Skip to contents

Ustekinumab (Aguiar 2021)

Source: vignettes/articles/Aguiar_2021_ustekinumab.Rmd
Aguiar_2021_ustekinumab.Rmd
library(nlmixr2lib)
library(rxode2)
#> rxode2 5.1.8 using 2 threads (see ?getRxThreads)
#>   no cache: create with `rxCreateCache()`
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)
library(ggplot2)
library(PKNCA)
#> 
#> Attaching package: 'PKNCA'
#> The following object is masked from 'package:stats':
#> 
#>     filter

Ustekinumab popPK-PD in Crohn’s disease (Aguiar 2021)

Replicate the population pharmacokinetic-pharmacodynamic model reported by Aguiar Zdovc et al. (2021) for ustekinumab in adults with Crohn’s disease. Ustekinumab is an IgG1k monoclonal antibody that binds the p40 subunit shared by interleukin-12 and interleukin-23. The structural model is a two-compartment quasi-equilibrium target-mediated drug disposition (QE-TMDD) model for the unbound drug and the unbound IL-12/IL-23 p40 target, both distributing in two-compartment systems and binding only in the central compartment, linked to fecal calprotectin via an indirect-response model in which the unbound target stimulates FC production.

  • Citation: Aguiar Zdovc J, Hanzel J, Kurent T, Sever N, Kozelj M, Smrekar N, Novak G, Stabuc B, Dreesen E, Thomas D, Vovk T, Ostanek B, Drobne D, Grabnar I. Ustekinumab Dosing Individualization in Crohn’s Disease Guided by a Population Pharmacokinetic-Pharmacodynamic Model. Pharmaceutics. 2021;13(10):1587. doi:10.3390/pharmaceutics13101587.
  • Article: https://doi.org/10.3390/pharmaceutics13101587

Population

The Aguiar 2021 cohort comprised 57 adults with active Crohn’s disease starting ustekinumab at a single Slovenian tertiary referral center, followed for 32 weeks. Baseline demographics (Aguiar 2021 Table 1): 56% female; median age 49 years (IQR 32-56); median weight 70 kg (IQR 59-84); median fat-free mass 45 kg (IQR 39-62) derived per subject via the Janmahasatian 2005 model; median disease duration 14 years (IQR 7-22); 66.7% had been exposed to a prior biologic (38/57 prior anti-TNF; 10/57 prior vedolizumab); 77.2% had endoscopically active luminal disease at baseline; median serum CRP 3 mg/L (IQR 3-11), albumin 43 g/L (IQR 41-44), fecal calprotectin 134 mg/kg (IQR 53-213). FCGR3A rs396991 genotype distribution: V/V 8.8% (5/57), V/F 54.4% (31/57), F/F 36.8% (21/57).

Standard ustekinumab dosing (which the simulations below use) is weight-tiered IV induction (260 mg if WT ≤\le 55 kg, 390 mg if 55 < WT ≤\le 85 kg, 520 mg if WT > 85 kg) at week 0, followed by 90 mg subcutaneous maintenance every 8 weeks.

The same demographics are available programmatically via the model’s metadata (readModelDb("Aguiar_2021_ustekinumab")$population).

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/Aguiar_2021_ustekinumab.R. The table below collects them in one place. Concentrations are in nmol/L; the source paper converted ustekinumab concentrations to nmol/L using a molecular weight of 149 kDa (Aguiar 2021 Methods section 2.3).

