Skip to contents

Model and source

Kong 2025 gave a single oral microdose cocktail of five probe substrates to Chinese healthy volunteers (HV) and haemodialysis-dependent end-stage renal disease (ESRD) patients, and fitted one population PK model per drug to the pooled pre-dialysis concentrations. The five models were fitted independently, so they are packaged as five model files that share this article.

  • Citation: Kong W, Pan Y, Wu Y, Hu Y, Jiang Z, Tian X, Bi S, Wang S, Feng F, Jin Y, Li J, Li H, Wang Y, Liang H, Tang W, Liu D. Microdose Cocktail Study Reveals the Activity and Key Influencing Factors of OATP1B, P-Gp, BCRP, and CYP3A in End-Stage Renal Disease Patients. Clin Pharmacol Ther. 2025;117(5):1303-. doi:10.1002/cpt.3546.
  • Article: https://doi.org/10.1002/cpt.3546 (open access; PMC11993298)
Model Probe for Covariates on the final model
Kong_2025_pitavastatin OATP1B ESRD and gut Veillonella abundance on CL/F
Kong_2025_dabigatran intestinal P-gp (dose given as dabigatran etexilate) ESRD on CL/F
Kong_2025_rosuvastatin OATP1B / BCRP gut Clostridium XIVb abundance on CL/F; serum albumin on Vc/F
Kong_2025_midazolam CYP3A ESRD on CL/F
Kong_2025_atorvastatin CYP3A4 / OATP1B / P-gp / BCRP none

All five are two-compartment models with first-order absorption and first-order elimination, parameterised as apparent oral values (CL/F, Vc/F, Vp/F, Q/F). Doses are in ug and concentrations in pg/mL, the paper’s units.

Population

Fourteen HV (5 male, 9 female; mean age 35.6 years, BMI 22.4 kg/m^2, eGFR 106.6 mL/min/1.73 m^2) and ten ESRD patients on haemodialysis three times a week (7 male, 3 female; mean age 43.4 years, BMI 21.9 kg/m^2, eGFR 4 mL/min/1.73 m^2) were enrolled at Peking University Third Hospital (Kong 2025 Table 1). All ESRD patients had chronic glomerulonephritis; diabetes was an exclusion criterion. Serum albumin was 44.9 g/L (HV) and 42.1 g/L (ESRD). The cocktail (midazolam 10 ug, dabigatran etexilate 375 ug, pitavastatin 10 ug, rosuvastatin 50 ug, atorvastatin 100 ug) was taken fasted 24 h before dialysis. HV were sampled at 0.5-24 h and ESRD patients at 1, 4 and 24 h; the 28-h post-dialysis sample was not modelled. One ESRD patient was excluded from the atorvastatin analysis. The same information is available from readModelDb("Kong_2025_pitavastatin")()$population.

Source trace

Every ini() value carries an in-file source comment; this table collects them. “IIV” is the Table 3 IIV% column, converted to a log-normal variance with omega^2 = log(1 + CV^2).

