay, but percent total body fat is an assay-independent body-composition quantity and the second registered use (a whole-body PBPK adipose-compartment volume) is unrelated to how the value was measured; the assay method belongs in the per-model covariateData[[BODYFAT_PCT]]$notes, not in the scope. Bioimpedance- or skinfold-derived %BF from future papers ratifies under the same canonical on the same terms. Distinct from FFM (fat-free mass in kg) and from VISCERAL_ABDOMINAL_FAT (regional visceral adiposity in cm^2 / CT slice). The negative effect coefficient on glucose distribution volume captures the physiological observation that glucose distributes preferentially into lean rather than adipose tissue, so per-kg distribution volume decreases as fat fraction rises.

WT_BASE (canonical for per-subject baseline body weight (time-fixed, paired with time-varying WT))

• Description: Per-subject baseline body weight, time-fixed. Used when a source paper applies Wahlby 2004’s extended covariate-model split (Br J Clin Pharmacol 2004;58(4):367-377) and the data column carries the subject’s baseline weight as a constant alongside a time-varying WT column. Operationally, WT_BASE is equal to WT at the subject’s first observation and remains constant for all subsequent records; the within-subject delta is computed in model as (WT - WT_BASE).
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – typically used as (WT_BASE - ref) in an exponential or linear-deviation effect on a structural parameter, or as the ratio (WT_BASE / ref) in a power effect, where ref is the cohort median baseline weight.
• Source aliases:
  • BWT (baseline weight) – Wahlby 2004 source-column convention; also the Verrest 2023 $INPUT column name, annotated there as “BWT: baseline body weight (kg)”; used in Wahlby_2004_pefloxacin.R and Verrest_2023_paromomycin.R.
• Example models: Wahlby_2004_pefloxacin.R (reference 65 kg; saturating “up to median WT” qualifier from Wahlby 2004 Methods plateaus the effect above 65 kg, encoded as min(WT_BASE, 65) - 65), Verrest_2023_paromomycin.R (reference 27.5 kg, the cohort median; standard allometric power scaling, exponent 0.75 on CL/F and Q/F and 1.00 on Vc/F and Vp/F, in Eastern African children and adults with visceral leishmaniasis), Yang_2025_iruplinalkib.R (reference 63 kg, the pooled cohort median and the paper’s own typical-subject value; estimated power exponents 0.441 on CL/F and 1.36 on V1/F in Chinese healthy volunteers and patients with ALK/ROS1-positive solid tumors), Osawa_2018_daclatasvir.R (reference 56 kg, the cohort median; estimated power exponent 0.605 on V/F only, in Japanese adults with chronic genotype-1 HCV infection; the Verrest/Yang shape – the paper names the covariate BBWT and uses time-varying laboratory covariates elsewhere in the same analysis, but does not use Wahlby’s within-subject delta form), Bi_2018_testosteroneCypionate.R (reference 85 kg on CL and V of the testosterone popPK and 84.70 kg on the LH30 inhibitory potency – the two differ because the PK and PD layers were separate NONMEM runs; estimated power exponents 0.785 on CL/F, 1.71 on V/F and -1.14 on the LH30 IC50, paired with an explicit (WT - WT_BASE) delta effect on CL), Bi_2018_testosteroneCypionate_sperm.R (reference 85 kg; power exponent -1.27 on the inhibitory potency of the spermatogenesis indirect-response model).
• Notes: Specific scope because the BCOV/DCOV decomposition pattern is paper-defined: WT_BASE exists conceptually as a per-subject snapshot of baseline weight that the original modeller chose to enter separately from the time-varying weight column. Distinct from time-fixed adult-cohort weight (where the per-subject weight is just constant by data design); use WT_BASE only when the source paper deliberately separates the BCOV from the time-varying COV. Verrest 2023 is a second, weaker ratification: its dataset carries both a time-varying WT and a baseline BWT column and its control stream deliberately writes the allometry against BWT, but it does not use Wahlby’s within-subject delta form (WT - WT_BASE) – it simply prefers baseline weight as the size descriptor. (The paper’s own Table 3 equation prints the subscript as WT_i,t, which would imply the time-varying column; the control stream is the executed model and disagrees.) Yang 2025 is a third ratification of the same Verrest-shaped kind, and the clearest one on the question of deliberateness: that paper screened every laboratory covariate “at baseline and at the time of measurements” and retained three of them (ALB, CRCL, LDH) in their explicitly time-varying form, while naming the weight covariate BBWT – baseline body weight – in the same breath and in the same printed equations. The baseline-versus-time-varying distinction is therefore demonstrably a modelling choice the authors were making consciously, column by column, which is exactly the condition this canonical exists to record. It still does not use the delta form. Scope is left at specific; promote to general once a paper ratifies the column with Wahlby’s within-subject delta form (WT - WT_BASE), which is the semantic none of the three current examples exercises.

BSA_BASE (canonical for per-subject baseline body surface area (time-fixed))

• Description: Per-subject baseline body surface area, time-fixed. Used when a source paper enters the subject’s baseline BSA as a constant alongside a time-varying BSA column under Wahlby 2004’s extended covariate-model decomposition (Br J Clin Pharmacol 2004;58(4):367-377). Time-fixed at the subject’s first observation.
• Units: m^2
• Type: continuous
• Scope: specific
• Reference category: n/a – typically used as a linear scaling factor on a volume parameter.
• Source aliases:
  • BBSA (baseline BSA) – Wahlby 2004 source-column convention; used in Wahlby_2004_gentamicin.R and Garrett_2019_inotuzumab.R.
• Example models: Wahlby_2004_gentamicin.R (replaces time-varying BSA in the final-model V1 equation V1 = 8.63 * BSA_BASE * (ALB/34)^-0.41 because BBSA was the better predictor than time-varying BSA, RSE 2.5% vs 110% on delta-BSA), Garrett_2019_inotuzumab.R (power exponents 1.54 on cl_exp_inf and 1.64 on cl_exp_component, both centered on the 1.84 m^2 population median, plus a LINEAR effect on vc in the centered deviation, 1 + 0.774 * (BSA_BASE - 1.84); baseline BSA displaced body weight in that paper’s covariate screen and, because InO is dosed per m^2, the retained effect supports BSA-based dosing).
• Notes: Specific scope because the BCOV/DCOV split is a paper-defined modelling choice (Wahlby 2004 demonstrated that BBSA was the relevant predictor for gentamicin V1 because delta-BSA was uninformative). Garrett 2019 ratifies the same baseline-BSA column with consistent semantics in a second, unrelated therapeutic area (oncology ADC popPK), and is a reminder that the functional form attached to this covariate is not always a power model – read the source equation rather than assuming (BSA_BASE / ref)^theta. Promote to general if a third paper ratifies it.

BSA_AFFECTED_PCT (canonical for baseline percentage of body surface area affected by a skin disease)

• Description: Per-subject baseline percentage of total body surface area affected by a skin disease (psoriatic plaque, atopic-dermatitis lesion, vitiligo, burn). Time-fixed baseline value from the screening / baseline skin assessment. An extent-of-disease measure, not a body-size measure.
• Units: %
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (BSA_AFFECTED_PCT / <ref>)^exponent. Reference value observed: 31% (Li_2022_gumokimab.R, Li 2022 Eq 7).
• Source aliases:
  • BSA / BSA (%) / “body surface area (BSA) involvement” – Li 2022 gumokimab (AK111) population PK/PD; used in Li_2022_gumokimab.R.
  • TBSA (total body surface area burned) – the burn-literature form of the same quantity.
• Example models: Li_2022_gumokimab.R (power exponent -0.572 on kout, the first-order rate of psoriatic plaque loss in the PASI indirect-response model, normalized to 31%; cohort mean 33.3%, SD 14.7%, in Chinese adults with moderate-to-severe plaque psoriasis).
• Notes: NOT body surface area in m^2: papers in dermatology routinely abbreviate the affected percentage as “BSA”, which collides with the canonical BSA (m^2) and BSA_BASE. Read the units in the source’s demographics table before mapping a “BSA” column. Named without a _BL suffix by analogy with BMBLAST_PCT (a per-subject baseline percentage); a model needing a time-varying affected percentage should propose a separate column. Scope general because the same quantity recurs across skin diseases and burns.

BLOOD_GROUP_O (canonical for ABO blood group O indicator)

• Description: 1 = subject has ABO blood group O (either O-positive or O-negative Rh subtype); 0 = subject has ABO blood group A, B, or AB (non-O). Time-fixed per subject. Enters factor VIII (FVIII) and von Willebrand factor (VWF) population PK models as a surrogate for baseline VWF plasma concentration: blood group O individuals have ~25% lower circulating VWF than non-O individuals due to accelerated VWF clearance by the ASGR / CLEC4M receptor system (VWF’s ABH glycan antigens are recognized differently by clearance receptors depending on the ABO glycosyltransferase phenotype). Because VWF binds and stabilizes circulating FVIII, lower VWF in blood group O individuals translates to faster FVIII clearance.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (non-O ABO blood group: A, B, or AB).
• Source aliases:
  • blood group (paper-prose 0/1 indicator, 1 = O) – used in Hazendonk_2016_factor_viii.R (Hazendonk 2016 Table 5: 1.26 ^ blood_group on CL, so blood group O subjects have 26% higher FVIII clearance than non-O).
• Example models: Hazendonk_2016_factor_viii.R (multiplicative effect on CL: (1 + 0.26 * BLOOD_GROUP_O) – equivalent to 1.26 ^ BLOOD_GROUP_O when binary; reference category 0 = non-O ABO blood group).
• Notes: General scope because ABO blood group is a universally recorded clinical variable that recurs in FVIII / VWF / VWF-related-protein PK models (Hazendonk 2016 Discussion cites the well-established 25% lower baseline VWF in blood group O as the mechanism; also documented in Gill 1987, Miller 1985). The canonical is orthogonal to VWF (the actual plasma VWF concentration): when both are available, papers typically retain VWF and drop BLOOD_GROUP_O because VWF captures the effect more directly (e.g., Nestorov 2014 uses VWF directly; Hazendonk 2016 uses BLOOD_GROUP_O because VWF measurements were only available for ~50% of the cohort). Founding example: Hazendonk 2016.

WAIST (canonical for waist circumference)

• Description: Waist circumference measured by tape at baseline; an abdominal-adiposity anthropometric that is distinct from whole-body size. Enters population PK models as a normalised power term (WAIST / ref)^exponent.
• Units: cm
• Type: continuous
• Scope: general
• Reference category: n/a – normalised power. Reference value observed: 101 cm (Pan 2026, the cohort median from Table 1).
• Source aliases:
  • WAIST – Pan 2026 NONMEM $INPUT (Data S1 Code S1).
  • Waist (cm) – Pan 2026 Table 1 row label.
• Example models: Pan_2026_adalimumab.R (founding example; (WAIST/101)^0.888 on apparent clearance of subcutaneous adalimumab in psoriasis, the strongest single covariate on CL after body weight).
• Notes: Distinct from [[MUAC]] (mid-upper arm circumference), a nutritional-status proxy measured on the arm and stored in millimetres, used chiefly in paediatric and undernutrition cohorts. Distinct from [[VISCERAL_ABDOMINAL_FAT]], an imaging-derived cross-sectional area rather than a tape measurement. Distinct from [[BMI]], which is a whole-body ratio: the two are correlated, but they are not interchangeable, and Pan 2026 screened both and retained waist while rejecting BMI – so a model that retains a waist effect must use this column rather than substituting BMI. Where a source paper codes missing waist measurements with a sentinel (Pan 2026 uses -99 and sets the whole covariate term to 1), supply the model’s reference value instead of the sentinel: the reference reproduces the “no effect” branch exactly, whereas a negative sentinel raised to a fractional power is undefined. Ratified 2026-09-02 alongside the Pan 2026 adalimumab extraction.

RESIDENCE_RURAL (canonical for rural-versus-urban settlement type of the subject’s residence)

• Description: 1 = the subject resides in a rural area; 0 = urban. A subject-level baseline attribute describing the character of the settlement a subject lives in, not its location. Deliberately disease- and country-neutral: a rural/urban split recurs across indications and geographies, and burning the founding study’s disease or country into the name would wrongly imply the concept is specific to it. What each model must document in its own covariateData[[RESIDENCE_RURAL]]$notes is how the source study drew the line, since the rural/urban boundary is defined by national census convention and is not comparable across countries without comment.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (urban). Chosen to match the printed reference row of the founding source, so that the published hazard ratio for rural residence transcribes directly with no polarity flip. Note that urban is the minority stratum in the founding study (16 of 222 episodes, 7.2%), so the reference group there is the smaller one and the estimate is correspondingly imprecise.
• Source aliases:
  • Residence – Tchaparian 2016 Table 3 row label, with Rural / Urban as the two printed levels and Urban carrying the 1.00 reference.
  • Episodes involving urban-resident children, no. (%) – Tchaparian 2016 Table 1 row label, which reports the complement of this indicator.
• Example models: Tchaparian_2016_lumefantrine_recurrence.R (founding example; adjustment covariate of the adjusted Cox proportional-hazards model for 28-day recurrent Plasmodium falciparum parasitaemia after artemether-lumefantrine in Ugandan children, e_residence_rural_haz = log(1.21) on the log-hazard scale, Table 3).
• Notes: Orthogonal to the REGION_<PLACE> family (REGION_TANZANIA, REGION_MOZAMBIQUE, REGION_JAPAN, …), whose members name a geographic PLACE – a study site, enrollment country or continent. Settlement type is a cross-cutting attribute within any such place: a single-country study has one REGION_ value and can still vary on this column, which is exactly the case in the founding study. Do not fold a rural/urban split into a REGION_ indicator or vice versa. Also distinct from SITE-style study-centre identifiers, which index an administrative unit rather than describe it. General scope on first use because the concept carries no study-specific scale or units; only the national convention behind the split is study-specific, and that belongs in the per-model notes. Ratified 2026-09-21 alongside the Tchaparian 2016 lumefantrine extraction.

PAGE (canonical for postmenstrual age)

• Description: Postmenstrual age in months (GA_weeks / 4.35 + postnatal_months). Time-varying.
• Units: months
• Type: continuous
• Scope: general
• Example models: Clegg_2024_nirsevimab.R, Robbie_2012_palivizumab.R, Germovsek_2018_meropenem.R (weeks), Riccobene_2017_ceftaroline.R (weeks). Franzese_2022_ceftazidime.R / Franzese_2022_avibactam.R (weeks; Rhodin maturation TM50 47.7 weeks, Hill 3.4, active only for AGE <= 2 years; PMA taken as postnatal age + 40 weeks where unknown).
• Notes: Months is the register default, but the neonatal maturation literature (Rhodin 2009 and everything built on it) states postmenstrual age in WEEKS, and its reference constants – TM50 = 47.7 weeks, a 33-week pivot, a 5.51-week maturational half-life – are only meaningful on that scale. A model whose source equations are written in weeks therefore declares units = "weeks" in its own covariateData entry and uses weeks throughout; the canonical name is unchanged. Always read the model file’s covariateData$PAGE$units before supplying this column.

PNA (canonical for postnatal age)

• Description: Postnatal age (chronological since birth). Time-varying.
• Units: months
• Type: continuous
• Scope: general
• Source aliases:
  • PNA – used in Zhao 2018 (paper Methods ‘Population pharmacokinetic-pharmacogenetic modelling’ and Table 2 report PNA in DAYS; the canonical PNA carries months, so Zhao 2018’s F_PNA = (PNA_days / 38)^0.472 is reparameterised inside model() as F_PNA = (PNA_months / 1.249)^0.472 using the conversion PNA_months = PNA_days / 30.4375 and reference 1.249 months = 38 days / 30.4375).
  • Age (h) – used in Bardhi 2026 (Bardhi_2026_ampicillin_foal.R), whose neonatal-foal cohort spans 9-194.5 h of life and whose covariate is centred on the weighted-mean age of 61.8 h. Same reparameterisation as Zhao 2018 but from HOURS: model() recovers hours with PNA * 730.5 (1 month = 30.4375 days = 730.5 h) before forming the age ratio, so the centring value 61.8 h corresponds to 0.0846 months. Bardhi 2026 Table 2’s legend prints “61.8 years”, a typographical slip for hours.
• Example models: Hu_2026_clesrovimab.R, Zhao_2018_omeprazole.R (power effect on the formation clearance of 5-hydroxy-omeprazole: (PNA / 1.249)^0.472; PNA reference 1.249 months / 38 days from Zhao 2018 Table 2 cohort median), Bardhi_2026_ampicillin_foal.R (veterinary; power effects on both clearance, exponent 0.40, and peripheral volume, exponent 1.38, of (age_h / 61.8)).
• Notes: Sub-daily postnatal ages are legitimate members of this canonical – Bardhi 2026’s hospitalized neonatal foals are enrolled from 9 h of life – and are carried in months like every other PNA value, with the source unit recovered inside model.

GA (canonical for gestational age at birth)

• Description: Gestational age at birth. Time-fixed per subject.
• Units: weeks
• Type: continuous
• Scope: general
• Example models: Hu_2026_clesrovimab.R, Clegg_2024_nirsevimab.R (folded into PAGE).

WT_BIRTH (canonical for birth weight)

• Description: Body weight at birth. Time-fixed per subject. Distinct from current body weight (WT), which is time-varying.
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – used with linear-deviation (1 + e * (WT_BIRTH - ref)) or power scaling. Reference value observed: 2.59 kg (Voller 2017 newborn cohort).
• Source aliases:
  • BWEIGHT – used in Voller_2017_phenobarbital.R (Voller 2017 source data column for birth weight in kg).
  • Bwb – used in Wu_2025_paracetamol.R (Wu 2025 source-paper symbol for birth weight in kg; same biological quantity, no value transformation).
• Example models: Voller_2017_phenobarbital.R (linear-deviation effect on CL: clbw = 1 + 0.369 * (WT_BIRTH - 2.59)), Wu_2025_paracetamol.R (power-law scaling of PTNA CL_birth and GFR_birth terms with reference 1.75 kg).
• Notes: Time-fixed at birth; characterises pre-/term-newborn cohorts. Pairs with GA (gestational age at birth) when both are reported. The conventional clinical-PK abbreviation BWT is intentionally NOT used as the canonical name because it is already used across the codebase (Gandhi 2021, Li 2019, Chen 2022, Wojciechowski 2022, Lu 2019) as a source-name alias for body weight (WT). The WT_BIRTH form keeps the WT root consistent with the existing body-weight canonical and avoids the BWT ambiguity.

HT_BIRTH (canonical for birth length)

• Description: Body length measured at birth. Time-fixed per subject. Distinct from HT (baseline or current body height), on exactly the principle that distinguishes WT_BIRTH from WT: a neonatal dataset can carry both a daily body length and the value recorded at birth, and a model that fitted one must not be driven with the other.
• Units: cm
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (HT_BIRTH / ref)^exponent or with linear centering. Reference value observed: 33 cm (Padavia 2024, extreme-preterm cohort median 32.75 cm rounded to the value the published equation uses).
• Source aliases:
  • BL – Padavia 2024 source-paper symbol for birth length in cm (beta_V2_BL); same quantity, no value transformation.
• Example models: Padavia_2024_paracetamol.R (reference 33 cm; power exponent 6.46 on the paracetamol peripheral volume of distribution V2, the only size covariate the paper retained on a volume).
• Notes: Registered because the source dataset separately carries a time-varying daily body length – Padavia 2024 Methods 2.2 states that “Bodyweight and body length were also collected every day during treatment”, while Results 3.2 and the Conclusion both name the retained covariate specifically as birth length. Reusing HT here would make one canonical name mean two different measurements in the same dataset, so the fixed-at-birth value gets its own column just as birth weight does. Pairs with WT_BIRTH and GA in pre-/term-newborn cohorts. A model retaining HT_BIRTH alongside the time-varying WT (as Padavia 2024 does) should say so in covariateData, because the two size descriptors then differ in whether they are fixed at birth and a user joining the wrong column into either one gets a silently wrong model.

AGE_DPF (canonical for zebrafish-larval age in days post-fertilization)

• Description: Age of a zebrafish (Danio rerio) larva in days post-fertilization (dpf). Time-fixed per subject in destructive-sampling designs (each larva is harvested at exactly one observation, so its dpf is fixed at the value assigned at exposure-start).
• Units: days post-fertilization (dpf)
• Type: continuous
• Scope: specific
• Reference category: n/a – used in van Wijk 2019 with a step-form effect on the absorption rate (IF (AGE_DPF > 3) k12 = k12_3 * (1 + e_age_dpf_k12)) and a per-day power-form effect on the elimination rate (k25 = k25_3 * (1 + e_age_dpf_k25)^(AGE_DPF - 3)). Reference age is 3 dpf (the youngest cohort in the study).
• Source aliases:
  • AGE – van Wijk 2019 NONMEM column. Renamed to canonical AGE_DPF because the human-PK canonical AGE denotes subject age in years; the zebrafish dpf semantic is incompatible and would silently corrupt any future model that mixed them. Same orientation, no value transformation.
• Example models: vanWijk_2019_paracetamol.R.
• Notes: Specific scope because the meaning is bounded to zebrafish-larval-development PK studies. Distinct from canonical AGE (human age in years), PNA (postnatal age in months), PAGE (postmenstrual age in months), and GA (gestational age in weeks) – none of those are appropriate for a non-mammalian organism whose developmental clock is anchored at fertilization rather than birth. Integer values 3, 4, 5 in the van Wijk 2019 dataset, but treated as a continuous covariate in the elimination-rate power form. Future zebrafish-PK or other non-mammalian-developmental-age models should reuse this canonical only when the covariate is indeed dpf-anchored; other developmental-time conventions (e.g., somite-stage, hpf, dph) would warrant separate canonicals.

ASPHYXIA (canonical for perinatal asphyxia indicator (low 5-minute Apgar score))

• Description: Binary indicator of perinatal asphyxia at birth defined by a 5-minute Apgar score < 5; 1 = 5-minute Apgar score was less than 5 (i.e., the neonate met the paper’s perinatal-asphyxia criterion), 0 = 5-minute Apgar score was 5 or higher. Time-fixed per subject (an event at birth, not a time-varying physiologic state). Used by neonatal popPK models that report a structural-PK shift in subjects who experienced perinatal asphyxia relative to those who did not.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no perinatal asphyxia; 5-minute Apgar >= 5). Effect form in Grasela 1985 is a linear-deviation increment on the volume of distribution: V = V_typical * (1 + e_asphyxia_vc * ASPHYXIA) with e_asphyxia_vc = 0.13 (+13% V in subjects with 5-minute Apgar < 5; p < 0.05).
• Source aliases:
  • None registered (Grasela 1985 abstract names the covariate asphyxia in prose; the abstract does not report the source NONMEM column label).
• Example models: Grasela_1985_phenobarbital.R (Grasela & Donn 1985; preterm-neonate phenobarbital popPK; +13% on V when 5-minute Apgar < 5; no detected effect on CL).
• Notes: Distinct from the continuous Apgar score itself (which has not been registered as a canonical because no model in nlmixr2lib uses the continuous form as of this entry). The < 5 cutoff at 5 minutes is the perinatal-asphyxia convention Grasela 1985 used; future papers that adopt a different Apgar cutoff (e.g., < 7 at 1 minute, or a different time point) should register a separate canonical (e.g., APGAR1_LT7) rather than reusing ASPHYXIA with relaxed semantics.

WT_INFANT (canonical for breastfed-infant body weight in a mother-infant dyad model)

• Description: Body weight of the breastfed infant paired with the modelled lactating subject. Distinct from WT, which in a dyad model is the mother’s body weight.
• Units: kg
• Type: continuous
• Scope: general
• Reference category: n/a – used both to size the maternal breast-milk compartment (milk ingested per feed = daily milk intake per kg x WT_INFANT / feeds per day) and, in a full dyad model, to scale the infant’s clearances and volumes allometrically from the mother’s, (WT_INFANT / WT)^0.75 on clearance and (WT_INFANT / WT) on volume.
• Source aliases:
  • INFWT – NONMEM $INPUT column name in Wattanakul_2024_primaquine.R / Wattanakul_2024_primaquine_motherinfant.R (INFWT; INFANT BODY WEIGHT (KG)); same quantity in kg, no value transformation.
• Example models: Wattanakul_2024_primaquine.R (enters only through the breast-milk compartment volume V_M = (0.15 L/kg/day * WT_INFANT) / feeds-per-day; cohort median 6.8 kg, range 4.13-10.8, giving milk-compartment volumes of 0.062-0.162 L), Wattanakul_2024_primaquine_motherinfant.R (additionally scales the infant’s apparent clearances and volumes from the mother’s estimates).
• Notes: Baseline in the founding models, but time-varying in principle – infant weight changes materially over a multi-month lactation period, and dyad simulations that sweep infant age (Wattanakul 2024 uses WHO weight-for-age curves with z-scores -3 to +3 from birth to 24 months) pair each AGE_INFANT with the corresponding WT_INFANT. Verify baseline vs. time-varying against the source paper.

AGE_INFANT (canonical for breastfed-infant postnatal age in a mother-infant dyad model)

• Description: Postnatal age of the breastfed infant paired with the modelled lactating subject. Distinct from AGE, which in a dyad model is the mother’s age.
• Units: months
• Type: continuous
• Scope: general
• Reference category: n/a – typically enters an enzyme-maturation function on the infant’s clearance. When the maturation function is written in postmenstrual age, the model derives PMA internally by adding an assumed term gestation to AGE_INFANT rather than requiring a separate column, because individual gestational ages are rarely reported in lactation studies; the assumed gestation must be stated in the model file.
• Source aliases:
  • INFAGE – NONMEM $INPUT column name in Wattanakul_2024_primaquine_motherinfant.R (INFAGE; INFANT AGE (MONTHS)); same quantity in months, no value transformation.
• Example models: Wattanakul_2024_primaquine_motherinfant.R (drives monoamine-oxidase-A maturation on the infant’s primaquine clearance, MF = PMA / (TM50 + PMA) with PMA = AGE_INFANT + 9.2 months for assumed full-term gestation of 40 weeks and TM50 = 7.6 months; cohort median 5.0 months, range 1.6-21.7).
• Notes: Months, not years – lactation cohorts are dominated by infants under two years old, where a year scale loses resolution, and paediatric maturation functions are conventionally written in months or weeks of postmenstrual age. This matches the units of the general PAGE canonical. Distinct from PAGE: AGE_INFANT is postnatal, so a model that needs postmenstrual age must add the gestational term explicitly.

NAT2_SLOW_INFANT (canonical for breastfed-infant NAT2 slow-acetylator phenotype in a mother-infant dyad model)

• Description: 1 = the breastfed infant paired with the modelled lactating subject is an NAT2 (arylamine N-acetyltransferase 2) slow acetylator; 0 = the infant is an intermediate or rapid (fast) acetylator. Same phenotype definition, same slow-versus-nonslow pooling and same _SLOW = 1 orientation as the general NAT2_SLOW canonical, which in a dyad model denotes the MOTHER. Time-fixed per infant (germline genotype-derived phenotype).
• Units: (binary)
• Type: categorical
• Scope: general
• Reference category: 0 (nonslow infant: intermediate or rapid acetylator pooled).
• Source aliases:
  • metaboliser status (child) / child: fast metaboliser / child: slow metaboliser – prose form used by Garessus_2019_isoniazid_pbpk.R when it enumerates the four mother-child metaboliser pairs; NAT2_SLOW_INFANT = as.integer(child status == "slow").
• Example models: Garessus_2019_isoniazid_pbpk.R (selects the infant’s hepatic clearance between the two values observed by Rey 2001 in children under 6 months, 2.55 L/h fast and 0.76 L/h slow; crossed with NAT2_SLOW to produce the four dyad phenotype pairs of Figures 4 and 5).
• Notes: Member of the _INFANT dyad-partner family established by WT_INFANT and AGE_INFANT, and registered for the same reason: the mother and the infant each have their own acetylator phenotype, they are genetically independent, and a dyad model must carry both simultaneously, so one column name cannot serve both people. A lactation-transfer model in which only the mother’s phenotype matters needs NAT2_SLOW alone; this canonical is required only when the infant side of the dyad has its own metabolism. Do not use it for a standalone paediatric popPK model whose subject is the infant – that model has no second person to disambiguate against and uses the plain NAT2_SLOW. The companion NAT2_RAPID canonical has a dyad-partner form NAT2_RAPID_INFANT under the same rule, reserved here but not yet exercised; a dyad model that resolves all three infant acetylator levels should register it at that point.

BREASTFEED_EXCLUSIVE (canonical for exclusive-breastfeeding indicator)

• Description: Whether the modelled lactating subject was breastfeeding her infant exclusively (1) or mixing breast milk with formula or complementary food (0). Screened in lactation popPK analyses as a proxy for milk maturity and for the volume and composition of milk produced, both of which can influence drug concentration in milk. Unlike WT_INFANT and AGE_INFANT, this describes the feeding RELATIONSHIP rather than the dyad partner’s own demographics, so it carries no _INFANT suffix.
• Units: (binary)
• Type: categorical
• Scope: general
• Reference category: 0 (not exclusively breastfeeding, i.e. mixed feeding).
• Source aliases:
  • breastfeeding exclusivity – prose form used in Baklouti_2026_amoxicillin.R (Baklouti 2026 Methods section 2.3, “breastfeeding exclusivity (yes/no, categorical variable)”) and screened under the same wording by Melander_2025_cetirizine.R.
• Example models: Baklouti_2026_amoxicillin.R (screened in the stepwise covariate search and not retained; documented in covariatesDataExcluded).
• Notes: Registered by the maintainers 2026-09-02 (PMC13206287) from a documentation-only use, deliberately ahead of the first model that RETAINS it. Lactation popPK is now a recurring family in this library (Wattanakul_2024_primaquine.R, Wattanakul_2024_primaquine_motherinfant.R, Ojara_2024_lamivudine.R, Melander_2025_cetirizine.R, Deferm_2025_lactation_physiology.R, Baklouti_2026_amoxicillin.R) and exclusivity is a standard screened covariate across it, so registering the name now pre-empts a divergent coinage later. checkModelConventions() does not validate names appearing only in covariatesDataExcluded, so this entry is documentation rather than a lint target until a model references it in model().

MAL_NOURISH (canonical for malnutrition status indicator)

• Description: 1 = subject is malnourished at study entry per a paper-defined anthropometric criterion (e.g., WHO height-for-age and weight-for-age Z-scores both < -2.0 in Tikiso 2021); 0 = not malnourished. Subject-level baseline indicator; the time-decaying recovery during nutritional supplementation is carried by the paired T_NUT_SUPP column rather than as a time-varying value of MAL_NOURISH.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not malnourished).
• Source aliases:
  • MAL – used in Tikiso_2021_abacavir.R (the dataset’s paper-defined indicator, 1 = malnourished, 0 = not malnourished).
  • NUT – used in CatalanLatorre_2018_taurine_rat.R (the dataset’s paper-defined indicator, 1 = undernourished UN, 0 = well-nourished WN).
  • Group – used in Kir_2025_atenolol_rat_pbpk.R and Kir_2025_metoprolol_rat_pbpk.R (the Monolix regressor named Group that selects the control vs malnourished parameter set).
• Example models: Tikiso_2021_abacavir.R (gates the time-decaying malnutrition effect: mal_decay = MAL_NOURISH * exp(-T_NUT_SUPP * log(2) / 12.2), which then drives multiplicative shifts of +115% on F and -64% on CL at the start of nutritional supplementation, decaying with a 12.2-day half-life).
  • Tikiso_2021_abacavir.R (gates the time-decaying malnutrition effect: mal_decay = MAL_NOURISH * exp(-T_NUT_SUPP * log(2) / 12.2), which then drives multiplicative shifts of +115% on F and -64% on CL at the start of nutritional supplementation, decaying with a 12.2-day half-life).
  • CatalanLatorre_2018_taurine_rat.R (static baseline indicator – no T_NUT_SUPP pairing because there was no nutritional rehabilitation phase in the preclinical Wistar-rat study; MAL_NOURISH = 1 reduces the saturable tubular secretion Vmax of taurine by 9.4% relative to well-nourished animals).
  • Kir_2025_atenolol_rat_pbpk.R (static baseline indicator in a preclinical Sprague-Dawley mPBPK model; no T_NUT_SUPP pairing because there was no nutritional rehabilitation phase. Unusually, MAL_NOURISH here gates ABSORPTION rather than disposition: it switches all three sequential zero-order absorption rates, the presence and end time of the third absorption window, the blood-to-plasma ratio, and the proportional residual-error magnitude. The structural disposition parameters fd1, Kp1 and CL were deliberately shared across nutrition groups).
  • Kir_2025_metoprolol_rat_pbpk.R (companion to the atenolol model above, from the same paper; MAL_NOURISH switches both zero-order absorption rates, the end time of the second absorption window, the blood-to-plasma ratio, the additive residual-error magnitude, and the fraction of hepatic intrinsic clearance operating in the oral arm).
• Notes: Specific scope because the malnutrition definition (WHO Z-score thresholds in Tikiso 2021; end-of-adaptation body weight below 80% of the well-nourished mean AND serum albumin below 23 g/L in Catalan-Latorre 2018; a 5% protein isocaloric diet fed for 17-20 days, confirmed by significant falls in body weight, serum albumin and total cholesterol, in Kir 2025; mid-upper arm circumference, weight-for-height vs height-for-age, etc.) is paper-defined; per-model covariateData[[MAL_NOURISH]]$notes must document the criterion used. Pairs with T_NUT_SUPP (days on nutritional supplementation) when the model uses a time-decaying recovery function; otherwise MAL_NOURISH alone serves as a static baseline indicator. Distinct from generic body-weight Z-scores (which are continuous anthropometric metrics rather than a binarised malnutrition indicator). For a model that distinguishes malnutrition SEVERITY levels, use the graded MAL_NOURISH_MILD / MAL_NOURISH_MOD / MAL_NOURISH_SEV indicators instead of this pooled one.

MAL_NOURISH_MILD (canonical for mild malnutrition indicator)

• Description: 1 = mild malnutrition per a paper-defined grading, 0 = not malnourished or a different malnutrition level. One of three mutually exclusive severity indicators (MAL_NOURISH_MILD / MAL_NOURISH_MOD / MAL_NOURISH_SEV); all three 0 is the non-malnourished reference.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not malnourished, when MAL_NOURISH_MOD and MAL_NOURISH_SEV are also 0).
• Source aliases:
  • M (“mild nutritional impairment”) – used in Sjogren_2021_malnutrition_physiology.R (the Barac-Nieto et al. grading, body weight / height 89.5 % of standard).
• Example models: Sjogren_2021_malnutrition_physiology.R (selects the ‘Mild’ column of the physiological scaling parameters that transform a non-malnourished PBPK physiology into a malnourished one).
• Notes: Specific scope because the grading scheme is paper-defined (Barac-Nieto body weight / height bands in Sjogren 2021; Gomez weight-for-age or WHO z-score bands elsewhere); per-model covariateData[[MAL_NOURISH_MILD]]$notes must document it. Follows the HEPIMP_MILD / HEPIMP_MOD / HEPIMP_SEV severity-suffix family. Distinct from the pooled binary MAL_NOURISH, which does not separate levels; a model that estimates or tabulates per-level effects must use these graded indicators.

MAL_NOURISH_MOD (canonical for moderate / intermediate malnutrition indicator)

• Description: 1 = moderate (or ‘intermediate’) malnutrition per a paper-defined grading, 0 = not malnourished or a different malnutrition level. Mutually exclusive with MAL_NOURISH_MILD and MAL_NOURISH_SEV.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not malnourished, when MAL_NOURISH_MILD and MAL_NOURISH_SEV are also 0).
• Source aliases:
  • I (“intermediate nutritional impairment”) – used in Sjogren_2021_malnutrition_physiology.R (the Barac-Nieto et al. grading, body weight / height 82.7 % of standard).
• Example models: Sjogren_2021_malnutrition_physiology.R (selects the ‘Intermediate’ column of the physiological scaling parameters).
• Notes: The middle level of a three-level grading, whatever the paper calls it (‘moderate’, ‘intermediate’); the _MOD suffix follows the HEPIMP_MOD / RENALIMP_MOD family. See MAL_NOURISH_MILD for scope and the distinction from MAL_NOURISH.

MAL_NOURISH_SEV (canonical for severe malnutrition indicator)

• Description: 1 = severe malnutrition per a paper-defined grading, 0 = not malnourished or a different malnutrition level. Mutually exclusive with MAL_NOURISH_MILD and MAL_NOURISH_MOD.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not malnourished, when MAL_NOURISH_MILD and MAL_NOURISH_MOD are also 0).
• Source aliases:
  • S (“severe nutritional impairment”) – used in Sjogren_2021_malnutrition_physiology.R (the Barac-Nieto et al. grading, body weight / height 73.9 % of standard).
• Example models: Sjogren_2021_malnutrition_physiology.R (selects the ‘Severe’ column of the physiological scaling parameters; the level the paper uses for every pediatric PK evaluation, giving a virtual weight-for-height z-score of about -3).
• Notes: See MAL_NOURISH_MILD for scope and the distinction from MAL_NOURISH. WHO ‘severe acute malnutrition’ (WHZ < -3 and/or MUAC < 115 mm) is one clinical definition of this level; where a model uses the continuous measurements themselves, carry WHZ / MUAC instead.

MUAC (canonical for mid-upper arm circumference)

• Description: Mid-upper arm circumference, the continuous anthropometric measurement of the circumference of the upper arm at its midpoint. A standard WHO nutritional-status measure in children aged 6-59 months; the WHO severe-acute-malnutrition threshold is MUAC < 115 mm. Time-fixed at admission in the models that use it, though it is measurable repeatedly during nutritional rehabilitation.
• Units: mm
• Type: continuous
• Scope: general
• Reference category: n/a
• Source aliases:
  • MUAC – used in Chotsiri_2019_lumefantrine.R (Chotsiri 2019 Table 1 and Table 2 report MUAC in mm and the covariate coefficient per cm; the model therefore converts MUAC / 10 inside model).
• Example models: Chotsiri_2019_lumefantrine.R (continuous exponential effect on the relative bioavailability of oral lumefantrine: fmuac = exp(e_muac_f * (MUAC / 10 - muac_ref)) with e_muac_f = -log(1 - 0.254) per cm and muac_ref = 13.0 cm, i.e. bioavailability falls 25.4% per 1 cm reduction in MUAC).
• Notes: Store the raw measurement in millimetres – the unit WHO uses for the SAM threshold and the unit source tables almost always tabulate – and convert inside model when a paper publishes its coefficient per centimetre. Distinct from the binary MAL_NOURISH: MAL_NOURISH is a paper-defined malnourished/not-malnourished indicator, whereas MUAC is the underlying continuous measurement and is one of the criteria papers use to set MAL_NOURISH. A model that retains a continuous MUAC effect must use MUAC; binarising it onto MAL_NOURISH would discard the dose-response. Where a paper reports both, carry both columns. Also distinct from the weight-for-height / weight-for-age / height-for-age z-scores (WHZ / WAZ / HAZ families), which are population-standardised scores rather than a raw circumference.

WAZ (canonical for weight-for-age z-score (age- and sex-standardised))

• Description: Age- and sex-standardised weight-for-age z-score: the number of standard deviations a child’s body weight lies above or below the median weight of a reference population of the same age and sex. Unitless and centred at 0 in the reference population. The WHO classifies WAZ < -2 as underweight and WAZ < -3 as severely underweight, and papers routinely make the -3 cut-off a breakpoint rather than using the score linearly across its whole range. Distinct from the raw body weight WT (kg), which is carried separately and is what allometric scaling uses – a model that adjusts for both size (WT) and nutritional status (WAZ) needs both columns. Baseline in the founding model, but time-varying in principle during nutritional rehabilitation; document baseline-vs-time-varying status in covariateData[[WAZ]]$notes.
• Units: unitless (z-score; standard-deviation units)
• Type: continuous
• Scope: general
• Reference category: n/a – the reference is 0 (the reference-population median for the child’s age and sex). Where a paper applies the effect only below a threshold (commonly the WHO severe-underweight cut-off -3), the effect multiplier must evaluate to exactly 1 at and above that threshold; encode the below-threshold depth as max(0, threshold - WAZ) so the covariate is inert for adequately-nourished children.
• Source aliases:
  • ZWFA – used in SemereGebreyesus_2024_rifabutin.R (the source paper’s column name; “weight-for-age z score”, computed from WHO growth charts at ages <= 10 years and US CDC growth charts above 10 years).
• Example models: SemereGebreyesus_2024_rifabutin.R (threshold effect on rifabutin oral bioavailability, active only below the WHO severe-underweight cut-off: waz_below3 <- max(0, -3 - WAZ) then fdepot *= (1 + e_waz_fdepot)^waz_below3 with e_waz_fdepot = -0.26, i.e. bioavailability falls 26.0% per z-score unit below -3 and is unchanged at WAZ >= -3; Semere Gebreyesus 2024 Table 2 and Results “Covariate model”), Wallender_2021_piperaquine.R (time-varying; linear-deviation effect on relative oral bioavailability, (1 + 0.113 * (WAZ - (-0.5))), centred on the cohort median -0.5, so each 1 z-score decrease lowers F by 11.3%), Chotsiri_2019_lumefantrine.R (screened alongside MUAC and WHZ and found significant, but not retained in the final model; carried as documentation).
• Notes: Belongs to the same anthropometric z-score family as BMIZ (body-mass-index z-score) and, when future papers register them, HAZ (height/length-for-age, reported as ZHFA by some papers) and WHZ (weight-for-height). Record which growth reference the source paper used – WHO Child Growth Standards, the WHO 2007 Growth Reference for school-aged children, or CDC 2000 – in covariateData[[WAZ]]$notes, because the same measured weight maps to materially different z-scores across references, and paediatric papers frequently switch reference at age 5 or 10 years. Distinct from the binary MAL_NOURISH (a paper-defined malnourished/not-malnourished indicator, which a WAZ threshold is often used to set) and from MUAC (a raw circumference in mm): a model that retains a continuous or threshold-graded z-score effect must use WAZ, since binarising it onto MAL_NOURISH would discard the dose-response. Where a source paper states a per-unit percentage change without printing the equation, prefer the compounding form (1 + e)^depth over the linear form (1 + e * depth): it matches the literal reading of “X% for each unit”, it keeps a bioavailability or clearance multiplier strictly positive at the extreme z-scores reachable in severely malnourished cohorts, and it is the form the register already uses for the closely analogous per-unit nutritional-anthropometry effect on bioavailability in Chotsiri_2019_lumefantrine.R (MUAC).

T_NUT_SUPP (canonical for time on nutritional supplementation)

• Description: Time elapsed since the start of nutritional / refeeding supplementation, in days. 0 at the start of supplementation; increases with time. Used by population PK models that describe the time-varying recovery of PK parameters during nutritional rehabilitation in malnourished cohorts.
• Units: days
• Type: continuous
• Scope: specific
• Reference category: n/a – enters as the argument of an exponential decay function exp(-T_NUT_SUPP * log(2) / T_half) whose half-life T_half is an estimated parameter (12.2 days in Tikiso 2021).
• Source aliases:
  • TNUTRI – used in Tikiso_2021_abacavir.R (the dataset’s paper-defined column for days since start of nutritional supplementation; same orientation as the canonical, 0 = start of supplementation, increasing with time on supplementation).
• Example models: Tikiso_2021_abacavir.R (paired with MAL_NOURISH; drives the recovery decay exp(-T_NUT_SUPP * log(2) / 12.2) of the malnutrition effect on F and CL).
• Notes: Specific scope because the underlying recovery dynamics are tied to the nutritional-rehabilitation protocol of the source study. For non-malnourished subjects (MAL_NOURISH == 0) the value is irrelevant because the malnutrition effect is gated by MAL_NOURISH; supply 0 as a default. For fully-recovered malnourished subjects, supply a large value (e.g., >= 100 days, well beyond the 12.2-day Tikiso 2021 half-life) so the decay function reaches near zero and the effect vanishes.

DRINK_OGTT (canonical for oral-glucose-tolerance-test-only drink indicator)

• Description: Binary indicator that the postprandial test drink is glucose-only (oral glucose tolerance test, OGTT). 1 = the drink is glucose-only with no fat content (e.g., 25 / 75 / 125 g OGTT); 0 = otherwise (water, or any drink containing fat). Per-occasion (per-test-drink-administration) covariate – a single subject in a crossover challenge protocol receives different drink types across occasions, so the indicator varies per dose-record rather than per subject.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (water or fat-containing drink). Mutually exclusive with DRINK_FAT – both indicators cannot be 1 simultaneously.
• Source aliases: paper narrative “glucose solution” / “OGTT” cohort labels driving the gastric-emptying-onset T50OGTT selection in Guiastrennec 2016.
• Example models: Guiastrennec_2016_gastric_emptying.R (selects the OGTT-specific half-onset time T50OGTT = 15.7 min for the gastric-emptying delay Hill function; water is recovered when DRINK_OGTT = DRINK_FAT = 0 with the onset factor pinned to 1).
• Notes: Specific scope because the OGTT / fat-containing partition is tied to the Guiastrennec 2016 postprandial-challenge design (Studies B and C OGTT arms). Set to 1 for Study B’s 25 / 75 / 125 g OGTT drinks and the Study C 75 g OGTT arm; set to 0 for Study A water and for all fat-containing drinks. Pairs with DRINK_FAT: the two indicators jointly select the appropriate gastric-emptying-delay T50 parameter (T50OGTT vs T50Fat) for the Hill onset function.

DRINK_FAT (canonical for fat-containing test-drink indicator)

• Description: Binary indicator that the postprandial test drink contains fat (any nonzero fat content). 1 = the drink contains fat (e.g., the Study C low / medium / high-fat isocaloric drinks, or the Study D medium-high-fat drink); 0 = otherwise (water or glucose-only OGTT drinks). Per-occasion covariate – a single subject in a crossover challenge protocol receives different drink types across occasions.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (water or glucose-only drink). Mutually exclusive with DRINK_OGTT – both indicators cannot be 1 simultaneously.
• Source aliases: paper narrative “low-fat” / “medium-fat” / “high-fat” / “medium-high-fat” cohort labels driving the gastric-emptying-onset T50Fat selection in Guiastrennec 2016.
• Example models: Guiastrennec_2016_gastric_emptying.R (selects the fat-specific half-onset time T50Fat = 23.1 min for the gastric-emptying delay Hill function).
• Notes: Specific scope because the fat-containing partition is tied to the Guiastrennec 2016 postprandial-challenge design. Pairs with DRINK_OGTT: the two indicators jointly select the appropriate gastric-emptying-delay T50 parameter (T50OGTT vs T50Fat) for the Hill onset function.

GE_T10_BL (canonical for baseline (drug-free) gastric emptying t10)

• Description: Per-subject time, in minutes, for 10% of a standardized test meal to leave the stomach, measured before treatment (drug-free). In Kim 2021 it comes from the 13C-Spirulina gastric emptying breath test (GEBT): the baseline kPCD values are converted to gastric emptying fractions at fixed timepoints with the Szarka 2008 regressions (which also use sex and BMI), and t10 is interpolated from those fractions. A short t10 means fast early gastric emptying.
• Units: min
• Type: continuous
• Scope: specific
• Reference category: n/a – enters as a power term normalised to a reference, (GE_T10_BL / 30)^theta; Kim 2021 uses 30 min. Record the per-model reference in covariateData[[GE_T10_BL]]$notes.
• Source aliases:
  • HT10 – used in Kim_2021_YH12852.R (supplementary NONMEM control stream, SLP = THETA(9) * EXP(ETA(8)) * (HT10/30)**THETA(11)); same orientation and units, no transformation.
• Example models: Kim_2021_YH12852.R (power effect on the linear slope of plasma YH12852 on the stomach emptying rate, exponent 3.57; Kim 2021 Table 1).
• Notes: Member of the <X>_BL baseline family (QTC_BL, MAP_BL, OFFTIME_BL, HAERATE_BL). Not <X>_BASE, which is reserved for a time-fixed baseline paired with a time-varying column of the same quantity; no time-varying gastric emptying column exists. A gastric emptying half-time (t50) covariate, screened but not retained in Kim 2021, would be a sibling GE_T50_BL. Time-fixed per subject. Distinct from the test-drink indicators DRINK_OGTT / DRINK_FAT, which describe the meal rather than the subject. Specific scope until a second gastric emptying model ratifies the name.

D25OH_BL (canonical for baseline (pre-supplementation) serum 25-hydroxyvitamin D concentration)

• Description: Per-subject serum concentration of 25-hydroxyvitamin D (25(OH)D, calcifediol) measured before the first vitamin D dose. 25(OH)D is the circulating storage metabolite of vitamin D and the analyte on which vitamin D status is clinically defined (deficiency below 50 nmol/L, severe deficiency below 25 nmol/L per the Institute of Medicine); every subject carries a substantial endogenous pool sustained by cutaneous synthesis and dietary intake, so a supplemental dose is superimposed on a large pre-existing baseline rather than starting from zero. In vitamin D PK models the baseline therefore plays a dual role: it is the initial condition of the 25(OH)D compartment(s), and it identifies the otherwise-unobservable per-subject endogenous vitamin D input rate, which is back-solved so that formation and elimination balance exactly at the measured baseline. Because the elimination of 25(OH)D is concentration-dependent, the baseline also sets where on the nonlinear clearance curve each subject starts and hence how much a given dose raises them.
• Units: nmol/L
• Type: continuous
• Scope: specific
• Reference category: n/a – not a centred covariate effect. Used directly as an initial condition and as the argument of the endogenous-input back-calculation, so no reference or normalisation value applies. Papers reporting ng/mL must be converted (1 ng/mL = 2.496 nmol/L for 25(OH)D3, molar mass 400.64 g/mol); record the conversion in covariateData[[D25OH_BL]]$notes.
• Source aliases:
  • D25BASE – used in Muhamad_2025_cholecalciferol_pbpk.R (Muhamad 2025 Figures S2/S3 equation symbol) and Huang_2021_cholecalciferol_pbpk.R (Huang 2021 Methods and supplement Final Model section); same orientation and units, no transformation.
• Example models: Muhamad_2025_cholecalciferol_pbpk.R (initial condition of all five 25(OH)D3 compartments, each scaled by that compartment’s volume and partition coefficient, and the argument of the per-child endogenous input ENDOG = CLmax * D25BASE^gamma / (C50^gamma + D25BASE^gamma) * D25BASE / Fm), Huang_2021_cholecalciferol_pbpk.R (the healthy-adult model Muhamad 2025 descends from: the same two roles, initial condition of all four 25(OH)D3 compartments and argument of the arm-level ENDOG, which also sets the pre-dose vitamin D3 steady state).
• Notes: Member of the <ANALYTE>_BL baseline-concentration family (GLN_BL, HGB_BL, FERRITIN_BL, INS_BL, PKK_BL, TRACP5B_BL, TRAST_BL, AT_BL_UDL, PLAQUE_BL). The closest structural precedents are GLN_BL and AT_BL_UDL, which likewise carry the endogenous baseline of the very substance being dosed and likewise anchor the modelled scale. Distinct from the dosed amount, which belongs in the DOSE_<drug>_<units> family. The canonical names the measured metabolite 25(OH)D, not parent vitamin D (cholecalciferol / ergocalciferol) and not the active hormone 1,25-dihydroxyvitamin D; a model needing a baseline for either of those should ratify a separate name rather than overload this one. Time-fixed per subject in Muhamad 2025, which sampled once at baseline, and per study arm in Huang 2021, which fitted arm means; note that seasonal variation in 25(OH)D is real (Muhamad 2025 Section 4.3 infers an 11.5 nmol/L sinusoidal amplitude in the Cape Town control arm), so a study that resamples the baseline across seasons should carry it as time-varying and say so. Specific scope, although two vitamin D models now share the name.