Parameter / equation Value Source
ka 0.381 /day Table 2 final-model Ka
CL (typical, reference covariates) 0.277 L/day Table 2 final-model CL
Vc (reference FFM = 45 kg) 3.57 L Table 2 final-model Vc
Vp (reference FFM = 45 kg) 3.30 L Table 2 final-model Vp
Q 1.89 L/day Table 2 final-model Q
F (V/F or F/F) 0.710 (logit 0.8954) Table 2 final-model F (V/F, F/F)
F (V/V) 0.888 (logit 2.0705) Table 2 final-model F (V/V)
FFM effect on CL power exponent 0.598, reference 45 kg Table 2 footnote a
FFM effect on Vc, Vp power exponents 0.590 (Vc), 0.586 (Vp), reference 45 kg Table 2 footnotes b, c
ALB linear effect on CL (1 - 0.0165 * (ALB - 43)) Table 2 footnote a
Bio-naive effect on CL (1 - 0.227 * bio_naive); bio_naive = 1 - PRIOR_BIO Table 2 footnote a
FCGR3A V/V effect on F logit shift 1.1751 Table 2 final-model
Ksyn (typical, reference CRP) 9.86e-9 nmol/L/day Table 2 final-model Ksyn
Kdeg 9.26e-10 /day Table 2 final-model Kdeg
CRP linear effect on Ksyn (1 + 0.0846 * (CRP - 3)) Table 2 footnote d
Vc-target 2.44 L Table 2 final-model Vc-target
Qtarget 0.493 L/day Table 2 final-model Qtarget
Vp-target 11.0 L Table 2 final-model Vp-target
Kint 2.83e-6 /day Table 2 final-model Kint
Kd 0.168 nmol/L Table 2 final-model Kd
Kout 0.0581 /day Table 3 final-model Kout
FC0 (no ulcers) 102 mg/kg Table 3 final-model FC0
FC0 (ulcers) 213 mg/kg Table 3 final-model FC0 with ulcers
Emax (FC stim) 2.19 (= 219%) Table 3 final-model Emax
C50 (FC stim) 2.46 nmol/L Table 3 final-model C50
Drug ODEs n/a, see model file Figure 1 + Table 2 footnotes
Target ODEs n/a, see model file Figure 1
Indirect-response PD dFC/dt = (FC0 * Kout / stim_baseline) * stim_fc - Kout * FC Table 3 footnote a
IIV CL omega^2 = log(0.18^2 + 1) = 0.03188 Table 2 (18.0% CV)
IIV Vc omega^2 = log(0.0979^2 + 1) = 0.00954 Table 2 (9.79% CV)
IIV Vp omega^2 = log(0.241^2 + 1) = 0.05645 Table 2 (24.1% CV)
IIV logit(F) omega^2 = 0.173^2 = 0.02993 Table 2 (SD = 0.173 on logit scale)
IIV Ksyn omega^2 = log(0.992^2 + 1) = 0.68522 Table 2 (99.2% CV)
IIV FC0 omega^2 = log(0.99^2 + 1) = 0.68309 Table 3 (99.0% CV)
Residual on Cc additive 4.55 nmol/L + proportional 7.77% Table 2
Residual on FC proportional 57.3% Table 3

Covariate column naming

Source column Canonical column used here Notes
FFM (kg) FFM Reference 45 kg (cohort median); derived per subject via Janmahasatian 2005.
Serum albumin ALB (g/L) Reference 43 g/L (cohort median).
C-reactive protein CRP (mg/L) Standard CRP assay; reference 3 mg/L (cohort median).
bio-naive (1 = no prior biologic) PRIOR_BIO (1 = previously exposed) Inverted: model derives bio_naive <- 1 - PRIOR_BIO to preserve the paper’s coefficient.
FCGR3A-158 V/V FCGR3A_VV 1 = V/V, 0 = V/F or F/F.
Endoscopically active disease at baseline ENDO_ULCER 1 = mucosal ulcers at baseline ileocolonoscopy, 0 otherwise.

FFM, PRIOR_BIO, FCGR3A_VV, and ENDO_ULCER are added to the canonical register in inst/references/covariate-columns.md alongside this model. ALB and CRP are pre-existing canonical columns.

Virtual cohort

Original subject-level data are not publicly available. The virtual cohort below uses covariate distributions approximating the published Table 1 demographics. Covariates are sampled independently (the paper does not publish joint distributions); FCGR3A genotype frequencies match the cohort’s Hardy-Weinberg distribution (V/V 8.8%, V/F + F/F 91.2%).

set.seed(20260425)

n_subj <- 200
mw_ust <- 149  # ustekinumab molecular weight in kDa, per Aguiar 2021 Methods.