Drug Parameter Value Source
PTV ka (1/h) 18, fixed Table 3 (RSE NA; bootstrap 18 (18, 18))
PTV CL/F (L/h) 16.5; IIV 9.60% Table 3
PTV ESRD on CL/F -0.381, as (1 + DIS) Table 3 ‘DISPTV on CL/F’; Eq. 1
PTV Veil on CL/F -0.677, as (exp(Veil)/1.08)^theta Table 3 ‘Veil on CL/F’; Eq. 1
PTV Vc/F, Vp/F (L) 25.2, 76.1 Table 3
PTV Q/F (L/h) 8.01; IIV 7.60% Table 3
DAB ka (1/h) 1.22; IIV 21.5% Table 3
DAB CL/F (L/h) 146; IIV 22.5% Table 3
DAB ESRD on CL/F -0.869, as (1 + DIS) Table 3 ‘DISDZB on CL/F’; Eq. 2
DAB Vc/F (L) 794; IIV 25.9% Table 3
DAB Vp/F (L) 429; IIV 86.0% Table 3
DAB Q/F (L/h) 92.5 Table 3
RSV ka (1/h) 0.30 Table 3
RSV CL/F (L/h) 43.1; IIV 50.8% Table 3
RSV Clos on CL/F -3.26, as (exp(Clos)/1.08)^theta Table 3 ‘Clos on CL/F’; Eq. 4
RSV Vc/F (L) 289; IIV 22.4% Table 3
RSV ALB on Vc/F -6.42, as (ALB/44)^theta Table 3 ‘ALB on Vc/F’ (form assumed, see below)
RSV Vp/F (L), Q/F (L/h) 2740, 48.6 Table 3
MDZ ka (1/h) 13, fixed Table 3 (RSE NA; bootstrap 13 (13, 13))
MDZ CL/F (L/h) 67.6; IIV 37.9% Table 3
MDZ ESRD on CL/F +0.94, as (1 + DIS) Table 3 ‘DISMDZ on CL/F’; Eq. 3
MDZ Vc/F (L) 195; IIV 42.1% Table 3
MDZ Vp/F (L), Q/F (L/h) 333, 34.7 Table 3
ATV ka (1/h) 7.53 Table 3
ATV CL/F (L/h) 363; IIV 48.0% Table 3; Eq. 5
ATV Vc/F (L) 835; IIV 99.3% Table 3
ATV Vp/F (L) 5830; IIV 62.6% Table 3
ATV Q/F (L/h) 3920; IIV 30.2% Table 3
all residual error propSd fixed at 0 not reported
all structure 2-compartment, first-order absorption Results, ‘PPK model’

Virtual cohort

Each drug is simulated in 100 HV and 100 ESRD subjects. Serum albumin is drawn from a normal distribution matching the Table 1 group mean and CV. The microbiome abundances are not reported per subject, so both are set to the value at which their effect equals 1 (log(1.08) = 0.077%); the Veillonella effect is explored separately below.

set.seed(20250601)
rxode2::rxSetSeed(20250601)

n_per_arm <- 100L
obs_times <- c(
  0, 0.25, 0.5, 0.75, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 16, 20, 24,
  30, 36, 48, 60, 72, 96, 120, 150
)
drugs <- tibble::tribble(
  ~drug, ~model, ~dose_ug,
  "PTV", "Kong_2025_pitavastatin", 10,
  "DAB", "Kong_2025_dabigatran", 375,
  "RSV", "Kong_2025_rosuvastatin", 50,
  "MDZ", "Kong_2025_midazolam", 10,
  "ATV", "Kong_2025_atorvastatin", 100
)

subjects <- tibble(
  id = seq_len(2 * n_per_arm),
  group = rep(c("HV", "ESRD"), each = n_per_arm),
  RENALIMP_ESRD = rep(c(0L, 1L), each = n_per_arm),
  ALB = c(
    rnorm(n_per_arm, 44.9, 0.08 * 44.9),
    rnorm(n_per_arm, 42.1, 0.04 * 42.1)
  ),
  GUT_VEILLONELLA_PCT = log(1.08),
  GUT_CLOSTRIDIUM_XIVB_PCT = log(1.08)
)

make_events <- function(dose_ug) {
  doses <- subjects |>
    mutate(time = 0, evid = 1L, amt = dose_ug, cmt = "depot")
  obs <- tidyr::crossing(subjects, time = obs_times) |>
    mutate(evid = 0L, amt = 0, cmt = "central")
  bind_rows(doses, obs) |>
    arrange(id, time, desc(evid))
}

Simulation

sim <- lapply(seq_len(nrow(drugs)), function(i) {
  ev <- make_events(drugs$dose_ug[i])
  rxode2::rxSolve(
    readModelDb(drugs$model[i]),
    events = ev,
    keep = c("group")
  ) |>
    as.data.frame() |>
    mutate(drug = drugs$drug[i])
}) |>
  bind_rows() |>
  mutate(
    drug = factor(drug, levels = drugs$drug),
    group = factor(group, levels = c("HV", "ESRD"))
  )
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ parameter labels from comments will be replaced by 'label()'