GLN_BL (canonical for baseline (pre-dose) endogenous plasma L-glutamine concentration)

• Description: Per-subject pre-dose plasma concentration of endogenous L-glutamine, measured before the first study dose. L-glutamine is the most abundant free amino acid in plasma and is present at several hundred umol/L in every subject, so an exogenous oral dose is superimposed on a large endogenous pool. In popPK models of administered L-glutamine the baseline plays a dual role: it is subtracted from every measured concentration to isolate the exogenous increment being modelled (so it defines the zero of the modelled concentration scale), and it enters the clearance covariate equation, because a higher endogenous pool competes with exogenous glutamine for metabolism and other elimination pathways.
• Units: umol/L
• Type: continuous
• Scope: specific
• Reference category: n/a – enters as a power (Eq. 5) term normalised to a standard baseline, (GLN_BL / GLN_BL_standard)^theta. Sadaf 2024 uses GLN_BL_standard = 683 umol/L. Record the per-model normalisation constant in covariateData[[GLN_BL]]$notes, since it is a modelling choice rather than a physiological constant.
• Source aliases:
  • Glu_BSL – used in Sadaf_2024_glutamine.R (Sadaf 2024 Table 2; the paired standard value is written Glu_BSLstardard in the table footnote, a typographical variant of “standard”). Note the Glu stem here abbreviates glutamine, not glutamate: Sadaf 2024 measured both analytes and screened baseline L-glutamate separately (it was not retained), so a future extraction of that paper’s sibling analyte must not reuse this alias.
• Example models: Sadaf_2024_glutamine.R (power effect on apparent oral clearance: cl = 78.5 * (WT/70)^0.75 * (GLN_BL/683)^-0.96, Sadaf 2024 Table 2 theta2 = -0.96; the same per-subject value is the quantity subtracted from each measurement before fitting).
• Notes: Specific scope until a second L-glutamine or amino-acid supplementation model ratifies the canonical. Member of the <ANALYTE>_BL baseline-concentration family (HGB_BL, FERRITIN_BL, INS_BL, PKK_BL, TRACP5B_BL, TRAST_BL); the closest structural precedent is AT_BL_UDL, which likewise carries the endogenous baseline of the very substance being dosed and likewise serves both as a power covariate normalised to a reference and as the endogenous anchor of the observation scale. Distinct from AA_* general amino-acid panel entries (none registered at the time of writing) and from the dosed amount, which is DOSE_GLN_GKG. Time-fixed per subject in Sadaf 2024, which verified that the pre-dose baseline did not drift across the four study visits; a study in which the endogenous pool does shift over time should carry it as time-varying and say so in the model’s covariateData notes.

EGA (canonical for maternal estimated gestational age during pregnancy)

• Description: The mother’s estimated gestational age at the time of an observation or dose, in weeks. Time-varying across a pregnancy (though usually effectively constant within a single treatment cycle). EGA = 0 is the non-pregnant anchor: semi-physiological pregnancy models are constructed so that every gestational relation collapses to its non-pregnant value at EGA = 0, which makes the covariate a continuous pregnant-vs-non-pregnant contrast rather than a within-pregnancy-only stratifier.
• Units: weeks
• Type: continuous
• Scope: general
• Reference category: n/a – EGA = 0 is the non-pregnant reference value, not a category. Enters as the argument of the gestational physiology polynomials (serum albumin, alpha-1-acid glycoprotein, GFR, haematocrit, CYP3A4 activity, plasma volume, extracellular water, total body water), each of which is written so that it returns the non-pregnant value at EGA = 0.
• Source aliases:
  • EGA – Janssen 2023 symbol for estimated gestational age; same orientation and units, no transformation.
  • GAWK, GESTAGE, GAWEEK – common NONMEM column spellings for maternal gestational age in weeks; same orientation, no transformation.
• Example models: Janssen_2023_docetaxel.R, Janssen_2023_paclitaxel.R, Janssen_2023_doxorubicin.R, Janssen_2023_epirubicin.R (semi-physiological enriched cytotoxic models in which EGA drives all ten gestational physiology relations reprinted as Janssen 2023 Eqs 1-17).
• Notes: Distinct from GA (gestational age at birth), which is a time-fixed maturity descriptor of the neonate used in paediatric models – EGA describes the mother and varies over her pregnancy. Also distinct from PREG (binary pregnancy status) and from TPP (time postpartum, which is 0 during pregnancy and increases after delivery). PREG gives a step contrast; EGA gives the continuous within-pregnancy trajectory that PREG explicitly defers to (see the PREG entry’s note that “trimester or gestational-age stratification within the pregnant cohort should use a separate canonical… ratified separately when needed”). When a source paper reports trimesters rather than weeks, record the week value used per trimester in covariateData[[EGA]]$notes rather than introducing a trimester-indicator canonical.

TPP (canonical for time postpartum (time after delivery))

• Description: Time elapsed since delivery, in weeks. 0 during pregnancy and at delivery; increases after delivery. Used by popPK models that describe the time-varying postpartum recovery of pregnancy-induced PK changes (e.g., renal blood flow, hepatic enzyme activity) that return gradually to the prepregnant baseline over weeks to months rather than instantaneously at delivery.
• Units: weeks
• Type: continuous
• Scope: general
• Reference category: n/a – enters as the argument of a sigmoidal recovery function (TPP^gamma) / (TPP^gamma + T50^gamma) whose T50 (weeks-to-50%-recovery) and gamma (shape parameter) are estimated. For pregnant visits (TPP = 0) the sigmoid evaluates to 0 and the pregnancy-state PK parameter is unchanged; for far-postpartum visits (TPP >> T50) the sigmoid approaches 1 and the parameter reaches its asymptotic non-pregnant value.
• Source aliases:
  • TPP – de Kock 2017 NONMEM column for time after delivery in weeks; same orientation, no transformation; assigned 0 for samples collected during pregnancy.
• Example models: deKock_2017_sulfadoxinePyrimethamine.R (sigmoidal effect on sulfadoxine CL with asymptotic fractional change -0.757, T50 = 6.35 weeks, gamma = 4.90; the sigmoid approaches its asymptote ~13 weeks postpartum, consistent with the literature for return of GFR and renal blood flow to prepregnant values within 6-12 weeks postpartum).
• Notes: Paired with PREG (pregnancy status indicator) when the source paper models pregnancy as a step contrast on one PK parameter and as a sigmoidal time-decay on another. During pregnancy TPP = 0; after delivery TPP > 0. Document the postpartum sampling window in covariateData[[TPP]]$notes per model. Distinct from T_NUT_SUPP (time on nutritional supplementation, days) and GA (gestational age at birth, weeks) – those are different timescale covariates anchored to different events.

TERM_BIRTH (canonical for term-vs-preterm birth indicator)

• Description: Binary indicator of term-vs-preterm birth status; 1 = term birth, 0 = preterm birth. Time-fixed per subject. The gestational-age cutoff separating the two levels is the source paper’s own: 37 weeks is the conventional obstetric definition (Allegaert 2015), but papers do use other cutoffs (Knebel 2011 defines preterm as gestational age < 38 weeks) and the cutoff actually used must be recorded per-model in covariateData[[TERM_BIRTH]]$notes. In Allegaert 2015 (paracetamol PK in young women) the indicator selects between two typical-value clearances for the sulphate-formation pathway; in Knebel 2011 it selects between two values of the maturation half-time AG50.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: varies by effect form; document per-model. Allegaert 2015 uses the symmetric CL = TERM_BIRTH * theta_term + (1 - TERM_BIRTH) * theta_preterm selection – so neither category is “the multiplicative reference”, and both per-stratum clearances are estimated parameters. Knebel 2011 instead makes term birth the reference and multiplies the maturation half-time by an estimated factor raised to (1 - TERM_BIRTH), so 0 (preterm) is the effect-carrying level.
• Source aliases:
  • TERM (Allegaert 2015 NONMEM column; same orientation, no transformation) – used in Allegaert_2015_paracetamol.R.
  • a preterm indicator (Knebel 2011: 1 = gestational age < 38 weeks) – orientation is INVERTED relative to the canonical, so TERM_BIRTH = 1 - PRETERM and the published coefficient applies to (1 - TERM_BIRTH). Used in Knebel_2011_pantoprazole.R.
• Example models: Allegaert_2015_paracetamol.R, Knebel_2011_pantoprazole.R (multiplicative 1.38-fold inflation of the sigmoid-Emax maturation half-time AG50 for preterm infants: mat_age50 * 1.38^(1 - TERM_BIRTH), taking AG50 from 0.153 to 0.211 years; 77 of 202 patients preterm).
• Notes: Distinct from GA (continuous gestational age in weeks): TERM_BIRTH is the binarized version. Use GA when the source paper carries gestational age as a continuous covariate; use TERM_BIRTH only when the paper itself dichotomizes. Do not derive TERM_BIRTH from GA programmatically inside model – the cutoff convention belongs in data assembly, not the model file, and papers disagree on the cutoff (37 vs 38 weeks). Scope promoted from specific to general on 2026-08-26 alongside the Knebel 2011 pantoprazole extraction, a second and independent paper dichotomizing gestational age at birth into term vs preterm.

CONMED_BIRTHCONTROL (canonical for oral-contraceptive use indicator)

• Description: Binary indicator of oral hormonal contraceptive use; 1 = currently taking an oral contraceptive (estrogen-progestin or progestin-only pill), 0 = not on hormonal contraception. Time-varying as women cycle on/off contraception across study occasions.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (no oral contraceptive). Effect form in Allegaert 2015 is multiplicative: CL_glucuronide *= theta_CONMED_BIRTHCONTROL when CONMED_BIRTHCONTROL == 1, with theta_CONMED_BIRTHCONTROL = 1.46 (estrogen-driven UGT2B7 induction).
• Source aliases:
  • BC (Allegaert 2015 NONMEM column; same orientation, no transformation) – used in Allegaert_2015_paracetamol.R.
  • OC (Csajka 2005 NONMEM column; same orientation, no transformation) – used in Csajka_2005_ephedrine_caffeine.R.
• Example models: Allegaert_2015_paracetamol.R, Csajka_2005_ephedrine_caffeine.R, Angeli_2016_iron_hepcidin.R (multiplicative effect on iron elimination kout_iron; women on oral contraception have ~18% lower iron elimination, consistent with the paper’s narrative that contraception limits menstrual blood loss; e_conmed_birthcontrol_kout_iron = -0.20).
• Notes: Specific scope because the canonical encoding pools all oral contraceptive types (combined / progestin-only) into a single binary; future models that need to distinguish formulations should register a finer-grained canonical (e.g., CONMED_BIRTHCONTROL_COMBINED, CONMED_BIRTHCONTROL_PROGESTIN). The full-word canonical name was chosen over a shorter BC_USE form for clarity in source traces.

CONMED_TICLOPIDINE (canonical for ticlopidine co-medication / study-phase indicator)

• Description: Binary indicator that the record belongs to a ticlopidine pre-dosing phase; 1 = ticlopidine phase, 0 = placebo phase. TIME-VARYING in principle (a crossover subject contributes records in both phases), time-fixed within one phase of one study period.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (placebo phase). In Ashraf_2018_s_ketamine_ticlopidine.R the column selects between the two phase-specific proportional residual SDs for S-ketamine and for norketamine, sd = propSd * (1 - CONMED_TICLOPIDINE) + propSd_ticlo * CONMED_TICLOPIDINE; it carries no structural effect.
• Source aliases:
  • TICLO – the $INPUT data item of the Ashraf 2018 deposited NONMEM control stream (Supplementary Information S3), where it both keys $ERROR on to EPS(1)/EPS(2) versus EPS(3)/EPS(4)/EPS(5) and gates the $DES interaction block with IF (TICLO.EQ.1).
• Example models: Ashraf_2018_s_ketamine_ticlopidine.R (founding example; doi:10.1002/psp4.12346).
• Notes: Member of the CONMED_<INN> family, auto-approved as a well-formed member during the Ashraf 2018 extraction (PMC6202471). Deliberately distinct from DOSE_TICLOPIDINE_MG, which carries the administered ticlopidine dose magnitude that drives the interaction term: the indicator marks the study phase (and stays 1 through the washout that follows the last ticlopidine dose), whereas the dose column falls to 0 as soon as dosing stops. A consumer who uses the indicator to gate the interaction will never see the enzyme-activity recovery the model predicts. The structural gate in the source control stream is redundant – with no ticlopidine in the system the mechanistic-static interaction term evaluates to exactly 1 – so the founding model uses this column only for the residual-error switch.

CONMED_CYPROTERONE (canonical for cyproterone acetate co-medication indicator)

• Description: Binary indicator of concomitant cyproterone acetate (CPA), the steroidal antiandrogen given as flare prophylaxis around the start of GnRH-agonist therapy in prostate cancer; 1 = the CPA effect is in force at this record’s time, 0 = not. TIME-VARYING: the column marks the window over which the model assumes CPA is suppressing endogenous GnRH activity, not merely whether the subject ever received CPA, so it opens before the first GnRH-agonist dose and closes when the assumed effect duration elapses.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no cyproterone acetate effect in force). Effect form in Snelder 2019 is a divisor on the endogenous-agonist activity ratio: agonist_kd_ratio / (1 + cpa * CONMED_CYPROTERONE) with cpa = 4.13, i.e. an 80.5% reduction in endogenous GnRH activity and a consequent 67.3% reduction in typical testosterone.
• Source aliases:
  • ICPA / CPA – Snelder 2019 Supplement 4 NONMEM control stream, where ICPA is the per-subject CPA flag and the $DES block gates CPA = CPA0 to a 336 h window opening 168 h before the leuprorelin dose.
• Example models: Snelder_2019_leuprorelin.R (founding example; gates the constant CPA suppression of endogenous GnRH activity in the modified Romero testosterone model).
• Notes: Member of the CONMED_<INN> family. General scope because CPA (and the equivalent nonsteroidal antiandrogens used the same way) is standard flare prophylaxis across the GnRH-agonist literature that this model family draws on, so the same column will recur in triptorelin, goserelin and leuprorelin models. The window is supplied as data rather than derived inside model() because its start and duration are protocol facts, not fitted parameters: EC403 gave CPA 7 to 3 days before the first leuprorelin dose (Methods 2.1) and the analysis assumed a two-week effect (Methods 2.5). Record the window actually used in covariateData[[CONMED_CYPROTERONE]]$notes. Distinct from a DOSE_CPA_MG style column, which would carry the administered magnitude; register that separately if a future model estimates a dose-dependent rather than constant CPA effect. Auto-approved as a well-formed CONMED_<INN> member during the Snelder 2019 leuprorelin extraction (PMC6533438).

DIS_EOPE (canonical for early-onset pre-eclampsia indicator)

• Description: Binary indicator of early-onset pre-eclampsia (eoPE); 1 = eoPE diagnosed before 34 weeks gestation, 0 = not eoPE. Time-fixed per subject within the gestational PK study window. Used by population PK models that compare drug disposition in pregnant women with vs without early-onset pre-eclampsia.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no eoPE). Effect form in Schoenmakers 2025 is multiplicative on CL: CL_eoPE = CL * ThetaPE^DIS_EOPE with ThetaPE = 0.617 (38% reduction in betamethasone CL when eoPE is present); encoded in nlmixr2 as the log-additive shift cl <- exp(lcl + etalcl + e_eope_cl * DIS_EOPE) with e_eope_cl = log(0.617).
• Source aliases:
  • PE – common abbreviation in obstetric pharmacology papers when the cohort restriction is to early-onset PE only; used in Schoenmakers_2025_betamethasone.R (paper notation: eoPE / ThetaPE).
• Example models: Schoenmakers_2025_betamethasone.R (multiplicative effect on CL, encoded via the log-additive form on lcl; reduces apparent betamethasone clearance from 15.6 L/h to 9.6 L/h).
• Notes: Distinct from a broader PREECL indicator that would pool early-onset, late-onset and postpartum pre-eclampsia. The “early-onset” specifier corresponds to diagnosis before 34 weeks gestation, the conventional clinical cutoff (Phipps 2019 Nat Rev Nephrol). Future papers that enrol mixed early-/late-onset cohorts or that report PE status without the 34-week stratification should register a separate canonical (e.g., PREECL for any-onset PE, or LOPE for late-onset PE) rather than reusing DIS_EOPE with relaxed semantics. Distinct from PREG (pregnancy status indicator): DIS_EOPE is a complication-of-pregnancy stratifier within a pregnant cohort, whereas PREG discriminates pregnant-vs-non-pregnant subjects.

SCORE_HIGHAM (canonical for Higham pictorial blood-loss assessment chart score)

• Description: Higham’s score (Pictorial Blood-loss Assessment Chart, PBAC), a semi-quantitative pictorial scoring of menstrual blood loss summed across all sanitary protection items used during a menstrual period. Higher scores correspond to heavier menstrual bleeding; a per-menstrual-period score >= 100 is the conventional clinical threshold for menorrhagia (Higham 1990 BJOG). Time-fixed per subject when computed as the mean of the last few cycles preceding the study (the Angeli 2016 use), time-varying when scored per cycle.
• Units: (score; unitless)
• Type: continuous
• Scope: specific
• Reference category: n/a – enters as a power-law multiplier (SCORE_HIGHAM / tHIGHAM)^exponent on hepcidin synthesis and elimination rate constants in Angeli 2016, with the reference value tHIGHAM set near the population median (96.6 in the Angeli 2016 HEPMEN cohort).
• Source aliases:
  • HIGHAM – prior canonical name (pre-2026-06-19 standardization audit).
  • HIGHAM – prior canonical name (pre-2026-06-19 SCORE_ family standardization).
  • HiS – Angeli 2016 paper notation; same orientation, no transformation.
• Example models: Angeli_2016_iron_hepcidin.R (per-subject mean of the three most recent cycles’ Higham scores; enters ksynH and koutH via power-law effects with exponents 0.66 and 0.83 respectively).
• Notes: Specific scope because the Higham / PBAC score is a reproductive-health-specific instrument. Promote to general if a second menstrual-cycle or iron-status paper registers the same instrument. Distinct from generic blood-loss quantities reported in mL (which would warrant a separate MENSTRUAL_BLOOD_LOSS_ML canonical). The threshold for menorrhagia is 100 (Higham 1990 BJOG 97:734); the Angeli 2016 cohort had a mean of 96.6 with SD 60.5 (Table I), so most subjects sat near the menorrhagia threshold.

DLOSS (canonical for menstrual-period duration)

• Description: Per-subject length of the menstrual period (days of menstrual bleeding) used as a time bound on the increased-elimination phase in joint iron / hepcidin turnover models. Time-fixed per subject within a single menstrual cycle (the Angeli 2016 use); set to the individual’s observed menses length over the study cycle rather than estimated from the data.
• Units: days
• Type: continuous
• Scope: specific
• Reference category: n/a – enters the model as the upper bound of the loss-phase indicator window t < DLOSS. Reference values observed: Angeli 2016 inclusion criterion required menses length between 3 and 5 days (Methods p. 491); the per-subject value was fixed to the observed individual menses length.
• Source aliases:
  • dloss – Angeli 2016 paper notation; same orientation, no transformation.
• Example models: Angeli_2016_iron_hepcidin.R (per-subject menses length in days; bounds the loss phase during which kloss adds to both kout_iron and kout_hep).
• Notes: Specific scope because the menses-length covariate is meaningful only inside joint iron / hepcidin turnover models that include a discrete loss-phase indicator. Promote to general if a second reproductive-cycle PK / PD paper uses an equivalent per-subject menses-length time-bound. Distinct from TPP (time postpartum, weeks; unbounded postpartum time after delivery) and from a hypothetical CYCLE_LENGTH covariate (full menstrual cycle length in days, which would warrant a separate canonical).

DAY_DELIVERY (canonical for day-of-delivery indicator)

• Description: Binary indicator that the PK sample was collected on the day of delivery (the labour-and-childbirth occasion). 1 = sample taken on the day of delivery; 0 = otherwise (pregnant non-delivery, or non-pregnant). In Hirt 2007 the indicator is mutually exclusive with PREG = 1: per the paper, on the day of delivery the coding was PREG = 0 and DAY_DELIVERY = 1, so the two indicators jointly partition women into non-pregnant (PREG = 0, DAY_DELIVERY = 0), pregnant non-delivery (PREG = 1, DAY_DELIVERY = 0), and delivery (PREG = 0, DAY_DELIVERY = 1) cohorts. Time-fixed per sampling occasion.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not day of delivery).
• Source aliases:
  • DEL – Hirt 2007 NONMEM symbol for the day-of-delivery indicator; same orientation, no value transformation.
• Example models: Hirt_2007_nelfinavir.R (multiplicative effect 1 + 1.92 * DAY_DELIVERY applied to both maternal nelfinavir clearance CL_Nm_No and distribution volume V; gates a body-weight power effect (WT/73)^2.81 on CL_Nm_No within the delivery cohort only).
• Notes: Distinct from PREG (pregnant non-delivery, 1 = pregnant; see entry above), TPP (time postpartum, a continuous time-since-delivery covariate for postpartum recovery), and TERM_BIRTH (term-vs-preterm birth at any postnatal time). Use DAY_DELIVERY when the source paper carries a contrasted day-of-delivery cohort against pregnant non-delivery women in the same PK analysis (placental-transfer / labour-PK studies). Specific scope until a second model ratifies the name; promote to general when a future placental-transfer or labour-PK paper uses the same indicator.

SEX_HORMONE_BINDING_GLOBULIN (canonical for serum sex hormone-binding globulin)

• Description: Serum sex hormone-binding globulin (SHBG) concentration, the high-affinity binding protein for testosterone and estradiol (baseline or time-varying). Higher SHBG restricts the unbound sex-steroid fraction available for hepatic extraction, so it typically enters a sex-steroid clearance with a negative power exponent.
• Units: nmol/L. Document per-model via covariateData[[SEX_HORMONE_BINDING_GLOBULIN]]$units; papers reporting ug/dL or mg/L must be converted on data ingestion.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (SEX_HORMONE_BINDING_GLOBULIN / ref)^exponent. Reference values observed: 20 nmol/L (Pastuszak 2021, hypogonadal men; Table 1 mean 20.8 nmol/L).
• Source aliases:
  • SHBG – Pastuszak 2021 Table 1 and Table 2 (‘SHBG, CL/F (power)’); baseline value, no value transformation.
• Example models: Pastuszak_2021_testosteroneUndecanoate.R (baseline-only; power exponent -0.219 on the apparent clearance of total testosterone, referenced to 20 nmol/L; effect parameter e_sex_hormone_binding_globulin_cl).
• Notes: Spelled out rather than abbreviated, per the maintainers’ policy for new canonicals. A plain analyte name like ALB, used for both baseline and time-varying values; each model’s covariateData notes say which one it uses. Distinct from the shbg turnover-STATE compartment in compartment-names.md (Jensen 2023, Reinecke 2018), which is a simulated amount, not a data column.

HR (canonical for heart rate)

• Description: Subject heart rate, in beats per minute. Captured in popPK studies where hemodynamic state modifies hepatic blood flow and hence clearance of high-extraction-ratio drugs (e.g., propofol). May be time-varying when serial intra-operative or intensive-monitoring values are recorded; many studies summarise as the per-subject median across the observation window and treat the covariate as time-fixed. Document baseline-vs-time-varying status in covariateData[[HR]]$notes per model.
• Units: beats/min
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (HR / ref)^exponent or linear-deviation forms (1 + e * (HR - ref)). Reference values observed: 158 beats/min (Ngamprasertwong 2016; population reference encoded in the Table 2 equation CL = theta1 * (HR/158)^theta2).
• Source aliases:
  • HR – same orientation as the canonical, no value transformation; used in Ngamprasertwong_2016_propofol_sheep.R (per-subject median HR over the propofol-infusion observation window, treated as time-fixed in line with the cohort-typical sheep hemodynamic state).
  • Baseline HR – Mukker 2026 Table S1; the per-subject pre-dose (time-zero) heart rate, treated as a time-fixed covariate entering the C-DeltaHR linear-mixed-effects intercept as a centered deviation. Same orientation, no value transformation; the baseline-versus-observation-time reading is documented per-model per the instruction in this entry’s Description.
• Example models: Ngamprasertwong_2016_propofol_sheep.R (power effect on maternal propofol clearance: CL_indiv = theta1 * (HR/158)^theta2 with theta2 = 0.764; clearance increases with heart rate, plausibly reflecting heart-rate-driven increases in hepatic blood flow that govern propofol’s high hepatic-extraction-ratio elimination), Mukker_2026_tuvusertib_HR.R (per-subject baseline HR entering a centered linear-deviation term e_hr_bl_e0 * (HR - hr_bl_ref) on the concentration-DeltaHR intercept; note that the model’s observable is named d_hr, the change from baseline, so that it does not shadow this covariate column).
• Notes: General scope because heart rate is a universally applicable vital sign suitable for any model where hemodynamic state modulates clearance. Future models can use a different reference HR (typical adult human is ~70 beats/min vs the sheep cohort 158 beats/min); document the reference in covariateData[[HR]]$notes. Distinct from HR_BAND or HRV (not yet registered) which would be a heart-rate-band stratifier or heart-rate variability metric, respectively.

SBP (canonical for systolic blood pressure)

• Description: Subject systolic blood pressure, in mmHg. Captured at baseline or serially during a study. Universal vital sign used in popPK / disease-progression models where hemodynamic state serves as either (a) a prognostic covariate on non-PK endpoints (overall survival, tumor dynamics), or (b) a covariate on clearance for drugs whose disposition is sensitive to blood pressure. Document baseline-vs-time-varying status per model in covariateData[[SBP]]$notes.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (SBP / ref)^exponent, linear-deviation forms (1 + e * (SBP - ref)), or additive-log forms beta * log(SBP). Reference values observed: 120 mmHg (Terranova 2022 avelumab JAVELIN Gastric 100 cohort median; also the population-adult clinical normal midpoint).
• Source aliases:
  • Log systolic blood pressure – Terranova 2022 Table S2 OS TTE coefficient label; enters as log(SBP) * beta (an additive-log form on log-median OS).
  • SBP – source-paper column name; same orientation, no value transformation.
• Example models: Terranova_2022_TGD_OS_gastric.R (additive-log effect on log-median OS: log_median_OS += 0.532 * log(SBP), screened by ML and retained in the parametric TTE model despite the credible interval including zero, in keeping with the paper’s hypothesis-generating scope).
• Notes: General scope because systolic BP is a universally applicable vital sign. Sibling of HR (heart rate, beats/min) and DBP_REL (relative change in diastolic BP from baseline, unitless). The two BP siblings are physiologically related but independent covariate columns: some source papers report SBP only, some report DBP or DBP-change only, some report both. SBP here is the raw absolute value; if a future model needs a relative-change (SBP_REL) or a diastolic sibling (DBP) it should register a new sibling canonical rather than overloading this entry.

BODYTEMP (canonical for body temperature)

• Description: Subject body temperature (typically axillary or oral) at the relevant clinical observation. Captured at study admission in acute-infection PK studies (fever as a marker of acute illness severity); may be time-varying when serial temperature measurements are recorded across visits. Document baseline-vs-time-varying status in covariateData[[BODYTEMP]]$notes per model.
• Units: degC
• Type: continuous
• Scope: general
• Reference category: n/a – used with linear-deviation forms (1 + e * (BODYTEMP - ref)) or exponential forms exp(e * (BODYTEMP - ref)). Reference values observed: 36.9 degC (Kloprogge 2013 lumefantrine; pooled-cohort median in Ugandan pregnant + non-pregnant women with uncomplicated P. falciparum malaria), 37.2 degC (Kloprogge 2014 quinine; pregnant Ugandan women with uncomplicated P. falciparum malaria, cohort median at admission).
• Source aliases:
  • TEMP – common short form in malaria / infectious-disease NONMEM control streams; used in Kloprogge_2013_lumefantrine.R and Kloprogge_2014_quinine.R (same orientation as the canonical, no value transformation).
  • Temp – rearing-water temperature in Yang_2020_enrofloxacin_rainbowTrout_pbpk.R (acslXtreme code constant Temp). For a poikilothermic animal the body temperature is the ambient water temperature, so the water-temperature column is carried as BODYTEMP with no value transformation.
• Example models: Kloprogge_2013_lumefantrine.R (linear-deviation effect on mean absorption transit time MTT: MTT_indiv = TVMTT * (1 + e_bodytemp_mtt * (BODYTEMP - 36.9)) with e_bodytemp_mtt = 0.165 per degC; mean transit time increases ~16.5% per degC over 36.0-39.8 degC, plausibly reflecting reduced gut motility / prolonged absorption in feverish malaria patients), Kloprogge_2014_quinine.R (exponential effect on elimination clearance: CL_indiv = TVCL * exp(e_bodytemp_cl * (BODYTEMP - 37.2)) with e_bodytemp_cl = -0.243 per degC, centered at the cohort median; clearance decreases ~21.6% per degC increase in admission body temperature over 36.0-38.9 degC, reflecting reduced metabolic CYP3A4 activity during acute febrile malaria), Hahn_2019_sufentanil.R (exponential effect on systemic clearance: CL = 37.8 * exp(e_bodytemp_cl * (BODYTEMP - 36.9)) with e_bodytemp_cl = +0.207 per degC, centered at the cohort median 36.9 degC; clearance INCREASES ~23.0% per degC over the 33-38.7 degC range observed in critically ill VA-ECMO patients – the OPPOSITE sign to Kloprogge 2014, and the two are not in conflict: Kloprogge’s range is febrile malaria at or above normothermia, whereas Hahn’s is dominated by therapeutic and incidental hypothermia in ECMO patients, where reduced total hepatic blood flow and slowed CYP3A4 activity both cut the clearance of a high-extraction drug), Yang_2020_enrofloxacin_rainbowTrout_pbpk.R (rainbow trout; not centred – drives cardiac output through the linear relationship CO (mL/min/kg) = 3.95 * BODYTEMP - 12.9, so every tissue blood flow rises with water temperature over the 5-17 degC calibration range; CO reaches zero at 3.27 degC, so the model cannot be used below about 3.3 degC).
• Notes: General scope because body temperature is a universally applicable vital sign; the Kloprogge 2013 reference value 36.9 degC is cohort-specific (Ugandan malaria cohort median) and future models should document their own reference in covariateData[[BODYTEMP]]$notes. Units are degrees Celsius; convert from Fahrenheit (degF) at data-assembly time, not inside model(). Distinct from BODYTEMP_FEBRILE (not yet registered) which would be a binary fever indicator if a future paper dichotomises at the conventional 37.5 / 38.0 degC threshold. For fish and other poikilotherms, whose body temperature equals the water (or ambient) temperature, record that temperature as BODYTEMP; the physiological role is the same (temperature-dependent blood flow and enzyme activity), only the measurement differs.

PACO2 (canonical for arterial blood carbon dioxide tension (PaCO2))

• Description: Arterial blood partial pressure of carbon dioxide, a routinely-measured blood-gas variable. Elevated in hypercapnia and respiratory acidosis (chronic obstructive pulmonary disease, restrictive lung disease, hypoventilation, severe asthma exacerbation). Used in population-PK analyses of theophylline and other drugs whose hepatic metabolism is sensitive to acid-base status / pulmonary disease severity. Typically captured at study admission via an automatic blood-gas analyzer on an arterial puncture sample; document baseline-vs-time-varying status in covariateData[[PACO2]]$notes per model.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – used as a continuous covariate in exponential / log-linear effect forms exp(e_paco2_param * PACO2) or centered exp(e_paco2_param * (PACO2 - ref)). Reference values observed: 42.5 mmHg (Yano 1993 stable chronic airway obstruction cohort mean; also the laboratory normal-range midpoint of 35-45 mmHg in adults).
• Source aliases:
  • PaCO2 – the standard physiological abbreviation; used directly in Yano 1993 Table I and Eqs 7-8.
• Example models: Yano_1993_theophylline.R (log-linear effect on both CL and Vd of theophylline: CL = exp(lcl + e_hepimp_cl * HEPIMP + e_paco2_cl * PACO2 + etalcl) with e_paco2_cl = 0.0233 per mmHg, and Vd = exp(lvc + e_paco2_vc * PACO2 + etalvc) with e_paco2_vc = 0.00934 per mmHg; both estimated by forward-selection likelihood-ratio testing).
• Notes: Companion to arterial PaO2, arterial blood pH, serum bicarbonate, and base excess in respiratory-physiology covariate panels (not yet registered; future models retaining these as final covariates should propose them as new canonicals at extraction time). Yano 1993 found PaCO2 to be the most informative blood-gas covariate when entered in a forward-selection competition against PaO2, pH, hematocrit, and albumin. General scope because PaCO2 is a routinely-measured laboratory variable suitable for any model where pulmonary disease severity or acid-base status modulates drug disposition.

PEEP (canonical for positive end expiratory pressure)

• Description: Positive end expiratory pressure (PEEP) applied at the end of the expiratory phase during positive-pressure mechanical ventilation. Airway pressure setting maintained above atmospheric to prevent alveolar collapse; range 0-20 mmHg (0-25 cmH2O) in adult ICU cohorts. Time-varying within an ICU stay as ventilator settings are adjusted; source papers may use it as an admission value or a time-varying covariate – document per-model in covariateData[[PEEP]]$notes. Higher PEEP raises intrathoracic pressure, decreases venous return and cardiac output, and can compromise hepatic blood flow (total hepatic blood flow decreases 4%, 12%, 32% at PEEP 10, 15, 20 mmHg respectively).
• Units: mmHg (the unit reported in Swart 2004 Table 3). Some ICU literature uses cmH2O (1 cmH2O = 0.736 mmHg); when the source paper uses cmH2O the convention should be documented in covariateData[[PEEP]]$notes and values converted to mmHg for consistency across models.
• Type: continuous
• Scope: specific
• Reference category: n/a – typically used with centred linear scaling. Reference values observed: 5 mmHg (Swart 2004 lorazepam CL formula centring value; representative of routine “protective” PEEP in mechanically-ventilated ICU adults).
• Source aliases:
  • PEEP – printed column label in Swart 2004 Table 3 and formula label in Table 4; used in Swart_2004_lorazepam.R.
• Example models: Swart_2004_lorazepam.R (centred linear effect on CL for non-alcohol-abuse subjects with reference 5: cl = 4.13 - (PEEP - 5) * 0.417 L/h; each 1 mmHg above 5 lowers CL by 0.417 L/h, capturing decreased hepatic blood flow at higher PEEP; alcohol-abuse subjects instead take a flat CL = 0.74 L/h with no PEEP effect).
• Notes: Specific scope because PEEP is meaningful only for mechanically-ventilated cohorts. Distinct from MECH_VENT (binary invasive-mechanical-ventilation status) which is a treatment-status flag, not a quantitative pressure – MECH_VENT cannot substitute for PEEP because every subject in a PEEP-covariate model already has MECH_VENT = 1. The Swart 2004 rationale (Discussion) links PEEP to hepatic blood flow, so PEEP as a covariate is expected to matter most for low-hepatic-extraction-ratio drugs (lorazepam ER ~ 0.06). Future ICU popPK papers reusing PEEP with the same quantitative encoding can extend Example models; scope may be promoted to general after multiple papers. Per-model covariateData[[PEEP]]$notes must document whether the value is admission PEEP, time-varying PEEP, or a cohort summary.

STO2 (canonical for tissue oxygen saturation)

• Description: Regional tissue (as opposed to arterial or venous) oxygen saturation, the fraction of haemoglobin in the microvasculature of a named tissue bed that is oxygen-saturated, expressed as a percent. Measured non-invasively by near-infrared spectroscopy (NIRS), which is the instrument; STO2 names the quantity, on the same principle that a drug concentration is not named after the assay that measures it. Used in trauma / critical-care popPK analyses as a marker of peripheral perfusion and shock severity: a falling tissue oxygen saturation signals inadequate microvascular oxygen delivery even while arterial saturation and blood pressure remain near normal. Document the tissue bed (skeletal muscle, cerebral, somatic / renal) and whether the value is an admission reading, a nadir, or time-varying in each model’s covariateData[[STO2]]$notes – these are materially different quantities and are not interchangeable.
• Units: % (percent of haemoglobin saturated; 0-100)
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (STO2 / ref)^exponent or linear-deviation forms. Reference values observed: 88 % (Stitt 2026; median admission skeletal-muscle NIRS StO2 in the TAMPITI adult severe-trauma cohort).
• Source aliases:
  • NIRS – the instrument abbreviation, used as the column label in Stitt 2026 Table 1 / Table 2 and Equation 1. Same orientation, no value transformation.
  • SmO2 – skeletal-muscle oxygen saturation, the sports-physiology and trauma-monitoring spelling.
  • NIRS StO2, StO2, rSO2 – punctuation / capitalisation variants seen in publication tables and device output.
• Example models: Stitt_2026_tranexamicAcid.R (median admission skeletal-muscle NIRS StO2, reference 88 %; exponent -0.29 on TXA clearance, so lower tissue oxygen saturation predicts faster clearance).
• Notes: The canonical names the measured quantity rather than the measurement modality, consistent with the analyte-naming style of IL6, CRP and HGB; NIRS is retained as a source alias so paper tables and NONMEM control streams using the instrument abbreviation resolve without value transformation. This opens a clean family alongside future SPO2 (peripheral pulse-oximetry saturation) and SAO2 (arterial blood-gas saturation) entries, which are different quantities measured at different sites and must be registered separately rather than aliased onto this one. General scope because tissue oxygen saturation is a routine critical-care measurement expected to recur across trauma, sepsis, cardiac-surgery and ECMO popPK models. Tissue-bed and summary-statistic mismatches are a real hazard, not a theoretical one: Stitt 2026 pairs an adult reference of 88 % (median value at admission) with a paediatric virtual-cohort range of 51-80 % taken from the literature as the median lowest recorded value, and the paper itself flags (Discussion, S262) that the two are not the same quantity and that the mismatch inflates the extrapolated paediatric clearance. Any model reusing this column must record which of the three – admission, nadir, or time-varying – its values are.

DBP_REL (canonical for relative change in diastolic blood pressure from baseline)

• Description: Relative (fractional) change in diastolic blood pressure from the per-subject baseline, unitless (e.g., 0.10 = +10% above baseline). Typically a time-varying driver simulated from an upstream diastolic-blood-pressure indirect-response model as DBP_REL(t) = (dbp(t) - dbp0) / dbp0, then consumed as a covariate by a downstream model. Distinct from a raw diastolic-BP value (mmHg); this is the normalized deviation.
• Units: fraction (unitless; positive when dBP is elevated above baseline)
• Type: continuous
• Scope: specific
• Reference category: n/a – enters a log-linear hazard term exp(e_dbprel_haz * DBP_REL).
• Source aliases:
  • DBPREL – Hansson 2013 e85 NONMEM column for the relative dBP change; same orientation, no value transformation.
• Example models: Hansson_2013_sunitinib_os.R (time-varying relative dBP change, simulated from Hansson_2013_sunitinib_dbp and entered as an overall-survival hazard modulator: a greater relative dBP increase lowers the hazard).
• Notes: Specific scope because the value semantics are tied to the upstream Hansson 2013 indirect-response dBP model’s relative-change definition; a future model needing the same relative-dBP-change driver should add itself to the Example models list (and consider promotion) rather than reuse the name with different semantics. Structurally analogous to RCFB1MAX (relative change from baseline in SUVmax driving the Schindler 2016 OS hazard). The related raw-state PD output is the dbp compartment in compartment-names.md; the placebo-arm indicator paired with that model family is PLACEBO.

QTC_BL (canonical for subject’s heart-rate-corrected QT interval at pre-dose baseline)

• Description: Subject’s pre-dose baseline QT interval corrected for heart rate, treated as a per-subject time-fixed covariate. Used in concentration-DeltaQTc PD models to capture each subject’s intrinsic QTc level entering the linear-mixed-effects intercept as a centered term e_qtc_bl_e0 * (QTC_BL - QTC_BL_ref). The QT correction method (Bazett, Fridericia, individual, study-population-specific) is model-specific and documented per-model in covariateData[[QTC_BL]]$notes; the canonical name does NOT carry the correction method, on the same precedent as HGB_BL (units / convention documented per-model) and IGG (units documented per-model). Distinct from HR (heart rate at observation, beats/min) and BODYTEMP (vital-sign covariate): QTC_BL is a derived per-subject ECG summary statistic, not a per-observation vital-sign reading.
• Units: ms (the standard ECG / QT interval unit; document the correction method per-model in covariateData[[QTC_BL]]$notes).
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a centered linear deviation e_qtc_bl_e0 * (QTC_BL - QTC_BL_ref). Reference values observed: 390 ms (Darpo 2014; rounded standard reflecting the male-dominated 28 men / 11 women cohort with day-0 QTcI ranges 384-394 ms (men) / 398-412 ms (women) and similar QTcF ranges; the specific cohort median was not quoted in the paper). Future popPK-DeltaQTc models centering on a different reference (e.g., 400 ms in older / sicker cohorts where QTc is elevated) should document the per-model reference in covariateData[[QTC_BL]]$notes.
• Source aliases:
  • baseline QTcI – used in Darpo_2014_racSotalol_QTcI.R (day-0 individually-corrected QT interval, ms; values transformed via centering against the cohort-median reference inside model).
  • baseline QTcF – used in Darpo_2014_racSotalol_QTcF.R (day-0 Fridericia-corrected QT interval, ms; same covariate canonical, correction method documented per-model).
  • QTcF_i0 / Baseline QTcF – Mukker 2026 Equation 1 and Table 2; the per-subject time-zero QTcF, entering as theta3 * (QTcF_i0 - QTcF0) where QTcF0 is the cohort mean baseline. No value transformation.
• Example models: Darpo_2014_racSotalol_QTcI.R, Darpo_2014_racSotalol_QTcF.R (PD-only linear mixed-effects E-R model for DeltaQTcI / DeltaQTcF after single 160 mg oral rac-sotalol; QTC_BL is each subject’s pre-dose day-0 mean QTc by the corresponding correction method, time-fixed for the analysis), Mukker_2026_tuvusertib_QTcF.R (Garnett LME C-DeltaQTcF model in patients with advanced solid tumors; centered on the published cohort mean baseline QTcF of 422 ms rather than a rounded standard – Mukker 2026 defines QTcF0 in the Equation 1 legend as “the overall mean QTcF_i0, that is, the mean of all the baseline (time zero) QTcF values” and Table 1 reports that mean (SD) as 422 (21.0) ms). Sasaki_2022_delamanid_QTc_dm6705.R and Sasaki_2022_delamanid_QTc_parent.R (pediatric DeltaQTcB models; Bazett correction; the unpublished cohort-mean baseline centering constant is taken as 421 ms from the Figure S5 baseline QTcB-RR regression at the reported ~100 bpm mean heart rate).
• Notes: General scope because pre-dose QTc baseline is a universally applicable ECG-derived per-subject covariate in any concentration-DeltaQTc model. The QT correction method (Bazett, Fridericia, individual, study-population-specific) is model-specific and is the per-model documentation responsibility of covariateData[[QTC_BL]]$notes; do NOT register parallel canonicals like QTCI_BL or QTCF_BL for different correction methods (the unit, role, and centering pattern are identical – only the precise QT-to-QTc conversion formula differs upstream of the model). Future concentration-DeltaQTc papers using a non-zero centering reference different from the Darpo 2014 rounded standard 390 ms should document the per-model reference in covariateData[[QTC_BL]]$notes.

The _PRESURG / _BL haemodynamic anchor pair

Perioperative haemodynamic models frequently need a subject’s vital signs at two distinct pre-treatment anchors, and the distinction is load-bearing rather than cosmetic: the two anchors bracket the physiological insult the model is about to describe, and a model can carry both as covariates on the same structural parameter with opposite signs.

• _PRESURG – the last value recorded before the anaesthetic / surgical intervention, i.e. the subject’s own undisturbed physiological set-point.
• _BL – the value at the moment treatment starts (here, at hypotension diagnosis, immediately before the first antihypotensive bolus). This keeps _BL’s established at-first-dose meaning used throughout this register (QTC_BL, HGB_BL, FERRITIN_BL).

Do not collapse the two onto one column, and do not use a single _PRE suffix for both – _PRE reads as either anchor and has blurred them in draft extractions. When a paper reports only one pre-treatment value, use _BL unless it is explicitly pre-intervention.

MAP_PRESURG (canonical for pre-surgery mean arterial pressure)

• Description: Mean arterial pressure recorded before the anaesthetic / surgical intervention – the subject’s undisturbed baseline set-point. See the _PRESURG / _BL anchor-pair note above.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (MAP_PRESURG / ref)^exponent. Reference values observed: 100 mmHg (Dings 2026 population median; cohort mean 101 mmHg in both treatment arms).
• Source aliases:
  • MAP (pre-surgery row of Dings 2026 Table 2) – same orientation, no value transformation. The source distinguishes the two anchors by table row rather than by column name, so the row label must be carried into the column name at data-assembly time.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (power effect on the maximum attainable MAP: rmax_map = 119 * (MAP_PRESURG/100)^0.499 * (MAP_BL/64)^-0.53; a higher undisturbed set-point raises the attainable ceiling).
• Notes: General scope – a pre-intervention vital-sign anchor applies to any perioperative or procedural model. Sibling of SBP_PRESURG and HR_PRESURG; paired with MAP_BL. The corresponding model output state is map in compartment-names.md – this column is an input, not a prediction.

MAP_BL (canonical for mean arterial pressure at treatment baseline)

• Description: Mean arterial pressure at the moment treatment starts – in the founding model, at hypotension diagnosis, immediately before the first antihypotensive bolus. See the _PRESURG / _BL anchor-pair note above.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling and as the additive baseline anchor of an Emax equation. Reference values observed: 64 mmHg (Dings 2026 population median; cohort means 64.2 and 64.5 mmHg).
• Source aliases:
  • MAP (baseline row of Dings 2026 Table 2) – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (serves two distinct roles: the additive baseline of the MAP Emax equation map = MAP_BL + (rmax_map - MAP_BL) * ..., and a power covariate on the ceiling with exponent -0.53, opposite in sign to MAP_PRESURG’s +0.499). Sang_2021_doxorubicin_rat_qsp.R, Sang_2021_doxorubicin_human_qsp.R (baseline MAP sets baseline TPR = MAP_BL / (HR * SV_BL) and scales the MAP feedback constant as (MAP_BL / 106.596)^-1.98).
• Notes: General scope. The opposite-signed pairing with MAP_PRESURG on one parameter is the clearest demonstration of why the two anchors must stay separate columns – collapsing them would cancel most of the covariate model. Sibling of SBP_BL, HR_BL, DBP_BL.

SBP_PRESURG (canonical for pre-surgery systolic blood pressure)

• Description: Systolic blood pressure recorded before the anaesthetic / surgical intervention. See the _PRESURG / _BL anchor-pair note above.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – power scaling. Reference values observed: 139 mmHg (Dings 2026 population median; cohort means 139 and 138 mmHg).
• Source aliases:
  • SBP (pre-surgery row of Dings 2026 Table 2) – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (rmax_sbp = 169 * (SBP_PRESURG/139)^0.577 * (SBP_BL/92)^-0.494).
• Notes: General scope. The absolute-value pre-intervention sibling of the already-registered SBP (which is the unqualified / time-varying systolic pressure).

SBP_BL (canonical for systolic blood pressure at treatment baseline)

• Description: Systolic blood pressure at the moment treatment starts. See the _PRESURG / _BL anchor-pair note above.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – power scaling and additive Emax baseline anchor. Reference values observed: 92 mmHg (Dings 2026 population median; cohort means 92.3 and 93 mmHg).
• Source aliases:
  • SBP (baseline row of Dings 2026 Table 2) – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (additive baseline of the SBP Emax equation and a power covariate on the ceiling, exponent -0.494).
• Notes: General scope.

HR_PRESURG (canonical for pre-surgery heart rate)

• Description: Heart rate recorded before the anaesthetic / surgical intervention. See the _PRESURG / _BL anchor-pair note above.
• Units: beats/min
• Type: continuous
• Scope: general
• Reference category: n/a – power scaling. Reference values observed: 92 beats/min (Dings 2026 population median; cohort means 92.6 and 91.7 beats/min).
• Source aliases:
  • HR (pre-surgery row of Dings 2026 Table 2) – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (rmax_hr = 77.8 * (HR_PRESURG/92)^0.399 * (1 + 0.15 * TRT_EPHEDRINE)).
• Notes: General scope. The pre-intervention anchor sibling of the already-registered HR (unqualified / time-varying heart rate used as a covariate on clearance).

HR_BL (canonical for heart rate at treatment baseline)

• Description: Heart rate at the moment treatment starts. See the _PRESURG / _BL anchor-pair note above.
• Units: beats/min
• Type: continuous
• Scope: general
• Reference category: n/a – additive Emax baseline anchor. Reference values observed: 84 beats/min (Dings 2026 population median; cohort means 85.4 and 87.5 beats/min).
• Source aliases:
  • HR (baseline row of Dings 2026 Table 2) – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (additive baseline of the HR Emax equation, hr = HR_BL + (rmax_hr - HR_BL) * conc/(conc + ec50_hr) + ...; no covariate effect is estimated on it). Sang_2021_doxorubicin_human_qsp.R (initial condition of the HR turnover state; 70 beats/min typical, Sang 2021 Table 1).
• Notes: General scope. Completes the _BL anchor set with MAP_BL, SBP_BL and DBP_BL. HR_BL was not in the ratified name list of, but Dings 2026 Eq. A12 requires the at-diagnosis heart-rate anchor as the additive baseline of the HR Emax term, so it is registered here as the mechanical fourth member of the same ratified family rather than as a new concept.

DBP_BL (canonical for diastolic blood pressure at treatment baseline)

• Description: Diastolic blood pressure at the moment treatment starts. See the _PRESURG / _BL anchor-pair note above.
• Units: mmHg
• Type: continuous
• Scope: general
• Reference category: n/a – power scaling. Reference values observed: 50 mmHg (Dings 2026 typical-parturient value; see notes).
• Source aliases:
  • DBP (baseline row) / DBPBL – same orientation, no value transformation.
• Example models: Dings_2026_cafedrine_theodrenaline_ephedrine.R (power effect on the K/PD apparent volume: vc = 1 * (DBP_BL/50)^-1.49; a lower baseline diastolic pressure – a more profound sympathetic block – inflates the apparent distribution volume and so blunts the effective driving concentration).
• Notes: General scope. The absolute-value counterpart of the already-registered DBP_REL (a relative change from baseline, unitless), and the diastolic member of the _BL anchor set. Dings 2026 does not tabulate baseline DBP in Table 2, so the 50 mmHg centring value is taken from the typical-parturient definition in the paper’s Section 3.4, which Section 2.3 defines as the population median covariate values; it is internally consistent with the tabulated baseline MAP and SBP via MAP = DBP + (SBP - DBP)/3 (50 + (92 - 50)/3 = 64 mmHg, matching the reported baseline MAP of 64 mmHg).