# 1 mg ustekinumab = 1000 / mw_ust nmol = 6.711 nmol.
mg_to_nmol <- function(mg) mg * 1000 / mw_ust

# Build a virtual Crohn's-disease cohort approximating Aguiar 2021 Table 1.
cohort <- tibble(
  ID  = seq_len(n_subj),
  # Weight: median 70 kg, IQR 59-84; lognormal covers the right skew.
  WT  = pmin(pmax(rlnorm(n_subj, log(70), 0.22), 40), 130)
) |>
  mutate(
    # FFM: median 45 kg, IQR 39-62; correlated with weight in the source cohort.
    # Approximated as 0.65 * WT for males (44%) and 0.55 * WT for females (56%).
    SEXF = rbinom(n_subj, 1, 0.56),
    FFM  = pmin(pmax(WT * if_else(SEXF == 1, 0.55, 0.65) +
                       rnorm(n_subj, 0, 3), 25), 90),
    # Albumin: median 43 g/L, IQR 41-44; tight distribution in CD remission/active.
    ALB = pmin(pmax(rnorm(n_subj, 42.5, 3.5), 28), 52),
    # CRP in CD skews right; median 3 mg/L, IQR 3-11; log-normal shape.
    CRP = pmin(pmax(rlnorm(n_subj, log(4), 1.0), 0.3), 80),
    # Prior biologic exposure: 66.7% in cohort (Table 1).
    PRIOR_BIO = rbinom(n_subj, 1, 0.667),
    # FCGR3A genotype: 8.8% V/V (Table 1).
    FCGR3A_VV = rbinom(n_subj, 1, 0.088),
    # Endoscopically active disease: 77.2% at baseline (Table 1).
    ENDO_ULCER = rbinom(n_subj, 1, 0.772)
  )

Standard ustekinumab dosing regimen

Aguiar 2021 simulated the licensed regimen: weight-tiered IV induction at week 0 (260 mg if WT ≤\le 55 kg, 390 mg if 55 < WT ≤\le 85 kg, 520 mg if WT > 85 kg) followed by 90 mg subcutaneous every 8 weeks (Methods section 2.4 scenario a). The dosing dataset is in nmol because the model is parameterized in nmol / nmol/L; the helper mg_to_nmol converts.

week  <- 7  # days
ind_t <- 0
sc_times <- c(8, 16, 24) * week  # SC maintenance at 8, 16, 24 weeks

induction_dose_mg <- function(wt) {
  ifelse(wt <= 55, 260,
         ifelse(wt <= 85, 390, 520))
}

iv_induction <- cohort |>
  mutate(
    TIME = ind_t,
    AMT  = mg_to_nmol(induction_dose_mg(WT)),
    EVID = 1L,
    CMT  = "central",
    DV   = NA_real_,
    phase = "induction_IV"
  )

sc_maintenance <- cohort |>
  tidyr::crossing(TIME = sc_times) |>
  mutate(
    AMT  = mg_to_nmol(90),
    EVID = 1L,
    CMT  = "depot",
    DV   = NA_real_,
    phase = "maintenance_SC"
  )

# Observation grid: dense early, weekly through 32 weeks.
obs_days <- sort(unique(c(
  c(0, 0.05, 1, 2, 3, 5, 7),
  seq(7, 32 * week, by = 3.5)
)))

obs <- cohort |>
  tidyr::crossing(TIME = obs_days) |>
  mutate(
    AMT  = 0,
    EVID = 0L,
    CMT  = "central",
    DV   = NA_real_,
    phase = NA_character_
  )

# Use the variable-name CMT for observations: "Cc" for ustekinumab
# concentration, "fc" for fecal calprotectin. Aguiar 2021's PD output is on
# the fc compartment; observing only "Cc" plus "fc" gives both endpoints.
obs_cc <- obs |> mutate(CMT = "Cc")
obs_fc <- cohort |>
  tidyr::crossing(TIME = c(0, 8, 16, 24, 32) * week) |>
  mutate(AMT = 0, EVID = 0L, CMT = "fc", DV = NA_real_, phase = NA_character_)

events <- bind_rows(iv_induction, sc_maintenance, obs_cc, obs_fc) |>
  arrange(ID, TIME, desc(EVID)) |>
  select(ID, TIME, AMT, EVID, CMT, DV,
         FFM, ALB, CRP, PRIOR_BIO, FCGR3A_VV, ENDO_ULCER, phase)

stopifnot(!anyDuplicated(unique(events[, c("ID", "TIME", "EVID", "CMT")])))

Simulate

By default rxSolve() samples between-subject random effects per subject when the model defines omega-style etas; the stochastic simulation is what the VPC plots below use. The published-value comparison further down uses rxode2::zeroRe() to remove between-subject variability so the typical-value trajectory can be compared head-to-head with Aguiar 2021’s reported point estimates.