Replicate Figure 1

Kong 2025 Figure 1 shows the observed concentration-time curves by group over 0-24 h. The simulated median and 90% prediction interval follow.

sim |>
  filter(time > 0, time <= 24) |>
  group_by(drug, group, time) |>
  summarise(
    Q05 = quantile(Cc, 0.05),
    Q50 = median(Cc),
    Q95 = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50, colour = group, fill = group)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.2, colour = NA) +
  geom_line() +
  facet_wrap(~drug, scales = "free_y") +
  scale_y_log10() +
  labs(
    x = "Time after dose (h)", y = "Plasma concentration (pg/mL)",
    colour = NULL, fill = NULL,
    caption = "Replicates Figure 1 of Kong 2025 (median and 90% PI, 100 subjects per group)."
  )

PKNCA validation

AUC0-24 and Cmax are computed over 0-24 h (the window of Kong 2025 Table 2, which reports geometric means of the observed data). AUCinf is computed over the full 150-h simulation, the window of Kong 2025 Table S7, in which the authors simulated each participant with the final models.

sim_nca <- sim |>
  filter(!is.na(Cc)) |>
  mutate(treatment = paste(drug, group)) |>
  select(id, time, Cc, treatment)
sim_nca <- bind_rows(
  sim_nca,
  sim_nca |> distinct(id, treatment) |> mutate(time = 0, Cc = 0)
) |>
  distinct(id, treatment, time, .keep_all = TRUE) |>
  arrange(treatment, id, time)

dose_df <- tidyr::crossing(drugs, subjects |> select(id, group)) |>
  mutate(treatment = paste(drug, group), time = 0, amt = dose_ug) |>
  select(id, time, amt, treatment)

conc_obj <- PKNCA::PKNCAconc(sim_nca, Cc ~ time | treatment + id)
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | treatment + id)
intervals <- data.frame(
  start = c(0, 0),
  end = c(24, Inf),
  cmax = c(TRUE, FALSE),
  auclast = c(TRUE, FALSE),
  aucinf.obs = c(FALSE, TRUE)
)
nca_res <- PKNCA::pk.nca(
  PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals)
)
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in log(data$conc): NaNs produced
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in log(conc.2/conc.1): NaNs produced
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in log(data$conc): NaNs produced
#> Warning in assert_conc(conc, any_missing_conc = any_missing_conc): Negative
#> concentrations found
#> Warning in log(conc.2/conc.1): NaNs produced

Comparison against published NCA

Cmax and AUC0-24 references are the observed geometric means of Kong 2025 Table 2; AUCinf references are the geometric means of the authors’ own per-participant simulations (Table S7). Simulated values are medians.

published <- tibble::tribble(
  ~treatment, ~cmax, ~auclast, ~aucinf.obs,
  "PTV HV", 222, 500, 636,
  "PTV ESRD", 246, 922, 1107,
  "DAB HV", 303, 2233, 2533,
  "DAB ESRD", 384, 6872, 19847,
  "RSV HV", 73.0, 591, 1184,
  "RSV ESRD", 58.6, 656, 1656,
  "MDZ HV", 48.8, 121, 145,
  "MDZ ESRD", 16.6, 34.0, 73.0,
  "ATV HV", 23.6, 141, 255,
  "ATV ESRD", 22.0, 181, 282
)