LVEDV_BL (canonical for left ventricular end-diastolic volume at treatment baseline)

• Description: Pre-dose left ventricular end-diastolic volume (echocardiographic), used as the initial condition and steady-state anchor of an LVEDV (preload) turnover state in cardiovascular-systems models.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – enters as the initial condition of the edv state and in its steady-state production rate. Reference values observed: 0.385 mL (typical rat, Sang 2021 Table S5 from Snelder 2014); 113 mL (cardiovascular-healthy human) and 141 mL (LVEDV-enlarged human) in Sang 2021 Table 1.
• Source aliases:
  • LVEDVbase – regressor name in the Sang 2021 deposited Mlxtran code.
  • LVEDV0 – Sang 2021 Table 1 / Table S5 row name.
• Example models: Sang_2021_doxorubicin_rat_qsp.R, Sang_2021_doxorubicin_human_qsp.R (initial condition of the LVEDV turnover state; baseline SV = LVEDV_BL - LVESV_BL).
• Notes: General scope. Member of the _BL haemodynamic anchor set (MAP_BL, HR_BL, SBP_BL, DBP_BL). Units are mL even for rodents (a rat LVEDV is about 0.3 mL); convert microlitre echocardiography values (Sang 2021 Table S2 reports uL) by dividing by 1000. Distinct from the edv model state, which is the predicted time course.

LVESV_BL (canonical for left ventricular end-systolic volume at treatment baseline)

• Description: Pre-dose left ventricular end-systolic volume (echocardiographic). Together with LVEDV_BL it fixes the baseline stroke volume (LVEDV_BL - LVESV_BL) and baseline ejection fraction of a cardiovascular-systems model.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – used to derive baseline stroke volume and to anchor an LVESV feedback term. Reference values observed: 0.085 mL (typical rat, Sang 2021 Mlxtran LVESV0); 48 mL (cardiovascular-healthy human, Sang 2021 Table 1 LVEDV0 - SV0 = 113 - 65).
• Source aliases:
  • LVESVbase – regressor name in the Sang 2021 deposited Mlxtran code.
• Example models: Sang_2021_doxorubicin_rat_qsp.R, Sang_2021_doxorubicin_human_qsp.R (baseline SV and the steady-state production rate of LVEDV through the (1 - FB_LVESV * LVESV) dissipation feedback).
• Notes: General scope; sibling of LVEDV_BL. When a source reports baseline LVEF instead, derive LVESV_BL = LVEDV_BL * (1 - LVEF/100).

DUR_MAP_BELOW_PRESURG (canonical for cumulative duration of mean arterial pressure below the pre-surgery anchor)

• Description: Cumulative minutes for which a subject’s mean arterial pressure stays below their own MAP_PRESURG value, over a defined window (in the founding model, from antihypotensive treatment until umbilical-cord clamping). A per-subject exposure-to-hypotension metric rather than an instantaneous measurement: it integrates both the depth-independent time spent hypotensive and, implicitly, the adequacy of treatment. Typically derived from a simulated or observed MAP-time profile rather than recorded directly.
• Units: min
• Type: continuous
• Scope: general
• Reference category: n/a – enters linear regressions directly with a per-minute coefficient. Reference values observed: cohort means 10.1 min (cafedrine/theodrenaline) and 10.2 min (ephedrine) in Dings 2026 Table 2.
• Source aliases:
  • TIMEMAP – Dings 2026 Section 2.3 symbol; same orientation and units, no transformation.
  • Duration MAP<pre-surgery MAP – Dings 2026 Tables A2 / A3 row label.
• Example models: Dings_2026_neonatal_acidosis.R (the single strongest predictor of both neonatal umbilical-arterial pH, -0.00177 per minute, and base excess, -0.105 mmol/L per minute).
• Notes: General scope – a duration-below-own-baseline exposure metric generalises to any endpoint driven by cumulative hypotension or hypoperfusion. Named against the _PRESURG anchor because that is the comparator that defines it; a variant measured against a different anchor should be registered as a separate canonical (e.g. DUR_MAP_BELOW_BL) rather than silently reusing this one, since the two are numerically very different. The founding paper’s convention when MAP never recovers within the window is documented in covariateData[[DUR_MAP_BELOW_PRESURG]]$notes of the founding model.

ORGVOL_SALGLAND (canonical for measured total salivary gland volume)

• Description: Measured total volume of the salivary glands. The founding source delineates the left plus right parotid glands only; a model that includes the submandibular or sublingual glands should say so in covariateData[[ORGVOL_SALGLAND]]$notes rather than registering a separate canonical.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – used directly as a physiologic input that sets a compartment’s sub-volumes, its perfusion and its permeability-surface-area product; not a covariate-effect coefficient.
• Source aliases:
  • `Salivary Glands (measured volume) – Golzaryan 2025 Table S1 column header; reported in mL, no value transformation.`
• Example models: Golzaryan_2025_lu177psmaIT_pbpk.R.
• Notes: Founding member of the ORGVOL_<ORGAN> family. The salivary glands are a dose-limiting organ at risk in PSMA-targeted radioligand therapy, so their volume is measured per patient rather than scaled from body size. Cohort range 17-54 mL in five mCRPC patients.

ORGVOL_TUMOR1 (canonical for measured individually delineated tumour lesion 1 volume)

• Description: Measured volume of the first of the individually delineated tumour lesions that a model carries as its own compartment. Lesion numbering follows the source paper.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – used directly as a physiologic input that sets a compartment’s sub-volumes, its perfusion and its permeability-surface-area product; not a covariate-effect coefficient.
• Source aliases:
  • `Tumor 1 (measured volume) – Golzaryan 2025 Table S1 column header; reported in mL, no value transformation.`
• Example models: Golzaryan_2025_lu177psmaIT_pbpk.R.
• Notes: Founding member of the ORGVOL_<ORGAN> family; see the family note above for the distinction from TUM_VOL. Lesion indices are positional labels from the source, not anatomical sites, so a model that names its lesions anatomically should register named members (ORGVOL_TUMOR_BONE,…) instead of numbered ones. Cohort range 0.5-4 mL in five mCRPC patients.

ORGVOL_TUMOR2 (canonical for measured individually delineated tumour lesion 2 volume)

• Description: Measured volume of the second of the individually delineated tumour lesions that a model carries as its own compartment.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – used directly as a physiologic input that sets a compartment’s sub-volumes, its perfusion and its permeability-surface-area product; not a covariate-effect coefficient.
• Source aliases:
  • `Tumor 2 (measured volume) – Golzaryan 2025 Table S1 column header; reported in mL, no value transformation.`
• Example models: Golzaryan_2025_lu177psmaIT_pbpk.R.
• Notes: Founding member of the ORGVOL_<ORGAN> family. Extend the family with ORGVOL_TUMOR3 and beyond as models resolve more lesions. Cohort range 1-34 mL in five mCRPC patients.

ORGVOL_TUMORREST (canonical for measured lumped remaining-tumour-lesion volume)

• Description: Volume assigned to the single compartment that represents every tumour lesion a model does not delineate individually. Unlike the other members of this family this is an assumption rather than a measurement, because the lesions it stands for were not imaged individually; record how the source arrived at it in covariateData[[ORGVOL_TUMORREST]]$notes.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – used directly as a physiologic input that sets a compartment’s sub-volumes, its perfusion and its permeability-surface-area product; not a covariate-effect coefficient.
• Source aliases:
  • `R_TU,Rest,0 / [R_TU,Rest,0] – Golzaryan 2025 Table S2 states the lumped compartment through its total binding sites at an assumed 266 nmol/L density, which is the same quantity divided by that density.`
• Example models: Golzaryan_2025_lu177psmaIT_pbpk.R.
• Notes: Founding member of the ORGVOL_<ORGAN> family. The founding source assumes either 10 mL or 50 mL per patient.

URINE_FLOW (canonical for instantaneous urine flow rate)

• Description: Instantaneous urine flow rate (mL/h) measured over the urine collection interval that includes the current observation. Time-varying.
• Units: mL/h
• Type: continuous
• Scope: general
• Reference category: n/a. Two registered roles, both keyed on the same URINE_FLOW == 0 sentinel for “no urine collected during the interval”:
  • Covariate-effect role – a centered-linear effect CL_renal = base + theta_URINE_FLOW * (URINE_FLOW - URINE_FLOW_ref) with URINE_FLOW_ref = 100 mL/h in Allegaert 2015. The linear-effect term is gated by URINE_FLOW > 0 and not extrapolated below the centering reference.
  • Flow-denominator role – the column is not a covariate effect at all but the physical flow that converts a renal excretion RATE into the urine CONCENTRATION observable, C_Ur = CL_renal * C_p / URINE_FLOW, and the same zero sentinel gates the renal arm off entirely in a subject with no diuresis.
• Source aliases:
  • UF (Allegaert 2015 NONMEM column; same orientation, no transformation) – used in Allegaert_2015_paracetamol.R.
  • UVol – Butragueno-Laiseca 2022 supplementary material, “the measured volume of urine excreted in each urine recovery interval”. Reported as a per-interval VOLUME (mL); divide by the interval duration on ingestion to obtain the mL/h this column carries.
• Example models: Allegaert_2015_paracetamol.R (covariate-effect role), ButraguenoLaiseca_2022_piperacillin.R (flow-denominator role; Curine <- cl_renal * Cc / (URINE_FLOW / 1000) reproduces supplementary eq. 3’s A_Ur/UVol in rate form, and (URINE_FLOW > 0) supplies the paper’s “CL_Renal … absent in patients without diuresis” gate).
• Notes: Promoted from specific to general alongside the Butragueno-Laiseca 2022 piperacillin extraction. The earlier version of this entry warned that a second model using a different EFFECT FORM should register its own canonical rather than reuse URINE_FLOW; the maintainers ruled that this second model may reuse it. That warning still stands for a genuinely different COVARIATE-EFFECT parameterization (e.g. direct URINE_FLOW / URINE_FLOW_ref proportional scaling on a clearance), which should register a sibling such as URINE_FLOW_PROP. Distinct from URINE_VOL_24H, which is a 24-hour cumulative volume used as a residual-renal-function descriptor rather than a per-interval rate; a paper that fits urine concentrations needs the per-interval rate this column carries, not the daily summary. The full-word canonical name was chosen over the bare UF source-data abbreviation for clarity in source traces.

URINE_VOL_24H (canonical for 24-hour residual diuresis (total urine volume per day))

• Description: Total urine volume collected over a 24-hour interval, in mL per 24 h. Used as an indicator of preserved native renal function in CRRT-dependent / oliguric / anuric critically ill patients, where it characterises the residual non-CRRT renal-excretion pathway. Time-fixed per study day in the founding example (a single 24-hour summary collected by nursing staff during the day of PK sampling). Distinct from URINE_FLOW (instantaneous urine flow rate in mL/h measured per collection interval).
• Units: mL/24h
• Type: continuous
• Scope: general
• Reference category: n/a – used with an additive linear effect CL = CL_base + e_URINE_VOL_24H_cl * (URINE_VOL_24H / 100) centered at 100 mL/24h (the anuria cutoff), or as a binary preserved-diuresis gate (URINE_VOL_24H > 100) multiplying a renal-clearance arm. Clinical cutoffs used in the founding example: anuria < 100 mL/24h, oliguria 100-500 mL/24h, preserved diuresis > 500 mL/24h.
• Source aliases:
  • residual diuresis (Ulldemolins 2015 prose); typical NONMEM column abbreviation DIUR. Same value orientation, no transformation.
• Example models: Ulldemolins_2015_meropenem.R (critically ill adults with septic shock and continuous renal replacement therapy; reference 100 mL/24h; additive linear effect 0.22 L/h per (URINE_VOL_24H / 100) on meropenem total CL on top of the CRRT-mediated baseline 3.68 L/h), Huppe_2023_fosfomycin.R (critically ill adults with renal insufficiency on continuous venovenous hemodialysis; used as the binary preserved-diuresis gate (URINE_VOL_24H > 100) that switches the renal clearance arm off entirely in anuric patients, per the source’s own definition of preserved diuresis as residual diuresis exceeding 100 mL/24h), Yu_2023_vancomycin.R (critically ill adults on CVVH or CVVHDF; the sole retained covariate, entering vancomycin CL as the multiplicative power term 1.90^(log10(URINE_VOL_24H + 10) / 2.3) where 2.3 is the cohort median-normalising constant log10(160 + 10) and the +10 mL offset keeps the term finite for the 46.5% of the cohort at or below 100 mL/24h; the anuric typical CL of 1.39 L/h is the CRRT-mediated clearance and a subject producing 3 L/day reaches 2.77 L/h).
• Notes: Promoted from specific to general alongside the Huppe 2023 fosfomycin extraction, the second model to retain a 24-hour urine-volume covariate with consistent semantics. The two registered effect forms are both multiplicative-or-additive in the same 100 mL/24h anuria cutoff and so share this entry: Ulldemolins 2015 uses the additive linear base + slope * (URINE_VOL_24H / 100), and Huppe 2023 uses the binary gate (URINE_VOL_24H > 100) on a renal-clearance arm. Distinct from URINE_FLOW because (a) URINE_VOL_24H is a 24-hour cumulative volume (mL/24h), not an instantaneous rate (mL/h), and (b) the additive-linear effect form base + slope * (URINE_VOL_24H / 100) is structurally different from the centered-linear-effect-with-sentinel-zero form used by URINE_FLOW. A future model that uses a proportional or multiplicative effect form on a 24-hour volume should evaluate whether the semantics still match this entry (proportional / multiplicative forms are compatible); a centered-linear-with-sentinel form (analogous to the URINE_FLOW convention) would diverge and should register a sibling canonical. Yu 2023 vancomycin exercises that multiplicative branch: it is a power model in log10(URINE_VOL_24H + offset) normalised at the cohort median rather than at the 100 mL/24h anuria cutoff, which stays within this entry’s semantics (a monotone multiplicative effect of 24-hour urine volume on a total clearance whose zero-diuresis limit is the extracorporeal clearance). Note that the offset used to keep the log finite (+10 in Yu 2023) and the normalising volume are both paper-specific and belong in covariateData[[URINE_VOL_24H]]$notes of the model file, not in this entry. The full-word URINE_VOL_24H form was chosen over the shorter URINE_24H per the maintainers’ instruction.

ANURIA (canonical for binary anuria indicator)

• Description: 1 = the subject is anuric (produces no clinically meaningful urine and has no residual renal elimination); 0 = the subject has residual urine production. Bare per-subject binary, for source papers that ascertain and tabulate anuria as a yes/no patient characteristic rather than reporting a urine volume. Its operational role is to switch a renal-clearance arm off entirely, so that an anuric subject’s total clearance reduces to the non-renal and extracorporeal / peritoneal routes.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (residual urine production preserved). Note the orientation: the covariate is 1 for the subgroup in which the renal arm is ABSENT, so a model gates its renal clearance by (1 - ANURIA) rather than by ANURIA.
• Source aliases:
  • Anuric – Patel 2015 Table 1 column header, carried per patient as Yes / No. Maps directly onto the canonical with no value transformation (Yes -> 1).
• Example models: Patel_2015_oseltamivir.R (gates the renal arm of oseltamivir carboxylate elimination in end-stage-renal-disease patients on automated peritoneal dialysis: cl_renal_oselcarb <- (1 - ANURIA) * exp(lcl_renal_oselcarb + etalcl_renal_oselcarb). Five of the nine analysed patients were anuric and four had residual urine production; the paper reports the renal arm as the dominant elimination route in the urine producers, at 0.736 L/h/70 kg with 117% between-subject variability, and its dosing simulations partition the cohort into anuric / low-urine / high-urine strata precisely because the gate changes the achievable exposure by more than fourfold).
• Notes: Deliberately registered as a bare binary rather than derived from an existing urine-volume column, because the source papers this entry exists for do not report a volume. Distinct from URINE_VOL_24H, which is a CONTINUOUS 24-hour residual diuresis in mL/24h and whose own register entry already sanctions a (URINE_VOL_24H > 100) preserved-diuresis gate – when a paper reports the volume, use URINE_VOL_24H and its threshold gate, because the continuous form carries strictly more information and its anuria cutoff is explicit; use ANURIA only when the paper ascertained the binary and nothing finer. Deriving ANURIA from URINE_VOL_24H at ingestion time would be lossy in the other direction too: a paper’s clinical adjudication of “anuric” is not always the same as its measured volume falling below 100 mL/24h. Distinct from URINE_FLOW (instantaneous mL/h) and URINE_VOL_INTERVAL (per-collection-interval volume feeding an observation equation). Distinct from RENALIMP_ESRD and DIS_ARF, which are disease-state classifications: an end-stage-renal-disease cohort can contain both anuric and urine-producing patients, which is exactly the split this column encodes, so the two are complementary rather than substitutable. Scope: specific; promote to general if a second paper carries a bare anuria binary with the same renal-arm-gating semantics. Ratified by the maintainers 2026-09-02 (PMC4386947).

URINE_VOL_INTERVAL (canonical for urine volume collected in a single urine recovery interval)

• Description: Volume of urine actually collected during the urine recovery interval that a urinary observation belongs to, in mL. Time-varying: one value per collection interval, changing whenever the interval changes. Its role is to convert a modelled cumulative urinary AMOUNT into the urinary CONCENTRATION the assay reports – Curine = A_urine / URINE_VOL_INTERVAL – so it appears in an OBSERVATION equation, not as a covariate effect on a structural parameter. A value of 0 means no urine was produced in the interval (anuria) and there is consequently no urinary concentration to observe; models must guard the division.
• Units: mL
• Type: continuous
• Scope: general
• Reference category: n/a – a per-record divisor, not a centered or normalized covariate effect. No reference value applies.
• Source aliases:
  • UVol – Butragueno-Laiseca 2024 Methods, Base population model: “CUr was obtained as AUr/UVol, where UVol is the measured volume of urine excreted in each urine recovery interval”. Same quantity in mL, no value transformation.
• Example models: ButraguenoLaiseca_2024_meropenem.R (founding example; the urinary observable of a five-matrix meropenem model in critically ill children, Curine <- urine / (URINE_VOL_INTERVAL / 1000), converting mL to L so the modelled mg amount yields mg/L. Urine was collected before and 2, 4, 6 and 8 h after the start of the infusion, so the recovery intervals are roughly 2 h; the urine state is reset to zero at each interval boundary with an evid = 5, amt = 0 replacement record placed a numerical epsilon AFTER the boundary, because rxode2 applies dose-type records before same-time observations and a reset sitting exactly on the boundary would zero the state before the end-of-interval sample is taken), Trang_2021_meropenem.R, Trang_2021_vaborbactam.R (the urinary observables of the two co-fitted plasma-and-urine models for meropenem and vaborbactam in noninfected adults, Curine <- urine / (URINE_VOL_INTERVAL / 1000); urine was collected over 0-4, 4-8, 8-12, 12-24, 24-48 and 48-72 h intervals in the phase 1 studies, so the same interval-boundary reset applies), Nayak_2021_rivipansel.R (the urinary observable of a two-compartment rivipansel model fit jointly to plasma and urine, Curine <- urine / (URINE_VOL_INTERVAL / 1000); the source NONMEM stream sets S3 = UVOL on the ADVAN3 output compartment and observes it with CMT = -3 records, which read and then reset the compartment – the same per-interval reset as the evid = 5 idiom above).
• Notes: Deliberately NOT folded into either existing urine canonical, both of which were checked before ratification. URINE_VOL_24H is fixed to a 24-hour window and is used as a renal-function DESCRIPTOR (an anuria / oliguria / preserved-diuresis marker feeding a clearance covariate), whereas this column is the collection volume of one short PK recovery interval and feeds an observation equation. URINE_FLOW is an instantaneous RATE in mL/h whose own entry explicitly warns against reuse under different effect semantics. The two are related – URINE_VOL_INTERVAL divided by the interval duration is a urine flow – but the interval duration is not itself a column, so deriving one from the other at ingestion time would be lossy. Scope promoted to general once a second paper (Trang 2021) fitted urinary concentrations from interval collections.

FLUID_IN_24H (canonical for total daily fluid intake volume)

• Description: Total volume of fluid administered to (or ingested by) a subject over a 24-hour window – enteral plus parenteral, summed across all routes. A whole-body INTAKE volume: not weight-normalized, not a rate, and not a net balance (nothing is debited for urine, insensible or drain losses). Used as a renal-haemodynamic covariate on the clearance of renally eliminated, hydrophilic drugs, where a larger fluid load raises urine output and glomerular filtration and therefore raises clearance. Most natural in neonatal / paediatric intensive care, where daily fluid intake is prescribed and charted per patient per day and is a first-class clinical variable rather than a derived one.
• Units: mL/24h
• Type: continuous
• Scope: specific
• Reference category: n/a – used algebraically. Zhao 2025 enters it as a power term normalized to the cohort MEAN, CL = 0.14 * ... * (FLUID_IN_24H / 367.18)^0.14. Future papers may center on a median or use a linear form; record the per-paper parameterisation in covariateData[[FLUID_IN_24H]]$notes.
• Source aliases:
  • DFI (daily fluid input) – used in Zhao_2025_vancomycin.R (Zhao 2025 Microbiol Spectr Table 1 “Daily fluid input (mL)”; cohort median 364 mL, IQR 252.91-473.10, mean 367.18). Same quantity in mL per 24 h, no value transformation.
• Example models: Zhao_2025_vancomycin.R (founding example; vancomycin popPK in 126 non-extremely preterm NICU neonates, where daily fluid input is one of the two covariates the paper contributes as novel. Enters clearance as (FLUID_IN_24H / 367.18)^0.14 alongside body weight, serum creatinine and a concomitant-diuretic indicator; univariate dOFV -20.894 on CL, entered at forward-inclusion step 7 and survived backward elimination).
• Notes: Specific scope until a second paper ratifies the name; at that point promote to general. Distinct from URINE_VOL_24H, which is the OUTPUT counterpart over the same 24-hour window: the two are not interconvertible (they differ by insensible losses, stool water and net fluid retention), they can move in opposite directions in a fluid-overloaded or oliguric patient, and the founding paper screened BOTH and retained only the intake – Zhao 2025 Discussion records that daily urine volume “was not a significant covariate affecting clearance”, so collapsing the two would destroy the paper’s result. Distinct from PFA (perioperative intra-operative fluid administration volume), which is a gross cumulative volume over one surgical procedure rather than a per-day intake, and from PFA_NET_RATE, which is a net hourly balance that explicitly debits urine output. Distinct from the extracorporeal circuit flows (RRT_CRRT_EFFLUENT_FLOW, Q_CVVH, DFR), which are machine settings rather than a patient’s charted intake. Deliberately named FLUID_IN_24H rather than a bare FLUID_24H so the direction of flow is unambiguous against the registered urine-output family; a future paper reporting a charted net fluid BALANCE per day (intake minus output) should register a sibling FLUID_NET_24H rather than overloading this entry, and one reporting weight-normalized intake (mL/kg/24h, the more common neonatal prescribing unit) should record the normalisation in per-model notes rather than silently rescaling this column.

CSF_DRAIN_VOL_24H (canonical for the volume of cerebrospinal fluid removed through a CSF drain over 24 hours)

• Description: Total volume of cerebrospinal fluid drained out of a patient through an external ventricular drain, a lumbar cistern drain or an equivalent CSF diversion device over a 24-hour window. A charted clinical observation, not a derived quantity: neurosurgical units record the drain output per day and titrate it against the patient’s intracranial pressure and post-operative symptoms. Its modelling role is mechanistic rather than covariate-effect: for a drug with a CSF compartment, the drained fluid carries drug out of that compartment, so the column enters model() directly as the CSF compartment’s drainage clearance (CL_csf = CSF_DRAIN_VOL_24H / 24, converted from mL/h into the model’s volume units) rather than as a coefficient on a structural parameter. This is the same “measured prescription quantity enters model() directly” role that RRT_PERIT_DIAL_FILL_VOLUME already sanctions. A patient with no drain in situ takes the value 0, which removes the drainage term entirely; the column is therefore safe to default to 0 for a non-drained cohort.
• Units: mL/24h
• Type: continuous
• Scope: general
• Reference category: n/a – used algebraically, not as a centered effect. Lu_2016_meropenem.R divides it by 24 to obtain the CSF drainage clearance in mL/h. Clinically relevant values from the founding example: 0 (no drainage), 50, 150 and 250 mL/24h are the Monte Carlo scenarios, and the paper’s dosing recommendation is conditioned on keeping drainage below 150 mL/24h.
• Source aliases:
  • CSF daily drainage vol – Lu 2016 Table 1 row label (“CSF daily drainage vol, ml”). Same quantity in mL per 24 h, no value transformation.
  • daily CSF drainage volume – Lu 2016 Results prose form.
  • QEVD – Chauzy 2022 (Antibiotics 11:1293) Eq. 6-8 and Table S3, the EVD flow in L/h measured per 0.5-1 h collection interval and supplied to the model as a time-varying covariate. Multiply by 24000 to obtain mL/24h (QEVD = CSF_DRAIN_VOL_24H / 1000 / 24).
• Example models: Lu_2016_meropenem.R (founding example; meropenem plasma + CSF popPK in 82 adults with post-neurosurgical bacterial meningitis, cohort mean 126 +/- 81 mL/24h and range 0-350. The CSF compartment’s drainage clearance is CSF_DRAIN_VOL_24H / 1000 / 24 L/h, which at the cohort mean removes roughly a third as much drug from the CSF compartment as back-transfer to plasma does, and which is the mechanism behind the paper’s headline recommendation to limit drainage); Chauzy_2022_metronidazole_pbpk.R (minimal CNS PBPK of unbound metronidazole in 8 brain-injured adults, 4 with an EVD. The drain flow removes drug from the cranial CSF compartment and displaces the physiological CSF sink flow, Qsink = max(0, 0.024 - QEVD) L/h; per-interval flows of 0-0.040 L/h, i.e. 0-960 mL/24h, are tabulated in Table S3. Time-varying within a dosing interval, unlike the founding example’s per-day value).
• Notes: Named by composing three shapes the register already uses: the CSF_<quantity> prefix from CSF_TPRO (a quantity measured in, or pertaining to, the CSF matrix), the <FLUID>_VOL_24H form from URINE_VOL_24H, and the direction-explicit naming that FLUID_IN_24H‘s entry insists on. The maintainers’ ruling (PMC5075067, 2026-09-21) chose the full-word CSF_DRAIN_VOL_24H over the shorter CSF_DRAIN_24H and over a rate-valued CSF_DRAIN_RATE in mL/h, for consistency with URINE_VOL_24H (whose own Notes record the same full-word preference) and because the source paper charts, tabulates and simulates the quantity strictly per day. Distinct from URINE_VOL_24H and FLUID_IN_24H, which are whole-body renal output and whole-body intake respectively and are renal-function / haemodynamic descriptors feeding covariate effects; this column is a matrix-specific efflux from one modelled compartment and is not interconvertible with either. Distinct from CSF_TPRO, which is a concentration measured in CSF rather than a volume removed from it. A future paper reporting an instantaneous CSF drain flow rate in mL/h rather than a per-day total should record the conversion in covariateData[[CSF_DRAIN_VOL_24H]]$notes if the semantics are otherwise identical, or register a sibling rate canonical if it is used with different effect semantics – the same boundary that URINE_FLOW and URINE_VOL_24H already draw. general scope: used by two independent papers, one with a per-day value and one with a time-varying per-interval flow converted to the 24-hour basis.

TER_MAG3 (canonical for tubular extraction rate measured by 99mTc-MAG3 renography)

• Description: Tubular extraction rate of 99mTc-mercaptoacetyltriglycine (MAG3), the whole-kidney tubular secretion clearance measured by MAG3 renography. Not BSA-normalized and not a filtration measurement: MAG3 is handled predominantly by active proximal-tubular secretion, whereas creatinine clearance and 51Cr-EDTA / iohexol clearance measure glomerular filtration. Papers that use TER as their renal-function input typically convert it to a filtration rate with a published regression before it enters the model.
• Units: mL/min
• Type: continuous
• Scope: general
• Reference category: n/a – used as a direct model input, not as a covariate-effect coefficient.
• Source aliases:
  • TER – the bare symbol used by Golzaryan 2025 Table S1 and by the upstream Kletting 2016 PLoS One Table 1 (“TER = tubular extraction rate as determined with the 99mTc Mag3 method”); same quantity in mL/min, no value transformation.
• Example models: Golzaryan_2025_lu177psmaIT_pbpk.R (Golzaryan 2025 whole-body 177Lu-PSMA I&T PBPK; per-patient TER of 136-252 mL/min is converted to the 51Cr-EDTA glomerular filtration rate as GFR = TER / 3 * 20 / 15 per Table S2, then scaled to peptide molecular size by the sieving ratio phi = 0.66 to give the filtration flow used by the kidney sub-model – founding example).
• Notes: Deliberately NOT folded into CRCL. CRCL pools creatinine-based and tracer-measured estimates of glomerular filtration, normalized to 1.73 m^2; TER_MAG3 is non-normalized tubular extraction, a physiologically distinct process measured by a different tracer, and the two are related only through a cohort-specific regression. Reusing CRCL would require an undocumented conversion at data-ingestion time and would silently mix two renal processes under one column. The tracer is named in the canonical because the number is uninterpretable without it – a tubular extraction rate measured with 131I-OIH is not numerically interchangeable with a MAG3 one.

CRCL (canonical for BSA-normalized renal function (creatinine-based estimate OR tracer-measured GFR))