mod <- readModelDb("Aguiar_2021_ustekinumab")
sim <- rxode2::rxSolve(mod, events = events, returnType = "data.frame")
#> ℹ parameter labels from comments will be replaced by 'label()'

Concentration-time profile (free ustekinumab) 

sim_cc <- sim |> filter(CMT == 7L)  # CMT 7 = Cc observations

sim_summary <- sim_cc |>
  filter(time > 0) |>
  group_by(time) |>
  summarise(
    Q05 = quantile(Cc, 0.05, na.rm = TRUE),
    Q50 = quantile(Cc, 0.50, na.rm = TRUE),
    Q95 = quantile(Cc, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  mutate(Cc_ugmL_05 = Q05 * mw_ust / 1000,
         Cc_ugmL_50 = Q50 * mw_ust / 1000,
         Cc_ugmL_95 = Q95 * mw_ust / 1000)

ggplot(sim_summary, aes(x = time / week, y = Cc_ugmL_50)) +
  geom_ribbon(aes(ymin = Cc_ugmL_05, ymax = Cc_ugmL_95),
              alpha = 0.25, fill = "steelblue") +
  geom_line(colour = "steelblue", linewidth = 0.8) +
  scale_y_log10() +
  scale_x_continuous(breaks = seq(0, 32, by = 4)) +
  labs(
    x = "Time (weeks)",
    y = expression("Free ustekinumab ("*mu*"g/mL)"),
    title = "Standard ustekinumab regimen: weight-tiered IV induction + 90 mg SC Q8W",
    subtitle = paste0("Median and 90% prediction interval (N = ", n_subj,
                      " virtual CD patients)"),
    caption = "Replicates Aguiar 2021 Figure 2A (free ustekinumab VPC structure)."
  ) +
  theme_bw()

Target and FC profiles

tfree (unbound target in central, nmol/L) and fc (fecal calprotectin, mg/kg) are emitted by the model alongside Cc and can be read directly from the simulation output.

sim_fc <- sim |> filter(CMT == 6L)  # CMT 6 = fc observations

sim_pd <- sim_fc |>
  filter(time >= 0) |>
  group_by(time) |>
  summarise(
    target_Q05 = quantile(tfree, 0.05, na.rm = TRUE),
    target_Q50 = quantile(tfree, 0.50, na.rm = TRUE),
    target_Q95 = quantile(tfree, 0.95, na.rm = TRUE),
    fc_Q05 = quantile(fc, 0.05, na.rm = TRUE),
    fc_Q50 = quantile(fc, 0.50, na.rm = TRUE),
    fc_Q95 = quantile(fc, 0.95, na.rm = TRUE),
    .groups = "drop"
  )

p_target <- ggplot(sim_pd, aes(x = time / week, y = target_Q50)) +
  geom_ribbon(aes(ymin = target_Q05, ymax = target_Q95),
              alpha = 0.25, fill = "darkorange") +
  geom_line(colour = "darkorange", linewidth = 0.8) +
  scale_x_continuous(breaks = seq(0, 32, by = 4)) +
  labs(x = "Time (weeks)", y = "Free target (nmol/L)",
       title = "Unbound IL-12/23 p40 target") +
  theme_bw()

p_fc <- ggplot(sim_pd, aes(x = time / week, y = fc_Q50)) +
  geom_ribbon(aes(ymin = fc_Q05, ymax = fc_Q95),
              alpha = 0.25, fill = "firebrick") +
  geom_line(colour = "firebrick", linewidth = 0.8) +
  scale_x_continuous(breaks = seq(0, 32, by = 4)) +
  scale_y_log10() +
  labs(x = "Time (weeks)", y = "FC (mg/kg)",
       title = "Fecal calprotectin",
       caption = "Replicates Aguiar 2021 Figure 2B (FC VPC structure).") +
  theme_bw()

if (requireNamespace("patchwork", quietly = TRUE)) {
  patchwork::wrap_plots(p_target, p_fc, ncol = 1)
} else {
  print(p_target)
  print(p_fc)
}

PKNCA validation

PKNCA is run on the steady-state SC dosing interval (last 8-week interval, weeks 24-32) and on the IV induction dose (week 0 - week 8, the first SC maintenance dose at week 8 closes the interval). Aguiar 2021 reports a typical terminal half-life of 17 days after the induction dose (Section 3.3).