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = nca_res,
  reference = published,
  by = "treatment",
  units = c(cmax = "pg/mL", auclast = "h*pg/mL", aucinf.obs = "h*pg/mL"),
  tolerance_pct = 20
)
knitr::kable(
  cmp,
  caption = "Simulated (median) vs. published NCA. * differs from the reference by >20%."
)
Simulated (median) vs. published NCA. * differs from the reference by >20%.
NCA parameter treatment Reference Simulated % diff
Cmax (pg/mL) PTV HV 222 325 +46.3%*
Cmax (pg/mL) PTV ESRD 246 341 +38.8%*
Cmax (pg/mL) DAB HV 303 299 -1.3%
Cmax (pg/mL) DAB ESRD 384 366 -4.8%
Cmax (pg/mL) RSV HV 73 68.9 -5.6%
Cmax (pg/mL) RSV ESRD 58.6 50 -14.6%
Cmax (pg/mL) MDZ HV 48.8 46.9 -3.8%
Cmax (pg/mL) MDZ ESRD 16.6 40.2 +142.2%*
Cmax (pg/mL) ATV HV 23.6 45.2 +91.6%*
Cmax (pg/mL) ATV ESRD 22 45 +104.8%*
AUC0-∞ (obs) (h*pg/mL) PTV HV 636 585 -8.0%
AUC0-∞ (obs) (h*pg/mL) PTV ESRD 1110 959 -13.3%
AUC0-∞ (obs) (h*pg/mL) DAB HV 2530 2580 +1.9%
AUC0-∞ (obs) (h*pg/mL) DAB ESRD 19800 19400 -2.2%
AUC0-∞ (obs) (h*pg/mL) RSV HV 1180 1220 +3.0%
AUC0-∞ (obs) (h*pg/mL) RSV ESRD 1660 1120 -32.7%*
AUC0-∞ (obs) (h*pg/mL) MDZ HV 145 147 +1.5%
AUC0-∞ (obs) (h*pg/mL) MDZ ESRD 73 77.2 +5.7%
AUC0-∞ (obs) (h*pg/mL) ATV HV 255 280 +9.9%
AUC0-∞ (obs) (h*pg/mL) ATV ESRD 282 314 +11.3%
AUClast (h*pg/mL) PTV HV 500 539 +7.9%
AUClast (h*pg/mL) PTV ESRD 922 824 -10.7%
AUClast (h*pg/mL) DAB HV 2230 2270 +1.7%
AUClast (h*pg/mL) DAB ESRD 6870 6040 -12.1%
AUClast (h*pg/mL) RSV HV 591 630 +6.6%
AUClast (h*pg/mL) RSV ESRD 656 564 -14.0%
AUClast (h*pg/mL) MDZ HV 121 132 +8.9%
AUClast (h*pg/mL) MDZ ESRD 34 73.8 +117.2%*
AUClast (h*pg/mL) ATV HV 141 189 +33.8%*
AUClast (h*pg/mL) ATV ESRD 181 174 -3.8%

The starred rows are explained by known features of the source rather than by the model code:

  • Cmax of pitavastatin and atorvastatin. The fixed ka of 18 1/h for pitavastatin and the ka of 7.53 1/h for atorvastatin give an early, sharp peak that the HV sampling schedule (first sample at 0.5 h) cannot capture. The authors’ own simulations show the same overshoot (Table S7 Cmax 313 and 317 pg/mL for pitavastatin, 45.1 and 52.8 pg/mL for atorvastatin).
  • ESRD Cmax and AUC0-24 of midazolam. The observed ESRD exposure is lower than the model predicts. The ESRD patients were sampled only at 1, 4 and 24 h, and the Table S7 simulations (AUCinf 73.0 h*pg/mL) agree with the model, not with the sparse observed AUC0-24.
  • HV AUC0-24 of atorvastatin. The model gives a higher exposure than the observed NCA (Table 2, 141 hpg/mL); the authors’ own per-participant simulations show the same (Table S7 AUClast 253 hpg/mL in HV).
  • ESRD AUCinf of rosuvastatin. CL/F depends on the per-subject Clostridium XIVb abundance, which the paper does not report; this cohort holds it at the reference value. The Table S7 ESRD CL/F (30.2 L/h vs 42.2 L/h in HV) implies higher abundance in the ESRD patients. The ESRD AUCinf of pitavastatin is low for the same reason (Veillonella), though within 20%.