• Description: BSA-normalized renal function expressed in mL/min/1.73 m^2. Accepts either (a) a creatinine-based estimate – MDRD- or CKD-EPI-estimated glomerular filtration rate, or a measured creatinine clearance that has been BSA-normalized as 1.73 x CrCl / BSA – or (b) a tracer-measured glomerular filtration rate using an exogenous filtration marker (iohexol clearance, inulin clearance, 99mTc-DTPA clearance, 51Cr-EDTA clearance), which is the clinical gold standard for measured GFR. All variants enter popPK models with the same operational role (a covariate on clearance) and the same units; the per-model covariateData[[CRCL]]$description and notes must state which assay the source paper used.
• Units: mL/min/1.73 m^2
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (CRCL / ref)^exponent or with linear centering (1 + slope * (CRCL - ref)). Reference values observed: 76 mL/min/1.73 m^2 (Hamren 2008, iohexol-clearance-measured GFR), 80 mL/min/1.73 m^2 (Cirincione 2017, MDRD eGFR), 90 mL/min/1.73 m^2 (Li 2019, calculated GFR; Bajaj 2017, CKD-EPI eGFR), 100 mL/min/1.73 m^2 (Xu 2019, measured-CrCl BSA-normalized), 141 mL/min/1.73 m^2 (Jung 2024, bedside Schwartz eGFR in critically ill children, whose renal function skews supranormal), 96 mL/min/1.73 m^2 (Shiau 2026, CKiD U25 eGFR in critically ill children with multiple organ dysfunction syndrome).
• Source aliases:
  • eGFR – estimated glomerular filtration rate; the estimating equation varies by cohort and must be documented per model. MDRD variant used in Cirincione_2017_exenatide.R and Kotani_2022_astegolimab.R; creatinine-based CKD-EPI variant in Bajaj_2017_nivolumab.R; the cystatin-C-based CKD-EPI variant in Patel_2025_eteplirsen.R, where creatinine is explicitly rejected as a renal marker because it is a by-product of the muscle breakdown that defines Duchenne muscular dystrophy (cystatin C is produced by all nucleated cells and is therefore independent of muscle mass – see CYSC); the paediatric bedside Schwartz variant 0.413 x height (cm) / Scr (mg/dL) in Jung_2024_vancomycin.R; the combined creatinine-cystatin C CKD-EPI 2021 variant in Taylor_2026_methotrexate.R, where the paper’s whole purpose was to show that the combined equation beats both single-marker CKD-EPI equations and either raw laboratory value as a predictor of clearance (dOFV 117.1 against the base model, versus 81.5 for cystatin C alone and 42.9 for creatinine alone). The same combined-equation variant, under the source column name eGFRCre-Cys, is used in Tseng_2026_piperacillin.R, which ran the same three-way comparison in the opposite population – adults with LOW body weight, where creatinine-based equations OVERestimate filtration because serum creatinine is a product of skeletal muscle metabolism – and reached the same ranking (dOFV 12.13 for the combined equation, versus 10.50 for cystatin C alone and 6.48 for Cockcroft-Gault creatinine clearance).
  • EGFR – all-caps variant.
  • CECYS – the CKD-EPI cystatin C eGFR in mL/min/1.73 m^2 (BSA-normalized; same units and orientation, no value transformation). Used in Kim_2022_piperacillin.R and Kim_2022_tazobactam.R (critically ill adults with and without ECMO) as the exponential effect exp(theta * (CRCL - 52.77)) on CL; selected over Cockcroft-Gault, MDRD, creatinine CKD-EPI, combined creatinine-cystatin C CKD-EPI and the BSA-de-normalized (mL/min) variants of each.
  • U25, GFR (CKiD U25) – the CKiD “Chronic Kidney Disease in Children Under 25” serum-creatinine eGFR equation, BSA-normalized to mL/min/1.73 m^2. An age- and sex-specific paediatric/young-adult successor to the bedside Schwartz equation (whose variant is recorded above under eGFR for Jung_2024_vancomycin.R), developed to give continuous estimates across the paediatric-to-adult transition rather than breaking at age 18. Used in Marianski_2026_ceftriaxone.R under the source column name GFR, where it enters clearance as the through-origin linear ratio (CRCL / 120). Note that the 120 reference there is an external normal-renal-function anchor, NOT the cohort median of 100 – so unlike most entries in this register the printed typical CL is not the typical patient’s CL. Same units and orientation as every other member of this canonical, so no value transformation is needed.
  • CLCRN / NCrCL (Franzese 2022) – BSA-normalized creatinine clearance, mL/min/1.73 m^2, computed with the updated bedside Schwartz equation in children and as Cockcroft-Gault creatinine clearance x 1.73 / BSA in adults, so one column spans a pediatric-to-adult cohort under two estimating equations on a single normalized scale. Used in Franzese_2022_ceftazidime.R (two-segment linear spline hinged at 100, capped at 150) and Franzese_2022_avibactam.R (power below / linear above an 80 hinge, capped at 150); in both it is read only for AGE > 2 years, a postmenstrual-age maturation function replacing it at younger ages. Note the parent adult models Li_2019_ceftazidime.R / Li_2019_avibactam.R used raw Cockcroft-Gault mL/min, so the two generations are not interchangeable on the same input.
  • CRCL_BSA – BSA-normalized creatinine clearance (measured CrCl / BSA x 1.73); used in Xu_2019_sarilumab.R.
  • GFR (Chinese-modified MDRD-type eGFR, raw mL/min) – serum-creatinine eGFR with a 0.742 female factor and a 1.233 Chinese correction, reported in mL/min and NOT BSA-normalized. Used in Ma_2020_vancomycin.R (adult kidney transplant recipients; power term (CRCL / 36.67)^0.698 on CL, where 36.67 is not the Table 1 per-patient median of 39.91 mL/min). The source’s typeset estimating equation has lost its exponent formatting, so the equation cannot be reconstructed from that paper; users supply the eGFR value directly.
  • eGFR (creatinine CKD-EPI as printed with a 144 multiplier for both sexes, raw mL/min) – Wei 2022 Eq. (4) writes 144 x (Scr/a)^b x 0.993^age (Scr in mg/dL; a = 0.7 female / 0.9 male) with no sex-specific 141 multiplier and no 1.73 m^2 term, and labels the result mL/min. Used in Wei_2022_vancomycin.R (postoperative neurosurgical adults; power term (CRCL / 115.2)^0.80 on CL, 115.2 mL/min being the cohort median). Supply the value exactly as that equation computes it.
  • 1.73*CrCl/BSA – the formula form appearing in Xu 2019 Eq. for Vm.
  • cGFR – calculated/estimated GFR, BSA-normalized; used in Li_2019_abatacept.R.
  • eGFR-LMR18rel / LMR18rel – the Lund-Malmo Revised 2018 creatinine-based estimating equation, BSA-normalized (mL/min/1.73 m^2). A Scandinavian-derived alternative to MDRD and CKD-EPI; used in Beijer_2026_cloxacillin.R. The companion LMR18abs / eGFR-LMR18abs is the SAME equation de-normalized back to absolute mL/min via a Du Bois BSA, so the two are not interchangeable – see the Notes on size normalisation below.
  • CRCL3 (CKD-EPI / bedside Schwartz eGFR de-indexed by BSA) – raw mL/min, NOT BSA-normalized. Nayak 2021 computes CKD-EPI eGFR for adults and bedside Schwartz eGFR for children younger than 18 years, then multiplies by the Du Bois BSA over 1.73 (the paper’s ‘normalized by the BSA’); Table 2 fixes the direction, since children aged 6-11 years with a Schwartz eGFR near 140 mL/min/1.73 m^2 are tabulated at 91.6 mL/min. Used in Nayak_2021_rivipansel.R (reference 119 mL/min, the control stream’s median, on the renal clearance arm; the exponent rises by 0.413 below 60 mL/min).
  • CLCR – source-paper column name; underlying assay form varies. Used in Delattre_2010_amikacin.R (raw Cockcroft-Gault, NOT BSA-normalized; median 55.5 mL/min in critically ill septic adults), in MedellinGaribay_2015_gentamicin.R (Schwartz formula CLCR = K * length / SCr with K in {0.33, 0.45, 0.55}, BSA-normalized to mL/min/1.73 m^2), and in Zhou_2019_vancomycin.R (raw Cockcroft-Gault, NOT BSA-normalized; reference 56.28 mL/min, the cohort mean, in Chinese geriatric patients with pulmonary infections – note that the reference here is a MEAN rather than the median most entries in this register use, because the source describes its normally distributed continuous covariates as mean +/- SD), and in Chauzy_2022_ceftaroline.R (creatinine clearance measured from a 24 h urine collection, raw mL/min, NOT BSA-normalized; time-varying, measured on each of two PK occasions; reference 180 mL/min, the cohort median, in ventilated ICU adults with augmented renal clearance, 83-309 mL/min), and in Sima_2022_ciprofloxacin.R (MEASURED 24-h urine creatinine clearance UCR x V / SCR, NOT BSA-normalized, reported in mL/s; the model divides the mL/min column by 60 and applies an uncentered exponential effect on the parent-to-metabolite transfer rate constant). Document the assay form per model in covariateData[[CRCL]]$description.
  • CRCL (Cockcroft-Gault on total body weight, serum creatinine in umol/L with a 0.818 divisor) – raw mL/min, NOT BSA-normalized. Jin 2022 Table 1 footnote: [(140 - Age) x weight (kg)] / [0.818 x Scr (umol/L)] x k, k = 1 for men and 0.85 for women. Used in Jin_2022_lithium.R as the median-centred power (CRCL / 116)^0.186 on lithium CL/F (cohort median 116 mL/min, range 61.7-226, in Chinese patients with bipolar disorder).
  • CrCL (Cockcroft-Gault >= 19 years, modified Schwartz < 19 years, both raw mL/min, UNCAPPED in pregnancy) – not BSA-normalized. Used in Goyal_2022_vancomycin.R (hospitalized pregnant women; reference 175 mL/min with exponent 1 on CL; cohort median 176, range 43-389 mL/min). The source deliberately does not cap the estimate at the conventional 120-150 mL/min because glomerular filtration rises in pregnancy: capping at 120 or 150 mL/min raised the OFV and biased clearance upward. A user supplying a capped value to this model will under-predict clearance in the high-CRCL tail.
  • CrCL (Cockcroft-Gault on a BMI-SELECTED body weight) – raw mL/min, NOT BSA-normalized. Used in Jin_2026_colistinSulfate.R (reference 116.3 mL/min, the cohort median, in critically ill adults with carbapenem-resistant Gram-negative pneumonia). The distinguishing feature is that the weight entering the Cockcroft-Gault numerator is chosen by body-mass-index band rather than being total body weight throughout: actual body weight if BMI < 18.5 kg/m^2, ideal body weight (IBW = Constant + 0.91 x (height_cm - 152.4), Constant 50 for men and 45.5 for women) if BMI is 18.5-24.9 kg/m^2, and adjusted body weight (ABW = IBW + 0.4 x (total body weight - IBW)) if BMI >= 25 kg/m^2. The source’s serum creatinine is in umol/L, so its equation carries the standard 0.814 divisor. Note that the typeset equation names “ideal body weight” in its numerator while the surrounding prose specifies the three-way BMI rule; the prose is operative because it is the description that defines all three cases. A model carrying this alias is NOT interchangeable with one built on a total-body-weight Cockcroft-Gault column in an obese cohort, where the ABW branch can run well below the total-weight value.
  • COGBWA (Cockcroft-Gault on body weight at ICU admission) – raw mL/min, NOT BSA-normalized. Used in Swartling_2022_cefotaxime.R (critically ill adults; centred at the cohort median 94 mL/min and truncated at 120 mL/min, above which the linear CL effect is flat). The weight in the Cockcroft-Gault numerator is the ICU-ADMISSION weight, which includes weight gained from resuscitation fluids, not the pre-admission weight.
  • CG / eCrClCG – Cockcroft-Gault estimated creatinine clearance, raw mL/min, NOT BSA-normalized; the body-weight convention inside the Cockcroft-Gault equation is not stated by the source. Used in Gijsen_2022_meropenem.R (reference 111.7 mL/min, printed in the covariate equation without saying which cohort statistic it is; power exponent 0.637 on clearance in critically ill septic adults with preserved or increased renal function).
  • CrCL (Cockcroft-Gault on total body weight, raw mL/min) – (140 - age) x weight / (72 x SCr in mg/dL), times 0.85 for women, NOT BSA-normalized. Used in Zhou_2022_simurosertib.R (adults with advanced solid tumors; power exponent 0.325 on CL/F centred on 90.45 mL/min, the analysis-data-set median printed in the covariate equation, whereas the Table 2 per-patient median is 89.9 mL/min).
  • CLcr (Bjornson Rcr / SCr, INCOMPLETE) – the source computes Bjornson’s creatinine production rate Rcr (males 27 - 0.173 x age, females 25 - 0.175 x age, both mg/kg/24 h) and divides it by serum creatinine in mg/dL, then labels the result mL/min. The published Bjornson method finishes with CrCl = Rcr x weight / (SCr x 14.4); the weight multiplier and the 14.4 divisor are both omitted because the source dataset recorded no body weight, so the column runs about weight / 14.4 = 4.9-fold BELOW a Cockcroft-Gault clearance for the same patient and is dimensionally (mg/kg/24 h)/(mg/dL), not mL/min. Used in Ahmed_2024_vancomycin.R. This alias is a warning, not just a naming note: because the covariate enters as a power term, supplying a conventional mL/min creatinine clearance to such a model inflates typical CL by 4.9^exponent. Verify the arithmetic against the source’s own cohort median before reusing any model that carries this alias – for Ahmed 2024 the Table 1 median patient (male, 65 years, SCr 1.2 mg/dL) gives (27 - 0.173 x 65) / 1.2 = 13.1 against a reported median of 12.7, which identifies the form unambiguously. A cohort that looks uniformly and implausibly renally impaired is the tell.
  • CLCR,LBW – creatinine clearance computed on LEAN body weight rather than total body weight, raw mL/min and NOT BSA-normalized. Rolsma 2025 builds it with the Cockcroft-Gault equation substituting lean body weight for subjects over 12 years of age and the Schwartz bedside equation for subjects up to 12 years, so a single column spans a paediatric-to-adult cohort under two different estimating equations. Used in Rolsma_2025_cefepime.R (reference 77 mL/min) and Rolsma_2025_meropenem.R (reference 90 mL/min).
  • CrCL (Cockcroft-Gault >= 13 years in mL/min, Schwartz < 13 years in mL/min/1.73 m^2, RRT from ultrafiltration rate and blood flow) – one column spanning a paediatric-to-adult hospitalized-influenza cohort under two estimating equations on two scales, like CLCR,LBW above. Used in Zuo_2020_zanamivir.R as a piecewise-linear hinge on CL: factor 1 at or above an estimated knot of 97 mL/min and 1 + 0.00929 x (CRCL - 97) below it.
  • CrCL (Cockcroft-Gault on total body weight, UNCENTRED exponential) – raw mL/min, NOT BSA-normalized. Used in AlQurain_2022_tramadol.R (older inpatients, cohort mean 65.1 mL/min, range 16-351) as exp(beta * CRCL) on both tramadol CL/F (beta 0.00498) and V2/F (beta 0.0119) with no reference value, the Monolix default; the printed typical CL/F and V2/F are therefore the values at CrCL = 0, not at the typical patient.
  • Clcr (measured 8-h urinary creatinine clearance) – raw mL/min, NOT BSA-normalized. Used in BeraldiMagalhaes_2021_ethambutol.R as the through-origin ratio (CRCL / 101) on clearance (reference 101 mL/min; cohort medians 92.3 mL/min in ICU patients and 113.88 mL/min in outpatients, Table 1). The paper’s simulation methods label the same quantity mL/min/1.73 m^2, which conflicts with its Table 1; the Table 1 unit is taken as authoritative.
  • CLCRurinary (measured urinary creatinine clearance) – raw mL/min, NOT BSA-normalized. Used in NgougniPokem_2022_temocillin.R as the through-origin ratio (CRCL / 39.9) on the clearance from the central compartment (reference 39.9 mL/min, the cohort median, range 20.55-149.3, in critically ill adults with septic shock and ascitic fluid effusion; Supplementary Table S1 Ke = CLi*(CLCRurinary/39.9)/V).
  • CLiohexol – plasma iohexol clearance, the clinical gold-standard tracer-measured glomerular filtration rate; used in Hamren_2008_tesaglitazar.R (tesaglitazar / acyl glucuronide interconversion popPK in subjects with varying renal function). Iohexol is an exogenous contrast agent cleared exclusively by glomerular filtration (no tubular secretion or reabsorption); its clearance is a direct measurement of GFR.
  • CLinulin – plasma inulin clearance, the historical gold standard for measured GFR; analogous to iohexol clearance but requires constant intravenous infusion of inulin.
  • CL_DTPA – plasma 99mTc-DTPA clearance, a radiolabelled tracer of glomerular filtration commonly used in clinical practice.
  • CL_EDTA – plasma 51Cr-EDTA clearance, another radiolabelled glomerular-filtration tracer.
  • CrCl1 – pre-dose creatinine clearance on a scale whose printed unit label is WRONG. Zhang 2015 Table 1 heads the row ml/min with median 1.88 and range 0.94-4.64, magnitudes that are anuric as mL/min and are instead consistent with the SI unit mL/s (1.88 mL/s = 113 mL/min). Used in Zhang_2015_methotrexate.R. This alias is a warning, not just a naming note, and belongs with the CLcr (Bjornson) alias above as the second registered case of a renal-function column that is NOT interchangeable with a conventional mL/min value: the coefficient is calibrated per unit on the source’s own numeric scale and centred at 1.89 on that scale, so supplying mL/min inflates the covariate deviation about 60-fold. Two independent tells identify it. First, the same Table 1 mislabels serum creatinine as mg/dl with median 49 and range 19-81, values that are only sane as umol/L – when one row of a demographics table is demonstrably in SI units, check every other row before trusting its unit column. Second, a Cockcroft-Gault estimate for the paper’s own median patient gives about 178 mL/min against a tabulated median equivalent to 113 mL/min, the expected direction and magnitude of Cockcroft-Gault bias when creatinine production is low, which is consistent with a MEASURED 24 h clearance (the protocol collected 24 h urine). A weight-normalized reading (mL/min/kg, as in Zhang_2026_tebipenem.R) gives essentially the same absolute clearance for the median patient and cannot be excluded from the paper alone; what matters for reuse is that the model be driven on the source’s numeric scale either way.
  • CREACL – ‘creatinine clearance (calculated by CKD-EPI equation)’ reported in mL/min with no statement of whether the CKD-EPI value was de-normalised from mL/min/1.73 m^2; used in Michalickova_2020_riociguat.R (reference 70 mL/min, the cohort median; enters the riociguat non-M1 clearance through the origin as CLe,r/F = 0.66 * (CRCL / 70)).
• Example models: Cirincione_2017_exenatide.R (MDRD eGFR), Xu_2019_sarilumab.R (measured CrCl BSA-normalized), Kotani_2022_astegolimab.R (MDRD eGFR), Li_2019_abatacept.R (cGFR), Bajaj_2017_nivolumab.R (CKD-EPI eGFR, reference 90 mL/min/1.73 m^2), NA_NA_lidocaine.R (DDMODEL00000281; binary stratification at threshold 52.7 mL/min adding -0.319 to the GX rate constant K30 in the CRCL <= 52.7 cohort; the source .ctl does not state the BSA-normalisation method), Delattre_2010_amikacin.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 55.5 mL/min population median; additive linear effect 1.42 L/h per (CRCL/55.5) on CL), MedellinGaribay_2015_gentamicin.R (Schwartz BSA-normalized CLCR; reference 75 mL/min/1.73 m^2 (population mean 76.7); additive linear effect 0.06 L/h per (CRCL/75) on CL in infants 1-24 months), Georges_2009_ceftazidime.R (raw MDRD-eGFR mL/min, NOT BSA-normalized; mean 121 mL/min; additive linear effect 0.024 L/h per mL/min on CL, no centering), Hamren_2008_tesaglitazar.R (iohexol-clearance-measured GFR; reference 76 mL/min/1.73 m^2; linear centered slope 0.99 %/(mL/min/1.73 m^2) on renal clearance of parent tesaglitazar AND simple linear normalised scaling (CRCL / 76) on the saturable Michaelis-Menten Vmax of acyl-glucuronide renal elimination; founding tracer-GFR example), Chen_2023_nemonoxacin.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; cohort 114.4 +/- 29.2 mL/min, median 116.2, range 50.7-200.7; additive linear effect 0.026 L/h per mL/min on a 10.3 L/h intercept with no centering, per Chen 2023 Eq. 8 CL = (10.3 + 0.026 x CLcr) x (BW/70)^0.75; source column CLcr), Wada_2023_sparsentan.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 112 mL/min overall cohort median, range 26-361; power effect (CRCL / 112)^0.222 on CL/F, so lower renal function raises sparsentan exposure even though urinary excretion accounts for only 2.18% of the dose – the paper notes the association may be partly confounded by sex, which is a Cockcroft-Gault input and also a retained covariate on CL/F), Shu_2024_posaconazole.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 103.81 mL/min cohort median, range 14.44-240.64 in Chinese HSCT recipients spanning ages 3-65; power effect (CRCL / 103.81)^0.68 on CL/F for posaconazole oral suspension, of which only about 13% of a dose is renally excreted – as with Wada 2023 the effect is more plausibly a marker of general physiological reserve than a mechanistic renal-elimination pathway, and the authors flag it as a departure from earlier posaconazole popPK models), Suzuki_2024_mycophenolic_acid.R (Cockcroft-Gault standardised to a fixed 70 kg body weight – i.e. mL/min/70 kg, NOT BSA-normalised and NOT a raw value; cohort 2.5th / 50th / 97.5th percentiles 6.6 / 52.8 / 148.8 in renal-transplant recipients at a single Japanese centre; enters as the ratio CRCL / 100 multiplying the renal arm of an additive renal-plus-non-renal apparent clearance, with body size handled separately by an allometric (WT/70)^0.75 term), Jung_2024_vancomycin.R (bedside Schwartz paediatric eGFR, BSA-normalized; reference 141 mL/min/1.73 m^2; power effect (CRCL / 141)^0.5259 on CL in critically ill children – a less-than-proportional renal effect, and the sole retained covariate on clearance), Nakai_2025_tranexamicAcid.R (raw Cockcroft-Gault mL/min computed with actual body weight, NOT BSA-normalized; reference 61.0 mL/min cohort median, range 21.8-147.5 in adults on cardiopulmonary bypass; power effect (CRCL / 61.0)^0.752 on CL; source column CLcr, and note that the paper’s summary equation misannotates the covariate as “CLcr (L/h)” while dividing by the mL/min median – the Discussion’s worked example resolves the ratio to mL/min), Rolsma_2025_cefepime.R, Rolsma_2025_meropenem.R (alias CLCR,LBW; raw Cockcroft-Gault-on-lean-body-weight / Schwartz-bedside mL/min, NOT BSA-normalized; references 77 and 90 mL/min respectively, both being the cohort typical value; through-origin linear normalisation (CRCL / ref) on the RENAL arm of an additive renal-plus-non-renal clearance, with the non-renal arm carrying the body-size covariate instead), Patel_2025_eteplirsen.R (cystatin-C-based CKD-EPI eGFR, BSA-normalized; reference 145 mL/min/1.73 m^2 cohort median, range 85.5-180 in boys with Duchenne muscular dystrophy; power effect (CRCL / 145)^1.60 on CL. Two features are unusual and worth carrying forward: the cohort is uniformly supranormal – no subject fell below 85.5 – so the model carries no information about renal impairment and the paper says so explicitly; and the estimating equation was chosen because creatinine is invalid in this disease, not merely as a convenience), Lee_2025_levofloxacin.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 105.71 mL/min cohort median, observed range 74.8-113 in 12 healthy Korean adults; power effect (CRCL / 105.71)^0.901 on CL, the paper having tested and rejected an additional non-CrCl clearance component as negligible – note that the paper’s own Monte Carlo dosing simulations extrapolate this term from the fitted 74.8-113 range out to 10-170 mL/min), Taylor_2026_methotrexate.R (CKD-EPI creatinine-cystatin C 2021 eGFR, BSA-normalized; reference 76 mL/min/1.73 m^2; power effect (CRCL / 76)^0.80 on CL in adults with lymphoma receiving high-dose methotrexate. Two features are worth carrying forward: the covariate is genuinely LONGITUDINAL – creatinine and cystatin C were drawn alongside every methotrexate sample and re-converted to eGFR at each timepoint, so the 76 is the median of the longitudinal record and sits BELOW the baseline medians of 82.5 and 80.2 mL/min/1.73 m^2 because renal function declines across the post-infusion monitoring window; and the source paper’s Table 1 heads the renal block “Estimated kidney function (mL/min)”, pooling the BSA-normalized CKD-EPI equations with a raw Cockcroft-Gault estimate, so that header alone is not a safe unit source – Methods Sect. 2.2.2 is the authority that the CKD-EPI values are mL/min/1.73 m^2), Zhang_2026_tebipenem.R (WEIGHT-normalized mL/min/kg, Sato 2008 paediatric predictive equations from serum creatinine; cohort mean 3.72 +/- 0.97, median 3.73, range 1.46-6.97; additive linear effect 0.104 (L/h/kg) per (mL/min/kg) on a 0.363 L/h/kg intercept with no centering, the whole bracket then multiplied by body weight to give CL/F in L/h – the founding weight-normalized example), Mouksassi_2015_thrombomodulinAlfa.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized, and additionally TRUNCATED at 150 mL/min – the source truncated the upper tail for the covariate analysis “in order to better represent upper physiological range of CrCL”, which bound on 1.6% of Phase I and 9.5% of Phase IIb subjects, so a user supplying real data must apply the same truncation or the term extrapolates past what was fitted; reference 72.6 mL/min; power effect (CRCL / 72.6)^0.27 on the CL of renally-eliminated thrombomodulin alfa, pooled healthy adults plus adults with sepsis and suspected DIC spanning 5-150 mL/min; BASELINE value, the paper having screened baseline and time-varying CrCL separately and retained baseline; subjects on renal replacement therapy for ESRD or acute kidney injury were excluded, so the model carries no information about dialysis), Chandorkar_2015_ceftolozane.R, Chandorkar_2015_tazobactam.R (raw Cockcroft-Gault mL/min on ACTUAL body weight, NOT BSA-normalized; BASELINE, the paper having verified that serum creatinine was stable over the short treatment course; power effects (CRCL / 109)^0.715 and (CRCL / 115)^0.67 on the clearance of the two analytes of one fixed-ratio combination. The two references differ because the two analytes were fitted to different subject sets – 376 subjects for ceftolozane against 243 for tazobactam, the cUTI study having dosed ceftolozane alone – so a single paper can legitimately carry two centring constants for what is nominally the same column, and they are not interchangeable. Both models are also a rare instance of a paper publishing enough of its own answer key to verify the transcription: the Figure 1 tornado panels print relative CL at six CrCL values per analyte, and (CrCL/109)^0.715 reproduces all six ceftolozane labels exactly), Tseng_2026_piperacillin.R (2021 CKD-EPI creatinine-cystatin C eGFR, BSA-normalized; power effect (CRCL / 104.368)^1.12 on CL in hospitalized adults with a body mass index at or below 18.5 kg/m^2. Three points are worth carrying forward. First, the reference constant 104.368 in the paper’s printed Table 2 formula does NOT equal the Table 1 cohort median of 109.5 mL/min/1.73 m^2 even though the prose calls it “the median eGFR value”; the printed constant is the one this model uses, and replicating the paper’s own Figure 3 target-attainment table confirms it (RMSE 2.06 percentage points against 4.40 for 109.5). Second, this is the mirror image of the Patel_2025_eteplirsen.R rationale: both papers reject creatinine as a renal marker because their cohort’s muscle mass is atypical, but in Duchenne muscular dystrophy creatinine is elevated by muscle breakdown while in low body weight it is depressed by low muscle mass, so the creatinine-based equation errs in opposite directions. Third, the cohort skews preserved-to-augmented (median 109.5, IQR 36) and patients who developed acute kidney injury or needed renal replacement therapy were excluded, so the model carries no information about unstable renal function; the paper’s own dosing simulations nevertheless extrapolate the term across 20-130 mL/min/1.73 m^2), Ma_2026_colistinSulfate.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 94.76 mL/min, cohort mean 100.42 +/- 68.87, observed range 12.14-246.91 in Chinese ICU adults with carbapenem-resistant infections; power effect (CRCL / 94.76)^0.232 on the CL of intravenous colistin sulfate, the sole retained covariate. Two features are worth carrying forward: the centring constant 94.76 appears ONLY inside the final-model equation printed in the Results text and is neither tabulated nor equal to the reported cohort mean, so a table-only read of the paper recovers the exponent but not the reference; and patients on CRRT or ECMO were excluded by design, so the term carries no information about renal replacement therapy even though the cohort is otherwise a renal-impairment-rich ICU population), Xie_2022_colistinSulfate.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 57.5 mL/min, printed only inside Eq. 3 and not the Table 1 cohort median of 48.8 (range 6.5-193.8) in Chinese ICU adults; power effect (CRCL / 57.5)^0.353 on the CL of intravenous colistin sulfate. Unlike Ma 2026, two of the 20 subjects were on CRRT with no separate dialysis term, and the paper’s own dosing tables are not reproduced in level by the printed parameters – see the model vignette), Ahmed_2024_vancomycin.R (Bjornson Rcr / SCr with the weight and 14.4 factors OMITTED – see the CLcr (Bjornson) alias above; reference 12.7, the cohort median, the source’s own Methods defining the reference as mean(CLcr) but never printing a mean; power effect (CRCL / 12.7)^0.49 on CL in Sudanese adult inpatients. Two features are worth carrying forward. First, this is the founding example of a renal-function column that is NOT interchangeable with any other member of this canonical: it is deflated about five-fold by two omitted factors, so it is the one case where reusing the model with a standard CrCl silently biases clearance by more than two-fold. Second, the paper prints the covariate model twice in mutually inconsistent forms – the Methods give the centered (CLcr / mean(CLcr))^beta that was fitted, while Equation 5 and the Table 3 footnote drop the denominator; the uncentered reading overshoots the paper’s own Table 4 simulated clearances roughly three-fold, which settles it), Peng_2025_meropenem.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; the lowest-renal-function cohort registered against this canonical – median 13.6 mL/min, IQR 6.9-22.2, in Chinese critically ill adults all of whom were on CVVH, so the column is RESIDUAL native function sitting alongside the extracorporeal QEFF arm rather than the whole of renal elimination. Enters as a median-CENTERED EXPONENTIAL exp(0.0183 * (CRCL - 13.6)) on the endogenous clearance arm – note this is the log-linear form, not the power-of-a-ratio form that most entries above use, so the coefficient carries units of per mL/min and is not a dimensionless exponent. The paper’s dosing simulations extrapolate the term to 10-50 mL/min, roughly twice the upper end of the fitted range), Bai_2024_imipenem.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized, computed in SI creatinine units – the Table 1 footnote prints the exact form used, CrCl = [(140 - age) x weight (kg)] / [0.818 x Cr (umol/L)], multiplied by 0.85 for women, so the 0.818 divisor plays the role of the familiar 72 when serum creatinine is in mg/dL; reference 99.896 mL/min; power effect (CRCL / 99.896)^0.473 on CL in critically ill Chinese ICU adults with sepsis, the sole covariate retained after forward inclusion and backward elimination, dropping the objective function from 776.148 to 734.330 and the CL between-subject CV from 38.562% to 35.748%. Two points are worth carrying forward. First, this is a third instance of the centring-constant-versus-tabulated-median mismatch already recorded for Tseng_2026_piperacillin.R and Ma_2026_colistinSulfate.R: the text below Eq. 8 calls 99.896 “the median value of CrCl” while Table 1 gives the cohort median as 99.34, and the equation constant is the one the fitted typical CL of 11.357 L/h is conditioned on. The discrepancy is immaterial here – at the exponent 0.473 it moves CL by 0.26% – but the pattern is now frequent enough that the printed equation, not the demographics table, should be treated as the authority for the centring value. Second, the paper states an explicit applicability window, “the predication model can only be used within the range (17.80-256.22 mL/min)”, and unlike several of the models above its own dosing simulations stay inside that window), Beijer_2026_cloxacillin.R (Lund-Malmo Revised 2018 creatinine-based eGFR, BSA-normalized – the founding LMR18 example; reference 67 mL/min/1.73 m^2, cohort median 72, IQR 61-84, range 18-142 in Swedish adults undergoing hip or knee arthroplasty; power effect (CRCL / 67)^0.51 on the UNBOUND clearance of cloxacillin, alongside a (WT / 84)^0.67 body-weight term. Three points are worth carrying forward. First, this is the one entry so far that registers the estimating-equation choice as a result: the paper’s Table S1 correlates seven eGFR variants against the empirical-Bayes eta_CL and LMR18 wins in both relative (r = 0.32) and absolute (r = 0.43) form, beating CKD-EPI 2021 (0.15 / 0.29) and Cockcroft-Gault (0.21 weight-adjusted 0.18), so a user substituting a CKD-EPI value is supplying a measurably worse predictor and not merely a different flavour of the same column. Second, the BETTER-correlating absolute LMR18 estimate was deliberately REJECTED in favour of the relative one, because absolute eGFR left every anthropometric covariate (adjusted / ideal / total body weight, BMI, BSA) independently correlated with clearance at p < 0.001 – the relative estimate plus a separate weight term is the decollinearized parameterisation, and the two arms of that pair must be supplied together or body size is double-counted. Third, this is a fourth instance of the centring-constant-versus-tabulated-median mismatch recorded for Tseng_2026_piperacillin.R, Ma_2026_colistinSulfate.R and Bai_2024_imipenem.R: Table S2’s printed CL equation divides by 67 while Table 1 gives the cohort median as 72, and here the Figure 1 caption independently names “eGFR 67 mL/min/1.73 m2” as the typical patient, corroborating the equation over the demographics table), GarciaHervalejo_2026_teicoplanin.R (CKD-EPI eGFR, BSA-normalized; reference 92.15 mL/min/1.73 m^2, the development-cohort median, observed range 26-141 in adults with haematological malignancies treated for febrile neutropenia; power effect (CRCL / 92.15)^0.35 on the CL of a one-compartment teicoplanin model. Two points worth carrying forward. First, this is another instance of the Table-1-header-versus-Methods unit hazard already recorded for Taylor_2026_methotrexate.R, but running the opposite way: here Table 1 heads the row mL/min/1.73m2 while the Table 2 abbreviation footnote writes in mL/min., and Table 1 governs because CKD-EPI is BSA-normalized by construction and the Methods never mention de-normalising it. Second, the renal term is the WEAKEST and least precisely estimated fixed effect in that model – RSE 47%, bootstrap 95% CI 0.06-0.67, a lower bound near zero – and it is comprehensively dominated by the co-retained IBW term at exponent 3.2; across the full observed eGFR range the renal multiplier spans only 0.64-1.16. Patients with end-stage renal disease or on renal replacement therapy were excluded by design, so the model carries no dialysis information), Hadigol_2026_talazoparib_enzalutamide.R (raw un-normalised mL/min, source column BCCL, reference 86.85 mL/min; power effect (CRCL / 86.85)^0.455 on talazoparib apparent BASE clearance, i.e. on the interaction-free intercept of a drug-drug-interaction model rather than on total clearance. Two cautions specific to this entry: the estimating equation is NOT named anywhere in the paper – the raw-mL/min renal-disease cutoffs of its Table 4 are the only evidence for the normalisation – and the reference 86.85 mL/min is presented only as the value for the typical patient, never as the cohort median, with the demographics table not open-access, so it must not be cited as a median), Komatsu_2016_penicillinG.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; cohort mean 82.52 +/- 33.29, range 11-144 in Japanese adults with infective endocarditis. The most extreme functional form registered against this canonical: a bare THROUGH-ORIGIN product with no intercept and no centring, CL (L/h) = 0.21 x CRCL (mL/min), so the covariate is not a multiplier on a typical clearance but the whole of it and the parameter carries the compound unit L/h per mL/min. Two consequences a reuser must know. First, there is no non-renal clearance arm at all, so the model predicts zero elimination at zero renal function – which is false for penicillin G – and it must not be used at or near anuria even though the paper’s own nomogram recommends a regimen for CLcr below 20 mL/min and its Monte Carlo simulations extrapolate to 5 mL/min, below the observed minimum of 11. Second, the absence of a reference value means there is no centring constant to get wrong, which is the one hazard recorded for Tseng_2026_piperacillin.R, Ma_2026_colistinSulfate.R, Bai_2024_imipenem.R and Beijer_2026_cloxacillin.R that this entry is structurally immune to. The form is verifiable from the paper’s own Discussion, which states a half-life of 0.79 h “calculated by keeping the CLcr fixed at 120 mL/min”: 0.693 x 28.9 / (0.21 x 120) = 0.795 h, which pins the slope, the volume and the mL/min unit simultaneously), Kuchimanchi_2022_diroximelFumarate.R (MDRD eGFR denormalized to absolute mL/min with each participant’s body surface area, per ESM Table S2 footnote b, so the per-model units are mL/min rather than the register default; power (CRCL / 111.9)^0.547 on 2-hydroxyethyl succinimide clearance only, 111.9 mL/min being the cohort median), Wang_2021_mycophenolic_acid.R (MDRD eGFR modified for Chinese patients, Ma 2006; linear centring 1 + 0.00791 x (CRCL - 57) on CL/F), He_2020_phosphocreatine.R (bedside Schwartz paediatric eGFR, with the original Schwartz 0.45 constant for infants under 1 year, BSA-normalized, source column GFR; reference 127.78 mL/min/1.73 m^2, the cohort median, range 66.33-224.01 in Chinese children with myocarditis; power effect (CRCL / 127.78)^0.311 on the clearance of the METABOLITE creatine formed from infused phosphocreatine, not on the parent drug), Royer_2021_palbociclib.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; reference 78.9 mL/min, the cohort MEAN (the paper’s Methods normalize every continuous covariate by the population mean; median 72.1, range 23.4-282.3) in French women with metastatic breast cancer sampled during routine therapeutic drug monitoring; power effect (CRCL / 78.9)^0.419 on palbociclib CL/F, retained over body weight and age in backward deletion although renal excretion is under 8% of palbociclib clearance – the authors read CRCL as a composite of the weight, age and creatinine it is computed from), WerumeusBuning_2021_ceftazidime.R (CKD-EPI 2009 eGFR, BSA-normalized; reference 73 mL/min/1.73 m^2, the cohort median, range 6-153 in critically ill adults; power effect (CRCL / 73)^0.772 on the ceftazidime clearance of patients OFF continuous veno-venous hemofiltration only – the term is switched off together with the eGFR-driven branch when RRT_CRRT_STATUS = 1, so a CVVH patient’s CRCL value is ignored but must still be finite. Table 1 prints the unit as mL/min/m 2, a dropped 1.73), BilbaoMeseguer_2021_levetiracetam.R (creatinine clearance MEASURED from a timed urine collection, raw mL/min, NOT BSA-normalized; cohort median 117 mL/min, range 54-239, in Spanish ICU adults, 37% with augmented renal clearance; enters as a POWER term with no multiplying coefficient added to a fixed non-renal arm, CL = 3.5 + (CRCL / 120)^2.5 L/h, so the renal arm is 1 L/h at the 120 mL/min divisor. Two points worth carrying forward. First, the exponent is a typeset superscript in both the Table 3 header and the Results equation, and a plain-text extraction flattens it into what reads as a linear slope 3.5 + 2.5 x CrCl/120; the paper’s Discussion settles the form, quoting mean CL of 4.5 L/h at 120 and 9.2 L/h at 240 mL/min, which only the power reading reproduces. Second, with an exponent of 2.5 the term is steeply supra-proportional, so a Cockcroft-Gault or BSA-normalized value substituted for the measured urinary clearance moves CL far more than it would under the near-proportional exponents most entries above carry; CrCl below 50 mL/min was an exclusion criterion), NgougniPokem_2022_temocillin.R (creatinine clearance MEASURED from a urine collection, raw mL/min, NOT BSA-normalized; reference 39.9 mL/min, the cohort median, range 20.55-149.3, in Belgian ICU adults in septic shock with intra-abdominal infection and ascitic fluid effusion; through-origin linear ratio (CRCL / 39.9) on the unbound temocillin clearance from the central compartment, with no non-renal arm on that clearance – the only other elimination route is the abdominal drain from the ascitic fluid compartment, so the model must not be used near anuria), Duong_2022_gentamicin_rea.R (MDRD eGFR 186.3 x (Scr/88.4)^-1.154 x Age^-0.203 x 1.212 [black] x 0.742 [female], Scr in umol/L, labelled mL/min by the source with no BSA de-normalisation described; enters gentamicin clearance as the sigmoid clmax * CRCL^1.2 / (crcl50^1.2 + CRCL^1.2) with crcl50 = 129 mL/min, the Rea 2008 structure refit to a Quebec ICU cohort), Duong_2022_gentamicin_bos.R (raw Cockcroft-Gault (140 - Age) x WT x 1.23 x 0.85 [female] / Scr, Scr in umol/L, mL/min, NOT BSA-normalised; linear-centred 1 + 0.00925 * (CRCL - 92) on gentamicin CL, centred at the 92.2 mL/min cohort mean – the Bos 2019 structure refit to the same Quebec cohort), Yu_2022_colistinSulfate.R (raw Cockcroft-Gault mL/min, NOT BSA-normalized; cohort 79.54 +/- 53.99 mL/min in Chinese ICU adults with carbapenem-resistant infections; additive linear effect CL = 0.994 + 0.525 x CRCL / 66.47 L/h on intravenous colistin sulfate, the same divisive-normalization form as Delattre_2010_amikacin.R. The divisor 66.47 appears only in the printed equation and is not the tabulated cohort mean; concentrations drawn during renal replacement therapy or ECMO were excluded).
• Notes: All estimation methods (creatinine-based MDRD / CKD-EPI / measured CrCl, and tracer-based iohexol / inulin / DTPA / EDTA clearance) produce values in the same units and are operationally interchangeable as a covariate on clearance. Three distinct size normalisations appear across the example models and must be documented per model, because they are not interchangeable: BSA-normalised (mL/min/1.73 m^2, the canonical default), raw un-normalised (mL/min, e.g. Delattre_2010_amikacin.R, Chen_2023_nemonoxacin.R), and per-70-kg-body-weight (mL/min/70 kg, e.g. Suzuki_2024_mycophenolic_acid.R). Supplying a value on the wrong normalisation silently rescales the renal-function term, and in a model that also carries an allometric weight term it double-counts body size. Document the method explicitly in each model’s covariateData[[CRCL]]$description so future reviewers can trace the source assay; the assay choice affects accuracy of the absolute GFR value but does not change the covariate’s role as a linear / power scalar on renal clearance. Tracer-measured GFR is the clinical gold standard (unaffected by muscle mass, malnutrition, and tubular-secretion confounders that perturb creatinine-based estimates) but is rarely available outside research and renal-impairment-cohort studies; creatinine-based estimates dominate routine clinical popPK datasets. Tracer-measured-GFR scope codified on 2026-06-17 alongside the Hamren 2008 tesaglitazar extraction (by the maintainers’ decision).

RENALFUNC_REL (canonical for relative renal function, 1 = normal)

• Description: Dimensionless per-subject renal function expressed as a FRACTION OF NORMAL, so that 1 = normal renal function, 0.5 = half of normal, and values above 1 describe supranormal filtration. This is the organ-function scalar of the Groningen / Medimatics MwPharm and Edsim++ modelling tradition, where it lives in a dedicated organ-function object (the ORG / P object) and multiplies the renal arm of a decomposed clearance: CLi = CL * (fe * RENALFUNC_REL + (1 - fe) * HEPFUNC_REL), with the paper-named parameter fe supplying the renal share of total clearance. Compute it as measured GFR / reference normal GFR (or as the ratio of any consistent renal-function measure to its own normal reference) – the ratio, not the absolute measurement, is what this column carries.
• Units: (dimensionless)
• Type: continuous
• Scope: general
• Reference category: n/a – the column IS the multiplier and enters the model directly; 1 denotes normal renal function. No centering or power transform is applied.
• Source aliases:
  • RF – Visscher 2025 Methods section 2.3 and Edsim++ ORG object symbol for relative renal function. Same orientation and scale, no value transformation. NOTE: the bare two-letter name RF must NOT be used as the canonical, because it is already recorded in this register as a rejected alias of the rheumatoid-factor column.
  • RENAL_FUNCTION_FRACTION, GFR_REL, GFR/GFR_normal – descriptive forms of the same ratio.
• Example models: Visscher_2025_parathyroidHormone.R (founding example; single-patient rhPTH(1-84) PK-PD in postsurgical hypoparathyroidism. Multiplies the renal fraction fe = 0.3 of the 39.93 L/h population clearance per Visscher 2025 Eq. 3. Held at 1 in the validation vignette because the source states that the equation corrects for the variations in GFR but reports NO numeric RF values and NO reference-GFR denominator from which the ratio could be recovered – and Table S3 omits the 24-h urine volume that the paper’s own GFR equation needs, so the per-interval ratio cannot be reconstructed from the supplement either.).
• Notes: Genuinely distinct from every pre-existing renal-function entry, which is why it needed its own canonical rather than an alias: CRCL and CRCL_BASE are ABSOLUTE renal function in their own units (mL/min/1.73 m^2) and converting one to a relative multiplier requires a reference denominator that a source paper often does not state; KBF is an absolute organ blood flow; and RENALIMP_MILD / RENALIMP_MOD / RENALIMP_SEV are CATEGORICAL impairment bands that cannot express a continuous multiplier. A model that reports an absolute GFR should use CRCL; use RENALFUNC_REL only when the source’s own parameterisation is the dimensionless fraction-of-normal. Sister canonical to HEPFUNC_REL; the two are normally used together, since the point of the parameterisation is to let the renal and non-renal arms of clearance vary independently. Expect reuse: the relative-organ-function scalar is the standard parameterisation of the MwPharm / Edsim++ family (Edsim++ author N.C. Punt is a co-author of the founding paper, whose reference list cites a sibling model from the same group and framework).

HEPFUNC_REL (canonical for relative liver function, 1 = normal)

• Description: Dimensionless per-subject hepatic function expressed as a FRACTION OF NORMAL, so that 1 = normal liver function and 0.5 = half of normal metabolic capacity. Companion to RENALFUNC_REL in the MwPharm / Edsim++ organ-function (ORG / P) object, multiplying the non-renal (hepatic-metabolic) remainder of a decomposed clearance: CLi = CL * (fe * RENALFUNC_REL + (1 - fe) * HEPFUNC_REL).
• Units: (dimensionless)
• Type: continuous
• Scope: general
• Reference category: n/a – the column IS the multiplier and enters the model directly; 1 denotes normal liver function.
• Source aliases:
  • LF – Visscher 2025 Methods section 2.3 and Edsim++ ORG object symbol for relative liver function. Same orientation and scale, no value transformation.
  • HEP_FUNCTION_FRACTION, LIVERFUNC_REL – descriptive forms of the same ratio.
• Example models: Visscher_2025_parathyroidHormone.R (founding example; multiplies the hepatic remainder 1 - fe = 0.7 of total clearance. Held at 1 because Visscher 2025 Methods states explicitly that “LF is the relative liver function (equal to 1 in this model)” – the patient had normal hepatic function and no hepatic-impairment data inform the parameter. Retained in the model rather than folded into CL so the paper’s own renal / hepatic decomposition of clearance stays visible and user-perturbable.).
• Notes: Distinct from the HEPIMP / HEPIMP_MILD / HEPIMP_MOD / HEPIMP_SEV / HEPIMP_MODSEV family, which are binary Child-Pugh-style impairment INDICATORS, and from the underlying laboratory measures AST / ALT / BILI / ALB: this column is a continuous fraction-of-normal metabolic-capacity multiplier, not a category or a lab value. A source that reports a Child-Pugh class should use the HEPIMP_* family; use HEPFUNC_REL only when the source’s own parameterisation is the dimensionless fraction-of-normal. Sister canonical to RENALFUNC_REL. A value of exactly 1 is a legitimate and common entry – it records that the source’s structural decomposition of clearance exists and was exercised at normal hepatic function, which is information a downstream user needs in order to perturb it.

KBF (canonical for native kidney (renal) blood flow)

• Description: Whole-organ blood flow perfusing the kidneys, used as a physiological system input in mechanistic kidney / renal PBPK models, where it sets the peritubular vascular flow that carries drug past the tubular epithelium. May be measured (para-aminohippurate or isovalerylglycine renal clearance, phase-contrast MRI, Doppler ultrasound) or assumed from a population physiological default. Distinct from the extracorporeal-circuit flows BFR / DFR / Q_CVVH / QBL / QEFF, which describe dialysis or ECMO hardware rather than native renal perfusion, and from CARDIAC_OUTPUT (whole-body hemodynamic output, of which renal blood flow is roughly one fifth).
• Units: mL/min (raw whole-organ flow, NOT BSA-normalized; the Huang & Isoherranen mechanistic kidney framework consumes it internally as L/h via the 1 mL/min = 0.06 L/h conversion)
• Type: continuous
• Scope: general
• Reference category: n/a – enters the model directly as the peritubular vascular flow, not as a ratio or centered deviation. Healthy-adult physiological default in the founding framework: 60 L/h (= 1000 mL/min), scaled proportionally with GFR in the population version of the model (Huang & Isoherranen 2020 distributed MATLAB code, Q_kidney0 = 60 * Proportional_Reduction). Granda 2024 replaces that population default with a per-subject measurement, which is the paper’s central methodological contribution.
• Source aliases:
  • Isovalerylglycine CL_r – Granda 2024 Table 3 column heading; the renal clearance of the endogenous solute isovalerylglycine used as the KBF surrogate (high extraction ratio, low protein binding, clearance more than fourfold GFR). Same orientation and units, no value transformation.
  • Q_kidney / Qk / Q_kidney0 – Huang & Isoherranen 2018 / 2020 and Chang 2023 symbol for the same quantity, expressed in L/h (divide by 0.06 to obtain the canonical mL/min).
  • RBF / RPF – renal blood flow / renal plasma flow shorthand common in renal-physiology and PAH-clearance literature. RPF is PLASMA flow and must be converted to blood flow (KBF = RPF / (1 - haematocrit)) before use.
• Example models: Granda_2024_kynurenicacid_pbpk.R, Granda_2024_tenofovir_pbpk.R, Granda_2024_oseltamivircarboxylate_pbpk.R (all three: per-subject KBF estimated from measured isovalerylglycine renal clearance, mL/min; cohort mean 575 +/- 387, range 76-1692 across 27 adults spanning CKD stages 1-5; enters as the peritubular blood flow qk <- KBF * 0.06 in the 35-state mechanistic kidney PBPK).
• Notes: General scope because native kidney blood flow is a paper-independent physiological quantity that any renal-PBPK or renal-physiology model may use. The canonical name states the quantity rather than a paper symbol, per the standing canonical-naming rule. Note the physiological constraint the founding models enforce: secretory clearance cannot exceed kidney blood flow, so a subject whose measured secretory-marker clearance exceeds their measured KBF cannot be fitted (Granda 2024 participants 38, 41 and 45, whose fitted secretory clearance therefore railed at the 1000 L/h search bound); a model consuming KBF should surface that violation rather than silently saturating. The companion per-subject quantity in Granda 2024 – unbound intrinsic secretory clearance – is deliberately NOT registered as a covariate: it is a fitted model parameter (lclintsec), not a measurement, per the same ruling.

CREAT (canonical for serum creatinine)

• Description: Serum creatinine concentration (baseline or time-varying).
• Units: umol/L or mg/dL – document the unit used in each model via covariateData[[CREAT]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (CREAT / ref)^exponent.
• Source aliases:
  • CRE (umol/L, reference 70.73) – used in Thakre_2022_risankizumab.R.
  • SCR – common clinical-PK abbreviation; also Llanos-Paez 2020 source column for the patient’s individual serum creatinine.
  • Serum_creatinine – the NONMEM $INPUT column in Ding_2026_vancomycin.R (umol/L), abbreviated Scr in that stream’s $PK block.
• Example models: Thakre_2022_risankizumab.R, Hennig_2013_tobra.R (umol/L; paired with CREAT_REF for the SCR_mean / SCR ratio used in the Hennig 2013 renal-function factor), Llanos_2017_gentamicin.R (umol/L; standardized per-patient against CREAT_REF rather than a fixed cohort reference), LlanosPaez_2020_gentamicin.R (umol/L; used as the patient’s SCR_i in the renal-function ratio (CREAT_REF / CREAT)^0.58 on CL).
• Notes: CREAT chosen over the shorter CRE/SCR as the NONMEM/clinical-PK convention that is unambiguous. Per-model reference values must be documented in covariateData[[CREAT]]$notes.

CREAT_REF (canonical for sex/age/size-expected normal-mean serum creatinine)

• Description: Externally-computed reference serum creatinine for the individual (the expected normal SCR for a healthy person of the same sex, age and body size). Used as the numerator of a ratio against the patient’s measured CREAT to define a renal-function factor on clearance.
• Units: umol/L or mg/dL – must match the unit of the paired CREAT column. Document via covariateData[[CREAT_REF]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – used as (CREAT_REF / CREAT)^exponent so that a patient with measured SCR equal to the population-expected normal SCR has factor 1.
• Source aliases:
  • SCR_mean – used in Hennig 2013 (Eq. 5: f_SCR = (SCR_mean / SCR)^theta_SCR); also Llanos-Paez 2020 paper notation for the Ceriotti 2008 age/sex-matched physiological mean SCR.
  • Scrmean – Llanos-Paez 2017 paper notation; computed from Ceriotti et al. 2008 age- and sex-stratified medians (Clin Chem 54:559-566, doi:10.1373/clinchem.2007.099648).
  • SCR_standardised – Germovsek 2018 paper notation; PMA-adjusted standardisation of raw SCR per the paper’s reference 28 (a previously-developed Standing-style PMA stratification).
  • CCR,adj – Ruhs 2012 paper notation; age- and gender-adjusted reference creatinine derived from the paper’s reference [23] (a Schwartz-style paediatric maturation adjustment); the main text does not give the explicit formula.
• Example models: Hennig_2013_tobra.R, Llanos_2017_gentamicin.R (umol/L; computed externally per Ceriotti et al. 2008), LlanosPaez_2020_gentamicin.R (umol/L; ratio (CREAT_REF / CREAT)^0.58 multiplies the maturation-scaled CL), Germovsek_2018_meropenem.R (umol/L; ratio (CREAT_REF / CREAT)^0.40 multiplies the maturation-scaled CL for renal meropenem clearance in neonates and young infants; PMA-stratification reference per Germovsek 2018 Methods reference 28), Ruhs_2012_methotrexate.R (mg/dL; age- and gender-adjusted CCR,adj per the paper’s reference [23]; ratio (CREAT_REF / CREAT)^0.314 multiplies the BSA-scaled MTX CL in paediatric ALL patients).
• Notes: Specific scope because the formula used to derive the reference value is paper-defined (Hennig 2013 cites a combination of Ceriotti 2008, Junge 2004 and Johansson 2011 reference-interval relationships; Llanos-Paez 2017 and 2020 both use Ceriotti 2008); a future paper that uses a different reference-SCR derivation (e.g., a CKD-EPI-style adult-only reference, or a Schwartz-derived paediatric-only reference) should pin its formula in covariateData[[CREAT_REF]]$notes so that a user assembling a virtual cohort can reproduce it. When no covariate data are available to compute CREAT_REF, set CREAT_REF = CREAT so the renal-function factor evaluates to 1 (matching the Hennig 2013 ‘covariate set to 1 for missing data’ rule).

BUN (canonical for blood urea nitrogen)

• Description: Blood urea nitrogen concentration (baseline or time-varying). Reflects the nitrogenous waste burden being cleared by the kidneys; rises with reduced glomerular filtration, dehydration, increased protein catabolism, or GI bleeding. Distinct from CREAT (the other commonly-reported renal-function marker) because urea reabsorption is flow-dependent in the renal tubule, so BUN is sensitive to volume status as well as GFR.
• Units: mg/dL or mmol/L – document the unit used in each model via covariateData[[BUN]]$units (1 mmol/L urea ~= 2.80 mg/dL BUN).
• Type: continuous
• Scope: general
• Reference category: n/a – used either as a linear / hinge effect on a PK parameter, or with power scaling (BUN / ref)^exponent. Reference values observed: 7 mg/dL (Hall 2017 MARS hinge knot for ka; not a population median); 4.2 mmol/L (Chen 2017 cohort median for the tacrolimus power-of-ratio CL effect); 5 mmol/L (Wang 2026 cohort median for the ciprofol power-of-ratio V1 effect – not printed in that paper and recovered algebraically, see below).
• Source aliases: none known.
• Example models: Hall_2017_dapsone.R (mg/dL; population median 13 mg/dL [range 7-28]; enters the MARS-based covariate model on the absorption rate constant via the basis function BF1 = max(0, BUN - 7), which interacts with a weight hinge to drive Ka), Chen_2017_tacrolimus.R (mmol/L; cohort median 4.2 mmol/L [range 1.7-10.4]; enters CL/F via standard power-of-ratio scaling (BUN/4.2)^1.42 in low-dose oral tacrolimus for Chinese myasthenia-gravis patients), Wang_2026_ciprofol.R (mmol/L; cohort median 5 mmol/L [range 2.9-7.6]; enters the central volume via power-of-ratio scaling (BUN/5)^-0.821 in Chinese children aged 1-9 years given a single ciprofol bolus).
• Notes: Promoted to general scope on 2026-06-03 alongside the Chen 2017 tacrolimus extraction, the second model registering BUN. Hall 2017 enters BUN only through a piecewise-linear MARS hinge (max(0, BUN - 7) * max(0, 63.7 - WT)), not as a power scaling – the per-model covariateData[[BUN]]$notes documents this is part of a machine-learning-driven hinge model, not a standard popPK covariate transform. Chen 2017 uses the conventional centred power-of-ratio form (BUN/ref)^exponent referenced to the cohort median; the positive exponent there is interpreted via a urea-driven protein-carbamylation mechanism that reduces albumin binding of the ~99%-protein-bound tacrolimus. Wang 2026 is the first model to enter BUN on a distribution volume rather than a clearance, and the first with a negative exponent: higher urea is interpreted there as a marker of reduced effective plasma volume, shrinking the initial distribution space of a highly protein-bound anaesthetic. It is also a worked example of recovering an unprinted normalising median – that paper prints only the cohort mean (5.2 mmol/L), but quotes V1 at two BUN values, and solving 506 * (BUN/ref)^-0.821 against both quoted volumes returns ref = 5.00 twice over.

UACR (canonical for urine albumin-to-creatinine ratio)

• Description: Urine albumin-to-creatinine ratio (UACR), a renal-damage biomarker quantifying albuminuria. UACR is the spot-urine albumin concentration divided by the spot-urine creatinine concentration; it approximates 24-hour urinary albumin excretion (in mg/day) because creatinine excretion is roughly constant at ~1 g/day in an adult. UACR is the principal albuminuria measure in KDIGO chronic-kidney-disease (CKD) staging, with thresholds A1 (< 30 mg/g, normal-to-mildly increased), A2 (30-300 mg/g, moderately increased), and A3 (> 300 mg/g, severely increased). Distinct from the function markers CRCL / CREAT / BUN: UACR captures upstream glomerular albumin leak (kidney damage), whereas CRCL / CREAT / BUN capture downstream filtration rate (kidney function). Both axes are routinely reported together in diabetic-kidney-disease (DKD) and CKD popPK / PD studies. Time-fixed at baseline by default; document baseline-vs-time-varying status per model via covariateData[[UACR]]$notes.
• Units: mg/g (urine albumin in mg per gram of urine creatinine). Equivalent to mg/mmol scaled by 8.84 (1 mg/g ~= 0.113 mg/mmol). Document the unit used in each model via covariateData[[UACR]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used either as a centred linear effect on (UACR - ref) or as a power scaling (UACR / ref)^exponent. Reference values observed: 800 mg/g (Goulooze 2022 serum-potassium analysis centring value) and 850 mg/g (Goulooze 2022 UACR / eGFR analysis centring value); both are close to the FIDELIO-DKD cohort median of 852 mg/g.
• Source aliases:
  • UACR0 – Goulooze 2022 paper convention for baseline UACR (the 0 suffix denotes the baseline / pre-randomisation value).
• Example models: Goulooze_2022_finerenone.R (centred linear effects on Emax and on the disease-progression slope TSLOPE, both centred at 800 mg/g; coefficients 9.31e-5 g/mg on Emax and 1.14e-3 g/mg on TSLOPE per Goulooze 2022 Table 1), Goulooze_2022_finerenone_uacr.R (power scaling (UACR / 850)^0.877 on the baseline of the UACR disease-progression state), Goulooze_2022_finerenone_egfr.R (same scaling on the embedded UACR sub-model baseline, plus (UACR / 850)^0.306 on the magnitude of the inter-individual variability of the eGFR decline rate – a covariate on a variance term rather than on a typical value).
• Notes: Promoted from specific to general scope alongside the Goulooze 2022 UACR / eGFR extraction, which was the second paper to register the column (as the original entry anticipated). Distinct from the uacr COMPARTMENT, which is the modelled UACR time course; this column is the time-fixed observed baseline.

UPRO_BL (canonical for baseline 24-hour urinary protein excretion)

• Description: Baseline (pre-treatment) total urinary protein excretion over 24 hours, the “daily urinary protein excretion” or “proteinuria” reported by glomerular-disease trials. Measured from a timed 24-hour urine collection (or, where a paper says so, estimated from a spot urine protein-to-creatinine ratio) and expressed as mass per day. Time-fixed at baseline; in an MBMA it is the study arm’s median or mean baseline rather than a single patient’s value.
• Units: g/day. Document per-model via covariateData[[UPRO_BL]]$units; a source in mg/day converts by dividing by 1000.
• Type: continuous
• Scope: specific
• Reference category: n/a – continuous baseline. Reference values observed: 1.82 g/day (Yu 2022 centring value, close to the median arm baseline of its 40 IgA-nephropathy trials).
• Source aliases:
  • Baseline / daily urinary protein excretion at baseline – Yu 2022 Equation 7 and Results.
• Example models: Yu_2022_igaNephropathy_mbma.R (study-arm level; centred linear effect -0.63 * (UPRO_BL - 1.82) on the drug-arm Emax of the change from baseline in urinary protein excretion, none on the placebo Emax).
• Notes: Member of the <MARKER>_BL baseline family (FERRITIN_BL, BMD_BL, HGB_BL, QTC_BL). Distinct from UACR (urine ALBUMIN-to-creatinine ratio, mg/g) and from a protein-to-creatinine ratio (UPCR, g/g): total protein includes non-albumin proteins, and a ratio is dimensionally different from a daily excretion rate, so none of the three may be substituted for another without the paper’s own conversion. A future model using baseline UPCR should register a sibling UPCR_BL rather than overload this name. The paired change-from-baseline endpoint is the uprocfb output in compartment-names.md.

UNGALCR (canonical for urine NGAL-to-creatinine ratio)

• Description: Urinary neutrophil gelatinase-associated lipocalin (NGAL) concentration divided by the urine creatinine concentration of the same spot-urine sample. NGAL is a 25 kDa lipocalin released by injured renal tubular epithelium; it rises in urine hours to days before serum creatinine moves, so it is a marker of tubular kidney injury rather than of glomerular filtration. Normalising to urine creatinine removes the dependence on urine volume / dilution, which is what makes a spot sample interpretable. Belongs to the same “damage marker, not function marker” axis as [[UACR]], and is distinct from the function markers [[CRCL]] / [[CREAT]] / [[CYSC]]; a model can carry one of each, and Downes_2023_vancomycin_full.R does exactly that (eGFR-Hoek for filtration, UNGALCR for injury), reporting a correlation of only -0.27 between the two.
• Units: ng/mg creatinine. Document the unit used in each model via covariateData[[UNGALCR]]$units; urinary biomarker ratios are also reported as ng/mmol creatinine and as ug/g creatinine in the wider literature, and the three differ by more than a decimal shift.
• Type: continuous
• Scope: specific
• Reference category: n/a. Two forms are anticipated: an UNNORMALISED exponential on the log ratio, base^log(UNGALCR), in which case there is no reference value and the intercept parameter is the value at UNGALCR = 1 ng/mg (the Downes 2023 form); or a conventional median-normalised power, (UNGALCR / ref)^exponent. Reference values observed: none (Downes 2023 uses the unnormalised form); the Downes 2023 training-group median for reference is 60.3 ng/mg creatinine.
• Source aliases:
  • uNGAL/uCr – Downes 2023 Table 2 and Table S4 footnotes (“the natural logarithm of the urinary NGAL concentration normalized to urinary creatinine”).
  • UNGAL – Downes 2023 Table S2 step 3, where the same quantity is screened in an untransformed power form (50.1/UNGAL)^TH2 before the log form was preferred. Note the column is the RATIO in both rows despite the bare UNGAL spelling.
  • LUNGAL – Downes 2023 Table S2 step 3 name for log(UNGALCR), i.e. the transformed value rather than the column itself. Store the ratio on its natural scale in this column and take the logarithm inside model().
• Example models: Downes_2023_vancomycin_full.R (founding example; ng/mg creatinine, no reference normalisation; exponential factor 0.94^log(UNGALCR) on vancomycin clearance in critically ill children, so clearance falls about 22 percent at the cohort median of 60.3 ng/mg relative to the UNGALCR = 1 intercept. Two features are worth carrying forward. First, the covariate spans more than three orders of magnitude within one small cohort – 1.4 to 2809 ng/mg in the 30-subject training group and 7.8 to 10034 in the 20-subject testing group – so a simulated cohort must draw it on the log scale, and the authors themselves flag the spread as a plausible reason the biomarker did not improve individual-level AUC estimation. Second, the unnormalised form means the intercept parameter is anchored at a value far below anything observed, so CL0 here is not interpretable as a typical clearance and must not be compared directly against the CL0 of the paper’s two reduced sibling models.).
• Notes: Scope is specific until a second model ratifies the column; promote to general at that point. The sibling urinary biomarkers Downes 2023 screened on the same footing – urinary KIM-1, urinary cystatin C and urinary osteopontin, each also creatinine-normalised and log-transformed (Table S2 step 3) – are NOT registered here because none was retained in a final model. A future paper that retains one should add a parallel UKIM1CR / UCYSCCR / UOPNCR entry rather than widening this one, since the analytes are biologically distinct injury markers. Do not confuse this column with plasma NGAL, which Downes 2023 screened separately as (95/PNGAL)^TH1 and also did not retain; a future model needing plasma NGAL should register NGAL (a plasma concentration in ng/mL) as its own canonical, because the plasma and creatinine-normalised-urine forms have different units, different dynamic ranges and different biological meaning.

SOD (canonical for serum sodium concentration)

• Description: Serum (or plasma) sodium concentration. Most commonly captured at study entry as a baseline biochemistry value, but can be time-varying when serial electrolytes are recorded. Reflects fluid / electrolyte balance; hyponatraemia is common in severe acute illness, severe malnutrition, GI losses, SIADH and adrenal insufficiency, while hypernatraemia accompanies dehydration and excessive sodium intake.
• Units: mmol/L (equivalent to mEq/L for sodium).
• Type: continuous
• Scope: general
• Reference category: n/a – used with a centered-linear-deviation form (1 + e_sod_<param> * (SOD - ref)). Reference values observed: 136 mmol/L (Thuo 2011 ciprofloxacin; cohort median in Kenyan children with severe malnutrition, normal-range lower bound).
• Source aliases:
  • Na+ / NA / SODIUM – common source-paper printed forms; renamed to canonical SOD when assembling input data. Used in Thuo_2011_ciprofloxacin.R (the paper writes “Na+ (mmol/L)” in Table 1 and structural-model equations).
• Example models: Thuo_2011_ciprofloxacin.R (linear centered-deviation effects on apparent CL and apparent Vc: 1 + 0.0368*(SOD - 136) and 1 + 0.0291*(SOD - 136); reference 136 mmol/L is the cohort median).
• Notes: General scope because serum sodium is a universally applicable serum-electrolyte covariate. Reference value is paper-specific (cohort median); future models should document their own reference in covariateData[[SOD]]$notes. Distinct from any “sodium content of dosed formulation” concept (e.g., sodium-rich oral rehydration solution) – that would warrant a separate canonical (DOSE_NA_MGML, etc.) if a future model retains it.

POT (canonical for serum potassium concentration)

• Description: Serum (or plasma) potassium concentration. Most commonly captured at study entry as a baseline biochemistry value, but can be time-varying when serial electrolytes are recorded. Central to mineralocorticoid-receptor-antagonist, potassium-sparing-diuretic and RAAS-inhibitor programmes, where hyperkalaemia is the dose-limiting toxicity and drives titration and discontinuation rules; KDIGO and ESC hyperkalaemia thresholds are 5.5 and 6.0 mmol/L. Direct sibling of SOD (serum sodium). Distinct from the serum_k COMPARTMENT, which is the modelled potassium time course when potassium is the PD output rather than an explanatory covariate.
• Units: mmol/L (equivalent to mEq/L for potassium).
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (POT / ref)^exponent or linear centring (1 + slope * (POT - ref)). Reference values observed: 4.4 mmol/L (Goulooze 2022 FIDELIO-DKD, the cohort median).
• Source aliases:
  • K0 – Goulooze 2022 convention for observed baseline serum potassium (the 0 suffix denotes the baseline / pre-randomisation value).
  • K, K+, POTAS, SERUM_K – common source-paper printed forms.
• Example models: Goulooze_2022_finerenone_egfr.R (power scaling (POT / 4.4)^-1.38 on DSLOPE, the slope of the acute finerenone effect on eGFR; ESM Table S2 theta_K0,DSLOPE = -1.38, reproducing the paper’s Table 2 range of +31.9% to -18.4% over the 3.6-5.1 mmol/L 5th-95th percentile band).
• Notes: General scope because serum potassium is a universally applicable serum-electrolyte covariate, matching the scope granted to SOD. Reference value is paper-specific (cohort median); future models should document their own reference in covariateData[[POT]]$notes.