ind_window_start <- 0
ind_window_end   <- 8 * week

ss_window_start  <- 24 * week
ss_window_end    <- 32 * week

# Pull only Cc observations from the simulation.
sim_cc_only <- sim |> filter(CMT == 7L)

iv_conc <- sim_cc_only |>
  filter(time >= ind_window_start, time <= ind_window_end, !is.na(Cc)) |>
  transmute(ID = id, time_rel = time - ind_window_start, Cc,
            treatment = "IV_induction")

sc_ss_conc <- sim_cc_only |>
  filter(time >= ss_window_start, time <= ss_window_end, !is.na(Cc)) |>
  transmute(ID = id, time_rel = time - ss_window_start, Cc,
            treatment = "SC_90mg_Q8W_ss")

nca_conc <- bind_rows(iv_conc, sc_ss_conc)

iv_doses <- iv_induction |>
  transmute(ID, time_rel = TIME - ind_window_start, AMT,
            treatment = "IV_induction")

sc_ss_dose <- sc_maintenance |>
  filter(TIME == 24 * week) |>
  transmute(ID, time_rel = TIME - ss_window_start, AMT,
            treatment = "SC_90mg_Q8W_ss")

nca_dose <- bind_rows(iv_doses, sc_ss_dose)
conc_obj <- PKNCA::PKNCAconc(nca_conc, Cc ~ time_rel | treatment + ID)
dose_obj <- PKNCA::PKNCAdose(nca_dose, AMT ~ time_rel | treatment + ID)

intervals <- data.frame(
  start     = 0,
  end       = 8 * week,
  cmax      = TRUE,
  tmax      = TRUE,
  cmin      = TRUE,
  auclast   = TRUE,
  half.life = TRUE
)

nca_data <- PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals)
nca_res  <- suppressWarnings(PKNCA::pk.nca(nca_data))

knitr::kable(
  summary(nca_res),
  digits  = 3,
  caption = "Simulated NCA: IV induction (0-8 weeks) and SC steady-state interval (weeks 24-32)."
)
Simulated NCA: IV induction (0-8 weeks) and SC steady-state interval (weeks 24-32).
start end treatment N auclast cmax cmin tmax half.life
0 56 IV_induction 200 8500 [25.7] 770 [16.7] 35.4 [78.7] 0.000 [0.000, 0.000] 17.9 [5.24]
0 56 SC_90mg_Q8W_ss 200 1710 [33.5] 60.7 [21.8] 10.3 [60.5] 3.50 [3.50, 7.00] 21.9 [4.57]

Comparison against published values

Aguiar 2021 reports the typical terminal half-life of ustekinumab after the induction dose at 17 days (Section 3.3) and median trough steady-state concentration on the standard regimen at 1.5 ug/mL (Section 4 discussion of Wang et al. comparison).

For the head-to-head comparison against published point estimates, run a typical-value simulation (zeroRe()) on a single subject with reference covariates so the result is deterministic.

ref_subj <- tibble::tibble(
  ID = 1L,
  FFM = 45, ALB = 43, CRP = 3,
  PRIOR_BIO = 1L, FCGR3A_VV = 0L, ENDO_ULCER = 1L
)

ref_iv <- ref_subj |>
  mutate(TIME = 0, AMT = mg_to_nmol(390), EVID = 1L,
         CMT = "central", DV = NA_real_)
ref_sc <- ref_subj |>
  tidyr::crossing(TIME = c(8, 16, 24) * week) |>
  mutate(AMT = mg_to_nmol(90), EVID = 1L,
         CMT = "depot", DV = NA_real_)
ref_obs <- ref_subj |>
  tidyr::crossing(
    TIME = c(0.5, 1, 7, 14, 21, 28, 42, 56,
             seq(56, 32 * week, by = 7))
  ) |>
  mutate(AMT = 0, EVID = 0L, CMT = "Cc", DV = NA_real_)

ref_events <- bind_rows(ref_iv, ref_sc, ref_obs) |>
  arrange(ID, TIME, desc(EVID))

mod_typical <- rxode2::zeroRe(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'
sim_typical <- rxode2::rxSolve(mod_typical, events = ref_events,
                               returnType = "data.frame") |>
  filter(CMT == 7L)
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etalogitfdepot', 'etalksyn', 'etalfc0'