RRT_HEMODIAL_STATUS (canonical for intermittent-hemodialysis treatment-status indicator)

• Description: 1 = the subject was undergoing intermittent hemodialysis during the modeled period; 0 = no intermittent hemodialysis. Treatment-status flag rather than a measured renal-function value; used as a multiplicative covariate on PK parameters that change with chronic dialysis (typically CL and Vc – intermittent hemodialysis decreases vancomycin-class CL and reduces interstitial volume overload, lowering Vc). Per-subject indicator in the source data; in Goti 2018 it is treated as time-fixed at the subject level (the cohort either was or was not receiving intermittent hemodialysis during the admission), and Goti 2018 explicitly notes that actual hemodialysis-session timing was not used because of documentation limitations.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no intermittent hemodialysis).
• Source aliases:
  • HEMODIAL – prior canonical name (pre-2026-06-19 RRT__ standardization).
  • DIAL – used in Goti_2018_vancomycin.R (binary indicator on CL and Vc in a 2-compartment vancomycin popPK model). Goti 2018 Methods notes the indicator was created for the routine-TDM cohort (n = 336 hemodialysis subjects of 1812 total) and that all hemodialysis procedures were intermittent and used high-flux membranes.
  • DIT1 – used in Knebel_2008_epoetinDelta.R (the first of two dialysis-modality switches in a three-level covariate; DIT1 = 1 selects the hemodialysis stratum, with predialysis rather than “no hemodialysis” as the reference – see the Notes below).
• Example models: Goti_2018_vancomycin.R (multiplicative factors on CL and Vc: 0.7^RRT_HEMODIAL_STATUS on CL and 0.5^RRT_HEMODIAL_STATUS on Vc, so dialysis subjects have 30% lower CL and 50% lower central volume than non-dialysis subjects), VanWart_2025_telavancin.R (used purely as a gate, not as an effect in its own right: it restricts the additive T_POST_HEMODIAL > 48 h central-volume increase to the 8 chronic-kidney-disease-stage-5 subjects, while the dialysis CLEARANCE arm of the same model is gated by the time-varying RRT_HEMODIAL_ACTIVE instead – the clearest available illustration of the STATUS-versus-ACTIVE split), Knebel_2008_epoetinDelta.R (exponentiated-switch factor 4.53^RRT_HEMODIAL_STATUS on central volume only, paired with PERIT_DIAL against a predialysis reference in a pediatric CKD erythropoietin popPK model), Li_2019_avibactam.R (selects an ABSOLUTE on-dialysis clearance of 20.8 L/h in place of the reference-clearance construction, Li 2019 Table 2 theta6; mutually exclusive in effect with that model’s RENALIMP_ESRD term, which supplies the much lower off-dialysis clearance).
• Notes: Specific to intermittent hemodialysis (IHD). Distinct from peritoneal dialysis (PD) and from continuous renal replacement therapy (CRRT), each of which has different drug-extraction kinetics and would warrant its own canonical (RRT_PERIT_DIAL_STATUS, RRT_CRRT_STATUS) if a future paper retains them as covariates. Three-level dialysis-modality cohorts: when a paper’s cohort spans predialysis, hemodialysis, and peritoneal dialysis and takes predialysis as the reference, encode the two non-reference strata as RRT_HEMODIAL_STATUS and PERIT_DIAL together, with both indicators 0 for the predialysis stratum (Knebel_2008_epoetinDelta.R). The 1-levels still mean exactly what the two canonical definitions say; what shifts is the reference the estimated effects are read against, which is a per-paper fact and must be documented in each model’s covariateData[[...]]$reference_category and $notes rather than inferred from the register. This differs from the two-level chronic-dialysis case (Takama_2007_darbepoetin.R) where the two indicators are perfectly anti-correlated and only one is needed. Goti 2018 treats RRT_HEMODIAL_STATUS as time-fixed per subject because session-level dialysis timing was not reliably documented in the source EHR data; a future paper that resolves drug clearance during versus between dialysis sessions would use a time-varying form (RRT_HEMODIAL_ACTIVE) or a separate per-session covariate. When pairing RRT_HEMODIAL_STATUS with CRCL, note that the Cockcroft-Gault CRCL of an anuric hemodialysis patient is by convention very low or set per institution to a small floor value (Goti 2018 truncated CRCL > 150 mL/min to 150 mL/min and corrected SCr < 1 mg/dL in elderly subjects); residual renal function in hemodialysis subjects is highly variable and the dialysis indicator captures the bulk PK shift on top of the CRCL covariate.

RRT_CRRT_STATUS (canonical for continuous / extended renal-replacement-therapy treatment-status indicator)

• Description: 1 = the subject was undergoing a continuous or extended (long-session) extracorporeal renal-replacement-therapy modality during the modeled period (continuous venovenous hemofiltration CVVH / CVVHF, continuous venovenous hemodiafiltration CVVHDF, sustained low-efficiency dialysis SLED, extended daily diafiltration EDD-f, or similar slow-clearance / long-duration extracorporeal therapies); 0 = no such therapy. Treatment-status flag rather than a measured renal-function value; used as a multiplicative or piecewise covariate on PK parameters that change when slow extracorporeal solute removal is active (typically CL). Per-subject indicator in the source data; in Shekar 2014 it is treated as time-fixed at the subject level (the cohort either was or was not receiving RRT during the entire PK sampling period; all RRT patients were on CVVH or EDD-f continuously / daily during sampling) – the indicator captures the subject-level RRT status, not session-level on/off timing.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no continuous / extended RRT).
• Source aliases:
  • CRRT_STATUS – prior canonical name (pre-2026-06-19 RRT__ standardization).
  • RRT – used in Shekar_2014_meropenem.R (binary indicator selecting between the RRT-fixed-CL term and the CrCL-driven non-RRT CL term in a 2-compartment meropenem popPK model). Shekar 2014 Methods describes the RRT cohort as mixed CVVH (control RRT subjects, true CRRT) and EDD-f (ECMO RRT subjects, extended daily diafiltration; pharmacokinetically CRRT-like for slow-clearance solutes such as meropenem) and the model treats the modalities as a single binary covariate without distinguishing them.
• Example models: Shekar_2014_meropenem.R (piecewise CL: TVCL = exp(lcl) * RRT_CRRT_STATUS + e_crcl_cl * CRCL_in_Lh * (1 - RRT_CRRT_STATUS), with CRCL in raw Cockcroft-Gault mL/min converted to L/h inside model(); 5/11 ECMO patients and 5/10 controls were on RRT), Wi_2017_teicoplanin.R (power-model multiplier on the peripheral volume, vp <- exp(lvp) * 1.50^RRT_CRRT_STATUS; 5/10 subjects on CVVHDF via Prismaflex), Sun_2025_teicoplanin.R (exponential effect on the CENTRAL volume, exp(-0.71 * RRT_CRRT_STATUS), i.e. CRRT reduces V by 50.8%; 20/86 sepsis patients, modality mix 14 CVVH / 5 CVVHD / 1 both), WerumeusBuning_2021_ceftazidime.R (the same piecewise switch as Shekar 2014 on a one-compartment ceftazidime model: cl = cl_off * (1 - RRT_CRRT_STATUS) + exp(lcl_cvvh) * RRT_CRRT_STATUS, where the on-CVVH clearance is a single fixed 2.9 L/h carrying NO between-subject variability and the off-CVVH branch carries the eGFR and comorbidity effects and the eta; 20/96 critically ill adults on CVVH), Lin_2021_vancomycin.R (additive-linear term in an additive-linear CL equation, +0.41 L/h * CRRT-S with CRRT-S = 2 - RRT_CRRT_STATUS, i.e. CRRT LOWERS vancomycin CL by 0.41 L/h in adult ICU patients; the paper notes the direction contradicts earlier reports).
• Direction is not fixed by the canonical. Most published CRRT effects act on clearance and increase it, but the registry now holds examples going both ways and on either volume: Wi_2017_teicoplanin.R has CRRT raising the peripheral volume 1.5-fold, while Sun_2025_teicoplanin.R has it lowering the central volume by half, with no retained effect on clearance at all. Sun 2025 attributes the reduction to fluid overload being the indication for starting CRRT, so that CRRT-treated subjects had an expanded interstitial volume corrected by the therapy. Do not “correct” a source paper’s sign against an expected direction; transcribe it and record the authors’ stated mechanism in covariateData[[RRT_CRRT_STATUS]]$notes.
• Notes: Distinct from RRT_HEMODIAL_STATUS (intermittent hemodialysis IHD only) and from RRT_HEMODIAL_ACTIVE (per-time-point session gate in within-subject time-varying dialysis-clearance models such as Liesenfeld 2013 dabigatran). Anticipated as a future canonical RRT_PERIT_DIAL_STATUS for peritoneal dialysis. Shekar 2014 ratification uses a mixed CVVH + EDD-f cohort because the source paper treats them identically as a single binary RRT covariate; a future paper that retains modality as a separate covariate (e.g. CVVH vs SLED vs CVVHDF) would either reuse RRT_CRRT_STATUS with finer per-modality columns layered on top, or warrant its own modality-specific canonical (RRT_CVVH_STATUS, RRT_SLED_STATUS, etc.). When pairing RRT_CRRT_STATUS with CRCL, note that Cockcroft-Gault CrCL is conventionally not defined / not reported for RRT-dependent subjects; Shekar 2014 records CrCL only for non-RRT subjects and the model formula switches off the CrCL term when RRT_CRRT_STATUS = 1.

RRT_CVVHDF_STATUS (canonical for the continuous-venovenous-hemodiafiltration modality indicator within a continuous-RRT cohort)

• Description: 1 = the subject’s continuous renal-replacement modality is continuous venovenous hemodiafiltration (CVVHDF), which combines a diffusive dialysate stream with a convective ultrafiltrate / substitution stream; 0 = the subject is on the reference continuous modality, continuous venovenous hemodialysis (CVVHD), which is purely diffusive. This is a modality discriminator layered on top of RRT_CRRT_STATUS / RRT_CRRT_ACTIVE, not a treatment-status flag in its own right: it says which continuous modality, never whether a continuous modality is running. It is therefore meaningful only while the circuit is active, and the pair (RRT_CRRT_ACTIVE, RRT_CVVHDF_STATUS) is what distinguishes off-circuit from CVVHD – the modality indicator alone cannot. Use it when a paper estimates a separate extracorporeal clearance per modality over a cohort that shares one body-PK model; a paper that pools modalities under a single binary needs RRT_CRRT_STATUS instead, and a paper that derives the extracorporeal clearance from circuit flows needs RRT_CRRT_EFFLUENT_FLOW (which CVVHDF and CVVHD both use, the CVVHDF effluent simply being the larger by the ultrafiltration rate).
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (CVVHD, the purely diffusive continuous modality).
• Source aliases:
  • Table footnote markers distinguishing modality per subject (Broeker 2018 Table 1 footnotes a = CVVHD, b = CVVHDF); papers frequently encode the modality only this way rather than as a named data column.
  • CVVHDF_STATUS – the pre-RRT_<MODALITY>_<KIND> form of this name, as first proposed when the concept was ratified.
  • MODE, CRRT_MODE, RRT_MODE – multi-level modality columns appearing in the critical-care CRRT popPK literature. Decompose a multi-level modality column into one binary per non-reference modality rather than carrying an integer code.
• Example models: Broeker_2018_tigecycline.R (founding example; 11 critically ill adults with acute kidney injury, 8 on CVVHD and 3 on CVVHDF, sharing one two-compartment body-PK model while Table 2 estimates two separate dialysis clearances – 1.69 L/h for CVVHD and 2.71 L/h for CVVHDF. The column selects between them, cl_hemodialysis <- RRT_CRRT_ACTIVE * ((1 - RRT_CVVHDF_STATUS) * exp(lcl_hemodialysis_cvvhd + eta) + RRT_CVVHDF_STATUS * exp(lcl_hemodialysis_cvvhdf)). Only the reference CVVHD arm carries IIV; the CVVHDF arm’s IIV “tended to zero during estimation” at n = 3).
• Notes: Ratified as the first member of the anticipated modality family that the RRT_CRRT_STATUS entry names (“a future paper that retains modality as a separate covariate (e.g. CVVH vs SLED vs CVVHDF) would … warrant its own modality-specific canonical (RRT_CVVH_STATUS, RRT_SLED_STATUS, etc.)”); the maintainers ruled 2026-05-30 for a single model file carrying one new binary modality canonical, CVVHD as reference. Siblings RRT_CVVH_STATUS and RRT_SLED_STATUS remain anticipated-but-unratified; add each on the same pattern when a paper retains it. Mutual exclusivity across a future multi-modality set is a convention, not a data constraint – if a cohort spans three or more continuous modalities, carry one binary per non-reference modality with all of them 0 denoting the reference, exactly as FILT_SA_MED / FILT_SA_LARGE handle three filter sizes in ButraguenoLaiseca_2022_piperacillin.R, and record the per-paper reference modality in each model’s covariateData[[...]]$reference_category. Which modality is the reference is a per-paper fact: Broeker 2018 takes CVVHD because it carries 8 of its 11 subjects, but a cohort dominated by CVVHDF would reasonably invert that, and the canonical’s 1-level always means CVVHDF regardless.

RRT_CRRT_EFFLUENT_FLOW (canonical for total effluent flow rate of a continuous-RRT circuit)

• Description: Total effluent flow rate leaving a continuous / extended renal-replacement circuit – the sum of spent dialysate, substitution (replacement) fluid, and net ultrafiltration. This is the standard clinical “CRRT dose” metric, normally prescribed weight-normalized in mL/kg/h and recorded absolutely in mL/h. It is the circuit OUTPUT, and therefore distinct from the circuit INPUTS BFR (blood flow into the filter) and DFR (dialysate flow into the filter): in a hemodiafiltration modality the effluent exceeds the dialysate inflow by the ultrafiltration and substitution rates. Because solute removal by a continuous modality is effluent-flow-limited for small, minimally protein-bound drugs, the effluent flow rate is the natural covariate on an extracorporeal clearance arm; the sieving / saturation coefficient that converts it into a clearance is absorbed into the estimated clearance parameter (lcl_crrt) rather than being applied to this column.
• Units: mL/h
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a through-origin linear ratio (RRT_CRRT_EFFLUENT_FLOW / ref) or a power-centered ratio on the CRRT clearance arm, and is meaningful only while RRT_CRRT_ACTIVE = 1. Reference values observed: 2000 mL/h (2 L/h; Zurawska 2026, the value the CRRT clearance estimate is standardized to).
• Source aliases:
  • FLOW – Zurawska 2026 Eq. 6 notation.
  • phi_Effl, phi_Effluent – Butragueno-Laiseca 2022 supplementary material eq. 5 and the companion design-optimisation paper (Garcia-Orueta 2026, doi:10.1007/s40262-025-01594-1); same orientation, no transformation.
  • Effluent flow, Q_eff, Q_effluent – variant abbreviations used across the critical-care CRRT popPK literature.
  • EFFLUENT_FLOW, Q_EFFLUENT, EFR, QEFFL – printed forms appearing in adjacent critical-care CRRT popPK literature. None was adopted as the canonical: the RRT section-header policy requires the RRT_<MODALITY>_<KIND> shape (maintainers’ ruling 2026-08-28).
  • Weight-normalized prescriptions (mL/kg/h) are NOT this column – multiply by body weight on ingestion and record the conversion in covariateData[[RRT_CRRT_EFFLUENT_FLOW]]$notes.
• Example models: Zurawska_2026_piperacillin.R (through-origin linear ratio on the estimated CRRT clearance arm, cl_crrt <- exp(lcl_crrt + etalcl_crrt) * (RRT_CRRT_EFFLUENT_FLOW / 2000) * RRT_CRRT_ACTIVE, so the 3.4 L/h population CRRT clearance is the value at a 2 L/h effluent flow; the cohort mean effluent was 2.75 +/- 1.1 L/h, i.e. 32 +/- 7.8 mL/kg/h), ButraguenoLaiseca_2024_meropenem.R (sibling model from the same group; supplies the denominator of the effluent-concentration observable of a five-matrix meropenem model in critically ill children on CKRT, Ceffluent <- cl_hemodialysis * Cc / (RRT_CRRT_EFFLUENT_FLOW / 1000), the division by 1000 converting mL/h to the L/h in which the clearance is expressed. Cohort total ultrafiltration rate median 1,253 mL/h, IQR 335-1,870 (Table 2); per-filter mean effluent flows 44 / 60 / 46 mL/kg/h for the low / medium / high surface-area filters (Table 3)), ButraguenoLaiseca_2022_piperacillin.R (founding example; critically ill children on continuous venovenous haemodiafiltration. Normalises the effluent observable Ceffluent <- cl_hemodialysis * Cc / (RRT_CRRT_EFFLUENT_FLOW / 1000) per supplementary eq. 5. Supplementary Table 3 reports the flow per filter surface area as 78 (SD 24) mL/kg/h for the 0.2 m^2 filter, 54 (5) for 0.6 m^2 and 60 (15) for 1.2 m^2; the absolute values used in the validation vignette are these multiplied by the weight-band body weight).
• Notes: Distinct from QBL / QEFF, which are extracorporeal CLEARANCES in L/h – Zhang_2025_fluconazole.R supplies QEFF as effluent flow ALREADY multiplied by a sieving coefficient, whereas this column is the raw prescribed flow with the sieving absorbed into the estimated parameter. Distinct from Q_CVVH, which is scoped to a CVVH filter placed parallel to an ECMO circuit (Ahsman 2010 cefotaxime). Distinct from URINE_FLOW (native urine output). Pair with RRT_CRRT_ACTIVE as the on/off gate; a model that instead derives the extracorporeal clearance from circuit flows via the Michaels equation should use BFR / DFR and an estimated lkoa rather than this column. Note that BFR and DFR are canonically mL/min because intermittent-hemodialysis circuits are prescribed in mL/min, whereas continuous modalities are prescribed in mL/h or mL/kg/h – each canonical follows its own modality’s clinical convention, so converting between them requires care.

RRT_PERIT_DIAL_FILL_VOLUME (canonical for the instilled fill volume of a peritoneal-dialysis exchange)

• Description: Volume of peritoneal dialysate instilled into the peritoneal cavity at the start of a dwell – the “fill volume” or “exchange volume” of a CAPD / APD exchange. This is the circuit INPUT (VOL IN in a peritoneal-dialysis model schematic) and is distinct from the drained volume at the end of the dwell (VOL OUT), which exceeds it by net ultrafiltration. Its usual role is not a covariate effect on a structural parameter at all: for a drug administered intraperitoneally the peritoneal cavity is a compartment whose volume is a measured prescription quantity rather than something the data can identify, so the fill volume IS the peritoneal compartment volume and enters model() directly as v1 <- RRT_PERIT_DIAL_FILL_VOLUME. Time-varying in principle (a patient’s prescription can change between exchanges) but time-fixed per subject in the founding example.
• Units: L
• Type: continuous
• Scope: general
• Reference category: n/a – used as an absolute volume, not a centered or normalized covariate effect. No reference value applies.
• Source aliases:
  • V1 – Hartinger 2023 Table 2, whose entry for the peritoneal compartment volume reads FIXED * with the footnote “Fixed to the actual volume of peritoneal solution used (range 1-2 L)”.
  • VOL IN – Hartinger 2023 Figure 1 schematic label for the instilled volume.
  • exchange volume, fill volume, dwell volume – prose forms across the peritoneal-dialysis literature. Same orientation, no transformation; convert mL to L on ingestion.
• Example models: Hartinger_2023_vancomycin.R (founding example; intraperitoneal vancomycin in adults on CAPD treated for PD-associated peritonitis. The column supplies the volume of the peritoneum compartment directly – v1 <- RRT_PERIT_DIAL_FILL_VOLUME, Cip <- peritoneum / v1 – and no IIV was estimated on it, the source having excluded the peritoneal volume from IIV testing precisely because it is a measured quantity. Observed prescriptions: 2000 mL in 56% of patients, 1500 mL in 39%, 1200 mL and 1000 mL in one patient each; the paper’s simulations use 2 L at 75 and 100 kg and 1.5 L at 50 kg. The value is verifiable against the paper’s own figure axes – a 1500 mg intraperitoneal loading dose reads exactly 750 mg/L on the Figure 4 peritoneal-concentration axis.).
• Notes: Family-conforming member of the RRT_<MODALITY>_<KIND> shape defined by the section-header policy above, with PERIT_DIAL as the modality and a spelled-out measured circuit quantity as the kind – the same construction as RRT_CRRT_EFFLUENT_FLOW, and continuous rather than binary for the same reason. Distinct from the binary PERIT_DIAL (a treatment-modality indicator answering whether a subject is on peritoneal dialysis) and from the anticipated RRT_PERIT_DIAL_STATUS; this column answers how much fluid an exchange instils. A model that needs the drained volume as well – e.g. one that resolves ultrafiltration, or that computes drug removal as Cip * VOL_OUT rather than emptying the compartment – should register a sibling RRT_PERIT_DIAL_DRAIN_VOLUME rather than overload this entry, since the two differ by the ultrafiltrate and are not interchangeable. Note that the end-of-dwell drain and the next instillation belong in the EVENT table (a replacement event zeroing the peritoneal compartment followed by the next dose), not in the ODEs; this column sizes the compartment, it does not empty it.

ALB (canonical for serum albumin)

• Description: Serum albumin concentration.
• Units: g/L (SI; canonical as of 2026-06-19). US-convention papers report in g/dL; convert via ALB_gL = ALB_gdL * 10 on data ingestion. Models calibrated to g/dL values must apply an inline conversion in model() (e.g., alb_gdL <- ALB * 0.1 # SI g/L -> US-convention g/dL) so the structural coefficients stay aligned with their original calibration.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (ALB / ref)^exponent.
• Source aliases:
  • BALB (baseline albumin) – used in Zhou_2021_belimumab.R. Maps directly to ALB; baseline-vs-time-varying status documented in per-model notes.
  • HSA (human serum albumin) – used in Fauchet_2015_lopinavir_unbound.R where the column name follows the paper’s protein-binding-equation notation distinguishing serum albumin from alpha-1 acid glycoprotein. Maps directly to canonical ALB; no value transformation. Reported in g/L; converted to umol/L inside model() via molecular weight 66500 g/mol for the K_HSA linear-binding term.
• Example models: Fasanmade_2009_infliximab.R (g/dL, reference 4.1), Thakre_2022_risankizumab.R (g/L, reference 45), Chua_2025_mirikizumab.R, Moein_2022_etrolizumab.R, Tiraboschi_2025_amlitelimab.R, Yamada_2025_zolbetuximab.R, Li_2019_abatacept.R (g/dL, reference 4.0; the Li 2019 Methods states ‘mg/dL’ which is a publication typo – see the model’s covariateData[[ALB]]$notes), Quartino_2019_trastuzumab.R (g/dL, reference 4; source column ALBU; negative exponent -0.998 on linear CL), Wang_2020_ontamalimab.R (g/L, reference 39), Zhou_2021_belimumab.R (g/L, reference 40; baseline-only, source column BALB), Okada_2025_rocatinlimab.R (g/L, reference 44; source column ALBU; power exponent -1.30 on linear CL), Xu_2020_daratumumab.R (g/L, reference 37.0; power exponent -1.149 on linear CL), Struemper_2017_belimumab.R (g/L, reference 41; baseline-only, source column BALB; power exponent -0.736 on linear CL), Fauchet_2015_lopinavir_unbound.R (g/L; source column HSA; enters the saturable-binding submodel via a linear K_HSA * [ALB] * Cunbound term with K_HSA = 0.036 L/umol, not as a power scaling on CL), Roepcke_2023_rezafungin.R (g/dL, reference 3.2; power exponent -0.708 on the shared peripheral volume V23 rather than on CL; model() applies the inline alb_gdL <- ALB * 0.1 SI-to-US conversion required by this entry’s Units note. Roepcke 2023 Table 2 prints the exponent magnitude without a sign and the footnote-a equation supplies the minus; the negative orientation is corroborated by Table 3, where the low-albumin patient cohort has the larger steady-state volume of distribution), Zhang_2025_abemaciclib_qsp.R (g/L, reference 31 = the source paper’s 3.1 g/dL mean in breast-cancer patients; ALB does NOT scale a clearance or a volume here – it scales the fraction unbound of four analytes through the paper’s Eq 2, fup = 1/(1 + B*ALB), so the effect is on the free concentration driving target engagement rather than on a disposition parameter), Bi_2018_testosteroneCypionate.R (g/dL calibration with the mandated inline ALB * 0.1 conversion; enters ONLY as the within-subject delta (ALB - ALB_BASE) under the Wahlby 2004 decomposition, in a LINEAR multiplier (1 + -0.27 * (dALB_gdL + 0.2)) on the apparent central volume, with the baseline half carried by the paired ALB_BASE column. This is the register’s clearest case of ALB appearing in a model in its delta role only – a consumer supplying a constant albumin column contributes nothing through this term, which is the intended behaviour, not a bug), Jiang_2021_ivosidenib.R (g/L; enters ONLY through the within-subject albumin ratio ALB / ALB_BASE, with power exponents 0.99 on CL/F and 1.1 on Vc/F; the baseline half is carried by ALB_BASE), KhanAsa_2020_voriconazole.R (g/dL calibration with the inline ALB * 0.1 conversion; LINEAR term 1 + 0.249 * (ALB_gdL - 3.2) on voriconazole CL/F, centred on the cohort median).
• Notes: Canonical units standardized to g/L (SI) on 2026-06-19. The per-model covariateData[[ALB]]$units field is load-bearing for model-file source-trace; effect-coefficient magnitude is meaningless without the unit. Per-model conversion to US-convention g/dL (where the structural coefficients were calibrated) is required in model() via an inline alb_gdL <- ALB * 0.1 line - see the per-model notes for the conversion factor used. Conversion factor: 1 g/dL = 10 g/L.

ALB_LT35 (canonical for serum albumin below 35 g/L (3.5 g/dL) indicator)

• Description: Binary indicator dichotomising serum albumin at 35 g/L (equivalently 3.5 g/dL), 1 = albumin below 35 g/L (hypoalbuminaemia; the poor-prognosis category), 0 = albumin at or above 35 g/L. Time-fixed per subject when derived from a baseline albumin measurement. Distinct from the continuous canonical ALB (g/L), which captures albumin magnitude and is used with power or linear-centred scaling; ALB_LT35 is the dichotomised form used by source papers whose published coefficient is defined only on the binary contrast, so the continuous value cannot reconstruct the reported effect.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (serum albumin >= 35 g/L, i.e. >= 3.5 g/dL – normal albumin, the better-prognosis category).
• Source aliases:
  • Albumin < 3.5 (g/dL) – Claret 2014 (J Clin Pharmacol 54(3):253-257) Table 2 covariate row label; the source study reports albumin in US-convention g/dL and dichotomises at 3.5 g/dL, which is 35 g/L in the canonical SI unit. Coded 1/0.
  • ALBLOW / HYPOALB – common NONMEM column-name variants for a baseline hypoalbuminaemia flag dichotomised at the same clinical threshold.
• Example models: Claret_2014_DCR_OS_nsclc.R (additive effect on the log-normal AFT location parameter of overall survival: mu_os += e_alb_lt35_mu_os * ALB_LT35 with e_alb_lt35_mu_os = -0.270, i.e. median OS is exp(-0.270) = 0.763 times shorter – 23.7% shorter – in patients with baseline albumin below 3.5 g/dL).
• Notes: Derive from a continuous albumin column with ALB_LT35 = as.integer(ALB < 35) when ALB is in canonical SI g/L, or as.integer(ALB_gdL < 3.5) when the source column is in US-convention g/dL. Register both ALB and ALB_LT35 in a dataset when downstream users may want either form; a model file should declare only the one its published coefficient is defined on. 3.5 g/dL is the conventional clinical-chemistry lower limit of normal for serum albumin and is the threshold used across the oncology-prognosis literature, so this canonical is expected to be reused rather than re-thresholded; a paper dichotomising at a different cut-point should register a sibling (ALB_LT30, ALB_LT40) following the AGE_GE70 / AGE_GT65 and MET_GE3 / MET_GE4 precedent for co-existing thresholds on one underlying continuous variable.

ALB_BASE (canonical for per-subject baseline serum albumin (time-fixed))

• Description: Per-subject baseline serum albumin, time-fixed. Used when a source paper enters the subject’s baseline albumin as a constant column alongside the time-varying ALB column under Wahlby 2004’s extended covariate-model decomposition (Br J Clin Pharmacol 2004;58(4):367-377). The within-subject delta is computed in model as (ALB - ALB_BASE).
• Units: g/L (SI), matching the paired ALB column. US-convention papers report in g/dL; convert on ingestion via ALB_BASE_gL = ALB_BASE_gdL * 10 and apply the inline alb_base_gdL <- ALB_BASE * 0.1 conversion in model() when the published coefficients were calibrated in g/dL. Document per-model via covariateData[[ALB_BASE]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – used either as the ratio (ALB_BASE / ref) in a power effect or as (ALB_BASE - ref) in a linear or linear-exp effect. Reference value observed: 4.60 g/dL = 46.0 g/L (Bi 2018 control-stream constant).
• Source aliases:
  • ALBUMIN (baseline albumin, paired with a separate CALBUMIN change-from-baseline column) – Bi 2018 $INPUT convention; used in Bi_2018_testosteroneCypionate.R.
• Example models: Bi_2018_testosteroneCypionate.R (g/dL calibration, reference 4.60 g/dL; power exponent -1.55 on the apparent central volume of a depot-testosterone popPK model, paired with the delta effect -0.27 dL/g on (ALB - ALB_BASE) centred at -0.2 g/dL. The centring constants appear only in the Supplementary S6 control stream, not in the main article), Jiang_2021_ivosidenib.R (g/L, reference 37 g/L; power exponents 0.82 on CL/F and 0.73 on Vc/F, paired with the albumin ratio ALB / ALB_BASE (exponents 0.99 and 1.1) – the ratio form of the same baseline-plus-within-subject decomposition, and a second paper using ALB_BASE with consistent semantics), Nguyen_2022_cemiplimab.R (g/L, reference 38.6 g/L; power exponent -0.217 on the central volume V1, alongside a separate time-varying ALB column carrying exponent -1.11 on CL; the paper keeps the baseline and time-varying albumin columns separate without a delta decomposition).
• Notes: Specific scope because the BCOV/DCOV decomposition is a paper-defined modelling choice. Promote to general if a second paper ratifies the column with consistent semantics. Distinct from time-fixed adult-cohort ALB (where the per-subject albumin is constant merely by data design); use ALB_BASE only when the source paper deliberately separates the baseline snapshot from a time-varying albumin column. Also distinct from ALB_LT35, which is a dichotomised prognostic indicator rather than a methodologic baseline-snapshot column.

TPRO (canonical for total serum protein)

• Description: Total serum protein concentration (sum of albumin + globulins; baseline or time-varying).
• Units: g/L (SI; canonical as of 2026-06-19). US-convention papers report in g/dL; convert via TPRO_gL = TPRO_gdL * 10 on data ingestion. Models calibrated to g/dL must apply an inline conversion in model (e.g., tpro_gdL <- TPRO * 0.1). 1 g/dL = 10 g/L.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (TPRO / ref)^exponent. Reference value observed: 74 g/L (Frey 2010 pooled-cohort median).
• Source aliases:
  • PROT – Frey 2010 abbreviation in the final-model equation.
  • TP – common clinical-chemistry abbreviation.
• Example models: Frey_2010_tocilizumab.R (g/L, reference 74; exponent -1.1 on V1), Hahn_2019_sufentanil.R (g/L, reference 45 = the source paper’s 4.5 g/dL cohort median; exponent +2.46 on the PERIPHERAL volume V2 rather than on V1 or a clearance. The positive orientation is unusual – higher total protein predicts a larger peripheral volume, the opposite sign to Frey 2010 – and Hahn 2019’s Discussion acknowledges the discrepancy explicitly, proposing that sufentanil binding is governed mainly by alpha-1 acid glycoprotein rather than by total protein, and that low total protein may instead mark impaired hepatic function. The same Discussion cautions that V2 shrinkage was 33%, so the effect should be treated as observational), Jeong_2020_tiropramide.R (paper in g/dL; the model converts inline via TPRO / 10 and applies a LINEAR centred form 1 + (TP - 7.6) * slope on both CL/F (-0.804 per g/dL) and V/F (-1.049 per g/dL), the first linear rather than power use of this column).
• Notes: Distinct from ALB (serum albumin, the largest single component of total protein). Frey 2010 retains both TPRO and ALB on V1 as separate covariates with opposite signs (TPRO negative, ALB positive) and notes there is no clear mechanistic explanation; the joint effect may reflect serum-volume modifications. TPRO ratified canonically on 2026-04-28 alongside the Frey 2010 extraction.

FU (canonical for fraction unbound of parent drug in plasma supplied as a data column)

• Description: Fraction of the parent drug unbound in plasma, supplied to the model as a data column rather than as an internal parameter. Two established usages: (a) an individual subject’s measured fraction unbound, typically determined by ultrafiltration of a per-subject ex-vivo plasma sample (occasionally by equilibrium dialysis) – the founding Hamren 2008 usage; and (b) a scenario-level fraction unbound that a mechanistic / PBPK model varies across simulated patient types, where the paper assigns one fu value per scenario rather than per subject – the Abdullah-Koolmees 2024 usage. In both cases the value is an input the user supplies, not a quantity the model derives. Distinct from binding-protein concentrations (ALB / AAG / TPRO), which would derive fu indirectly through a binding equation – FU is the fraction itself.
• Units: % (percent; e.g., 0.1 = 0.1% fraction unbound = 0.001 expressed as a fraction) OR unitless fraction, depending on the source paper’s convention. The per-model covariateData[[FU]]$units field documents which, and MUST be set; effect-coefficient magnitudes and Kp scalings are meaningless without the unit.
• Type: continuous
• Scope: general
• Reference category: n/a – used with linear centering (1 + slope * (FU - ref)), with power scaling (FU / ref)^exponent, or directly as a multiplicative physiological factor (PBPK Kp = Kpu * FU and unbound-drug clearance terms). Reference values observed: 0.1% (Hamren 2008 tesaglitazar cohort median imputed for three subjects with missing fu measurements; range 0.06-0.2%); 0.42 unitless (Abdullah-Koolmees 2024 voriconazole, healthy / DDI scenarios at plasma albumin 40 g/L, with 0.49 in the ICU scenarios at albumin 30 g/L).
• Source aliases:
  • f_u – Hamren 2008 paper notation (italicised fu).
  • FRAC_UNBOUND – common NONMEM column-name form when the value is reported as a unitless fraction.
  • fup – Abdullah-Koolmees 2024 supplement R source (fup, fup_icu, fup_ddi); unitless fraction, same orientation as the canonical.
• Example models: Hamren_2008_tesaglitazar.R (%, reference 0.1%; linear centered effect 555 %/unit fu on the parent metabolic clearance CLmt – per-subject fu measurements at day 42 by ultrafiltration drive the realised CLmt range from approximately 1.5 to 3.0 L/h across the cohort fu range 0.06-0.2%; founding example), AbdullahKoolmees_2024_voriconazole_pbpk.R (unitless fraction, scenario-level; enters twice in a whole-body PBPK – once converting every tissue Kpu to a tissue:plasma Kp via Kp = Kpu * FU, and once selecting the unbound drug presented to saturable hepatic CYP metabolism and to renal clearance), Panday_2025_sert_tremor_cavg_mbma.R, Panday_2025_sert_tremor_cmax_mbma.R (unitless fraction, per-DRUG rather than per-subject: the human plasma fraction unbound of each of 20 SERT inhibitors, 0.02 for sertraline to 0.8 for tramadol and tapentadol, converting the reported total plasma concentration to an unbound one in Eq. S1).
• Notes: Rare in popPK literature – only papers that do per-subject ultrafiltration measurements report the per-subject form. Most popPK papers either (a) assume a single fixed population-typical fu, in which case use a parameter-side fu paper-named parameter rather than this covariate, or (b) parameterise the effect through measured binding-protein concentrations (ALB, AAG, TPRO). Use FU when the fu value is a per-record model input the user is expected to supply – either because the paper measured it per subject, or because the paper varies it across named simulation scenarios. Promoted from specific to general alongside the Abdullah-Koolmees 2024 voriconazole whole-body PBPK extraction, the second model to use FU with consistent “fraction unbound of parent in plasma, supplied as data” semantics across two distinct mechanistic uses.

CSF_TPRO (canonical for cerebrospinal-fluid total protein)

• Description: Total protein concentration measured in cerebrospinal fluid (baseline or time-varying). CNS-compartment analogue of TPRO; used in popPK models of CSF-penetrating drugs as a surrogate for blood-brain-barrier integrity (elevated CSF protein indicates inflammation or barrier breakdown and typically correlates with increased CNS penetration of small-molecule drugs).
• Units: g/L (SI; canonical as of 2026-06-19). US-convention reporting in mg/dL converts via CSF_TPRO_gL = CSF_TPRO_mgdL * 0.01. 1 g/L = 100 mg/dL.
• Type: continuous
• Scope: general
• Reference category: n/a – enters either as an additive term on the logit-scale CSF uptake / barrier parameter, or as a power scaling (CSF_TPRO / ref)^exponent on a penetration fraction. Reference value observed: 1.2 g/L (Germovsek 2018 typical-infant value; sick-neonate cohort median).
• Source aliases:
  • CSF_protein – Germovsek 2018 paper notation.
  • CSFPROT / CSF_PROT – compact column-name forms common in NONMEM control streams.
• Example models: Germovsek_2018_meropenem.R (g/L, reference 1.2; additive on the logit CSF barrier parameter with coefficient theta_CSFproteins = -0.17 per g/L deviation from 1.2; ratified canonically on 2026-05-21 alongside the Germovsek 2018 meropenem extraction), Svensson_2020_rifampicin.R (mg/dL in the source, converted inside model() as CSF_TPRO * 100; rifampicin CSF partition coefficient multiplied by 1 + (log10(prot_mgdl) - log10(165)) / log10(165) * 0.631, centred on the cohort median 165 mg/dL).
• Notes: Distinct from TPRO (serum total protein) – the two are biologically independent because the blood-brain barrier prevents free equilibration of serum protein into CSF. Normal CSF protein is approximately 0.15-0.45 g/L in healthy adults; sick neonates and meningitis patients can reach several g/L. The covariate is typically time-varying because CSF protein evolves over the course of CNS inflammation; missing values are commonly imputed to the cohort median when the source paper does not report a per-sample CSF protein measurement.

IGG (canonical for serum immunoglobulin G)

• Description: Serum total immunoglobulin G concentration (baseline or time-varying). Used in mAb PK analyses as a competition-for-FcRn-recycling covariate on therapeutic-mAb clearance – high endogenous IgG is hypothesized to displace the therapeutic mAb from FcRn salvage and increase its catabolic clearance.
• Units: g/L (SI; canonical as of 2026-06-19). US-convention papers report in mg/dL; convert via IGG_gL = IGG_mgdL * 0.01 on data ingestion. 1 g/L = 100 mg/dL. Models calibrated to mg/dL must apply an inline conversion in model (e.g., igg_mgdL <- IGG * 100).
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (IGG / ref)^exponent. Reference values observed: 14.8 g/L (Zhou 2021), 9.65 g/L (Yang 2021).
• Source aliases:
  • BIGG (baseline IgG) – used in Zhou_2021_belimumab.R.
  • IGGBL (baseline IgG) – used in Yang_2021_cemiplimab.R.
• Example models: Zhou_2021_belimumab.R (g/L, reference 14.8; baseline-only; exponent 0.293 on CL), Yang_2021_cemiplimab.R (g/L, reference 9.65; small positive exponent 0.184 on shared CL/Q), Struemper_2017_belimumab.R (g/L, reference 13.7; baseline-only; exponent 0.347 on CL).
• Notes: Mechanistically meaningful for monoclonal-antibody PK because endogenous IgG competes with the therapeutic mAb for FcRn-mediated recycling. The per-model covariateData[[IGG]]$units field is load-bearing (1 g/L ~= 100 mg/dL). Baseline-vs-time-varying status documented in covariateData[[IGG]]$notes. Distinct from lIgG0 / IgG-as-a-state in mechanistic FcRn-competition TMDD models (e.g., Valenzuela_2025_nipocalimab.R), where IgG is a dynamic state, not a baseline covariate; use IGG only when the source paper treats IgG as a static (baseline) covariate column.

IGM (canonical for serum immunoglobulin M)

• Description: Serum total immunoglobulin M (IgM) concentration (baseline). Used in IgRT population-PK analyses as a proxy for B-cell antibody-producing capacity / humoral function – IgM is the first antibody produced after B-cell activation, so circulating IgM reflects ongoing B-cell activity prior to class-switching to IgG.
• Units: g/L (SI; canonical as of 2026-06-19). US-convention reporting in mg/dL converts via IGM_gL = IGM_mgdL * 0.01. 1 g/L = 100 mg/dL.
• Type: continuous
• Scope: specific
• Reference category: n/a – used with power scaling (IGM / ref)^exponent. Reference values observed: 0.21 g/L (Cheng 2026, pooled PID + SAD pediatric cohort median).
• Source aliases: none known.
• Example models: Cheng_2026_immunoglobulin.R (g/L, reference 0.21; baseline-only; power exponent 0.11 on baseline IgG (CBAS) – IgM enters as a humoral-capacity proxy that informs the endogenous-IgG baseline rather than directly modifying clearance).
• Notes: IgM is the immune-globulin class produced by activated B cells before class-switching, so it remains detectable in patients with hypogammaglobulinaemia who still have residual B-cell function. Scope: specific because the relevance of IgM as a covariate depends on the paper’s mechanistic interpretation (in Cheng 2026 it acts on the endogenous-IgG baseline; future use cases may differ). Promote to general if a second paper retains IgM with consistent semantics.

TBILI (canonical for total bilirubin)

• Description: Total serum bilirubin concentration.
• Units: umol/L (SI; canonical as of 2026-06-19). US-convention papers report in mg/dL; convert via TBILI_umolL = TBILI_mgdL * 17.1 on data ingestion. 1 mg/dL = 17.1 umol/L. Models calibrated to mg/dL must apply an inline conversion in model() (e.g., tbili_mgdL <- TBILI / 17.1 # SI umol/L -> US-convention mg/dL).
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (TBILI / ref)^exponent.
• Source aliases:
  • BIL (legacy NONMEM short label for total bilirubin) – used in NA_NA_lidocaine.R (DDMODEL00000281; binarised at threshold 0.53 mg/dL with BIL_HIGH = as.integer(BIL > 0.53)).
  • BILT (Urien 2005 capecitabine paper’s NONMEM short label for “total bilirubin”) – used in Urien_2005_capecitabine.R (umol/L, reference 8.8; power scaling on the capecitabine non-transformation CL10 and on the 5’-DFUR -> 5-FU rate constant K34).
• Example models: Yamada_2025_zolbetuximab.R (mg/dL, reference 0.38; small positive exponent 0.0347 on V1), NA_NA_lidocaine.R (mg/dL, source column BIL; binary effect at threshold 0.53 mg/dL on the GX elimination rate constant K30), Urien_2005_capecitabine.R (umol/L, reference 8.8; source column BILT; positive exponent +0.32 on capecitabine non-transformation CL10 and negative exponent -0.36 on the 5’-DFUR -> 5-FU rate constant K34), Cheng_2020_sirolimus.R (umol/L, reference 11.29 cohort median; source column TBIL; negative exponent -0.32 on sirolimus CL/F in children), Michalickova_2020_riociguat.R (mg/dL, reference 0.69 = cohort median, source column BILTOT; exponent -0.462 on the riociguat formation clearance to its M1 metabolite, CLf,M1/F).
• Notes: Hepatic-function marker. Unit varies by paper (US convention mg/dL, SI convention umol/L; 1 mg/dL ~= 17.1 umol/L). The per-model covariateData[[TBILI]]$units field is load-bearing.

DBIL (canonical for direct (conjugated) bilirubin)

• Description: Direct (conjugated) serum bilirubin concentration. Distinct from TBILI: direct bilirubin is the water-soluble glucuronide-conjugated fraction processed by hepatocytes and excreted in bile, so a rise in DBIL specifically flags impaired biliary excretion / cholestasis or intrahepatic shunting, whereas total bilirubin also captures unconjugated (indirect) hyperbilirubinaemia from haemolysis or Gilbert-type conjugation defects.
• Units: umol/L (SI; canonical as of 2026-06-19). US-convention papers report in mg/dL; convert via DBIL_umolL = DBIL_mgdL * 17.1 on data ingestion. 1 mg/dL = 17.1 umol/L. Models calibrated to mg/dL must apply an inline conversion in model (e.g., dbil_mgdL <- DBIL / 17.1).
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (DBIL / ref)^exponent. Reference values observed: 2.6 umol/L (Chen 2015 voriconazole Chinese ICU cohort population median), 2.82 umol/L (Li 2023 TQ-B3203 Chinese advanced-solid-tumor cohort population median).
• Source aliases: none known; source papers commonly write the abbreviation DBIL directly.
• Example models: Chen_2015_voriconazole.R (umol/L, reference 2.6; negative exponent -0.40 on CL: CL = TVCL * (DBIL / 2.6)^-0.40), Li_2023_TQB3203.R (umol/L, reference 2.82; negative exponent -0.24 on CL and -1.77 on the inter-compartmental clearance CL2: CL = TVCL * (BMI / 23.44)^0.78 * (DBIL / 2.82)^-0.24).
• Notes: Hepatic-function / cholestasis-specific marker. Unit varies by paper (US convention mg/dL, SI convention umol/L; 1 mg/dL ~= 17.1 umol/L). Distinct entry from TBILI because direct vs total are not interchangeable: total = direct + indirect, and the two fractions track different pathophysiologic processes. Promoted from Scope: specific to Scope: general alongside the Li 2023 TQ-B3203 extraction, which is the corroborating second model the prior note required: it uses DBIL with the same semantics as the founding example (SI umol/L, median-normalized negative power exponent on clearance, interpreted as impaired biliary excretion lowering clearance) in an unrelated drug and cohort. Both registered uses are median-normalized power effects on a clearance term, so the reference value is always a cohort median rather than a fixed clinical threshold – record the per-model median in covariateData[[DBIL]]$notes. Li 2023 additionally illustrates that a paper may retain DBIL on an inter-compartmental clearance as well as on elimination clearance; an inter-compartmental effect changes the concentration-time shape (and therefore Cmax) without changing AUC0-inf, which depends only on dose and elimination clearance.

AST (canonical for aspartate aminotransferase)

• Description: Serum aspartate aminotransferase activity (baseline or time-varying).
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[AST]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (AST / ref)^exponent.
• Source aliases:
  • SGOT (serum glutamic-oxaloacetic transaminase; the legacy clinical-chemistry name for AST) – used in Quartino_2019_trastuzumab.R.
• Example models: Lu_2014_trastuzumabemtansine.R (U/L, reference 27; small positive exponent 0.071 on CL), Quartino_2019_trastuzumab.R (IU/L, reference 24; source column SGOT; positive exponent 0.205 on linear CL), Xu_2026_caspofungin_optimalDesign.R (U/L, reference 52.5 = the fitted subset median; exponent 0.898 on the intercompartmental clearance Q rather than on CL – the only example on Q; the effect was retained in the paper’s supplemental design model but did NOT survive into its final model, so treat it as subset-specific rather than a class effect), Osawa_2018_asunaprevir.R (U/L, reference 52 = the cohort median; genuinely time-varying, exponent -0.382 on CL/F, and the only example paired with a separate AST_BASE term – exponent -0.598 – normalised to the SAME reference constant), Krishnatry_2021_molibresib.R (IU/L, reference 28; time-varying, exponent -0.194 on the clearance of the active metabolite composite GSK3529246 rather than of the parent).
• Notes: Hepatic-function marker. Commonly reported alongside ALT and TBILI; register a separate ALT canonical if a future paper requires it. SGOT is the older lab-reporting name; values and units are identical to AST. When a paper carries a time-fixed baseline AST column alongside this one and gives the two distinct effects, the baseline column is AST_BASE; do not collapse the pair onto this entry.

ALT (canonical for alanine aminotransferase)

• Description: Serum alanine aminotransferase activity (baseline or time-varying).
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[ALT]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (ALT / ref)^exponent.
• Source aliases:
  • SGPT (serum glutamic-pyruvic transaminase; the legacy clinical-chemistry name for ALT, paralleling SGOT -> AST) – used in NA_NA_lidocaine.R (DDMODEL00000281; binarised at threshold 11 with SGPT_HIGH = as.integer(SGPT > 11)).
• Example models: Nikanjam_2019_siltuximab.R (U/L, reference 19; small negative exponent -0.096 on CL), Melhem_2022_dostarlimab.R (U/L, reference 18; small negative exponent -0.0585 on CL, time-varying), NA_NA_lidocaine.R (source column SGPT; binary effects at threshold 11 on the GX rate constant K30 and on the 2,6-xylidide rate constant K40), Nguyen_2022_cemiplimab.R (IU/L, reference 19; baseline ALT, exponent -0.0729 on CL), Xie_2022_colistinSulfate.R (U/L, reference 37, cohort median 37.5, range 7-495 in critically ill adults; power exponent 0.635 on the PERIPHERAL volume of colistin sulfate rather than on a clearance, so it alters the time to steady state but not the steady-state average concentration).
• Notes: Hepatic-function marker. Commonly reported alongside AST and TBILI. SGPT is the older lab-reporting name; values and units are identical to ALT.

ALP (canonical for alkaline phosphatase)

• Description: Serum alkaline phosphatase activity (baseline or time-varying). Liver-function / cholestasis marker; often used in popPK covariate models either as a continuous concentration with power scaling or as a binary above/below upper-limit-of-normal (ULN) indicator. When binarized inline, document the ULN threshold used (typically ~120 U/L for adults; varies by lab, age, and sex).
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[ALP]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (ALP / ref)^exponent, with a linear-deviation form, or binarized inline as alp_high <- (ALP > uln) for a binary >ULN indicator.
• Source aliases: none known.
• Example models: Gupta_2016_lenvatinib.R (binarized inline as alp_high <- (ALP > 120); the source paper enters ALP as a 0/1 NONMEM indicator with ALP = 1 when the ratio ALP/ULN > 1; multiplicative effect on CL/F: 0.883^alp_high), Wada_2023_sparsentan.R (continuous power scaling on CL/F normalised to the 68 U/L overall cohort median: (ALP / 68)^-0.208; elevated alkaline phosphatase lowers apparent clearance, and ALKP was the only one of four hepatic markers – ALKP, ALT, AST, total bilirubin – retained after backward elimination in a study where hepatic impairment itself was not tested as a covariate), Lalovic_2020_lemborexant.R (power exponent -0.118 on CL/F referenced to the 71 U/L cohort median).
• Notes: Liver-function / cholestasis marker; routine clinical-chemistry covariate. Commonly tested alongside ALT / AST / GGT / TBILI.

PHOS (canonical for serum phosphate)

• Description: Serum inorganic phosphate concentration, reported clinically as elemental phosphorus (baseline or time-varying). A routine serum-chemistry analyte and the pharmacodynamic biomarker of fibroblast growth factor receptor (FGFR) inhibition: FGFR1 blockade suppresses FGF23 signalling in the proximal tubule, reducing urinary phosphate excretion and raising serum phosphate, so on-treatment hyperphosphatemia is an on-target class effect of the FGFR-inhibitor drug class. Used both as a baseline covariate (predicting the on-treatment rise) and, in its change-from-baseline form, as an exposure surrogate.
• Units: mmol/L (SI; canonical). US-convention papers report in mg/dL; convert via PHOS_mmolL = PHOS_mgdL * 0.3229 on data ingestion (elemental phosphorus, atomic mass 30.974 g/mol, so 1 mg/dL = 0.32285 mmol/L and 1 mmol/L = 3.0975 mg/dL). Models calibrated against mg/dL values must either apply an inline conversion in model (e.g., phos_mgdL <- PHOS * 3.0975) or carry the coefficient rescaled to per-mmol/L units, with the conversion shown in the model file – the same treatment ALB receives (see the ALB entry). Adult reference range is roughly 2.5-4.5 mg/dL (0.81-1.45 mmol/L).
• Type: continuous
• Scope: general
• Reference category: n/a – entered so far in an uncentered linear form theta * PHOS. Reference values observed: baseline serum phosphate was not tabulated by Gong 2023, but the observed quartile boundaries of the change from baseline at 13.5 mg once daily were 0.5, 2.1, 2.7, 3.5, and 6.3 mg/dL (0.16, 0.68, 0.87, 1.13, and 2.03 mmol/L; Gong 2023 Figure 3a).
• Source aliases:
  • PHOS – the NONMEM $INPUT column name in the Gong 2023 Appendix S1 control stream, in mg/dL. Maps directly to the canonical name; unit conversion to SI mmol/L is required.
  • PO4 / PHOSPHORUS / IPHOS (inorganic phosphorus) – common clinical-laboratory reporting names for the same analyte.
• Example models: Gong_2023_pemigatinib_phosphate.R (mmol/L; uncentered linear coefficient -0.5730 mg/dL per mmol/L on the change-from-baseline endpoint, rescaled from the published -0.185 per mg/dL; the model’s endpoint is retained in the paper’s mg/dL so Table S2 remains usable verbatim).
• Notes: Distinct from ALP (alkaline phosphatase, an enzyme activity) and from CPK (creatine phosphokinase) despite the shared word stem – PHOS is the electrolyte. Housed in the renal / hepatic function section because phosphate homeostasis is renally regulated and because source papers report it inside the same serum-chemistry covariate panel as ALB / ALP / ALT / AST / TBILI (Gong 2023 tests exactly that panel); the register has no separate electrolyte section. When a source paper models the change from baseline in serum phosphate as an endpoint rather than as a covariate, that is a model output, not a covariate column – register only the baseline (or time-varying) input value here.

GGT (canonical for gamma-glutamyltransferase)

• Description: Serum gamma-glutamyltransferase activity (baseline or time-varying); hepatic / cholestatic biliary-enzyme marker.
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[GGT]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with linear-deviation form 1 + theta * (GGT - ref) or power scaling (GGT / ref)^exponent. Reference values observed: 33 U/L (Retlich 2015 popPK linagliptin median), 32.3 U/L (Retlich 2015 popPK/PD linagliptin median).
• Source aliases: none known.
• Example models: Retlich_2015_linagliptin.R (U/L, reference 33; linear-deviation effect on linagliptin CL with coefficient -0.0339 % per U/L deviation. The PK/PD layer uses GGT (reference 32.3 U/L) as a piecewise covariate on baseline DPP-4 activity BSL with a linear-deviation effect below GGT = 175 U/L and a constant +21.3% effect above the threshold).
• Notes: Liver-function / cholestasis marker; routine clinical-chemistry covariate. Commonly tested alongside ALT / AST / ALP / TBILI. The piecewise above/below-threshold form in Retlich 2015 reflects empirical saturation of the GGT-vs-DPP-4-activity relationship at extreme values.

TBA (canonical for total serum bile acids)

• Description: Total serum (or plasma) bile-acid concentration, the summed pool of primary and secondary conjugated and unconjugated bile acids. A sensitive marker of hepatocellular uptake / excretory function and of cholestasis: because bile acids undergo near-complete first-pass hepatic extraction from portal blood, the systemic concentration rises early when sinusoidal uptake or canalicular secretion is impaired, often before the transaminases or bilirubin move. Both baseline and time-varying usages are covered; document per-model in covariateData[[TBA]]$notes which applies.
• Units: umol/L (the conventional clinical reporting unit; document per-model via covariateData[[TBA]]$units).
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (TBA / ref)^exponent or exponential effects exp(coef * (TBA - ref)). Reference values observed: 3.95 umol/L (Liu 2025, cohort median in an elderly Chinese voriconazole TDM cohort).
• Source aliases:
  • TBA – used in Liu_2025_voriconazole.R (Table 1 / Table 2 and the published final-model equation).
  • SBA, TSBA, BA – total-(serum)-bile-acid variants seen in clinical-chemistry panels and NONMEM $INPUT columns; the same analyte, no value transformation.
• Example models: Liu_2025_voriconazole.R (median-normalized power effect on apparent voriconazole clearance, (TBA / 3.95)^-0.158, so a cholestatic rise in bile acids predicts slower clearance and higher voriconazole exposure).
• Notes: Registered on the same terms as the established general-scope liver-panel analytes TBILI, DBIL, AST, ALT, ALP and GGT – a routine clinical-chemistry measurement that recurs across hepatically-cleared-drug population-PK analyses – and is not an alias of any of them: total bile acids report hepatic uptake and canalicular transport capacity, whereas the transaminases report hepatocyte injury and bilirubin reports conjugation plus excretion, so the four can move independently (bile acids commonly rise with normal transaminases in early or mild cholestasis). Closest sibling is GGT, the other cholestasis-weighted marker, but GGT is an induced membrane enzyme and TBA a transported substrate pool; they are separate assays. Values against this column must be a total bile-acid pool, not an individual species (cholic, chenodeoxycholic, ursodeoxycholic acid), and not a post-prandial-stimulated measurement unless the source paper states so – record the fasting-vs-postprandial status in per-model notes, since the postprandial pool can be several-fold the fasting one. Right-skewed in unselected inpatient cohorts (Liu 2025: median 3.95, IQR 2.50-7.45 umol/L), so median normalization rather than mean centring is the usual encoding.

LDH (canonical for serum lactate dehydrogenase)

• Description: Serum lactate dehydrogenase activity (baseline or time-varying). General-purpose marker of tissue / cellular turnover; in oncology PK analyses it is interpreted as a disease-burden / cell-turnover proxy.
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[LDH]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (LDH / ref)^exponent or with an additive linear-on-log form exp(coef * (log(LDH) - log(ref))) (algebraically equivalent to (log(LDH) / log(ref))^coef). Reference values observed: 217 U/L (Sanghavi 2020).
• Source aliases:
  • BLDH (baseline LDH) – used in Sanghavi_2020_ipilimumab.R.
• Example models: Sanghavi_2020_ipilimumab.R (linear-on-log form on CL with reference 217 U/L; coefficient 0.703), NA_NA_lidocaine.R (DDMODEL00000281; binary stratification at threshold 195 U/L switching the typical-value baseline of the 2,6-xylidide rate constant K40).
• Notes: Universal lab marker. Sanghavi 2020 log-transforms LDH because the distribution is heavily right-skewed (range 74-6,245 U/L over a median of 217); other papers may use a simple (LDH/ref)^exponent form. Document the functional form in covariateData[[LDH]]$notes. When a paper supplies LDH already normalized to each laboratory’s upper limit of normal (a unitless ratio rather than a U/L activity), use the sibling canonical LDHR instead – the two are NOT interchangeable, because LDH carries U/L units and LDHR is dimensionless.

LDHR (canonical for serum lactate dehydrogenase normalized to the laboratory’s upper limit of normal)

• Description: Baseline (or time-varying) serum lactate dehydrogenase activity divided by the reporting laboratory’s upper limit of normal (LDH_observed / ULN_LDH). Dimensionless. Used in multi-site oncology analyses so LDH from laboratories with different reference ranges can be pooled on a common scale, and because elevated-LDH-versus-normal is the clinically meaningful contrast in advanced melanoma rather than the absolute activity.
• Units: (unitless ratio)
• Type: continuous
• Scope: general
• Reference category: n/a – LDHR = 1 is the upper limit of normal, so a log-linear term (1 + log(LDHR) * coef) is centered at exactly the ULN and contributes nothing for a patient at the ULN. Also usable as a power term (LDHR / ref)^exponent. Feng 2019 reports a cohort median LDHR of 1.0 (range 0.4-40.5).
• Source aliases:
  • BLDHU – Feng 2019 Table 2 footnote f (“BLDHU is the normalized baseline LDH with upper limit of normal”).
  • LDH ratio – Feng 2019 Table 1 (“LDH ratio indicates patient’s actual value divided by the upper limit of normal”).
• Example models: Feng_2019_ipilimumab.R (log-linear form on both baseline tumor burden and the linear tumor-growth rate: TB0_TV = TB0_REF * (1 + log(LDHR) * 0.868) and TG_TV = TG_REF * (1 + log(LDHR) * 0.771) * (1 + CAV * -0.00342); note the linear-in-coefficient form (1 + log(LDHR) * coef), not the exponential exp(coef * log(LDHR)) form more common elsewhere in the register).
• Notes: Structurally parallel to ALBR (serum albumin normalized to ULN) – the same “analyte / ULN” construction applied to a different analyte. Distinct from LDH, which is the absolute activity in U/L; do not substitute one for the other, since a model calibrated on a ratio near 1 will be badly wrong if fed a U/L value near 200. Papers log-transform this ratio (as both LDH and LDHR users do) because the distribution is heavily right-skewed – Feng 2019 Table 1 footnote a: “Log-transformed LDH ratio was used in tumor growth dynamics and overall survival model development due to skewed distribution”, with an observed range spanning two orders of magnitude (0.4-40.5). General scope because ULN-normalized LDH is a standard multi-site oncology covariate and is not tied to any one analysis plan.

T4 (canonical for serum total thyroxine)

• Description: Serum total (protein-bound plus free) thyroxine concentration, the principal thyroid hormone. Baseline or time-varying. A member of the plain clinical-chemistry analyte family alongside ALB, TPRO, TBILI, AST, ALT, ALP, GGT and LDH.
• Units: ug/dL (the conventional clinical-chemistry unit for total T4; adult reference interval roughly 4.5-12 ug/dL). Papers reporting SI nmol/L convert via T4_ugdL = T4_nmolL / 12.87. Document per-model via covariateData[[T4]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – used with power scaling (T4 / ref)^exponent. Reference value observed: 7.40 ug/dL (Bi 2018 control-stream constant, near the middle of the adult reference interval).
• Source aliases:
  • THYROXIN – Bi 2018 $INPUT column name and Figure 1 forest-plot label; used in Bi_2018_testosteroneCypionate.R.
• Example models: Bi_2018_testosteroneCypionate.R (reference 7.40 ug/dL; power exponent 1.19 on the zero-order synthesis rate of LHRH-stimulated luteinizing hormone in a depot-testosterone HPG-axis PK/PD model. Bi 2018 retained the covariate in the final model on statistical grounds but reports that “No difference in simulated LH suppression was identified among volunteers with different thyroxin values”).
• Notes: Do not confuse with thyroxine-BINDING GLOBULIN (TBG), the transport protein, which is a different analyte on a different scale (adult reference roughly 12-30 ug/mL). Bi 2018 is the cautionary case: its Table 1 tabulates thyroxine-binding globulin (group medians 19.5 / 18.5 / 21 ug/mL) while its control stream’s covariate is THYROXIN referenced to 7.40, and the supplementary LH dataset carries thyroxin = 7.4 for its example subject – a value consistent with total T4 in ug/dL and inconsistent with TBG. A future extraction needing the binding protein should register a separate TBG canonical rather than reuse this one. Free thyroxine (FT4, ng/dL) is likewise a distinct analyte and needs its own sibling entry.

AMYL (canonical for blood amylase activity)

• Description: Blood (serum or plasma) total amylase activity (baseline or time-varying). Routine clinical-chemistry enzyme, conventionally read as a pancreatic / salivary marker, but amylase is also cleared renally and therefore accumulates as glomerular filtration falls. In population PK analyses of renally-impaired cohorts it is used in the latter role – as a surrogate for uremic burden rather than for pancreatic disease – so the direction of any effect should be interpreted against the paper’s stated mechanism, not assumed to be pancreatic.
• Units: U/L (SI; canonical, matching the AST / ALT / ALP / GGT / LDH enzyme family; IU/L is used interchangeably). Document per-model via covariateData[[AMYL]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (AMYL / ref)^exponent. Reference values observed: 59.5 IU/L (Kim 2025 healthy-subject median).
• Source aliases:
  • Amyl – Kim 2025 Table 1 / Table 2 and NONMEM $INPUT column abbreviation.
• Example models: Kim_2025_evogliptin.R (IU/L, reference 59.5; positive exponent 0.363 on relative bioavailability F1, paired with TRIG; both markers rise with worsening renal impairment and jointly express the paper’s finding that uremia inhibits CYP3A4-mediated first-pass metabolism and so raises oral bioavailability).
• Notes: Routine clinical-chemistry covariate; follows the bare-abbreviation naming of the established enzyme canonicals AST, ALT, ALP, GGT, and LDH. Distinct from lipase, which is the companion pancreatic enzyme and was recorded but not tested in Kim 2025 – register a separate LIPASE canonical if a future paper retains it. Because amylase can move for either pancreatic or renal reasons, covariateData[[AMYL]]$notes should state which interpretation the source paper intends.

HEPIMP_MILD (canonical for mild hepatic impairment indicator)

• Description: 1 = mild hepatic impairment per the National Cancer Institute Organ Dysfunction Working Group (NCI ODWG) criteria, 0 = normal hepatic function or non-mild category. NCI ODWG mild = total bilirubin <= ULN with AST > ULN, OR total bilirubin > 1.0xULN to <= 1.5xULN with any AST.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal hepatic function; the moderate / severe categories are typically pooled into the reference for population PK analyses where mild impairment is the only category with non-trivial sample size).
• Source aliases:
  • HEPIMP (with values 1 = mild / 0 = others) – used in Lin_2024_casirivimab.R.
  • Child-Pugh A – used in Desai_2016_isavuconazole.R and in the Luo_2024_*_pbpk.R CES1 semi-PBPK family, where the three HEPIMP_MILD / HEPIMP_MOD / HEPIMP_SEV indicators are mutually exclusive and select the Child-Pugh A / B / C physiology columns respectively, with all three 0 selecting the normal-hepatic-function column.
• Example models: Lin_2024_casirivimab.R (NCI ODWG mild; multiplicative fractional change on CL), Lu_2022_patritumab.R (NCI ODWG mild; paired with HEPIMP_MOD_OR_MISSING; multiplicative fractional effect 0.706 on CLDXd for mild impairment vs the normal-hepatic-function reference), Desai_2016_isavuconazole.R (Child-Pugh A; log-additive shift e_hepimp_mild_cl = log(1.55 / 2.54) = -0.494 on CL and e_hepimp_mild_q = log(38.8 / 33.678) = +0.142 on Q vs the healthy reference, paired with HEPIMP_MOD for the parallel Child-Pugh B stratum), Nguyen_2019_cabozantinib_liver_dysfunction.R (NCI ODWG mild; direct multiplicative factor 1.12^HEPIMP_MILD on CL/F and 1.04^HEPIMP_MILD on Vc/F, paired with HEPIMP_MODSEV so that both indicators zero is the normal-hepatic-function reference; 558 / 2023 = 28% of the cohort were mild. This is the rare case where the covariate’s scientific value is a NEGATIVE result: the 12% clearance increase was judged clinically negligible and the authors retained the model WITHOUT these covariates as final, which is why the companion Nguyen_2019_cabozantinib.R omits them).
• Notes: Use this column when a model dichotomizes hepatic-impairment status as “mild vs. others” (i.e., normal + the rare moderate/severe cases pooled into the reference). For models that test moderate or severe as separate categories, register additional canonicals HEPIMP_MOD / HEPIMP_SEV rather than overloading this entry.

HEPIMP_MOD_OR_MISSING (canonical for composite moderate-or-data-missing hepatic impairment indicator)

• Description: 1 = moderate hepatic impairment per the NCI ODWG criteria OR baseline hepatic-function data missing/unknown; 0 = normal hepatic function or any other (non-moderate, non-missing) category. Composite indicator used by source papers that pool the moderate-impairment subgroup with patients whose hepatic-function data are missing because both subgroups are individually too small to estimate as separate effects.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (normal hepatic function; mild impairment is typically captured by a separate HEPIMP_MILD indicator paired with this column, so all-zero corresponds to NCI ODWG group 1 = normal).
• Source aliases:
  • HEPIMP_MOD_MISSING – prior canonical name (pre-2026-06-19 OR-disambiguation standardization).
  • HEPATIC / HEPATIC_MOD_MISSING – informal NONMEM names for the composite group.
• Example models: Lu_2022_patritumab.R (paired with HEPIMP_MILD; multiplicative fractional effect 0.532 on CLDXd; the composite group pools n = 6 moderate-impairment patients with n = 6 missing/unknown patients per Lu 2022 Table S5).
• Notes: Specific scope because the composition of the “moderate or missing” group is paper-defined and the missing/unknown subgroup may have a different distribution of true hepatic-function status across studies. Use only when the source paper explicitly pools the moderate-impairment cases with missing-data cases under a single coefficient; for models that estimate moderate impairment separately (without pooling missing data), use the HEPIMP_MOD canonical instead.

B2M (canonical for serum beta-2-microglobulin)

• Description: Serum beta-2-microglobulin concentration. Low-molecular-weight (~12 kDa) protein freely filtered at the glomerulus and reabsorbed in the proximal tubule; serum levels rise with renal impairment, with increased plasma-cell turnover in multiple myeloma, and with broader lymphoid-cell turnover. Used in oncology PK analyses both as a renal-function proxy and as a tumor-burden / disease-severity covariate.
• Units: mg/L
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (B2M / ref)^exponent. Reference values observed: 3.90 mg/L (Fau 2020 multiple-myeloma cohort median).
• Source aliases: none; B2M is the universal abbreviation.
• Example models: Fau_2020_isatuximab.R (mg/L, reference 3.90; exponent 0.343 on the steady-state linear clearance CLinf), Ide_2022_elotuzumab.R (thresholded B2M >= 3.5 mg/L indicator, factor 1.11 on VC; source column B2MICG_Z in mg/dL thresholded at 0.35).
• Notes: In multiple myeloma B2M is part of the International Staging System (ISS); in routine PK analyses it is interpreted simultaneously as a renal-function and disease-burden marker. Document the interpretation per-model via covariateData[[B2M]]$notes.

CYSC (canonical for serum cystatin C)

• Description: Serum cystatin C concentration. Low-molecular-weight (~13 kDa) protease-inhibitor protein produced at a near-constant rate by all nucleated cells; freely filtered at the glomerulus and almost entirely reabsorbed and catabolized in the proximal tubule, so serum cystatin C reflects glomerular filtration rate (GFR) more directly than serum creatinine. Used as a renal-function covariate on the clearance of renally-eliminated drugs, especially in populations where creatinine-based estimates of GFR are unreliable (low muscle mass, normal-creatinine concentrations masking impaired GFR, ICU patients).
• Units: mg/L
• Type: continuous
• Scope: general
• Reference category: n/a – used with either power scaling (CYSC / ref)^exponent (Chung 2013) or centred-linear scaling on the reciprocal 1 + e * (1/CYSC - ref_inv) (Viberg 2006). Reference values observed: 0.91 mg/L (Chung 2013 vancomycin Korean adults with SCr <= 1.2 mg/dL; cohort median); 1.32 mg/L equivalent to 1/CYSC = 0.758 (mg/L)^-1 (Viberg 2006 cefuroxime adult patients with broad renal-function range; population-typical).
• Source aliases:
  • Cystatin C / cystatin – Chung 2013 paper narrative and Table 2 footnote.
  • CysC – Viberg 2006 paper narrative and Table 4 footnote.
  • CYSC – used in Wang_2019a_tacrolimus.R (Wang 2019 Methods ‘Patients and data collection’ and Table I; 0.85 +/- 0.25 mg/L, median 0.8, range 0.4-2.3 mg/L in Chinese children with refractory nephrotic syndrome). This is the first registered use of cystatin C as a disease-progression marker rather than as a renal-elimination covariate: tacrolimus is not appreciably renally cleared, and Wang 2019 interprets the negative coefficient on apparent oral clearance as nephrotic-syndrome severity rather than as reduced glomerular filtration. Serum creatinine was screened in the same model and was NOT retained.
• Example models: Chung_2013_vancomycin.R (mg/L, reference 0.91; power exponent -0.780 on CL: CL_pop * (CYSC / 0.91)^-0.780), Viberg_2006_cefuroxime.R (mg/L; centred-linear effect on 1/CYSC with coefficient 1.43 per (mg/L)^-1 and reference 0.758 (mg/L)^-1 on CL: CL_pop * (1 + 1.43 * (1/CYSC - 0.758))), Ding_2026_vancomycin.R (mg/L, reference 1.5; the only registered model in which cystatin C enters both disposition parameters, as two separately-estimated power exponents: -0.650 on CL and -0.294 on V. Ding 2026 attributes the volume effect to reduced excretion at higher CYSC producing a relative fall in apparent V, and notes it is the first report of the association), Huang_2025_colistinSulfate.R (mg/L, reference 2.31 – the cohort MEDIAN, explicitly distinguished in the Results text from the Table 1 mean of 2.87; power exponent -0.626 on the CL of colistin sulfate, observed range 1.05-5.11 in critically ill adults. This is the founding continuous-renal-replacement-therapy example, and the reason it matters is mechanistic rather than incidental: the cohort was entirely on CVVHDF, and creatinine clearance, serum creatinine, blood urea nitrogen and uric acid were ALL screened on CL and none was retained, while cystatin C was. The authors attribute this to molecular size – CVVHDF combines convection and diffusion and therefore clears creatinine (113 Da) efficiently, so serum creatinine and any Cockcroft-Gault estimate derived from it stop tracking residual renal function once the filter is running, whereas cystatin C (13.25 kDa) is not appreciably removed (cited clearance 17 mL/min, under 30% of its production rate) and so remains interpretable on dialysis. A model needing a renal-function covariate in a CRRT population should prefer this column over CRCL for that reason).
• Notes: Cystatin C is freely filtered at the glomerulus and is not secreted by the renal tubule (unlike creatinine), so it is less sensitive to muscle mass, body composition, and tubular-secretion blockers. Reference ranges 0.57-0.97 mg/L for adult females and 0.65-1.10 mg/L for adult males (Chung 2013 Methods; Roche Cobas 6000 particle-enhanced immunoturbidimetric assay). Distinct from CREAT (serum creatinine) – the two are commonly reported alongside each other and can enter the same model as separate covariates (as in Chung 2013, where CYSC explains 62% of CL variability vs SCr 13%). The functional form (power on CYSC vs centred-linear on 1/CYSC) is paper-specific and lives in the model file; the canonical column is the underlying biomarker concentration in mg/L.

NTPROBNP (canonical for serum N-terminal pro-B-type natriuretic peptide)

• Description: Serum (or plasma) concentration of N-terminal pro-B-type natriuretic peptide, the biologically inactive cleavage fragment co-secreted with BNP by cardiac ventricular myocytes in response to increased ventricular wall tension. The standard laboratory index of cardiac stress / ventricular dysfunction, and the covariate through which impaired cardiac function – and the reduced cardiac output and organ perfusion that accompany it – enters a pharmacokinetic model. Baseline or time-varying.
• Units: ng/mL. Read the source paper’s unit carefully: the near-universal clinical reporting unit is pg/mL (equivalently ng/L), and 1 ng/mL = 1000 pg/mL, so a paper reporting a cohort median of “1.095” in ng/mL and one reporting “1095” in pg/mL describe the same cohort. Document the unit actually used via covariateData[[NTPROBNP]]$units and convert the centring constant with it.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (NTPROBNP / ref)^exponent. Reference values observed: 10 ng/mL (Ding 2026 cardiac-surgery cohort; a round value well above the cohort median of 1.095 ng/mL, chosen in the region of the elevated-cardiac-stress scenarios the paper simulates rather than at the cohort centre).
• Source aliases:
  • NT_proBNP – the NONMEM $INPUT column name in Ding_2026_vancomycin.R; abbreviated BNP inside that stream’s $PK block (BNP=NT_proBNP).
  • NT-proBNP / NTproBNP – the common narrative and table spellings.
• Example models: Ding_2026_vancomycin.R (founding example; ng/mL, reference 10; power exponent -0.0823 on vancomycin CL, so clearance falls as cardiac stress rises. This is the paper’s headline finding: in cardiac-surgery patients, cardiac function measured by NT-proBNP is a covariate on vancomycin clearance independently of the two renal-function covariates CREAT and CYSC already in the model, and eliminating it costs 41.35 OFV points on 1 df).
• Notes: Distinct from LVEF (left ventricular ejection fraction), the imaging-based measure of systolic function: the two are related but not interchangeable, and Ding_2026_vancomycin.R is a worked case where NT-proBNP was retained while LVEF was rejected in the same covariate search, both as a continuous covariate (dOFV 1.074) and dichotomised at 0.4 and 0.3. An assay upper limit of detection is common and matters for the power form: Ding 2026’s assay saturated at 35 ng/mL and results reported as “> 35” were set to 35 before modelling, which truncates the covariate’s upper tail. A cut-off of 0.3 ng/mL is conventionally used to identify dyspnoea caused by heart failure. Because the power model is undefined at zero and NT-proBNP spans orders of magnitude within a single cohort (0.117-19.12 ng/mL at the 2.5-97.5 percentiles in Ding 2026), a simulated cohort must draw this covariate on the log scale.

HEPIMP (canonical for hepatic-impairment indicator (NCI ODWG classification))

• Description: Baseline hepatic-impairment indicator per the National Cancer Institute Organ Dysfunction Working Group (NCI ODWG) classification: 1 = mild or worse hepatic impairment (group >= 2 = mild, moderate, or severe), 0 = normal hepatic function (group 1).
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal hepatic function, NCI ODWG group 1).
• Source aliases:
  • BHPTGRPN (categorical: 1 = normal, 2 = mild, 3 = moderate, 4 = severe; 9999 = missing) – used in Lu_2019_polatuzumab.R. Decompose: HEPIMP = as.integer(BHPTGRPN > 1.5 & BHPTGRPN != 9999).
  • HEP_IMP – retired canonical name; replaced by HEPIMP for consistency with the HEPIMP_MILD family.
• Example models: Lu_2019_polatuzumab.R (multiplicative effect on FRAC_NS = 1.19, applied as 1.19^HEPIMP).
• Notes: NCI ODWG classification (Ramalingam SS et al., J Clin Oncol 2010;28:4507) groups subjects by total bilirubin and AST: group 1 = normal, group 2 = mild (TBILI <= ULN and AST > ULN, or TBILI > 1-1.5 x ULN), group 3 = moderate (TBILI > 1.5-3 x ULN), group 4 = severe (TBILI > 3 x ULN). Source papers typically pool groups 2-4 versus group 1 for a binary indicator because the impaired-liver subgroups are individually small. If a future model needs finer resolution (separate effects for mild vs moderate-or-worse), add a parallel HEPIMP_MOD canonical rather than overloading this one.

NASF (canonical for nonalcoholic steatohepatitis severity score combining NAS and fibrosis staging)

• Description: Integer composite severity score for nonalcoholic steatohepatitis (NASH), summing the NAFLD activity score (NAS: steatosis 0-3 + hepatocyte ballooning 0-2 + lobular inflammation 0-3, total 0-8) and the fibrosis staging score (0 = absent, 1 = perisinusoidal / pericellular, 2 = periportal, 3 = bridging, 4 = cirrhosis). Total range 0-12; healthy subjects without biopsy-confirmed NASH are assigned NASF = 0 by convention. Scores below 5 reflect a benign form of NAFLD; scores >= 5 reflect biopsy-confirmed NASH.
• Units: (count, 0-12)
• Type: count
• Scope: specific
• Reference category: n/a – used with a linear effect on log(NASF / 4) for NASF >= 4 and zero contribution for NASF < 4 (so NASF = 4 and any NASF < 4 reduce to the typical-value reference). The cutoff of 4 distinguishes patients with a benign form of NAFLD from those with biopsy-confirmed NASH (Pierre 2017 Methods ‘Covariate analysis’ and references 31 and 34).
• Source aliases:
  • NASF – used in Pierre_2017_morphine.R (Pierre 2017 Methods ‘Covariate analysis’).
• Example models: Pierre_2017_morphine.R (linear effect on log(NASF / 4) for NASF >= 4 with coefficient -0.628 on M3G clearance: CL_M3G_i = CL_M3G_pop * (1 + e_nasf_cl_m3g * log(NASF / 4)) for NASF >= 4 and CL_M3G_i = CL_M3G_pop for NASF < 4; higher NASF reduces M3G clearance via reduced biliary excretion and increased basolateral efflux of M3G into systemic circulation).
• Notes: The NAFLD activity score (NAS) component is the histology score described by Bondini 2007 / Kleiner 2005 and references therein; the fibrosis staging is the Brunt / NASH-CRN system. The combined NASF score is the noninvasive staging proposed by Santiago-Rolon 2015 (Proc R Health Sci J 34:189-194) and used by Angulo 2007 as the NAFLD Fibrosis Score cutoff. Scope: specific because the precise cutoff (NASF >= 4) and the linear-on-log functional form are Pierre 2017’s modeling choice; future papers may model NASF or its components differently. Distinct from HEPIMP* (NCI ODWG oncology-trial hepatic-impairment categories) and from continuous liver enzymes (ALT, AST, ALP) – NASF is a biopsy-derived disease-severity ordinal specific to NAFLD / NASH.

HEPIMP_SEV (canonical for severe hepatic impairment indicator)

• Description: 1 = severe hepatic impairment, 0 = normal hepatic function or less-than-severe category. The classification scheme that defines “severe” is paper-specific and must be documented in per-model covariateData[[HEPIMP_SEV]]$notes. Two schemes are commonly encountered:
  • NCI ODWG group 4: total bilirubin > 3 x ULN with any AST (Ramalingam SS et al., J Clin Oncol 2010;28:4507).
  • Child-Pugh Class C: composite score 10-15 across bilirubin, albumin, INR, ascites, and encephalopathy.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (any non-severe category: normal, mild, or moderate; the model typically uses HEPIMP_SEV alongside other severity-specific indicators that partition the non-severe pool further).
• Source aliases:
  • Child-Pugh Class C – used in vanderWalt_2013_dapagliflozin.R (covariate effects on CLP_M15 and V2M; the paper dichotomizes severe hepatic impairment per the Child-Pugh classification).
• Example models: vanderWalt_2013_dapagliflozin.R (Child-Pugh Class C; multiplicative fractional effects -0.422 on the dapagliflozin -> D3OG metabolic clearance and +1.33 on the D3OG central volume of distribution; paper text “With severe HI (Child-Pugh Class C), CLP M15 decreased by 41% and V2M increased by 134%”), Comisar_2025_rimegepant.R (multiplicative fractional effect on CL/F = -0.423 per Comisar 2025 Table 3, a 42.3% clearance decrease giving a 1.7-fold rise in simulated steady-state AUCtau, which supports the label recommendation to avoid rimegepant in severe hepatic impairment; paired with HEPIMP_MOD (-0.229) as two mutually exclusive indicators sharing a normal-OR-mild reference group; classification scheme not stated in the paper).
• Notes: Use this column when a model dichotomizes severe hepatic impairment as a separate indicator from milder categories. The classification scheme (NCI ODWG vs Child-Pugh vs other) is paper-specific and must be documented per-model. For composite “moderate-or-severe” pooled indicators, use the parallel HEPIMP_MODSEV canonical rather than overloading this entry. Companion to HEPIMP_MILD (mild only) and HEPIMP_MODSEV (moderate + severe pooled); the register anticipates each severity level as its own canonical when the source paper tests them as separate covariates.

HEPIMP_MODSEV (canonical for composite moderate-or-severe hepatic impairment indicator)

• Description: 1 = moderate or severe hepatic impairment, 0 = normal hepatic function or mild impairment. Composite indicator used by source papers that pool the moderate and severe subgroups because the severe subgroup alone is too small to support a separate covariate-effect estimate. Distinct from HEPIMP_MOD_OR_MISSING (which pools moderate cases with missing-data cases, not with severe cases). The classification scheme that defines the cut points is paper-specific and must be documented in per-model covariateData[[HEPIMP_MODSEV]]$notes. Two schemes are commonly encountered:
  • NCI ODWG groups 3-4 pooled: total bilirubin > 1.5 x ULN with any AST.
  • Child-Pugh Class B or C pooled: composite score >= 7.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal hepatic function or mild impairment; the indicator is mutually exclusive with HEPIMP_MILD, so all-zero on both indicators corresponds to the normal-function reference and HEPIMP_MILD=1 with HEPIMP_MODSEV=0 corresponds to mild-only).
• Source aliases:
  • Child-Pugh Class B,C – used in vanderWalt_2013_dapagliflozin.R (covariate effects on V3P and CLM; the paper dichotomizes moderate-or-severe hepatic impairment per the Child-Pugh classification).
• Example models: vanderWalt_2013_dapagliflozin.R (Child-Pugh Class B or C; multiplicative fractional effects -0.600 on the dapagliflozin peripheral volume of distribution V3P and -0.293 on the D3OG renal clearance CLM; paper text “Moderate or severe HI (Child-Pugh Class B or C) decreased CLM and the peripheral volume of distribution of dapagliflozin (V3P) by 29 and 60%, respectively”), Nguyen_2019_cabozantinib_liver_dysfunction.R (NCI ODWG groups 3-4 pooled; direct multiplicative factor 0.978^HEPIMP_MODSEV on CL/F and 1.06^HEPIMP_MODSEV on Vc/F, paired with HEPIMP_MILD. An extreme example of why the pooling exists and of its limits: only 15 moderate and 1 severe subject out of 2023, so both effects are indistinguishable from unity and the Vc/F 90% CI spans 0.658 to 1.71. The paper’s Conclusions state that the limited data “preclude providing any dosing recommendations for these subpopulations”).
• Notes: Use this column when a model pools moderate-and-severe hepatic impairment under a single coefficient (typically because the severe subgroup alone is too small to estimate as its own effect). The classification scheme (NCI ODWG vs Child-Pugh vs other) is paper-specific and must be documented per-model. Companion to HEPIMP_MILD (mild only) and HEPIMP_SEV (severe only). Distinct from HEPIMP_MOD_OR_MISSING (which pools moderate cases with subjects whose hepatic-function data are missing/unknown, not with severe cases). The composite mod-or-sev pooling is a different load-bearing convention than the mod-or-missing pooling, so the two canonicals must remain separate.

LIVER_RESECT_MAJOR (canonical for major hepatic-resection indicator)

• Description: 1 = major hepatic resection (three or more liver segments resected), 0 = minor hepatic resection (two liver segments resected). Time-fixed per subject, describing the surgical exposure rather than a chronic hepatic-impairment state. Distinct from the HEPIMP_* family (which classifies chronic liver disease by NCI ODWG / Child-Pugh criteria) because major hepatic resection produces a transient reduction in metabolic capacity that recovers within 2-3 weeks in patients with normal underlying liver function (Ollier 2015 Discussion citing Fortner et al., Ann Surg 1987).
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (minor resection; two liver segments)
• Source aliases:
  • NSH – used in Ollier_2015_ropivacaine.R (Ollier 2015 paper notation, dichotomised: NSH = 0 for 2 segments, NSH = 1 for 3, 4 or 5 segments per Methods ‘Covariate model’).
• Example models: Ollier_2015_ropivacaine.R (multiplicative exponential effect on free-ropivacaine clearance: Cl = exp(lcl + etalcl) * exp(beta_NSH * LIVER_RESECT_MAJOR) with beta_NSH = log(620/1310) = -0.7480; Cl drops from 1310 L/h at LIVER_RESECT_MAJOR = 0 to 620 L/h at LIVER_RESECT_MAJOR = 1, a 53% reduction).
• Notes: Specific scope pending a second liver-resection-popPK model to ratify the name. The dichotomisation at >= 3 segments reflects the Ollier 2015 analytical choice; a future model that parameterises resection extent as a continuous count of segments should introduce a separate canonical rather than repurpose LIVER_RESECT_MAJOR. Do not confuse with the HEPIMP_* family: HEPIMP_MILD / HEPIMP_MOD / HEPIMP_SEV etc. describe chronic hepatic impairment classified from labs (bilirubin, AST) or Child-Pugh scores, whereas LIVER_RESECT_MAJOR is a surgical-exposure covariate that acts on drug clearance for the transient postoperative window.

LSM (canonical for liver stiffness measured by transient elastography)

• Description: Baseline liver stiffness measurement (LSM) obtained by vibration-controlled transient elastography (FibroScan-type), the standard non-invasive surrogate for hepatic fibrosis stage in chronic liver disease. Time-fixed per subject.
• Units: kPa
• Type: continuous
• Scope: general
• Reference category: n/a – typically enters as a power term normalised to a cohort reference, (LSM / ref)^exponent. Reference value observed: 8.1 kPa (Ooi 2026 Supplementary Datafile S6 and the Figure S4 forest plots).
• Source aliases:
  • LIVSTBL (baseline liver stiffness) – the NONMEM $INPUT column in Ooi 2026 Supplementary Datafiles S5 and S6; same quantity in kPa, no value transformation.
• Example models: Ooi_2026_elafibranor_alptb.R (power exponent 0.333 on the typical baseline total bilirubin of the joint ALP / total-bilirubin indirect-response model; Ooi 2026 Table 2).
• Notes: A bare clinical-measurement abbreviation in the same group as ALB, ALT, ALP, TBILI and CRCL, and registered at general scope because liver stiffness is a standard hepatology measurement rather than a paper-defined construct. Distinct from the categorical fibrosis-stage and hepatic-impairment indicators (HEPIMP, HEPIMP_MILD, …), which classify rather than measure. When a source imputes a missing LSM by setting the covariate factor to 1 (as Ooi 2026 does), supply the reference value instead so the power term evaluates to 1.

RENALIMP (canonical for renal-impairment indicator (any degree))

• Description: 1 = renal impairment of any degree (mild, moderate, severe, or kidney failure / ESRD), 0 = normal renal function. Pooled parent of the RENALIMP_* severity family, for source papers that carry renal impairment as a single binary covariate rather than resolving the individual strata. The threshold separating “normal” from “impaired” is paper-specific and must be documented in per-model covariateData[[RENALIMP]]$notes; a common scheme (used by Zhang 2025 and mirrored in FDA / EMA renal-impairment labeling guidance) is eGFR < 90 mL/min/1.73 m^2 as impaired.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal renal function; matched healthy controls in the founding example).
• Source aliases:
  • Renal Impairment (binary, “0 representing controls and 1 indicating renal impairment”) – used in Zhang_2025_remdesivir.R (Zhang 2025 Methods 2.3 ‘Mixed effect model and covariant analysis’).
• Example models: Zhang_2025_remdesivir.R (multiplicative shift on the remdesivir-to-GS-704277 and GS-704277-to-GS-441524 metabolic conversion rate constants, from the healthy-control column to the severe-renal-impairment column of Zhang 2025 Table 1).
• Notes: Exact structural analogue of HEPIMP (the pooled “mild or worse hepatic impairment” indicator that sits alongside HEPIMP_MILD / HEPIMP_MOD / HEPIMP_SEV / HEPIMP_MODSEV), and completes the same symmetry for the renal family, whose members already state that they “follow the HEPIMP_* family precedent”. Use this column only when the source paper genuinely pools every impairment stratum under one coefficient – Zhang 2025 says so explicitly: “To maintain simplicity, varying degrees of renal impairment were not further distinguished as separate covariates.” If a future model needs finer resolution, use the existing RENALIMP_MILD / RENALIMP_MOD / RENALIMP_SEV / RENALIMP_ESRD indicators rather than overloading this one. Companion to continuous CRCL. Distinct from RRT_HEMODIAL_STATUS / RRT_HEMODIAL_ACTIVE, which record dialysis treatment rather than the underlying renal-function stratum.

RENALIMP_MOD (canonical for moderate renal impairment indicator)

• Description: 1 = moderate renal impairment, 0 = normal renal function, mild renal impairment, or any non-moderate category. Binary categorical indicator used by source papers that test renal-impairment status as a discrete covariate rather than via continuous creatinine clearance (CRCL). The classification scheme that defines “moderate” is paper-specific and must be documented in per-model covariateData[[RENALIMP_MOD]]$notes. Two schemes are commonly encountered:
  • Cockcroft-Gault CrCl 30-50 mL/min (Morris 2011 ZP-006 study eligibility criteria).
  • FDA / EMA labeling: CrCl 30-59 mL/min (US FDA Guidance for Industry: PK in Patients with Impaired Renal Function, 2020).
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal renal function or mild renal impairment; severe and end-stage categories are typically pooled with the reference or excluded from the source cohort).
• Source aliases:
  • RENAL (binary 1 = moderate / 0 = healthy or mild) – used in Morris_2011_telapristone.R (Morris 2011 Methods “Covariate Analysis”, paragraph after Equation for the proportional covariate model).
• Example models: Morris_2011_telapristone.R (proportional fractional effect on the telapristone absorption rate constant Ka; coefficient -0.744 yielding Ka = Ka_typ * (1 + e_renalimp_mod_ka * RENALIMP_MOD), i.e., a 74% decrease in Ka in moderate-renal-impaired subjects vs the reference healthy/mild-renal cohort), CarlssonPetri_2018_semaglutide.R.
• Notes: Use this column when a model dichotomizes moderate renal impairment as a separate indicator (typically because severe / ESRD subjects are not in the source cohort and the moderate group is the only renal-impairment stratum with non-trivial sample size). The classification scheme (Cockcroft-Gault vs MDRD vs CKD-EPI; mL/min vs mL/min/1.73 m^2) is paper-specific and must be documented per-model. Companion to CRCL (continuous Cockcroft-Gault clearance), RENALIMP_MILD and RENALIMP_SEV (parallel canonicals reserved for future extractions following the HEPIMP_* family precedent). The Morris 2011 Discussion notes that the effect is “not directly attributed to renal impairment (in terms of decrease glomerular filtration rate/changes in creatinine clearance)” but “attributed to some other disease precluding/resulting in renal impairment or other pathophysiologic states induced by renal impairment (e.g., delayed gastric emptying as a result of diabetes)”, so the binary indicator captures the cohort allocation rather than a mechanistic GFR effect.

RENALIMP_MILD (canonical for mild renal impairment indicator)

• Description: 1 = mild renal impairment, 0 = normal renal function or any non-mild category. Binary categorical indicator used by source papers that test renal-impairment status as a discrete covariate rather than via continuous creatinine clearance (CRCL). The classification scheme that defines “mild” is paper-specific and must be documented in per-model covariateData[[RENALIMP_MILD]]$notes. A common scheme (used by Carlsson Petri 2018 and mirrored in FDA / EMA labeling guidance for renal-impairment PK) is eGFR 60-89 mL/min/1.73 m^2 as mild versus eGFR >= 90 mL/min/1.73 m^2 as normal.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal renal function; the moderate / severe categories are typically encoded as separate RENALIMP_MOD / RENALIMP_SEV indicators alongside this one, with all three set to 0 for the normal-renal-function reference).
• Source aliases:
  • RENAL (multi-level categorical with mild as one level; decompose to binary indicators) – used in CarlssonPetri_2018_semaglutide.R (Carlsson Petri 2018 Methods ‘Renal impairment groups were categorized [by estimated glomerular filtration rate (eGFR)] as normal function (eGFR >= 90 ml/min/1.73 m^2), mild (eGFR = 60-89 ml/min/1.73 m^2), moderate (eGFR = 30-59 ml/min/1.73 m^2), or severe (eGFR < 30 ml/min/1.73 m^2) renal impairment’).
• Example models: CarlssonPetri_2018_semaglutide.R (multiplicative CL/F ratio 0.948 for mild renal impairment vs normal reference, encoded as e_renal_mild_cl^RENALIMP_MILD per Table S3).
• Notes: Sister canonical to RENALIMP_MOD and RENALIMP_SEV; the trio decomposes a four-level renal-function classification (normal / mild / moderate / severe) into three binary indicators with normal as the implicit reference. Follows the HEPIMP_* family precedent (mild / moderate / severe as parallel binary canonicals). Companion to continuous CRCL. Founded alongside the CarlssonPetri_2018_semaglutide extraction where all three renal-impairment categories were tested simultaneously.

RENALIMP_SEV (canonical for severe renal impairment indicator)

• Description: 1 = severe renal impairment, 0 = normal renal function or any non-severe category. Binary categorical indicator used by source papers that test renal-impairment status as a discrete covariate rather than via continuous creatinine clearance (CRCL). The classification scheme that defines “severe” is paper-specific and must be documented in per-model covariateData[[RENALIMP_SEV]]$notes. A common scheme (used by Carlsson Petri 2018 and mirrored in FDA / EMA labeling guidance for renal-impairment PK) is eGFR < 30 mL/min/1.73 m^2 as severe.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (normal renal function; typically paired with RENALIMP_MILD and RENALIMP_MOD so the trio of indicators is mutually exclusive and normal renal function is the shared reference).
• Source aliases:
  • RENAL (multi-level categorical with severe as one level; decompose to binary indicators) – used in CarlssonPetri_2018_semaglutide.R.
• Example models: CarlssonPetri_2018_semaglutide.R (multiplicative CL/F ratio 0.920 for severe renal impairment vs normal reference, encoded as e_renal_sev_cl^RENALIMP_SEV per Table S3; the Discussion notes the severe stratum came only from SUSTAIN 6 and might be confounded by trial effect, and that a dedicated renal-impairment clinical pharmacology trial found no clinically relevant effect).
• Notes: Sister canonical to RENALIMP_MILD and RENALIMP_MOD; the trio decomposes a four-level renal-function classification (normal / mild / moderate / severe) into three binary indicators with normal as the implicit reference. Follows the HEPIMP_* family precedent. Companion to continuous CRCL. Founded alongside the CarlssonPetri_2018_semaglutide extraction.

RENALIMP_ESRD (canonical for end-stage renal disease indicator)

• Description: 1 = end-stage renal disease (ESRD), 0 = any renal function above ESRD. Fourth and most severe member of the RENALIMP_* binary renal-function family, for source papers that carry ESRD as its own discrete covariate rather than letting continuous CRCL extrapolate into the anuric range. The threshold that defines ESRD is paper-specific and must be documented in per-model covariateData[[RENALIMP_ESRD]]$notes; a common scheme (used by Xie 2025 and mirrored in FDA / EMA renal-impairment labeling guidance) is CrCL < 15 mL/min.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (any renal function above ESRD; typically paired with RENALIMP_MILD / RENALIMP_MOD / RENALIMP_SEV so the set of indicators is mutually exclusive and normal renal function is the shared reference).
• Source aliases:
  • ESRD – used in Xie_2025_aztreonam_avibactam.R (Table S3 row ESRD on CL_AVI).
• Example models: Xie_2025_aztreonam_avibactam.R (proportional shift of -0.923 on avibactam clearance, leaving 7.7% of the reference value; aztreonam carries no ESRD term).
• Notes: Sister canonical to RENALIMP_MILD, RENALIMP_MOD and RENALIMP_SEV, completing the four-level family; general scope for the same reason those carry it, namely that the renal-impairment strata have a standard regulatory definition any future model can reuse. Two traps worth documenting per model. First, an ESRD indicator commonly replaces rather than multiplies the continuous CRCL relationship – Das 2024 prints its avibactam power arm as applying when “nCrCL < 80 mL/min and not ESRD”, and multiplying the two in Xie 2025 would give a physiologically impossible ~0.7% of reference clearance instead of 7.7%. Second, papers routinely define the ESRD stratum on raw CrCL (mL/min) while the model’s continuous renal covariate is the BSA-normalized CRCL (mL/min/1.73 m^2), so the column is generally not derivable from CRCL by a threshold; supply it from the raw value. Distinct from RRT_HEMODIAL_STATUS / RRT_HEMODIAL_ACTIVE, which record dialysis treatment rather than the underlying renal-function stratum – a model may carry both, and Xie 2025 makes them mutually exclusive because dialysis patients take a separate absolute clearance.

RENAL_ARC (canonical for augmented-renal-clearance indicator)

• Description: 1 = the subject meets the study’s definition of augmented renal clearance (ARC), a supraphysiologic glomerular filtration rate seen in younger, less comorbid critically ill and infected patients; 0 = not augmented. Counterpart at the opposite pole of the RENALIMP_* family, for source papers that carry ARC as its own discrete covariate rather than letting the continuous CRCL relationship extrapolate into the supranormal range. ARC is conventionally established from a measured clearance (a timed urine collection), not from a Cockcroft-Gault or MDRD estimate, so the column is generally not derivable from CRCL by a threshold and must be supplied from the measured value; the threshold and the assay must both be documented in per-model covariateData[[RENAL_ARC]]$notes.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (renal function not augmented). Where a source paper records the flag in only one contributing study, every subject outside that study takes the reference level, which must be stated per model.
• Source aliases:
  • ARC – data-set column name in the Li 2019 avibactam NONMEM control stream (supplementary Data S1, CLCLCR block); used in Li_2019_avibactam.R.
• Example models: Li_2019_avibactam.R (founding example; ARC defined as a measured creatinine clearance >= 140 mL/min by 8-hour urine collection and recorded only in study CXL-PK-04. The flag does two structurally distinct things: it scales the supra-80 CRCL slope on clearance by 0.992, and it EXCLUDES the subject from the sub-80 power arm, so an ARC subject with CRCL < 80 takes a renal factor of exactly 1).
• Notes: Scope specific until a second model ratifies it. Deliberately NOT folded into CRCL, and deliberately not expressed as a RENALIMP_* level: the RENALIMP_* family partitions reduced renal function into severity strata that share one reference, whereas ARC is a separate binary flag that coexists with the continuous renal covariate and modifies how that covariate acts rather than replacing it. Distinct from a high CRCL value. The two are not interchangeable even when a paper’s ARC threshold is expressed in mL/min, because ARC is established by a measured clearance while CRCL normally holds an estimate; a subject can have an estimated CRCL above the ARC threshold without being flagged, and a flagged subject’s estimated CRCL can fall below it. Where a source paper’s ARC flag interacts with the CRCL relationship (as in the founding model, where it both rescales one arm and gates the other), encode both behaviours – a model that reproduces only the rescaling will silently mis-predict ARC subjects at low estimated clearance.

CPK (canonical for serum creatine phosphokinase / creatine kinase)

• Description: Serum creatine phosphokinase (also called creatine kinase, CK) activity (baseline or time-varying). Skeletal-muscle / cardiac-muscle injury and turnover marker; in macrophage-targeted PK/PD analyses (axatilimab, anti-CSF-1R) it is interpreted as a Kupffer-cell / tissue-macrophage clearance surrogate because Kupffer cells participate in the elimination of circulating muscle-derived enzymes.
• Units: U/L (SI; canonical as of 2026-06-19; IU/L is used interchangeably). Document per-model via covariateData[[CPK]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (CPK / ref)^exponent. Reference values observed: 63 U/L (Yang 2024 axatilimab; pooled-cohort median).
• Source aliases:
  • BLCPK (baseline CPK) – informal usage in Yang 2024.
• Example models: Yang_2024_axatilimab.R (baseline-only covariate on baseline NCMC concentration BL_NCMC with power exponent 0.376; reference 63 U/L).
• Notes: Muscle-origin enzyme distinct from AST / ALT (hepatic) and LDH (general tissue turnover). Yang 2024 uses CPK alongside AST and LDH as tracked safety biomarkers. Per-model covariateData[[CPK]]$notes should document baseline-vs-time-varying status and the clinical interpretation in the source population (skeletal-muscle injury, macrophage-clearance surrogate, or both). Distinct from any model state variable representing CPK time-course dynamics – covariate column is the pre-dose laboratory observation.

MYO (canonical for plasma myoglobin)

• Description: Plasma or serum myoglobin concentration, an oxygen-binding haem protein of skeletal and cardiac muscle used clinically as a muscle-damage / muscle-mass marker. Baseline or time-varying; document which via per-model covariateData[[MYO]]$notes.
• Units: ng/mL (equivalently ug/L; SI)
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (MYO / ref)^exponent. Reference values observed: 25 ng/mL (Zhao 2025 paracetamol; pooled-cohort mean).
• Source aliases:
  • myoglobin – spelled-out form used in Zhao_2025_paracetamol.R (Table 1 baseline characteristics, Table 2 covariate rows).
• Example models: Zhao_2025_paracetamol.R (power exponent -1.10 on the paracetamol leftover clearance CL_p/F, normalised to the pooled mean of 25 ng/mL; baseline medians 34 ng/mL in healthy controls vs 17 ng/mL in spinal-muscular-atrophy patients, with the paper quoting a normal range of 15-75 ng/mL, and the effect carrying dOFV = 18.5 as the strongest covariate in that analysis).
• Notes: Sibling of the existing muscle / tissue-turnover markers CPK and LDH, and distinct from both – myoglobin is a protein mass concentration in ng/mL while CPK and LDH are enzyme activities in U/L, and Zhao 2025 screens all three independently (only myoglobin is retained). Scope: general because the concept is assay-standard and paper-independent, matching the other routine plasma clinical-chemistry analytes ALB, CRP, CREAT, HGB, ALT, AST, TBILI, ALP, LDH, and CPK. In muscle-wasting populations myoglobin is low because muscle mass is low, so a negative exponent on a clearance term should be read as a muscle-mass surrogate rather than as muscle injury; the direction of the clinical interpretation is paper-specific and belongs in per-model notes. Distinct from any model state variable representing a myoglobin time course – the covariate column is the laboratory observation.