# Terminal half-life from the IV induction phase (days 7-56 after dose):
typical_late <- sim_typical |>
  filter(time >= 7, time <= 56)
fit_thalf <- lm(log(Cc) ~ time, data = typical_late)
typical_thalf <- as.numeric(log(2) / -coef(fit_thalf)[2])

# Steady-state trough at week 32 (typical value, single reference subject):
typical_trough <- sim_typical |>
  filter(time == 32 * week) |>
  pull(Cc)
typical_trough_ugmL <- typical_trough * mw_ust / 1000

comparison <- tibble::tribble(
  ~Metric,                                          ~Published, ~Simulated,
  "Terminal half-life after IV induction (days)",   17,         round(typical_thalf, 1),
  "SC SS trough free ustekinumab (ug/mL)",          1.5,        round(typical_trough_ugmL, 2)
)

knitr::kable(
  comparison,
  caption = "Typical-value simulation vs published values (Aguiar 2021 Section 3.3 and Discussion comparison with Wang et al.)."
)
Typical-value simulation vs published values (Aguiar 2021 Section 3.3 and Discussion comparison with Wang et al.).
Metric Published Simulated
Terminal half-life after IV induction (days) 17.0 17.00
SC SS trough free ustekinumab (ug/mL) 1.5 1.29

The simulated terminal half-life and SC steady-state trough should agree with the published values within ~20%. Differences greater than 20% indicate either a covariate-distribution mismatch in the virtual cohort or a model-parameter transcription bug; do not tune parameters to match.

Assumptions and deviations

  • PK assay is for unbound (free) drug. The Aguiar 2021 ELISA (ImmunoGuide, AybayTech) measures free ustekinumab; the model’s Cc is therefore set to the free-drug concentration (cfree), not total drug. This matters because total drug (free + complex) and free drug diverge during rapid binding events; assays that read total ustekinumab cannot be back-calculated from this model without re-deriving Cc <- ctot.
  • FFM derivation. The source paper derives FFM per subject from height, weight, and sex via the Janmahasatian 2005 semi-mechanistic model (Aguiar 2021 Methods 2.2). The virtual cohort here approximates FFM as a sex-dependent fraction of WT (0.55 for women, 0.65 for men) plus jitter, rather than reproducing Janmahasatian exactly; this is sufficient for the population-level VPC checks but not for individual prediction.
  • Joint covariate distributions are not published. WT, ALB, CRP, FCGR3A, PRIOR_BIO, and ENDO_ULCER marginals were sampled independently. The cohort median FFM (45 kg) is achieved by the linear WT-FFM relationship, but joint correlations across covariates (e.g., albumin and CRP, or FFM and prior biologic exposure) are not preserved.
  • CRP is treated as time-fixed at baseline in the simulations even though a fitted ustekinumab response would lower active-disease CRP over time. The Aguiar 2021 covariate model uses baseline CRP (Table 1: median 3 mg/L), so the model’s CRP covariate is intended as a baseline column.
  • Initial target distribution at SS. Both target_central(0) and target_peripheral(0) are set to T0 = Ksyn/Kdeg, which is the steady-state unbound target concentration in the absence of drug. Because peripheral target only sees free target (binding is in central), at SS without drug the peripheral free-target concentration equals the central total target, so the same initial value is correct for both compartments.
  • No model-derived helper for Janmahasatian FFM is exported. Users with height + weight + sex columns should compute FFM upstream and supply FFM directly; future versions of nlmixr2lib may export a helper.
  • CV-to-omega convention. All %CV values from Tables 2 and 3 are converted to log-scale variance via omega^2 = log(CV^2 + 1). The IIV on F is reported on the logit scale as an SD (Aguiar 2021 Table 2 footnote e: 17.3); that value is squared directly without log transformation.
  • Non-canonical compartment names. total_target matches the canonical QSS / MM TMDD compartment name in inst/references/compartment-names.md. The companion target_peripheral (free peripheral target) and fc (fecal calprotectin) compartments are not in the canonical list; checkModelConventions() emits two warnings for them. Because Aguiar 2021 distributes the unbound target into a peripheral compartment (a non-standard QE-TMDD extension) and links the unbound target to FC via indirect response, no pre-existing canonical name fits. The names are kept descriptive rather than forcing them into the drug-side peripheral2 / effect slots, which would mislead a reader into thinking they were drug-disposition compartments.