BFR (canonical for blood flow rate through the extracorporeal circuit during dialysis)

• Description: Instantaneous blood flow rate through the extracorporeal circuit during an active dialysis session. Time-varying within subject; meaningful only when RRT_HEMODIAL_ACTIVE = 1 – in the interdialytic period the value is sentinel and the Michaels-equation term is gated off by RRT_HEMODIAL_ACTIVE.
• Units: mL/min
• Type: continuous
• Scope: general
• Reference category: n/a – enters the Michaels equation together with DFR and a hemodialyzer mass-transfer-area coefficient. Values investigated in the ratification source were 200, 300, and 400 mL/min (Liesenfeld 2013 Methods, Study Design; Table 1). A second registered role is as the plasma-flow denominator of a post-filter concentration observable, C_Post = C_Pre * (1 - CL_CRRT / (BFR * BPR)), where the blood-to-plasma ratio BPR converts the charted whole-blood pump flow to the plasma flow actually presented to the membrane.
• Source aliases:
  • phi_Blood – Butragueno-Laiseca 2022 supplementary material eq. 4 and the sibling Butragueno-Laiseca 2025 teicoplanin analysis; same orientation, no transformation.
• Example models: Liesenfeld_2013_dabigatran.R (Michaels-equation role), ButraguenoLaiseca_2022_piperacillin.R (post-filter plasma-flow denominator; paediatric CVVHDF, supplementary Table 3 blood flow 70 (SD 31) mL/min. Note that this paper’s main-text Results misprint the same quantity weight-normalised as “5 (SD 2) ml/kg/h” – the per-minute reading is the one that reproduces the paper’s own published extraction ratio, see the model file and vignette Errata).
• Notes: Pairs with RRT_HEMODIAL_ACTIVE (binary on/off gate) and DFR (dialysate flow rate).

DFR (canonical for dialysate flow rate through the extracorporeal circuit during dialysis)

• Description: Instantaneous dialysate flow rate through the extracorporeal circuit during an active dialysis session. Time-varying within subject; meaningful only when RRT_HEMODIAL_ACTIVE = 1.
• Units: mL/min
• Type: continuous
• Scope: general
• Reference category: n/a – enters the Michaels equation together with BFR. The ratification source fixed DFR at 700 mL/min throughout (Liesenfeld 2013 Methods, Study Design) and additionally simulated 500 mL/min (Methods, Simulations).
• Source aliases: none known.
• Example models: Liesenfeld_2013_dabigatran.R.
• Notes: Pairs with RRT_HEMODIAL_ACTIVE and BFR.

VASCACC_AVF1N (canonical for single-needle arteriovenous-fistula vascular-access indicator)

• Description: 1 = the haemodialysis session was performed through an arteriovenous fistula cannulated in a SINGLE-needle setup (one needle alternately draws and returns blood, so a fraction of already-dialysed blood is recirculated); 0 = any other access type. Per-subject (or per-session) indicator describing how the patient is cannulated for extracorporeal therapy, as distinct from the circuit-setting covariates BFR / DFR / Q_CVVH (flow rates) and the therapy-presence flags RRT_HEMODIAL_STATUS / RRT_HEMODIAL_ACTIVE. Access type matters pharmacokinetically because it changes the effective solute presentation to the dialyser: a single-needle fistula recirculates blood, so the nominal pump blood flow rate overstates the flow of undialysed blood reaching the membrane.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (tunnelled dialysis catheter, TDC), shared with the sister canonical VASCACC_AVF2N; the pair decomposes the three-level access classification (TDC / AVF single-needle / AVF double-needle) into two binary indicators with TDC as the implicit reference (both indicators 0).
• Source aliases:
  • CVC0SN1DN2 (three-level integer selector, 0 = TDC, 1 = single-needle AVF, 2 = double-needle AVF; decompose to binary indicators) – used in Kong_2025_piperacillin_tazobactam.R (ESM NONMEM $PK: LGT_TAZ = THETA(18); IF(CVC0SN1DN2.EQ.1) LGT_TAZ = THETA(19); IF(CVC0SN1DN2.EQ.2) LGT_TAZ = THETA(20)).
  • AVF 1N – Kong 2025 Table 1 / Table 3 row label (“AVF 1N single-needle arteriovenous fistula”).
• Example models: Kong_2025_piperacillin_tazobactam.R (selects the tazobactam dialyser extraction ratio: 73.9% for single-needle AVF versus the 80.1% TDC reference and 73.5% for double-needle AVF, Table 3; retained in backward elimination at p < 0.001, dOFV -22.88).
• Notes: Sister canonical to VASCACC_AVF2N. Decomposed-binary-indicator encoding follows the register’s settled convention for multi-level categoricals (RENALIMP_MILD / RENALIMP_MOD / RENALIMP_SEV, the parallel HEPIMP_* family, and RACE_<GROUP>); an integer selector column was explicitly rejected because it would imply an ordering between access types that does not exist. Kong 2025 additionally applied a time-averaged blood flow rate for single-needle sessions when computing BFR (Methods 2.4), so a model that carries both VASCACC_AVF1N and BFR should document in covariateData notes whether the recirculation correction has already been folded into the supplied BFR values. General scope because vascular access type is a universally recorded haemodialysis patient characteristic and is a plausible determinant of effective dialyser clearance in any ESKD antibiotic popPK.

VASCACC_AVF2N (canonical for double-needle arteriovenous-fistula vascular-access indicator)

• Description: 1 = the haemodialysis session was performed through an arteriovenous fistula cannulated in a DOUBLE-needle setup (separate arterial-draw and venous-return needles, so there is no single-needle recirculation); 0 = any other access type. Per-subject (or per-session) indicator describing how the patient is cannulated for extracorporeal therapy; see the sister entry VASCACC_AVF1N for the family rationale and the distinction from the circuit-flow and therapy-presence covariates.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (tunnelled dialysis catheter, TDC), shared with VASCACC_AVF1N; both indicators 0 selects the TDC reference level.
• Source aliases:
  • CVC0SN1DN2 (three-level integer selector, level 2; decompose to binary indicators) – used in Kong_2025_piperacillin_tazobactam.R.
  • AVF 2N – Kong 2025 Table 1 / Table 3 row label (“AVF 2N double-needle arteriovenous fistula”).
• Example models: Kong_2025_piperacillin_tazobactam.R (selects the tazobactam dialyser extraction ratio: 73.5% for double-needle AVF versus the 80.1% TDC reference, Table 3).
• Notes: Sister canonical to VASCACC_AVF1N; the two are mutually exclusive (a subject with both set to 1 is invalid). Kong 2025 found the access-type effect statistically significant but clinically limited (Discussion: “the difference in extraction ratio between the vascular access types were limited and therefore not expected to be clinically relevant”), and the analogous access-type effect on the PIPERACILLIN extraction ratio did NOT survive backward elimination (dOFV +11.26), so piperacillin keeps a single 64.0% extraction ratio.

ECMO_PUMP_SPEED (canonical for extracorporeal-membrane-oxygenation centrifugal-pump rotational speed)

• Description: Rotational speed of the extracorporeal-membrane-oxygenation (ECMO) centrifugal blood pump during VA-ECMO or VV-ECMO support. Continuous covariate; treated as time-fixed per subject in Yang 2017 (the per-subject pump speed reported in the source data was the prevailing speed during the PK sampling window; pump-speed adjustments during sampling were not modelled as time-varying). For future models that resolve session-level changes in pump speed, the covariate is naturally time-varying and the per-model covariateData[[ECMO_PUMP_SPEED]]$notes should document the time resolution.
• Units: RPM (revolutions per minute)
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a power-centered effect (ECMO_PUMP_SPEED / ref)^exponent. The reference value is paper-specific (median pump speed in the source cohort): Yang 2017 uses 2350 RPM (cohort median; Table 1 / Results: “median ECMO pump speeds of 2350 RPM”).
• Source aliases: none known.
• Example models: Yang_2017_remifentanil.R (power effect on remifentanil CL: (ECMO_PUMP_SPEED / 2350)^2.04; higher pump speed associated with higher CL, hypothesised mechanism is increased spontaneous drug degradation at high centrifugal-pump shear).
• Notes: Distinct from blood flow rate (BFR, mL/min) and from dialysate flow rate (DFR, mL/min) which characterise renal-replacement-therapy circuits. ECMO circuits use a centrifugal pump whose rotational speed sets the cardiac-output augmentation; the resulting blood flow rate (LPM) is a separate measured quantity that depends on circuit resistance and patient hemodynamics. Yang 2017 tested both ECMO pump speed (RPM) and ECMO flow rate (LPM) and only pump speed was retained as a significant covariate (ECMO flow rate was not significantly associated with PK parameters).

Q_CVVH (canonical for continuous venovenous hemofiltration circuit flow rate during ECMO)

• Description: Flow rate through the continuous venovenous hemofiltration (CVVH) filter when it is placed parallel to the ECMO circuit. The filter contributes solute removal that adds to the drug’s apparent CL whenever CVVH is active; the column is zero in subjects whose ECMO circuit has no CVVH filter or whose CVVH is currently off. Continuous covariate; naturally time-varying although the source paper treats the per-subject median as time-fixed (the median Q_CVVH per individual did not vary much during sampling).
• Units: mL/min
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a power-centered effect (Q_CVVH / ref)^exponent with the source paper additionally gating the effect to 1 when Q_CVVH = 0 (no CVVH active). The reference value is paper-specific (cohort median during CVVH operation): Ahsman 2010 uses 193 mL/min (cohort median; Table 1).
• Source aliases:
  • QCVVH – used in Ahsman_2010_cefotaxime.R (paper notation, subscript dropped).
• Example models: Ahsman_2010_cefotaxime.R (power effect on DACT metabolite CL: (Q_CVVH / 193)^0.72 when CVVH active, 1 otherwise).
• Notes: Distinct from ECMO_PUMP_SPEED (the centrifugal-pump rotational speed setting cardiac-output augmentation) and from BFR / DFR (blood-flow / dialysate-flow rates characterising standalone hemodialysis circuits): Q_CVVH is the slow-clearance flow through a CVVH filter integrated INTO an ECMO circuit. Distinct from RRT_CRRT_STATUS (binary on / off indicator for continuous RRT) which captures presence-of-therapy rather than its rate. Future ECMO + CVVH popPK extractions that quantify CVVH flow should reuse this canonical and extend the example list.

Q_ECMO (canonical for extracorporeal-membrane-oxygenation circuit blood flow rate)

• Description: Blood flow rate delivered through the extracorporeal-membrane-oxygenation (ECMO) circuit – the volume of blood per unit time actually passing through the pump head and oxygenator, as read from the circuit flow probe. This is the delivered flow, a measured quantity distinct from the pump-speed setting that produces it: for a given centrifugal-pump speed the resulting flow depends on circuit resistance, cannula size, preload, and patient hemodynamics. Continuous covariate and naturally time-varying, because flow is deliberately titrated during ECMO support and stepped down during weaning; a source paper that reports only a per-subject representative flow during the PK sampling window is treating it as time-fixed, and each model must say which applies in covariateData[[Q_ECMO]]$notes.
• Units: L/min
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a median-centered effect. Kim 2026 uses the median-normalized proportional form 1 + theta * (Q_ECMO - ref) / ref on 1-OH-midazolam clearance with ref = 2.7 L/min. The reference value is paper-specific (cohort median flow); note that Kim 2026 does not report its own median, so the 2.7 L/min there is back-solved from the paper’s Figure 1A rather than read from a table (see that model’s covariateData notes and vignette Errata).
• Source aliases:
  • ECMO flow rate – Kim 2026 paper notation (main text, Table 1 theta ECMO flow rate on CL1-OH MDZ, and Supplementary Methods “Covariate model development”).
  • ECMO flow rate (LPM) – Yang 2017 notation for the same quantity (tested, not retained).
• Example models: Kim_2026_midazolam_ecmo.R (median-normalized proportional effect on 1-OH-midazolam clearance: 1 + 0.336 * (Q_ECMO - 2.7) / 2.7; higher ECMO flow associated with higher metabolite clearance, so a higher midazolam infusion rate is needed to hold the 24-h concentration), Kang_2026_ticagrelor.R (power effect on the ticagrelor apparent central volume, 0.518^(Q_ECMO / 3), nested inside the ECMO_STATUS indicator so it applies only while cannulated; the 3 L/min reference is printed in the paper’s own final-model equation rather than being a cohort median, and because 0.518 < 1 a HIGHER flow gives a SMALLER volume – the opposite sign to Kim 2026’s clearance effect. The first register example where the flow term is gated by the ECMO indicator; the paper’s main text prints it ungated, which is algebraically identical only when off-ECMO flow is carried as 0).
• Notes: Named in the Q_ circuit-flow family (alongside Q_CVVH, Q_TOTAL_LPM, QBL / QEFF) rather than the device-prefixed ECMO_ family (ECMO_PUMP_SPEED, T_ECMO, T_POST_ECMO): Q_CVVH is already registered as the blood flow of a CVVH extracorporeal circuit, and ECMO circuit blood flow is that same physical quantity for a different circuit, so Q_ECMO is its exact sibling. The ECMO_ prefix groups by device, which is a different axis; a flow rate belongs with flow rates. Explicitly distinct from ECMO_PUMP_SPEED (RPM – the pump setting, whose own Notes already flag delivered flow as “a separate measured quantity”), from Q_CVVH (mL/min through a CVVH filter placed parallel to an ECMO circuit), and from BFR / DFR (mL/min through a standalone dialysis circuit). Both Q_ECMO and ECMO_PUMP_SPEED can appear in the same dataset and the same model; Yang 2017 recorded both and retained only pump speed, whereas Kim 2026 recorded both and retained only flow rate. Note the unit difference from the dialysis-circuit flows: ECMO flows are conventionally reported in L/min, dialysis flows in mL/min.

T_ABT (canonical for time elapsed on hormonal add-back therapy)

• Description: Time since the start of hormonal add-back therapy (ABT) – an oestrogen plus progestogen, or a progestogen alone, co-administered with a GnRH agonist or antagonist to partially replenish circulating oestrogen and so prevent bone-mineral-density loss and vasomotor symptoms during hypoestrogenaemia. 0 for subjects never receiving add-back therapy, which zeroes any term built on it.
• Units: weeks. Document per-model via covariateData[[T_ABT]]$units; days is the other common reporting unit.
• Type: continuous
• Scope: specific
• Reference category: 0 (no add-back therapy). In Pohl 2022 this is the entire EDELWEISS phase 2b population and therefore the whole dose-selection analysis; add-back therapy (1 mg oestradiol / 0.5 mg norethisterone acetate) was given only in the phase 1 studies 16-OBE2109-011 and 17-OBE2109-008, which used concurrent and delayed add-back arms respectively.
• Source aliases:
  • add-back therapy / ABT – Pohl 2022 Table 1 and Table 5 (“E2 increase rate on add-back therapy”).
• Example models: Pohl_2022_linzagolix_e2.R (additive oestradiol increment of 1.58 pg/mL per week on add-back therapy).
• Notes: Member of the T_<event> time-since-event family (T_NUT_SUPP, T_ECMO, T_POST_ECMO, T_POST_HEMODIAL). A time-since-event column rather than a binary flag because the published Pohl 2022 effect is a RATE (pg/mL per week), so the covariate must carry elapsed time; a bare on/off indicator cannot express it. This also handles the delayed-add-back arm of 17-OBE2109-008 correctly, where dosing of linzagolix and of add-back therapy start at different times. Specific scope because the column is meaningful only for GnRH-modulator models in endometriosis and uterine fibroids. CAUTION for anyone reusing this column from Pohl 2022: the functional form of the add-back term is not printed anywhere in that source – the Supporting Information equation omits the term entirely and only the tabulated pg/mL/week unit implies an additive linear-in-time increment.

T_ECMO (canonical for time since ECMO cannulation start (beginning of extracorporeal circulation))

• Description: Wall-clock time elapsed since the patient was cannulated onto ECMO (the beginning of extracorporeal circulation). Zero before cannulation; positive and monotonically increasing throughout the ECMO support period. Time-varying. Whether the value continues to increase after decannulation, resets at decannulation, or is censored at decannulation is paper-specific and must be documented in covariateData[[T_ECMO]]$notes; when the source model uses a saturating Hill / Emax form (Kleiber 2017) it does not matter whether T_ECMO continues past decannulation because the effect is already saturated.
• Units: hour
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a sigmoidal Emax multiplier 1 + Emax * T_ECMO^gamma / (T50_EC^gamma + T_ECMO^gamma) (Kleiber 2017 Eq 9) or a saturating half-life form (Ahsman 2010 Eq 10-style). At T_ECMO = 0 (pre-ECMO baseline) the Hill ratio evaluates to 0 and the multiplier collapses to 1.
• Source aliases:
  • tec / t_EC – used in Kleiber_2017_clonidine.R (paper notation, Eq 9 and Table 3).
• Example models: Kleiber_2017_clonidine.R (sigmoidal Emax effect on V: V x (1 + 0.55 * T_ECMO^18.5 / (51.7^18.5 + T_ECMO^18.5)); V approaches 1.55 * V_pop * (WT/70) on full ECMO and recovers to V_pop * (WT/70) before ECMO).
• Notes: Conceptually the on-ECMO mirror of T_POST_ECMO – T_ECMO measures time since cannulation, T_POST_ECMO measures time since decannulation. A model that uses BOTH typically gates each by an additional ECMO-on indicator or by the natural floor (T_ECMO = 0 before cannulation, T_POST_ECMO = 0 before-and-during ECMO). Kleiber 2017 uses only T_ECMO; Ahsman 2010 uses only T_POST_ECMO. Distinct from on-ECMO duration covariates (per-run summary of total ECMO hours) and from ECMO_PUMP_SPEED (instantaneous centrifugal-pump rotational speed). Anticipated by the existing T_POST_ECMO entry’s Notes (which referenced t_EC as “tested but not retained” in Ahsman 2010); ratified canonically alongside the Kleiber 2017 clonidine extraction where the covariate is retained.

T_POST_ECMO (canonical for time after ECMO decannulation (end of extracorporeal circulation))

• Description: Wall-clock time elapsed since the patient was decannulated from ECMO (the end of extracorporeal circulation). Zero before and during ECMO support; becomes positive after decannulation and increases linearly with time. Naturally time-varying. Used as a power-centred recovery-time covariate on PK parameters whose values change as the patient transitions off ECMO back to native circulation.
• Units: hour
• Type: continuous
• Scope: general
• Reference category: n/a – enters as a power-centered effect (T_POST_ECMO / ref)^exponent with the source paper additionally gating the effect to 1 when T_POST_ECMO = 0 (before or during ECMO). The reference value is paper-specific: Ahsman 2010 uses 100 h (Appendix).
• Source aliases:
  • tEND – used in Ahsman_2010_cefotaxime.R (paper notation, subscript dropped).
• Example models: Ahsman_2010_cefotaxime.R (power effect on parent CTX CL: (T_POST_ECMO / 100)^0.16 when post-ECMO, 1 otherwise; power effect on DACT metabolite CL: (T_POST_ECMO / 100)^0.53 when post-ECMO, 1 otherwise).
• Notes: Conceptually a time-since-event covariate similar to TPP (time postpartum) and T_ENTRY (per-subject study-entry time) but anchored to ECMO decannulation. Distinct from t_EC (time since ECMO START / beginning of extracorporeal circulation, which the source paper tested but did not retain) and from on-ECMO duration covariates. The gating-at-zero convention is structural: during ECMO the recovery-time covariate is undefined / not yet started, so the (0 / ref)^exponent expression would collapse pathologically and the implementer must short-circuit the multiplier to 1. Future ECMO popPK extractions that include a post-decannulation recovery phase should reuse this canonical and extend the example list.

T_HEMODIAL_INIT (canonical for time since the initiation of maintenance hemodialysis therapy (“dialysis vintage”))

• Description: Elapsed time since the subject was first established on maintenance intermittent hemodialysis for end-stage kidney disease – the quantity nephrology calls dialysis vintage. A per-subject baseline duration measured in months or years, NOT a within-record clock: it is fixed for the duration of a typical popPK study. Its pharmacokinetic role is as a surrogate for residual renal function, which decays progressively after dialysis is started and which cannot be measured directly in maintenance-dialysis patients because serum creatinine there is dominated by time since the last session. A drug with a residual-renal elimination route therefore shows FALLING interdialytic clearance as this covariate rises.
• Units: month. Document per-model via covariateData[[T_HEMODIAL_INIT]]$units when a paper reports years or days.
• Type: continuous
• Scope: general
• Reference category: n/a – supplied as a duration and normally entered as a centred inverse-power term against the cohort median. Duke_2024_cefazolin.R uses (59 / T_HEMODIAL_INIT)^0.28, where 59 months is that cohort’s median. Must be strictly positive whenever it enters a denominator.
• Source aliases:
  • TOH – “time on haemodialysis”, the Duke 2024 Methods and Table 1 label and the symbol used in its Table S3 Pmetrics model file. Read the definition, not the abbreviation: “time ON haemodialysis” here means months since therapy was STARTED, not minutes elapsed within the current session.
• Example models: Duke_2024_cefazolin.R (inverse-power effect on the interdialytic unbound-cefazolin clearance arm, CL = CLnHD * (59 / TOH)^0.28, exponent hard-coded in the Table S3 model file; reported inverse-power fit r^2 = 0.433; the paper’s PTA simulations sweep TOH = 6, 12, 24, 36 and 60 months and PTA rises monotonically with it).
• Notes: Third and slowest member of the intermittent-hemodialysis time family, and the one that must not be confused with the other two. RRT_HEMODIAL_ACTIVE is the seconds-to-hours gate saying whether a session is running right now; T_POST_HEMODIAL is the hours-scale interdialytic clock measuring time since the last session ENDED and resetting at every session; T_HEMODIAL_INIT is the months-to-years clock measuring time since the THERAPY began and never resets. The three are dimensionally and physiologically distinct – session gating, interdialytic fluid accumulation, and progressive loss of residual renal function respectively – so a value from one can never substitute for another. Named in the T_<event> elapsed-time family alongside T_NUT_SUPP (time since start of nutritional supplementation) and T_ABT (time since start of add-back therapy), both of which likewise measure time from the initiation of an ongoing therapy; the explicit _INIT token is what keeps it unambiguous against a “duration of the current session” reading, which a bare T_HEMODIAL would invite. Duke 2024 states this covariate had not previously been included in a pharmacokinetic model for patients requiring intermittent hemodialysis, so this entry has one founding example; a future paper reporting it should reuse this column and record its functional form in covariateData[[T_HEMODIAL_INIT]]$notes. A peritoneal-dialysis or continuous-RRT paper reporting the same quantity should register a modality-matched sibling rather than reusing this column, following the RRT_<modality>_<kind> precedent.

T_POST_HEMODIAL (canonical for time since the end of the last intermittent-hemodialysis session)

• Description: Wall-clock time elapsed since the END of the subject’s most recent intermittent-hemodialysis (IHD) session. Resets to zero at the end of each session and increases through the interdialytic interval, so it is naturally time-varying within subject. Undefined for subjects never dialysed; set to 0 for them and pair the effect with RRT_HEMODIAL_STATUS so the value is inert outside the dialysis-dependent stratum. Captures interdialytic physiology that the on/off session gate cannot – principally fluid accumulation between sessions, which expands the central volume of distribution.
• Units: hour
• Type: continuous
• Scope: general
• Reference category: n/a – enters either as a threshold indicator (T_POST_HEMODIAL > <cut>) or, if a future paper resolves it, as a continuous centred / power effect. VanWart_2025_telavancin.R uses the threshold form at 48 h.
• Source aliases: none known; Van Wart 2025 describes the covariate in prose (“PK samples were collected more than 48 hours after the last active IHD session”) rather than naming a data-set column.
• Example models: VanWart_2025_telavancin.R (48-hour threshold gating a fixed additive central-volume increase of +1.55 L in dialysis-dependent subjects: vc <- ... + e_t_post_hemodial_vc * RRT_HEMODIAL_STATUS * (T_POST_HEMODIAL > 48); Van Wart 2025 attributes the rise to “fluid depletion during IHD and accumulation between sessions”, telavancin plasma concentrations being higher immediately off IHD and falling thereafter).
• Notes: The INTERDIALYTIC time axis of the intermittent-hemodialysis covariate set: RRT_HEMODIAL_STATUS says whether the subject is dialysis-dependent at all, RRT_HEMODIAL_ACTIVE says whether a session is running right now, T_POST_HEMODIAL says how long it has been since the last one ended, and T_HEMODIAL_INIT says how many months the subject has been on the therapy altogether (dialysis vintage, a residual-renal-function surrogate). Do not substitute one for another: this entry resets at every session and is measured in hours, while T_HEMODIAL_INIT never resets and is measured in months. The RRT_HEMODIAL_STATUS entry anticipated exactly this column (“a future paper that resolves drug clearance during versus between dialysis sessions would use a time-varying form (RRT_HEMODIAL_ACTIVE) or a separate per-session covariate”). Named in the T_<event> time-since-event family alongside T_POST_ECMO (time after ECMO decannulation), T_ECMO, T_CPB and T_NUT_SUPP, which is the same shape: time measured from the end of an extracorporeal therapy. Distinct from FILT_AGE_HI, which also uses a 48-hour threshold but measures the age of the DIALYSER MEMBRANE while a therapy is running (declining dialysis efficiency from membrane fouling), not the time since a session ended (interdialytic fluid accumulation) – the two thresholds coincide numerically and mean opposite things, so do not substitute one for the other. General scope because time since the last dialysis session is a paper-independent quantity any IHD popPK model may use. A future paper that retains the effect as a continuous function rather than a threshold should reuse this column and record the functional form in per-model covariateData[[T_POST_HEMODIAL]]$notes.

FILT_AGE_HI (canonical for dialysis-filter-age above-48-hour indicator)

• Description: Binary indicator that the in-use extracorporeal renal-replacement-therapy (CVVHDF / CVVH / SLED / EDD-f / IHD) hemofilter or dialyser membrane has been in continuous use for more than 48 hours at the time of dosing / sampling. 1 = filter is older than 48 h; 0 = filter is within the first 48 h of use (the fresh-filter reference). Subject-level binary in the source data (Patel 2011 had 1 of 10 patients on a > 48 h filter at the start of CVVHDF treatment; the rest had fresh filters), but the covariate is naturally time-varying within subject as the filter ages and is replaced – record the per-model time resolution in covariateData[[FILT_AGE_HI]]$notes.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (fresh filter, less than 48 h in use). Models that apply a multiplicative reduction to the dialysis clearance arm should encode the effect as cl_dialysis * e_filt_age_hi_<arm>^FILT_AGE_HI so the FILT_AGE_HI = 0 case recovers the fresh-filter baseline.
• Source aliases: none known; Patel 2011 retains the effect via the parameter ffCL_CVVHDF (a multiplicative factor on the CVVHDF clearance arm) rather than naming a data-set column. When ratified the column was assigned the canonical name FILT_AGE_HI in the model file.
• Example models: Patel_2011_fluconazole.R (multiplicative factor e_filt_age_hi_cl_renal^FILT_AGE_HI on the CVVHDF clearance arm: e_filt_age_hi_cl_renal = 0.368, so > 48 h filters give 36.8% of fresh-filter CVVHDF efficiency; bootstrap 95% CI 0.326-0.426 from Patel 2011 Table 2; Delta-OBJ = -11.46 vs base model).
• Notes: Distinct from RRT_CRRT_STATUS (subject-level binary indicator that ANY continuous / extended renal-replacement therapy is occurring) and from RRT_HEMODIAL_ACTIVE (per-time-point on/off gate for the dialysis-clearance term in within-subject time-varying dialysis-clearance models). FILT_AGE_HI gates WITHIN the CRRT-on subset by membrane age – meaningful only when RRT_CRRT_STATUS = 1 or RRT_HEMODIAL_ACTIVE = 1. The > 48 h threshold is paper-specific (Patel 2011 Methods note that membrane fouling and clot formation in the hemofilter pores progressively reduce dialysis efficiency, with the inflexion typically observed near 48 h). Future extractions that retain a finer filter-age structure (continuous filter_age_hours or multiple thresholds) should reuse FILT_AGE_HI for the binary > 48 h case and add a separate canonical for the continuous form. Scope kept specific pending a second model that ratifies the same > 48 h threshold; promote to general once corroborated.

FILT_SA_MED (canonical for medium-haemofilter membrane-surface-area indicator)

• Description: Binary indicator that the in-use extracorporeal renal-replacement-therapy haemofilter is the manufacturer’s MEDIUM membrane-surface-area size, within a paper’s discrete filter-size range. 1 = medium filter; 0 = the small (reference) filter, the large filter, or no extracorporeal therapy. Membrane surface area is the physical determinant of the diffusive/convective solute-transfer capacity of the filter, so papers that stock only a few filter sizes retain size as a categorical covariate on the extracorporeal clearance arm rather than as a continuous area. Paired with FILT_SA_LARGE; the two indicators are mutually exclusive and both are 0 for the small reference filter.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 for the COLUMN, meaning the smallest filter the source paper stocked (0.2 m^2 in both registered sources). The MODEL’s reference level, by contrast, is paper-specific and must not be assumed: Butragueno-Laiseca 2025 fixes the small filter to a multiplier of 1 and estimates the medium and large multipliers above it, whereas Butragueno-Laiseca 2022 – the same group, the same three filter sizes, the same PICU – fixes the LARGE filter to 1 and estimates the medium (0.74) and small (0.28) multipliers below it. Keep the column encoding fixed (FILT_SA_MED and FILT_SA_LARGE both 0 => small filter) and absorb the paper’s choice of reference level into the effect expression. For a small-filter reference that is cl_hemodialysis * e_filt_sa_med_<arm>^FILT_SA_MED * e_filt_sa_large_<arm>^FILT_SA_LARGE; for a large-filter reference, derive the small indicator inside model and write cl_hemodialysis * e_filt_sa_med_<arm>^FILT_SA_MED * e_filt_sa_small_<arm>^(1 - FILT_SA_MED - FILT_SA_LARGE). Both forms let every multiplier the paper prints appear verbatim in ini.
• Source aliases:
  • FILT, theta_FILT – Butragueno-Laiseca 2025 Table 2 notation for the three-level categorical covariate that these indicators decompose.
  • Filter surface area (m2) – Butragueno-Laiseca 2025 Table 1 row label.
• Example models: ButraguenoLaiseca_2025_teicoplanin.R (multiplicative factor e_filt_sa_med_cl_hemodialysis = 3.58 on the CKRT clearance arm; Butragueno-Laiseca 2025 Table 2, RSE 13%, SIR 95% CI 2.78-4.40; small-filter reference), ButraguenoLaiseca_2024_meropenem.R (same three physical filter sizes – 0.2 / 0.6 / 1.2 m^2 – but the paper references the MEDIUM filter rather than the small one, so the model uses the column only through the derived low-filter indicator filt_sa_small <- 1 - FILT_SA_MED - FILT_SA_LARGE and the medium level contributes the reference multiplier of 1; Butragueno-Laiseca 2024 Table 4, theta_FILT_Low = 0.27, theta_FILT_Med = 1 (reference), theta_FILT_High = 1.5), ButraguenoLaiseca_2022_piperacillin.R (large-filter reference: e_filt_sa_med_cl_hemodialysis = 0.74 (bootstrap 95% CI 0.6-0.9) and e_filt_sa_small_cl_hemodialysis = 0.28 (0.2-0.4) against theta_CLCKRT = 1.34 L/h for the 1.2 m^2 filter, reproducing the Results text’s per-size clearances of 1.34, 1.01 and 0.38 L/h; Butragueno-Laiseca 2022 Table 2).
• Notes: The absolute surface areas that the size labels denote are paper-specific and MUST be recorded in per-model covariateData[[FILT_SA_MED]]$notes – in the founding source, small = 0.2 m^2, medium = 0.6 m^2, large = 1.2 m^2, assigned by patient weight band (small 3-10 kg, medium 10-30 kg, large 30-60 kg). Because size is assigned by weight in paediatric practice, filter size and body weight are strongly collinear and a paper may describe one as a surrogate for the other; record which the model actually uses. Distinct from BSA (body surface area, a patient property) and from FILT_AGE_HI (membrane age, not size). Meaningful only when RRT_CRRT_ACTIVE = 1 (or the equivalent hemodialysis gate); the whole extracorporeal arm is gated off otherwise. A source paper that reports a genuinely continuous membrane surface area, or more than three sizes, should use the continuous canonical FILT_SA rather than extending the indicator set indefinitely (that canonical was founded on this instruction by Berthaud_2025_cefazolin.R, whose six paediatric patients span six distinct areas in a power model). Butragueno-Laiseca 2025 also tested effluent flow and blood flow on the extracorporeal clearance arm and could not retain them alongside filter size because of strong mutual correlation, so filter size is doing the work those flows would otherwise do. Scope kept specific pending a second model that ratifies the same three-level decomposition.

FILT_SA_LARGE (canonical for large-haemofilter membrane-surface-area indicator)

• Description: Binary indicator that the in-use extracorporeal renal-replacement-therapy haemofilter is the manufacturer’s LARGE membrane-surface-area size. 1 = large filter; 0 = the small (reference) filter, the medium filter, or no extracorporeal therapy. The large-size partner of FILT_SA_MED; see that entry for the full rationale.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 for the COLUMN, meaning the smallest filter the source paper stocked (0.2 m^2 in both registered sources). The MODEL’s reference level is paper-specific – Butragueno-Laiseca 2022 makes the LARGE filter its reference (multiplier fixed to 1) even though FILT_SA_LARGE = 1 still denotes that filter. See the FILT_SA_MED entry for both encodings.
• Source aliases:
  • FILT, theta_FILT – Butragueno-Laiseca 2025 Table 2 notation.
  • Filter surface area (m2) – Butragueno-Laiseca 2025 Table 1 row label.
• Example models: ButraguenoLaiseca_2025_teicoplanin.R (multiplicative factor e_filt_sa_large_cl_hemodialysis = 5.04 on the CKRT clearance arm; Butragueno-Laiseca 2025 Table 2, RSE 14%, SIR 95% CI 3.80-6.26; only 2 of the 12 CKRT patients used a large filter), ButraguenoLaiseca_2024_meropenem.R (multiplicative factor e_filt_sa_large_cl_hemodialysis = 1.5 on the CKRT clearance arm – a much smaller effect than the sibling’s 5.04 because it is measured against the MEDIUM reference rather than the small one, not because the filters differ; Butragueno-Laiseca 2024 Table 4, theta_FILT_High = 1.5, RSE 23%; 2 of the 7 CKRT patients used the high filter), ButraguenoLaiseca_2022_piperacillin.R (the 1.2 m^2 filter is the REFERENCE level there – theta_FILT_High = 1 Fixed – so no e_filt_sa_large_* parameter appears in ini(); the indicator is still required, because it is what distinguishes the reference large filter from the small filter that both indicators being 0 would otherwise denote; 3 of the 13 CKRT patients used a large filter).
• Notes: Mutually exclusive with FILT_SA_MED. A source paper may reference any of the three sizes: the sibling teicoplanin model references the small filter (so FILT_SA_MED and FILT_SA_LARGE carry both non-reference multipliers), whereas ButraguenoLaiseca_2024_meropenem.R references the medium filter (so the LOW level carries a multiplier and the model derives its indicator as 1 - FILT_SA_MED - FILT_SA_LARGE). The data columns are identical either way – each indicator always means “this filter is the medium / large one” – and only the model parameterisation’s reference level differs; encode whichever reference the paper printed so its published thetas stay verbatim. See the FILT_SA_MED entry for the reference-category encoding, the requirement to record the paper-specific absolute surface areas in per-model notes, the distinction from BSA and FILT_AGE_HI, and the collinearity with body weight and circuit flows. Scope kept specific pending a second ratifying model.

FILT_SA (canonical for continuous haemofilter / dialyser membrane surface area)

• Description: Membrane surface area of the extracorporeal renal-replacement-therapy filter (dialyser / haemofilter) in use, as a continuous quantity in m^2. Membrane surface area is the physical determinant of the diffusive and convective solute-transfer capacity of the filter, so it is the natural driver of an extracorporeal clearance arm. This is the continuous counterpart of the discrete FILT_SA_MED / FILT_SA_LARGE indicator pair: use those when a paper stocks only two or three filter sizes and retains size as a categorical covariate, and use FILT_SA when a paper spans enough distinct areas to fit a continuous (typically power) relationship. The FILT_SA_MED entry anticipated this split explicitly (“A source paper that reports a genuinely continuous membrane surface area, or more than three sizes, should add a continuous canonical rather than extending the indicator set indefinitely”); Berthaud 2025 is that paper.
• Units: m^2
• Type: continuous
• Scope: specific
• Reference category: n/a – enters as a power-centred ratio (FILT_SA / ref)^e_filt_sa_<arm> on the extracorporeal clearance arm, and is meaningful only while RRT_HEMODIAL_ACTIVE = 1 (or the equivalent CRRT gate). Reference values observed: 1 m^2 (Berthaud 2025, the area the published dialysis-clearance estimate is standardized to).
• Source aliases:
  • DMSA, Dialysis membrane surface area – Berthaud 2025 abstract, Table 1 row label, and Table 2 covariate-effect row.
  • Filter surface area, dialyser surface area, membrane area – variant forms used across the intermittent-hemodialysis and CRRT popPK literature.
• Example models: Berthaud_2025_cefazolin.R (founding example; power effect on the dialysis-clearance arm, cl_hemodialysis <- exp(lcl_hemodialysis) * (FILT_SA / 1)^e_filt_sa_cl_hemodialysis with e_filt_sa_cl_hemodialysis = 1.26 (RSE 1.06%) against CLdialpop = 1.98 L/h at 1 m^2; Berthaud 2025 Table 2. Six paediatric maintenance-hemodialysis patients spanning DMSA 0.2, 0.3, 1, 1, 1.5 and 1.4/1.7 m^2. Adding DMSA to the dialysis arm made allometrically scaled body weight on that arm non-significant, decreased the BIC by 6 units and drove the BSV on the arm towards 0, so the final model carries no IIV there).
• Notes: Distinct from BSA (body surface area – a patient property, not a device property). The two are related in practice but must not be conflated: paediatric dialysis prescriptions commonly SIZE the membrane from the patient’s BSA, and Berthaud 2025 exploits this in its dosing-regimen simulations by approximating FILT_SA = 0.85 * BSA with BSA = (4 * BW + 7) / (BW + 90) when a measured area is unavailable. That approximation is a simulation convenience and does NOT reproduce the fitted data (the paper’s own Table 1 areas differ from it by up to a factor of two), so record the measured area in this column and keep any BSA-based imputation in the model’s validation vignette, flagged as an assumption. The Berthaud 2025 Discussion notes 0.85 sits mid-range of the recommended 75-100% of BSA and that dialysis clearance changes proportionally under a different coefficient. Because membrane size tracks patient size in paediatric practice, FILT_SA and WT are strongly collinear – record which the model actually uses, and expect an allometric weight term on the extracorporeal arm to lose significance once FILT_SA is added (exactly what happened in the founding source). Naturally time-varying within subject when a patient’s membrane is changed mid-course (Berthaud 2025 patient 6 switched between 1.4 and 1.7 m^2). Scope kept specific pending a second model that ratifies the continuous form; promote to general once corroborated.

FILT_FLUX_LOW (canonical for low-flux dialyser-membrane indicator)

• Description: Binary indicator that the in-use dialyser / haemofilter membrane is a LOW-FLUX membrane – small pore size, low water permeability (ultrafiltration coefficient below roughly 10-12 mL/h/mmHg), and therefore poor clearance of middle-molecular-weight solutes. 1 = low-flux membrane; 0 = high-flux membrane or membrane type not recorded. Paired with FILT_FLUX_HIGH; the two are mutually exclusive and both are 0 for the “not recorded” reference, which in the founding source is by far the largest stratum. Member of the FILT_<attribute> family alongside FILT_SA / FILT_SA_MED / FILT_SA_LARGE (membrane size) and FILT_AGE_HI (membrane age); flux is the membrane’s PERMEABILITY class, orthogonal to its surface area.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 for the COLUMN. The MODEL’s reference level in a decomposed pair is the all-zero state, which for the founding source means “dialysis membrane not available” rather than a physical membrane – 424 of 459 subjects (92%). Encode as cl_dialysis * e_filt_flux_low_<arm>^FILT_FLUX_LOW * e_filt_flux_high_<arm>^FILT_FLUX_HIGH so both printed multipliers appear verbatim in ini().
• Source aliases:
  • DIAM [Low Flux] – Xu 2017 supplementary Table S2 row label for theta 13; DIAM is the source’s dialysis-membrane covariate. Same orientation, no value transformation.
• Example models: Xu_2017_daptomycin.R (multiplicative factor e_filt_flux_low_cl = 0.96 on the dialysis clearance equation only; Xu 2017 Table S2, 14% RSE; 7 of 459 subjects).
• Notes: The flux term acts only inside the founding source’s DIALYSIS clearance equation (Eq 1), not in its not-on-dialysis equation (Eq 6), because a membrane exists only when an extracorporeal circuit is running – meaningful only when one of the dialysis-modality indicators is 1. Record the manufacturer’s membrane and its ultrafiltration coefficient per model in covariateData[[FILT_FLUX_LOW]]$notes when the paper gives them; the low/high-flux boundary is a clinical convention rather than a single numeric threshold and has drifted over the decades the pooled datasets span. Distinct from FILT_SA (surface area) and FILT_AGE_HI (membrane age). A paper reporting a genuinely continuous ultrafiltration coefficient should register a continuous canonical rather than extending this pair. Ratified by the maintainers’ ruling 2026-09-21 (PMC5306496): the decomposed pair with an all-zero ‘flux not recorded’ reference was preferred over a single high-vs-low binary, which cannot represent the 92%-of-cohort reference level the thetas were estimated against.

FILT_FLUX_HIGH (canonical for high-flux dialyser-membrane indicator)

• Description: Binary indicator that the in-use dialyser / haemofilter membrane is a HIGH-FLUX membrane – large pore size, high water permeability, and substantially better clearance of middle molecules. 1 = high-flux membrane; 0 = low-flux membrane or membrane type not recorded. The high-permeability partner of FILT_FLUX_LOW; see that entry for the full rationale.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (low-flux membrane or not recorded).
• Source aliases:
  • DIAM [High Flux] – Xu 2017 supplementary Table S2 row label for theta 14.
• Example models: Xu_2017_daptomycin.R (multiplicative factor e_filt_flux_high_cl = 1.36 on the dialysis clearance equation; Xu 2017 Table S2, 8% RSE; 28 of 459 subjects, including all 8 CVVHDF subjects and 2 of 9 CVVHD subjects).
• Notes: This factor is load-bearing when reproducing a source’s published exposures: in Xu 2017 the whole CVVHDF cohort carried a high-flux membrane, so the exposures printed in that paper’s Table 3 are generated at 0.528 x 1.36 = 0.72 L/h and cannot be recovered from the bare CVVHDF clearance estimate alone. Note also that the high- and low-flux multipliers are NOT symmetric about 1 in the founding source (1.36 vs 0.96): the reference is the unrecorded-membrane stratum, not the low-flux stratum, so the pair does not partition a two-level covariate.

RRT_HEMODIAL_ACTIVE (canonical for hemodialysis-active indicator (time-varying per-session gate))

• Description: Within-subject time-varying indicator for whether an extracorporeal renal-replacement-therapy session (intermittent hemodialysis, hemofiltration, or hemodiafiltration) is currently running. 1 during the session; 0 in the interdialytic interval and in non-dialysed subjects.

• Units: (binary)

• Type: binary

• Scope: general

• Reference category: 0 (no dialysis running). Models that compose an additional cl_hemodialysis term should gate it by RRT_HEMODIAL_ACTIVE = 1 so that the interdialytic clearance reduces to the intrinsic body clearance. Write the gated total into cl itself – see the encoding rule in Notes below; a separate cl_total variable silently disables the dialysis arm.

• Source aliases:

  • HEMODIALYSIS – prior canonical name (2026-06-09 – 2026-06-19); the 2026-06-19 RRT family standardization renamed it to RRT_HEMODIAL_ACTIVE to align with the sibling RRT_HEMODIAL_STATUS / RRT_CRRT_STATUS shape.
  • DIAL – legacy form (used in Liesenfeld 2013, Jacobs 2016, Veinstein 2013 prior to the 2026-06-09 rename to the more explicit HEMODIALYSIS).
  • IHD_ON, HD_ACTIVE, RRT_ACTIVE – variant abbreviations used in adjacent ESRD / CRRT popPK literature.

• Example models:

  • Veinstein_2013_gentamicin.R (additive arm with an estimated primary ini parameter; cl <- exp(lcl + etalcl) * WT + RRT_HEMODIAL_ACTIVE * cl_hemodialysis).
  • Liesenfeld_2013_dabigatran.R (Michaels-equation gate; the dialysis arm is derived from blood flow rate, dialysate flow rate, and a hemodialyzer mass-transfer-area coefficient rather than estimated directly. Retains a separate gated cl_total; its explicit ODE is honoured – verified by solving both gate states, not assumed – but do not copy the shape, for the reason in Notes).
  • Jacobs_2016_colistin.R (additive on/off gate of fixed device-level hemodialysis-clearance constants for CMS and colistin, in a parent-metabolite chain; likewise retains a separate gated cl_tot whose explicit ODE is honoured, and likewise is not a shape to copy).
  • Berthaud_2025_cefazolin.R (additive arm in a paediatric maintenance-hemodialysis cohort, with the arm driven by a continuous membrane surface area instead of a bare estimate: cl <- exp(lcl + etalcl) * (WT / 70)^0.75 + RRT_HEMODIAL_ACTIVE * cl_hemodialysis where cl_hemodialysis <- exp(lcl_hemodialysis) * (FILT_SA / 1)^1.26. Note that Berthaud 2025 does NOT print the gate in its final-equation block – the block states only CLtot_i = CL_i + CLdial_i – and the on/off structure is established by the Introduction (“during hemodialysis sessions, a dialysis clearance … is added to the patient residual elimination clearance”), by the Discussion’s contrast between an “interdialytic” and a “dialysis” half-life, and by Figure 4’s three-sessions-per-week simulation design).
  • VanWart_2025_telavancin.R (same additive-arm shape as Veinstein 2013, with IIV on the dialysis arm: cl_hemodialysis <- exp(lcl_hemodialysis + etalcl_hemodialysis) * RRT_HEMODIAL_ACTIVE, CL_DL = 1.77 L/h; Van Wart 2025 states CL_DL “was estimated only during those periods where intermittent hemodialysis (IHD) was active and was fixed to a value of zero when IHD was not operative”). Paired in the same model with the interdialytic T_POST_HEMODIAL on central volume.

• Notes: Distinct from a renal-impairment indicator (subject-level baseline class) and from RRT_HEMODIAL_STATUS (the subject-level treatment-status indicator) – RRT_HEMODIAL_ACTIVE is the per-time-point gate that turns the dialysis-clearance term on and off WITHIN a single hemodialysis subject’s record. Pair with BFR and DFR when the dialysis clearance depends on flow rates; pair with a filter-specific mass-transfer coefficient (estimated lkoa in the model, not a covariate) when the Michaels parameterisation is used. Pair with lcl_hemodialysis (canonical) when the dialysis arm is a primary estimated structural parameter rather than a derived expression.

  Encoding rule – write the gated total into cl, never into a separate cl_total. rxode2 does not always integrate the d/dt() a model file declares: for some model shapes it solves the linear-compartment system with its analytic kernel driven by variables named cl (or CL) and vc (or v), and the explicit right-hand side is discarded. When that happens, the idiom

  cl       <- <off-therapy arm>
  cl_total <- cl + RRT_HEMODIAL_ACTIVE * cl_hemodialysis
  kel      <- cl_total / vc
  d/dt(central) <- -kel * central

  produces a silently inert dialysis arm: the reported cl_total and kel output columns switch correctly with the gate while the simulated amounts decay at the off-therapy rate in BOTH states. Write it as a single assignment instead:

  cl <- <off-therapy arm> + RRT_HEMODIAL_ACTIVE * cl_hemodialysis
  kel <- cl / vc

  Under the REPLACEMENT rule (see the lcl_hemodialysis entry in parameter-names.md) the same principle applies: cl <- (1 - RRT_HEMODIAL_ACTIVE) * <off-therapy arm> + RRT_HEMODIAL_ACTIVE * cl_hemodialysis.

  You cannot tell by inspection whether a given model is affected – defining both cl and vc is necessary but not sufficient, and Liesenfeld_2013_dabigatran.R and Jacobs_2016_colistin.R both define the pair yet have their explicit ODEs honoured, while Eyler_2014_ertapenem.R (a plain two-compartment model) did not. So treat the single-assignment form as mandatory for new models rather than as a case-by-case judgement. Nothing else in the package catches a violation: checkModelConventions() does not, modeldb$linCmt reads FALSE for an affected model, and a validation vignette that simulates only one gate state cannot see it. The mechanical gate is tests/testthat/test-modeldb-active-gate.R, which solves every model carrying an RRT_*_ACTIVE covariate at gate 0 and gate 1 and fails if no ODE state moves. Three shipped models were found inert by that sweep and repaired: Veinstein_2013_gentamicin.R, Dohmann_2025_piperacillin.R and Eyler_2014_ertapenem.R. The same hazard applies verbatim to the sibling gates RRT_CRRT_ACTIVE, RRT_CAPD_ACTIVE and RRT_CCPD_ACTIVE.

RRT_CRRT_ACTIVE (canonical for continuous-renal-replacement-therapy-active indicator (time-varying gate))

• Description: Within-subject time-varying indicator for whether a continuous or extended extracorporeal renal-replacement-therapy modality is currently running – continuous venovenous hemodialysis (CVVHD), hemofiltration (CVVH), hemodiafiltration (CVVHDF), sustained low-efficiency dialysis (SLED), or extended daily diafiltration (EDD-f). 1 while the therapy is running; 0 while it is interrupted and in subjects never treated. Completes the RRT_<modality>_<kind> family as the continuous-modality counterpart of RRT_HEMODIAL_ACTIVE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no continuous RRT running). Models that compose an additional dialysis-clearance term should gate it by RRT_CRRT_ACTIVE = 1 so that clearance during an interruption reduces to the intrinsic body clearance.
• Source aliases:
  • CVVHD – Huppe 2023 prose and figure legends.
  • CRRT_ON, CVVH_ACTIVE – variant abbreviations used in adjacent critical-care popPK literature.
• Example models: Huppe_2023_fosfomycin.R (Michaels-equation gate on the dialysis clearance arm: cl_cvvhd <- RRT_CRRT_ACTIVE * 60 * Michaels(BFR, DFR, K0A); each of the 15 patients contributed one PK profile during CVVHD and one after CVVHD was interrupted for tubing-system replacement, so the covariate genuinely varies within subject), Zhang_2025_fluconazole.R (gates the covariate-supplied extracorporeal clearance: clcrrt <- QEFF * RRT_CRRT_ACTIVE, reproducing the supplement’s $PK block IF (CRRTYN.EQ.0) CL_crrt = 0 / IF (CRRTYN.EQ.1) CL_crrt = CLCRRT. Every subject in the model-building dataset was on CRRT, so the gate does not vary within that dataset; it carries the paper’s off-CRRT simulation arm – the 0 mL/kg/h row of Tables 3 and 4 – and the intermittent-session scenarios of the companion Shiny application).
• Notes: Distinct from RRT_CRRT_STATUS, which is the SUBJECT-LEVEL treatment-status flag for the same modality family (Shekar 2014 records whether a patient was on RRT at all during sampling, not when a session was running). Distinct from RRT_HEMODIAL_ACTIVE, which is the per-session gate for INTERMITTENT hemodialysis (Liesenfeld 2013, Jacobs 2016, Veinstein 2013). The intermittent-versus-continuous split is retained at the ACTIVE level for the same reason the register already retains it at the STATUS level: the two modality classes have materially different drug-extraction kinetics – intermittent hemodialysis runs high blood and dialysate flows for a few hours, whereas continuous modalities run flows an order of magnitude lower around the clock, so a coefficient or mass-transfer coefficient calibrated on one does not transfer to the other. Pair with BFR and DFR when the dialysis clearance is derived from circuit flow rates via the Michaels equation, and with lkoa as the estimated filter mass-transfer-area coefficient in model (not a covariate). Note that BFR and DFR are canonically both mL/min, while continuous-modality papers commonly report dialysate flow in L/h – convert on ingestion and document any in-model back-conversion, as Huppe_2023_fosfomycin.R must do.

HEPIMP_MOD (canonical for moderate hepatic impairment indicator)

• Description: 1 = moderate hepatic impairment, 0 = normal hepatic function or any other (non-moderate) category. The classification scheme that defines “moderate” is paper-specific and must be documented in per-model covariateData[[HEPIMP_MOD]]$notes. Two schemes are commonly encountered:
  • NCI ODWG group 3: total bilirubin > 1.5-3 x ULN with any AST (Ramalingam SS et al., J Clin Oncol 2010;28:4507).
  • Child-Pugh Class B: composite score 7-9 across bilirubin, albumin, INR, ascites, and encephalopathy.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (any non-moderate category: normal, mild, or severe; the model typically uses HEPIMP_MOD alongside HEPIMP_MILD and/or HEPIMP_SEV to partition the non-moderate pool further).
• Source aliases:
  • HEP2 – used in Desai_2016_isavuconazole.R (paper’s liver-function index HEP2 for moderate Child-Pugh B; same orientation as the canonical).
• Example models: Desai_2016_isavuconazole.R (Child-Pugh Class B; log-additive shift e_hepimp_mod_cl = log(1.326 / 2.54) = -0.650 on CL and e_hepimp_mod_q = log(63.554 / 33.678) = +0.635 on Q vs the healthy reference, paired with HEPIMP_MILD for the parallel Child-Pugh A stratum; encodes the paper’s three-typical-value form CL = theta_g * exp(eta_jCL) as a healthy-reference baseline plus group-specific log shifts), Comisar_2025_rimegepant.R (multiplicative fractional effect on CL/F = -0.229 per Comisar 2025 Table 3, a 22.9% clearance decrease giving a 1.3-fold rise in simulated steady-state AUCtau; paired with HEPIMP_SEV (-0.423) as two mutually exclusive indicators whose shared reference group is normal-OR-mild hepatic function, because mild impairment was screened and found not statistically significant; the paper does not state whether its mild / moderate / severe classification follows Child-Pugh or NCI ODWG).
• Notes: Use this column when a model dichotomizes moderate hepatic impairment as a separate indicator from milder or more-severe categories. The classification scheme (NCI ODWG vs Child-Pugh vs other) is paper-specific and must be documented per-model. Companion to HEPIMP_MILD (mild only) and HEPIMP_SEV (severe only). Distinct from HEPIMP_MOD_OR_MISSING (pools moderate cases with missing-data cases) and HEPIMP_MODSEV (pools moderate with severe). Anticipated by the HEPIMP_MILD entry’s Notes: “For models that test moderate or severe as separate categories, register additional canonicals HEPIMP_MOD / HEPIMP_SEV rather than overloading this entry.” HEPIMP_SEV was registered separately; HEPIMP_MOD completes the parallel.

CRCL_BASE (canonical for per-subject baseline creatinine clearance (time-fixed))

• Description: Per-subject baseline renal function (creatinine clearance or BSA-normalized eGFR), time-fixed. Used when a source paper enters the subject’s baseline renal-function value as a separate constant column alongside a time-varying CRCL column under Wahlby 2004’s extended covariate-model decomposition (Br J Clin Pharmacol 2004;58(4):367-377). The within-subject delta is computed in model as (CRCL - CRCL_BASE).
• Units: mL/min or mL/min/1.73 m^2 – match the units of the paired CRCL column and document per-model via covariateData[[CRCL_BASE]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used either as (CRCL_BASE - ref) in a linear or linear-exp effect, or as the ratio (CRCL_BASE / ref), where ref is a cohort or upstream-analysis median baseline renal function. Reference values observed: 71.7 mL/min (Wahlby 2004 BCLC median), 93 mL/min/1.73 m^2 (Decker 2024, median eGFR of the upstream adult rheumatoid-arthritis analysis).
• Source aliases:
  • BCLC (baseline creatinine clearance) – Wahlby 2004 source-column convention; used in Wahlby_2004_gentamicin.R (raw mL/min) and Wahlby_2004_pefloxacin.R (concept tested but not retained in final model).
  • baseline eGFR – Decker 2024 source-table wording for the time-fixed bedside-Schwartz pediatric eGFR entering the renal-clearance arm as (baseline eGFR / 93); used in Decker_2024_baricitinib.R.
• Example models: Wahlby_2004_gentamicin.R (raw mL/min; reference 71.7 mL/min = BCLC median; coefficient 0.0098 per mL/min on CL alongside paired DCLC effect 0.0068 per mL/min), Decker_2024_baricitinib.R (bedside-Schwartz pediatric eGFR in mL/min/1.73 m^2; enters the apparent renal clearance arm multiplicatively as the ratio (CRCL_BASE / 93) rather than as a centred linear deviation, paired with the delta effect 0.00586 per mL/min/1.73 m^2 on (CRCL - CRCL_BASE); the reference 93 is the median eGFR of the upstream adult analysis, not a statistic of the pediatric cohort, whose mean baseline eGFR is 119), Tsuchiwata_2026_tofacitinib.R (raw Cockcroft-Gault CrCl in mL/min, source column BCCL; enters CL/F as the power function (CRCL_BASE / 126)^0.233), Osawa_2018_daclatasvir.R (raw CrCl in mL/min, source column BCRCL; enters CL/F as the power function (CRCL_BASE / 86.48)^0.235; reference is the cohort median, printed as 86.48 in the equation text and 86.5 in Table 1 – the Figure 3 caption’s ‘88.48’ is a typo).
• Notes: Promoted to general scope on 2026-08-04 alongside the Decker 2024 baricitinib extraction, the second model registering the column with consistent Wahlby-style BCOV/DCOV semantics. The two ratifying papers differ in the functional form applied to the baseline arm – Wahlby 2004 uses a centred linear deviation (CRCL_BASE - 71.7), Decker 2024 a multiplicative ratio (CRCL_BASE / 93) – and in the underlying assay (Cockcroft-Gault-style raw CrCl vs bedside-Schwartz pediatric eGFR); both are within scope because the column’s role is identical (the time-fixed baseline half of a time-varying renal-function covariate). Document the assay, the units, and the functional form per model in covariateData[[CRCL_BASE]]$notes. Distinct from time-fixed adult-cohort CRCL (where per-subject CRCL is just constant by data design); use CRCL_BASE only when the source paper deliberately separates the BCOV from the time-varying COV.

TBILI_BASE (canonical for per-subject baseline total bilirubin (time-fixed))

• Description: Per-subject baseline total serum bilirubin, time-fixed. Used when a source paper enters the subject’s baseline TBILI as a constant column alongside the time-varying TBILI column under Wahlby 2004’s extended covariate-model decomposition (Br J Clin Pharmacol 2004;58(4):367-377). The within-subject delta is computed in model as (TBILI - TBILI_BASE).
• Units: mg/dL or umol/L – match the units of the paired TBILI column and document per-model via covariateData[[TBILI_BASE]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – typically used as (TBILI_BASE - ref) in a linear or linear-exp effect.
• Source aliases:
  • BBIL (baseline total bilirubin) – Wahlby 2004 source-column convention; used in Wahlby_2004_pefloxacin.R and Wahlby_2004_paclitaxel_myelosuppression.R.
• Example models: Wahlby_2004_pefloxacin.R (umol/L; centered at 25 umol/L; coefficient -0.0068 per umol/L on CL via exp[e * (TBILI_BASE - 25)]), Wahlby_2004_paclitaxel_myelosuppression.R (umol/L; used together with TBILI as the within-subject delta TBILI - TBILI_BASE entering the Slope-DBIL effect).
• Notes: Specific scope because the BCOV/DCOV decomposition is paper-defined. Promote to general if a second paper ratifies. Distinct from DBIL (direct/conjugated bilirubin, a clinical-chemistry concept) – TBILI_BASE is a methodologic baseline-snapshot column, not a separate biomarker.

ALP_BASE (canonical for per-subject baseline alkaline phosphatase (time-fixed))

• Description: Per-subject baseline serum alkaline phosphatase, time-fixed. Used when a source paper enters the subject’s baseline ALP as a constant column alongside the time-varying ALP column under Wahlby 2004’s extended covariate-model decomposition (Br J Clin Pharmacol 2004;58(4):367-377). The within-subject delta (or log-ratio) is computed in model from ALP and ALP_BASE.
• Units: U/L (IU/L; interchangeable). Match the units of the paired ALP column and document per-model via covariateData[[ALP_BASE]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – typically used inside a log-ratio log(ALP / ALP_BASE) or a linear delta (ALP - ALP_BASE).
• Source aliases:
  • BALKP (baseline alkaline phosphatase) – Wahlby 2004 source-column convention; used in Wahlby_2004_voriconazole.R.
• Example models: Wahlby_2004_voriconazole.R (IU/L; population median 136 IU/L; paired with time-varying ALP to form log(ALP/ALP_BASE), the paper’s log(DALKP) term, with coefficient 0.59 on CL inside Eq 7 Final-Model column).
• Notes: Specific scope because the BCOV/DCOV decomposition is paper-defined. Promote to general if a second paper ratifies.

AST_BASE (canonical for per-subject baseline aspartate aminotransferase (time-fixed))

• Description: Per-subject baseline serum aspartate aminotransferase activity, time-fixed. Used when a source paper enters the subject’s baseline AST as a separate constant column alongside the time-varying AST column, so that the two can carry distinct effects on the same structural parameter. Operationally equal to AST at the subject’s first observation and constant thereafter.
• Units: U/L (SI; IU/L is used interchangeably). Match the units of the paired AST column and document per-model via covariateData[[AST_BASE]]$units.
• Type: continuous
• Scope: specific
• Reference category: n/a – used as the ratio (AST_BASE / ref) inside a power function, or as a centred linear deviation (AST_BASE - ref), where ref is the cohort median baseline AST. Reference value observed: 52 U/L (Osawa 2018, the asunaprevir cohort median).
• Source aliases:
  • BAST (baseline aspartate aminotransferase) – Osawa 2018 source-column convention, defined in the Table 2 footnote as “BAST, baseline aspartate aminotransferase” against “AST, aspartate aminotransferase at each time”.
• Example models: Osawa_2018_asunaprevir.R (U/L, reference 52 = the cohort median; power exponent -0.598 on CL/F, paired with the time-varying AST exponent -0.382 on the same parameter; both terms normalised to the SAME constant 52 U/L).
• Notes: The hepatic-transaminase member of the baseline-snapshot family alongside WT_BASE, BSA_BASE, CRCL_BASE, TBILI_BASE and ALP_BASE, and registered for the same reason: the split between a baseline value and a time-varying value is a modelling choice the source authors made deliberately, and collapsing the two loses it. Osawa 2018 is explicit that the split earned its place – adding the time-varying AST term to a model already carrying baseline AST dropped the objective function by a further 139.192 units. One feature to watch when transcribing this family: Osawa 2018 normalises the time-varying term by BASTref as well, so a single reference constant serves both terms and a subject whose AST never moves carries the SUM of the two exponents. Do not assume a separate reference for the time-varying column without checking the printed equation. Specific scope because the baseline/time-varying decomposition is paper-defined; promote to general if a second paper ratifies the column with consistent semantics. Distinct from AST, which is the time-varying (or single-value) transaminase column.

RRT_CCPD_ACTIVE (canonical for continuous-cycler-assisted-peritoneal-dialysis-active indicator (time-varying per-exchange gate))

• Description: Within-subject time-varying indicator for whether a continuous cycler-assisted peritoneal dialysis (CCPD) exchange is currently running. 1 during a cycler-driven exchange; 0 during ambulatory exchanges, during dry periods, and in subjects not on peritoneal dialysis. Completes the RRT_<modality>_<kind> family as the cycler-assisted peritoneal member, alongside its ambulatory counterpart RRT_CAPD_ACTIVE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no cycler-assisted exchange running). Models that compose a CCPD clearance arm should gate it by RRT_CCPD_ACTIVE = 1 so that clearance outside a cycler exchange reduces to the other active routes.
• Source aliases:
  • CCPD – Patel 2015 prose, Figure 1 (CL_OC,CCPD/CAPD) and Table 2 (CL_OCCCPD). Same orientation, no value transformation.
• Example models: Patel_2015_oseltamivir.R (gates the cycler arm of a two-arm peritoneal dialysate clearance in end-stage-renal-disease patients on automated peritoneal dialysis: cldial <- RRT_CCPD_ACTIVE * cl_ccpd_oselcarb + RRT_CAPD_ACTIVE * cl_capd_oselcarb. The study regimen ran three CCPD exchanges over an 8 h daytime block followed by two CAPD exchanges over 16 h overnight, repeating daily, so the gate genuinely alternates within every subject; the paper’s Monte Carlo simulations vary the two gates’ schedules across three different automated-peritoneal-dialysis prescriptions, which is what makes the split load-bearing rather than cosmetic).
• Notes: Distinct from PERIT_DIAL, which is the SUBJECT-LEVEL chronic-modality flag (is this patient a peritoneal-dialysis patient at all) and cannot express which exchange type is running at a given time. Distinct from RRT_HEMODIAL_ACTIVE and RRT_CRRT_ACTIVE, which gate EXTRACORPOREAL circuits: peritoneal dialysis moves drug across the peritoneal membrane into instilled dialysate with no blood circuit, so neither a blood flow rate nor a hemodialyzer mass-transfer coefficient applies and a coefficient calibrated on an extracorporeal modality does not transfer. Registered as a pair with RRT_CAPD_ACTIVE rather than as one peritoneal gate plus a modality selector because the two arms carry separate estimated clearances (Patel 2015: 0.319 vs 0.170 L/h/70 kg, a 1.9-fold difference driven by exchange frequency, not membrane properties) and because a pair of binaries composes additively in model() without a nested conditional. In a schedule with dry periods both gates are 0 and only the residual routes clear drug. Pair with the clearance-component tokens _ccpd / _capd (R/conventions.R::clComponents). Ratified by the maintainers 2026-09-02 (PMC4386947).

RRT_CAPD_ACTIVE (canonical for continuous-ambulatory-peritoneal-dialysis-active indicator (time-varying per-exchange gate))

• Description: Within-subject time-varying indicator for whether a continuous ambulatory peritoneal dialysis (CAPD) exchange is currently dwelling. 1 during an ambulatory exchange; 0 during cycler-assisted exchanges, during dry periods, and in subjects not on peritoneal dialysis. Ambulatory counterpart of RRT_CCPD_ACTIVE.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no ambulatory exchange dwelling). Models that compose a CAPD clearance arm should gate it by RRT_CAPD_ACTIVE = 1.
• Source aliases:
  • CAPD – Patel 2015 prose, Figure 1 (CL_OC,CCPD/CAPD) and Table 2 (CL_OCCAPD). Same orientation, no value transformation.
• Example models: Patel_2015_oseltamivir.R (gates the ambulatory arm of the same two-arm peritoneal dialysate clearance; see the RRT_CCPD_ACTIVE entry for the shared model form and schedule).
• Notes: See the RRT_CCPD_ACTIVE Notes for the full family rationale, the extracorporeal-versus-peritoneal distinction, and why the two exchange types are registered as a pair of binaries. One caution specific to CAPD: the osmotic agent differs across exchanges in practice (dextrose versus icodextrin), and this column does NOT resolve it. Patel 2015 measured both and found them comparable for oseltamivir carboxylate (geometric mean CAPD clearance 0.195 L/h with icodextrin versus 0.177 L/h with dextrose), so a single CAPD gate was sufficient there; a drug whose peritoneal transport does depend on the osmotic agent would need a separate column rather than an overload of this one. Ratified by the maintainers 2026-09-02 (PMC4386947).

PERIT_DIAL (canonical for peritoneal-dialysis treatment-status indicator)

• Description: 1 = the subject was undergoing chronic peritoneal dialysis (PD) during the modeled period; 0 = not on peritoneal dialysis. Treatment-status flag rather than a measured renal-function value; used as a multiplicative covariate on PK parameters that change between dialysis modalities (typically central volume Vc – chronic peritoneal dialysis is associated with higher extracellular fluid volume than intermittent hemodialysis, so plasma volume markers are larger in PD subjects). Per-subject indicator in the source data; typically time-fixed at the subject level (chronic dialysis modality is a months-to-years-stable patient characteristic, not a per-session attribute).
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (not on peritoneal dialysis). In a chronic-dialysis cohort comparing PD vs HD modalities, the reference category corresponds to the HD subjects; PERIT_DIAL = 1 selects the PD subgroup.
• Source aliases:
  • DIA – used in Takama_2007_darbepoetin.R (dialysis-modality indicator: DIA = 0 for hemodialysis, DIA = 1 for peritoneal dialysis; maps directly onto PERIT_DIAL with no value transformation). Takama 2007 Table 4 and Methods describe the covariate equation V1 = theta_V1 * [1 + theta_V1_WT * (WT - 54) + theta_V1_DIA * DIA] where theta_V1_DIA = 0.170 increments V1 by +17% in PD subjects relative to HD subjects.
  • DIT2 – used in Knebel_2008_epoetinDelta.R (the second of two dialysis-modality switches in a three-level covariate; DIT2 = 1 selects the peritoneal-dialysis stratum, with predialysis rather than hemodialysis as the reference – see the Notes below).
• Example models: Takama_2007_darbepoetin.R (+17% multiplicative-deviation effect on central volume V1; chronic-dialysis cohort comparing 63 HD vs 68 PD Japanese adult patients), Knebel_2008_epoetinDelta.R (exponentiated-switch factor 2.48^PERIT_DIAL on central volume only, paired with RRT_HEMODIAL_STATUS against a predialysis reference in a pediatric CKD erythropoietin popPK model).
• Notes: Distinct from RRT_HEMODIAL_STATUS (intermittent-hemodialysis IHD indicator; RRT_HEMODIAL_STATUS = 1 selects HD subjects) and from RRT_CRRT_STATUS (continuous / extended RRT indicator). When a cohort spans three modalities (predialysis / HD / PD) and the paper takes predialysis as the reference, PERIT_DIAL and RRT_HEMODIAL_STATUS are both used, are no longer anti-correlated, and are both 0 for the predialysis stratum; see the three-level paragraph in the RRT_HEMODIAL_STATUS Notes. In a chronic-dialysis cohort where every subject is on either HD or PD, RRT_HEMODIAL_STATUS and PERIT_DIAL are perfectly anti-correlated (PERIT_DIAL = 1 - RRT_HEMODIAL_STATUS); the choice of which to use as the covariate is dictated by which modality the source paper treats as the reference category. Takama 2007 took HD as the reference and reports a +17% effect on V1 for PD, so PERIT_DIAL is the natural canonical for that paper (V1 increases when PERIT_DIAL = 1). Anticipated in the RRT_HEMODIAL_STATUS register-entry notes (“would warrant its own canonical (RRT_PERIT_DIAL_STATUS, RRT_CRRT_STATUS) if a future paper retains them as covariates”) and ratified canonically on 2026-06-12 alongside the Takama 2007 darbepoetin alfa extraction. Future audit may rename this to RRT_PERIT_DIAL_STATUS for consistency with the RRT__ family; left unchanged for now to avoid disturbing the Takama 2007 ingestion before broader peritoneal-dialysis coverage in the registry.

HEMOADSORB_ACTIVE (canonical for hemoadsorption-active indicator (time-varying cartridge-in-circuit gate))

• Description: Within-subject time-varying indicator for whether an extracorporeal hemoadsorption (hemoperfusion) cartridge is currently in the blood circuit. 1 while a sorbent cartridge is perfused; 0 before and after the session, and throughout for subjects whose extracorporeal circuit carried no cartridge (sham circuit) or who were never treated. Hemoadsorption is a third extracorporeal modality alongside renal replacement therapy and ECMO: solute is removed by binding to a porous polymer sorbent rather than by diffusion or convection across a membrane, so the removal is saturable – the arm it gates is normally a decreasing function of the amount already adsorbed rather than a constant clearance.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no hemoadsorption cartridge in circuit). Models that compose an additional cl_hemoadsorption term should gate it by HEMOADSORB_ACTIVE = 1 so that clearance outside the session reduces to the intrinsic body clearance.
• Source aliases:
  • CytoSorb, CS, HP_ON, HA_ACTIVE – variant abbreviations used in the hemoperfusion / blood-purification PK literature. Leber 2023 has no data column for the indicator at all: it encodes the paper’s intervention-versus-control group assignment (five animals per combination with a CytoSorb cartridge in the circuit versus three with an identical sham circuit and no cartridge), so the ratifying models set source_name = NULL.
• Example models: Leber_2023_tacrolimus_sheep.R, Leber_2023_ciclosporin_sheep.R, Leber_2023_mycophenolateMofetil_sheep.R, Leber_2023_everolimus_sheep.R, Leber_2023_methylprednisolone_sheep.R (founding examples; all five gate the saturable arm cl_hemoadsorption <- HEMOADSORB_ACTIVE * clmax_hemoadsorption * (1 - adsorbed / amax_hemoadsorption) over a single 6-hour session, and setting the covariate to 0 throughout reproduces the paper’s sham-circuit control arm).
• Notes: Named ACTIVE rather than STATUS on the same split the RRT family uses: hemoadsorption is administered as a discrete, usually short session (6 h in Leber 2023, and commonly 8-24 h in critical care), so the per-time-point gate is the useful form and a subject-level HEMOADSORB_STATUS would be registered separately if a paper ever retained one. Deliberately outside the RRT_<MODALITY>_<KIND> namespace: hemoadsorption is not a renal-replacement modality – it provides no solute clearance by dialysis or filtration, no fluid removal, and is frequently run in series WITH an RRT circuit or an ECMO circuit rather than instead of one, so a model may legitimately carry HEMOADSORB_ACTIVE alongside RRT_CRRT_ACTIVE or ECMO_STATUS. Distinct from ECMO_STATUS for the same reason. Pair with BFR when the extracorporeal blood flow rate is a covariate on the adsorption arm (Leber 2023 held it constant at 120 mL/min and it therefore does not appear in the ratifying models), and with the adsorbed compartment plus lclmax_hemoadsorption / lamax_hemoadsorption parameters that make up the saturable sub-model. Because the gated arm depends on a cumulative state, switching the covariate back to 0 does not reset the cartridge load: a model that simulates two successive sessions with a fresh cartridge must reset adsorbed explicitly.

APHERESIS_LIPO_ACTIVE (canonical for lipoprotein-apheresis-active indicator (time-varying per-session gate))

• Description: Within-subject time-varying indicator for whether a lipoprotein apheresis (LA) session is currently running. 1 while blood is being processed through the apheresis column; 0 between sessions and throughout for subjects never treated. Lipoprotein apheresis is a selective extracorporeal procedure that removes apolipoprotein-B-containing particles (LDL and Lp(a)) by adsorption, precipitation or filtration, and is used in homozygous familial hypercholesterolemia when drug therapy alone does not reach LDL-C goals. The arm it gates is normally an additional first-order loss from an LDL-C turnover pool, so the covariate belongs on the PD side of a lipid model rather than on drug clearance.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (no apheresis session running). Models that compose an additional LDL-C elimination term should gate it by APHERESIS_LIPO_ACTIVE = 1 so that the between-session turnover reduces to the intrinsic kout.
• Source aliases:
  • APHON – used in Bihorel_2025_evinacumab.R (Bihorel 2025 NM-TRAN $INPUT column, apheresis on/off).
  • LA_ON, APH_ACTIVE, LDLAPH – variant abbreviations used in the adjacent lipid-apheresis literature.
• Example models: Bihorel_2025_evinacumab.R (founding example; gates the second elimination arm of the LDL-C indirect-response pool, - kout_apheresis * APHERESIS_LIPO_ACTIVE * ldl, reproducing the supplement’s $DES term -KAPH*APHON*A(4). The estimated rate of 8.36 /day corresponds to roughly a 58% LDL-C reduction over the paper’s typical 2.5-hour session, matching its reported “approximately 60%”).
• Notes: Named ACTIVE rather than STATUS on the same split the RRT family uses – apheresis is delivered as discrete sessions of a few hours at weekly to monthly frequency, so the per-time-point gate is the useful form. A subject-level “receives lipoprotein apheresis at baseline” flag and a session-frequency column are distinct concepts and would be registered separately if a paper ever retained one as a covariate (Bihorel 2025 carries both as dataset columns, APH and APHFRQN, but neither enters the final model). Deliberately outside the RRT_<MODALITY>_<KIND> namespace, for the same reason as HEMOADSORB_ACTIVE: apheresis is not a renal-replacement modality and may legitimately be carried alongside one. Session duration is not part of this column – it is expressed by how long the indicator stays at 1 in the event table, so a model reusing this covariate must state the assumed session length (2.5 hours is the average Bihorel 2025 reports, and the value its supplement uses to impute missing start/stop times). Distinct from HEMOADSORB_ACTIVE: hemoadsorption gates a saturable sorbent arm driven by a cumulative adsorbed state, whereas lipoprotein apheresis is modelled here as a plain first-order loss that leaves no cumulative state to reset between sessions.

ECMO_STATUS (canonical for extracorporeal-membrane-oxygenation treatment-status indicator)

• Description: 1 = the subject was supported by extracorporeal membrane oxygenation (veno-venous VV-ECMO or veno-arterial VA-ECMO) during the modeled period; 0 = no ECMO support. Treatment-status flag rather than a measured circuit quantity; used as a multiplicative covariate on the PK parameters that shift when a patient is cannulated – most often the central volume, which expands because the extracorporeal circuit adds several litres of circulating fluid for a hydrophilic drug to distribute into, and sometimes clearance. Completes the extracorporeal-support covariate family as the presence-of-therapy counterpart of the circuit-quantity columns ECMO_PUMP_SPEED and Q_CVVH and of the time-since-event columns T_ECMO / T_POST_ECMO.
• Units: (binary)
• Type: binary
• Scope: general
• Reference category: 0 (not receiving ECMO).
• Source aliases:
  • ECMO – the bare source-column form used by both ratifying papers; same orientation (1 = on ECMO), no value transformation. Used in Kang_2020_cefpirome.R, Valadez_2025_cefepime.R, Xu_2026_caspofungin.R and Pressiat_2022_amikacin.R.
  • ECMO – the bare source-column form used by both ratifying papers; same orientation (1 = on ECMO), no value transformation. Used in Kang_2020_cefpirome.R, Valadez_2025_cefepime.R, Xu_2026_caspofungin.R, Kim_2022_piperacillin.R and Kim_2022_tazobactam.R (the last two select a separate typical central volume, each with its own IIV, per ECMO status rather than applying a multiplicative effect).
• Example models: Kang_2020_cefpirome.R (power-form multipliers on both parameters, CL = 5.71 * 0.487^(CREAT/1.6) * 1.41^ECMO_STATUS and V1 = 2.74 * 4.22^ECMO_STATUS, in VA-ECMO patients sampled both during support and after weaning), Valadez_2025_cefepime.R (exponential multiplier on the central volume only, Vd = V1 * (WT/70) * exp(1.043 * ECMO_STATUS), a 2.8-fold expansion; an ECMO effect on CL was tested and rejected for imprecision, and the cohort excluded renal-replacement patients so the indicator is not confounded by CRRT), Xu_2026_caspofungin.R (power-form multiplier on the central volume only, V1 = 2.22 * (BSA/0.79) * 18.2^ECMO_STATUS, the largest ECMO volume effect in the register at 18.2-fold; the only paediatric example, and the only one where the effect’s magnitude is barely identified – 4 of 29 patients on ECMO, 62% RSE, bootstrap 95% CI 4.00-298.30), Kang_2026_ticagrelor.R (power-form multipliers on both parameters in the SAME direction the mechanism predicts but OPPOSITE to Kang 2020 on clearance: CL/F = 12.2 * 0.35^ECMO_STATUS – a 65% clearance REDUCTION on ECMO, the only decrease in the register – and Vd/F = 157 * 2.74^ECMO_STATUS * (0.518^(Q_ECMO/3))^ECMO_STATUS, the only example that pairs the indicator with a circuit-flow term nested inside it; same first author, institution and VA-ECMO cohort design as Kang_2020_cefpirome.R, which makes the opposite clearance sign a drug-property contrast – renally cleared cefpirome speeds up, hepatically metabolised ticagrelor slows down – rather than a cohort difference), Pressiat_2022_amikacin.R (exponential multiplier on the PERIPHERAL volume only, V2 = 15.9 * exp(0.76 * ECMO_STATUS), a 2.14-fold expansion, in a V-A ECMO versus non-ECMO ICU control cohort; subject-level indicator for a single studied dose).
• Notes: Time resolution is paper-specific and must be documented in each model’s covariateData[[ECMO_STATUS]]$notes. Kang 2020 treats it as time-varying within subject – PK samples were drawn both during ECMO and after successful weaning, so the per-record indicator switches at decannulation. Valadez 2025 treats it as a subject-level indicator because the source does not resolve cannulation / decannulation timing. Distinct from ECMO_PUMP_SPEED (continuous centrifugal-pump rotational speed in RPM, which quantifies the degree of support rather than its presence) and from T_ECMO / T_POST_ECMO (time since cannulation / decannulation, which model a gradual onset or recovery rather than a step change); a paper that reports both presence and a circuit quantity should register both columns. Distinct from RRT_CRRT_STATUS – ECMO and continuous renal replacement therapy are separate extracorporeal circuits that frequently co-occur, and conflating them is the confounder Valadez 2025 was designed to remove by excluding RRT patients entirely. When pairing ECMO_STATUS with CRCL, note that the two ratifying papers disagree on whether ECMO alters clearance (Kang 2020 retains a 1.41-fold CL increase; Valadez 2025 rejects an ECMO-on-CL term as a null effect with CV > 200%), so the covariate’s effect on CL should never be assumed from class precedent.

RRT_CVVHD_STATUS (canonical for continuous veno-venous haemodialysis treatment-status indicator)

• Description: 1 = the subject was undergoing continuous veno-venous haemodialysis (CVVHD) – purely DIFFUSIVE continuous renal replacement, dialysate counter-current across the filter with no replacement fluid – during the modeled period; 0 = not on CVVHD. Treatment-status flag rather than a measured renal-function value. The modality-resolved sibling of RRT_CRRT_STATUS, for the case the RRT_CRRT_STATUS Notes anticipate: a source paper that retains individual continuous modalities as SEPARATE covariates because it estimates different disposition for each. Per-subject, time-fixed in the founding example.
• Units: (binary)
• Type: binary
• Scope: specific
• Reference category: 0 (not on CVVHD). In a paper that strata-codes every dialysis modality, the all-zero state across RRT_HEMODIAL_STATUS, PERIT_DIAL, RRT_CVVHD_STATUS and RRT_CVVHDF_STATUS is the not-on-dialysis reference.
• Source aliases:
  • CVVHD – Xu 2017 Table 2 and supplementary Table S2 row labels (CL CVVHD, V1 CVVHD, Q CVVHD, V2 CVVHD); same orientation, no value transformation.
• Example models: Xu_2017_daptomycin.R (REPLACEMENT rule on all four disposition parameters – the indicator selects lcl_cvvhd 0.936 L/h, lvc_cvvhd 5.736 L, lq_cvvhd 7.111 L/h and lvp_cvvhd 4.891 L in place of the not-on-dialysis / other-stratum values; 9 of 459 pooled subjects).
• Notes: Use RRT_CRRT_STATUS when a paper lumps every continuous / extended modality into one indicator, and these modality-resolved columns only when the paper itself estimates modality-specific parameters. The distinction is not cosmetic: in the founding source CVVHD clearance is 1.25-fold the not-on-dialysis typical value while CVVHDF clearance is 29% BELOW it, so collapsing the two onto a single CRRT indicator would average away a 1.8-fold contrast. A model carrying these columns should leave RRT_CRRT_STATUS out rather than setting all three, to avoid double-counting. Paired with RRT_CVVHDF_STATUS; mutually exclusive with it and with RRT_HEMODIAL_STATUS / PERIT_DIAL. The clinical procedure the estimate is conditioned on must be recorded per model in covariateData[[RRT_CVVHD_STATUS]]$notes – filter surface area, blood flow and dialysate flow determine the extracorporeal clearance, and Xu 2017 states explicitly that its dose recommendation applies only to comparable procedures (high-flux 1.4-1.8 m^2 filter, blood flow 100-200 mL/min, dialysis flow 30-40 mL/kg/h). Where the paper records membrane flux separately, pair with FILT_FLUX_LOW / FILT_FLUX_HIGH. Ratified by the maintainers’ ruling 2026-09-21 (PMC5306496): adopt the two modality siblings ONLY and do NOT additionally set RRT_CRRT_STATUS on a model carrying them.

AST_BL (canonical for per-subject baseline (pre-treatment) aspartate aminotransferase (time-fixed, paired with time-varying AST))

• Description: Serum aspartate aminotransferase activity at treatment baseline, time-fixed per subject. Register this column alongside the time-varying AST whenever a source paper enters BOTH the pre-treatment anchor and the on-treatment value as separate covariates – typically because the on-treatment value enters only as the ratio AST / AST_BL, which decomposes a hepatic-function covariate into a time-fixed severity component and a time-varying recovery component. Operationally AST_BL equals AST at the subject’s first record and stays constant thereafter. Carries _BL’s established at-first-dose meaning used throughout this register (QTC_BL, HGB_BL, FERRITIN_BL, INS_BL).
• Units: U/L (SI; IU/L is used interchangeably). Document per-model via covariateData[[AST_BL]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (AST_BL / ref)^exponent. Reference values observed: 60 U/L (Zhu 2018; the reference subject’s value, close to the 62 U/L cohort mean and above the 51 U/L median).
• Source aliases:
  • Baseline AST – the printed covariate-equation term in Zhu_2018_asunaprevir.R; same orientation, no value transformation.
• Example models: Zhu_2018_asunaprevir.R (U/L, reference 60; exponent -0.458 on CL/F, paired with the time-varying ratio term (AST / AST_BL)^-0.291 on the same parameter – higher baseline AST means lower apparent clearance and therefore higher asunaprevir exposure, and clearance then rises again as on-treatment AST falls toward baseline-normal during antiviral therapy).
• Notes: Hepatic-function marker. General scope because the baseline-versus-time-varying AST decomposition is a generic popPK construct rather than a paper-specific one, and because _BL is already an established register-wide suffix with exactly this meaning. Use plain AST when a paper enters only one AST column, whether that column holds a baseline or a time-varying value – AST_BL exists specifically for the two-column case, where using AST for both would be ambiguous about which one a covariate effect refers to. Semantically the same construction as WT_BASE (per-subject baseline paired with a time-varying WT), but named with the register’s _BL lab-value suffix rather than _BASE, matching QTC_BL / HGB_BL / FERRITIN_BL. A dataset with no on-treatment AST measurements should set AST = AST_BL on every record, which reduces the ratio term to 1.

HGB (canonical for hemoglobin)

• Description: Blood hemoglobin concentration.
• Units: g/L or g/dL – document the unit used in each model via covariateData[[HGB]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (HGB / ref)^exponent.
• Source aliases: none; HGB is the common NONMEM / clinical-PK abbreviation.
• Example models: Yamada_2025_zolbetuximab.R (g/L, reference 118; exponent -0.374 on V1), Zou_2020_leuprorelin.R (g/dL, reference 13.6; exponent 2.30 on the resistant-fraction parameter rp of a PSA disease-progression model).
• Notes: Unit varies by paper (SI g/L, US g/dL; 1 g/dL = 10 g/L). The per-model covariateData[[HGB]]$units field is load-bearing.

HGB_BL (canonical for per-subject baseline hemoglobin concentration (initial-condition use))

• Description: Pre-treatment (or per-subject anchor-time) hemoglobin concentration used as the initial condition for a hemoglobin state variable in indirect-response / turnover anaemia models. Distinct from HGB (which is the time-varying or baseline-only hemoglobin value when entering a covariate effect on a structural PK parameter): HGB_BL is a static per-subject covariate that supplies the steady-state baseline for an Hb-state ODE, typically together with a kin = kout * HGB_BL synthesis-rate-from-baseline coupling so the Hb state starts at steady state. Parallel to FERRITIN_BL (baseline ferritin) and WT_BASE (baseline weight).
• Units: g/L, g/dL, or mmol/L – document the unit used in each model via covariateData[[HGB_BL]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – subject-level baseline supplied as a covariate column. Reference values observed: 8.3 mmol/L (Mulder 2025 chronic-HEV SOT cohort median).
• Source aliases:
  • HBBASE – used in Mulder_2025_ribavirin.R.
  • Hbsw – used in Chanu_2010_methoxyPolyethyleneGlycolEpoetinBeta.R (“the Hb concentration at the switch from the previous ESA treatment to C.E.R.A.”).
• Example models: Mulder_2025_ribavirin.R (mmol/L; initial condition for the hb_state compartment; per-subject kin = kout * HGB_BL so the Hb state is at steady state pre-RBV; the source paper substitutes the cohort typical (median) baseline for the one subject without an individual baseline value), Chanu_2010_methoxyPolyethyleneGlycolEpoetinBeta.R (g/dL; initial condition of the hb state and the driver of the SESA = (Hb0 - Hbsw) / LS prior-ESA-withdrawal term in a red-blood-cell life-span model; read only for previously ESA-treated subjects, with ESA-naive subjects starting at their own drawn pre-ESA baseline parameter instead).
• Notes: Promoted from specific to general on the Chanu 2010 extraction, which is the second paper to use a per-subject baseline hemoglobin covariate as the initial condition of a hemoglobin state – the condition this entry originally named for promotion. The two founding models instantiate the same idiom through different structures: Mulder 2025 couples it to an explicit kin = kout * HGB_BL turnover balance, Chanu 2010 uses it directly as hb(0) in a life-span (delay differential) model where the anchor differs from the model’s own estimated pre-ESA baseline parameter. Distinct from HGB (time-varying or baseline hemoglobin entering a covariate effect on a structural PK parameter) and from a PD baseline PARAMETER such as lrbase (a fitted typical value with IIV, not a data column). The unit varies widely across papers (SI g/L, US g/dL, Dutch / European convention mmol/L; 1 g/dL = 10 g/L; 1 mmol/L of tetrameric haemoglobin ~ 16.114 g/L); the per-model covariateData[[HGB_BL]]$units field is load-bearing.

TRANSF_BURDEN_BL (canonical for baseline red-blood-cell transfusion burden)

• Description: Pre-treatment red-blood-cell (RBC) transfusion burden, the number of RBC units transfused over a fixed look-back window before the first dose. A disease-severity covariate for transfusion-dependent anemias (beta-thalassemia, lower-risk myelodysplastic syndromes). Time-fixed per subject.
• Units: RBC units/24 weeks – the look-back window varies between trials (12, 16 or 24 weeks); document the window in covariateData[[TRANSF_BURDEN_BL]]$units and rescale before reusing a model with data on a different window.
• Type: continuous
• Scope: specific
• Reference category: n/a – continuous. Reference values observed: centred at 14 RBC units/24 weeks (Chen 2021, the rounded dataset median 14.1; range 0-34.0).
• Source aliases:
  • RBCT / RBCT burden (units/24 weeks) – Chen 2021 Table 1, Table 2 and the final covariate equations.
• Example models: Chen_2021_luspatercept.R (exponential effects exp(-0.0118 * (TRANSF_BURDEN_BL - 14)) on CL/F and exp(-0.0141 * (TRANSF_BURDEN_BL - 14)) on V1/F; higher transfusion burden, lower CL/F and V1/F).
• Notes: Distinct from any time-varying transfusion EVENT (an RBC transfusion given during the study, which is a dosing-like input to a hemoglobin model, not a baseline covariate) and from a binary transfusion-dependence indicator; a model that uses only “transfusion-dependent yes/no” should register a separate binary canonical. Promote to general if a second drug’s model uses the same count covariate.

WBC (canonical for white blood cell count)

• Description: Total white blood cell count (baseline or time-varying). In chronic lymphocytic leukaemia (CLL) populations the value is elevated because circulating leukaemic B-cells make up the majority of the count, so WBC can serve as a biomarker of target-cell burden rather than general hematology.
• Units: 10^9 cells/L (equivalent to 10^3 cells/uL). Document per-model via covariateData[[WBC]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (WBC / ref)^exponent. Reference values observed: 10 x 10^9/L (Mould 2007, typical CLL Vmax normalization).
• Source aliases: none known (WBC is the universal clinical-PK abbreviation).
• Example models: Mould_2007_alemtuzumab.R (reference 10 x 10^9/L; exponent 0.194 on Vmax), Dirks_2008_cetuximab.R (reference 6.8 x 10^9/L; additive linear-deviation effect 0.0216 per 10^9/L on Vmax; baseline only – cetuximab does not deplete circulating leukocytes, so WBC is treated as a static covariate), Gibiansky_2021_rituximab.R (CLL; reference 100 x 10^9/L; baseline only; power exponent 0.223 on the time-dependent target-mediated clearance component cl_exp_component, consistent with leukaemic B cells carrying the CD20 target).
• Notes: Time-varying in treatment studies where the drug depletes the leukaemic clone (e.g., alemtuzumab in CLL): WBC must be supplied at every observation time in the event dataset. In diseases where WBC is not therapeutically targeted the column can be treated as a baseline-only covariate; record the per-model convention in covariateData[[WBC]]$notes.

NLR (canonical for neutrophil-to-lymphocyte ratio)

• Description: Ratio of absolute neutrophil count to absolute lymphocyte count from a complete blood count with differential. Used as a peripheral inflammation marker. May be reported as baseline only or as a time-varying covariate.
• Units: ratio (unitless)
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (NLR / ref)^exponent or exponential effects. Reference values observed: 2.11 (Lin 2024, median in pooled COVID-19 + non-infected cohort).
• Source aliases: none; NLR is the universal abbreviation in clinical-PK and inflammation-biomarker literature.
• Example models: Lin_2024_casirivimab.R (time-varying; reference 2.11; small positive exponent +0.029 on CL).
• Notes: Document baseline-vs-time-varying status in covariateData[[NLR]]$notes. Although it derives from WBC differential counts, register it as its own canonical because the ratio (not the absolute counts) is what the model uses.

HCT (canonical for hematocrit)

• Description: Hematocrit – packed red blood cell volume fraction (baseline or time-varying).
• Units: % (volume fraction times 100). Document per-model via covariateData[[HCT]]$units.
• Type: continuous
• Scope: general
• Reference category: n/a – used with power scaling (HCT / ref)^exponent. Reference values observed: 45 % (Nestorov 2014, study-population median for severe hemophilia A adults).
• Source aliases: none; HCT is the universal NONMEM / clinical-PK abbreviation. Note that source papers report hematocrit on either the percent scale (45) or the volume-fraction scale (0.45); the canonical column is percent, so a paper reporting fractions must have both its per-subject values and its reference value multiplied by 100 before they are recorded against this column (Zhou_2025_tacrolimus.R does exactly this: paper median 0.265 -> canonical reference 26.5 %).
• Example models: Nestorov_2014_factorviii.R (reference 45 %, exponent -0.419 on V1), Zhou_2025_tacrolimus.R (reference 26.5 % = the Zhou 2025 cohort median of 0.265 rescaled to percent, exponent -0.71 on tacrolimus CL/F; a negative exponent on clearance rather than on volume – see Notes).
• Notes: Higher HCT (more red-cell volume) leaves a smaller plasma fraction within total body volume; for plasma-restricted distribution (e.g., factor VIII activity, which circulates in plasma) the central volume of distribution decreases as HCT rises, so the exponent is negative. Document baseline-vs-time-varying status in covariateData[[HCT]]$notes. Distinct from HGB (mass concentration of hemoglobin); the two correlate but enter different mechanistic relationships. Two distinct mechanisms put HCT on different parameters with the same negative sign. (a) On volume, via the plasma-fraction argument above (Nestorov 2014 factor VIII). (b) On apparent clearance of an erythrocyte-partitioned drug: tacrolimus is extensively bound to red blood cells, so whole-blood clearance falls as HCT rises because a larger share of the measured whole-blood drug is sequestered in erythrocytes and unavailable to hepatic metabolism. Because the assay measures whole blood, the negative exponent on CL/F is an artefact-free consequence of the partitioning, not a change in intrinsic clearance (Zhou 2025 Discussion: ‘Recipients with low levels of HCT have more free tacrolimus in the plasma, resulting in a corresponding increase in clearance’). The exponent magnitude is reproducible across independent lung-transplant cohorts: -0.71 (Zhou 2025) versus -0.868 (Cai et al., cited in Zhou 2025 Discussion). Because the effect enters as a ratio (HCT / ref)^exponent, a model that mistakenly mixes the percent and fraction scales between the column and the reference is off by a factor of 100^exponent – a 24-fold error at exponent -0.71 – so the scale of both must be checked together, not separately.

THB_MASS (canonical for total hemoglobin mass)

• Description: Subject-level total hemoglobin mass in grams. Plasma-volume-independent quantity measured by the optimised CO-rebreathing method (Schmidt 2005, Pottgiesser 2008); distinct from HGB (mass concentration in plasma) and HCT (volume fraction). Used in erythropoiesis / RBC-regeneration models as the steady-state set point Base that drives the negative-feedback term and seeds the steady-state initial conditions of the precursor compartments.
• Units: g
• Type: continuous
• Scope: specific
• Reference category: n/a – subject-level baseline supplied as a covariate column. Reference values observed: 885.42 g (Tetschke 2018 Table 1 example subject); Pottgiesser 2008 cohort mean ~870 g across 28 estimable adult-male volunteers.
• Source aliases: Base (Tetschke 2018 paper symbol).
• Example models: Tetschke_2018_erythropoiesis.R (reference 885.42 g; Pottgiesser 2008 dataset of 29 healthy adult male volunteers).
• Notes: Specific scope because total hemoglobin mass requires the optimised CO-rebreathing method to obtain (Schmidt 2005), which is a specialised technique not present in routine clinical labs; promote to general if a second model registers this quantity. Distinct from HGB (g/L or g/dL plasma concentration) and HCT (RBC volume fraction): THB_MASS is the absolute body-pool mass and is not perturbed by short-term plasma-volume fluctuations (Pottgiesser 2008 Section 3.2 explicitly motivates the choice of mass over concentration). Sex-dimorphic: typical value in adult males is meaningfully higher than in adult females; document the sex composition of the population in covariateData[[THB_MASS]]$notes.

LYMPH_ABS (canonical for absolute peripheral-blood lymphocyte count (total))

• Description: Total absolute peripheral-blood lymphocyte count (all lymphocyte subsets pooled: T, B, and NK cells). Baseline or time-varying; document the time resolution per model via covariateData[[LYMPH_ABS]]$notes. Used as a continuous covariate on clearance and other PK parameters when the source paper carries the count as an exogenous exposure covariate rather than as a modelled PD state. Common in allo-HSCT / GvHD-prophylaxis popPK studies where circulating-lymphocyte burden is a mechanistic driver for lymphocyte-targeting biologics (integrin blockers, T-cell / B-cell-depleting mAbs).
• Units: cells/uL (equivalent to 10^9 cells/L = K/uL x 1000; document the per-paper reporting unit in covariateData[[LYMPH_ABS]]$units). Reporting conventions vary: some papers report as K/uL (i.e., 10^3 cells/uL, numerically identical to 10^9 cells/L) and some as cells/uL – both encode the same numerical value once the K/uL suffix is expanded. Prefer cells/uL as the canonical unit for consistency with CD4_ABS, CD19_ABS, and NEUT.
• Type: continuous
• Scope: general
• Reference category: n/a – typically enters as a power term (LYMPH_ABS / ref)^exponent against a paper-specific reference. Reference values observed: 100 cells/uL (Waterhouse 2024, 0.1 K/uL, exponent -0.0180 on CL; the paper further replaces any zero lymphocyte counts with 0.01 K/uL = 10 cells/uL to avoid a log-of-zero in the power form).
• Source aliases:
  • LYMPH – Waterhouse 2024 NM-TRAN column, time-varying absolute lymphocyte count in K/uL.
  • ALC – clinical laboratory abbreviation for “absolute lymphocyte count”; common in hematology / transplant papers. Same semantics; no value transformation.
• Example models: Waterhouse_2024_vedolizumab.R (time-varying absolute lymphocyte count on CL: cl *= (LYMPH_ABS / 100)^-0.0180; reference 100 cells/uL = 0.1 K/uL per paper Table 2; zeros replaced with 10 cells/uL to avoid log(0)).
• Notes: Distinct from CD4_ABS (a subset – CD4+ T-lymphocytes only), CD19_ABS (a subset – CD19+ B-lymphocytes only), and NLR (a ratio, neutrophil-to-lymphocyte). LYMPH_ABS is the total pooled lymphocyte count and is the correct canonical for papers that measure “absolute lymphocyte count” without further flow-cytometry subsetting. Distinct from WBC (total white blood cell count, of which lymphocytes are one differential fraction). Zero-value handling: allo-HSCT / lymphocyte-depleting-therapy cohorts routinely have observed lymphocyte counts of exactly 0 K/uL at some time points; source papers commonly substitute a small floor (e.g., Waterhouse 2024 uses 0.01 K/uL) before evaluating a power-form covariate effect, per Waterhouse 2024 Supplement Equation S2. Document any per-paper zero-floor convention in covariateData[[LYMPH_ABS]]$notes.