Typical-value check against the authors’ simulations

AUCinf after a single oral dose of a linear model equals dose / (CL/F). The typical-value AUCinf of each model is compared with the Table S7 geometric means where the covariates of this cohort fully determine CL/F: all HV groups (microbiome at the reference value) and the ESRD groups of dabigatran and midazolam, whose only covariate is ESRD.

typical <- lapply(seq_len(nrow(drugs)), function(i) {
  ev <- make_events(drugs$dose_ug[i]) |>
    filter(id %in% c(1L, n_per_arm + 1L)) |>
    mutate(ALB = 44)
  rxode2::rxSolve(
    rxode2::zeroRe(readModelDb(drugs$model[i])),
    events = ev,
    keep = c("group")
  ) |>
    as.data.frame() |>
    group_by(group) |>
    summarise(cl = first(cl), .groups = "drop") |>
    mutate(drug = drugs$drug[i], aucinf = 1000 * drugs$dose_ug[i] / cl)
}) |>
  bind_rows() |>
  mutate(treatment = paste(drug, group)) |>
  left_join(published |> select(treatment, ref = aucinf.obs), by = "treatment") |>
  mutate(pct_diff = 100 * (aucinf - ref) / ref)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalka', 'etalcl', 'etalvc', 'etalvp'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etalq'
#> Warning: multi-subject simulation without without 'omega'

knitr::kable(
  typical |>
    select(treatment, cl, aucinf, ref, pct_diff) |>
    rename(
      "Group" = treatment,
      "Typical CL/F (L/h)" = cl,
      "Typical AUCinf (h*pg/mL)" = aucinf,
      "Table S7 AUCinf (h*pg/mL)" = ref,
      "% diff" = pct_diff
    ),
  digits = 1
)
Group Typical CL/F (L/h) Typical AUCinf (h*pg/mL) Table S7 AUCinf (h*pg/mL) % diff
PTV ESRD 10.2 979.1 1107 -11.6
PTV HV 16.5 606.1 636 -4.7
DAB ESRD 19.1 19606.8 19847 -1.2
DAB HV 146.0 2568.5 2533 1.4
RSV ESRD 43.1 1160.1 1656 -29.9
RSV HV 43.1 1160.1 1184 -2.0
MDZ ESRD 131.1 76.3 73 4.5
MDZ HV 67.6 147.9 145 2.0
ATV ESRD 363.0 275.5 282 -2.3
ATV HV 363.0 275.5 255 8.0

gated <- typical |>
  filter(group == "HV" | drug %in% c("DAB", "MDZ"))
# Deterministic: typical values against the authors' per-participant
# geometric means. Observed spread is -5% to +8%; a mis-transcribed CL/F or
# ESRD shift moves a row by tens of percent.
stopifnot(all(abs(gated$pct_diff) < 15))

Gut Veillonella effect on pitavastatin clearance

The Discussion of Kong 2025 states that, together with the direct ESRD effect, the Veillonella term reduces OATP1B-mediated pitavastatin clearance by 35% at the lowest observed abundance (0.015%), by 54% at 0.52%, and by up to 75% at the highest (1.42%). Those figures follow from Eq. 1 with the abundance in percent, which confirms the covariate units.

mod_ptv <- rxode2::rxode2(readModelDb("Kong_2025_pitavastatin"))
#> ℹ parameter labels from comments will be replaced by 'label()'
theta <- mod_ptv$theta
veil_reduction <- function(veil) {
  100 * (1 - (exp(veil) / 1.08)^theta[["e_gut_veillonella_pct_cl"]] *
    (1 + theta[["e_renalimp_esrd_cl"]]))
}
claims <- tibble(
  veil = c(0.015, 0.52, 1.42),
  paper = c(35, 54, 75),
  model = veil_reduction(veil)
)
knitr::kable(
  claims |>
    rename(
      "Veillonella (%)" = veil,
      "Paper reduction (%)" = paper,
      "Model reduction (%)" = model
    ),
  digits = c(3, 0, 1)
)
Veillonella (%) Paper reduction (%) Model reduction (%)
0.015 35 35.4
0.520 54 54.1
1.420 75 75.1
stopifnot(all(abs(claims$model - claims$paper) < 1.5))

tibble(veil = seq(0, 1.5, by = 0.01)) |>
  mutate(reduction = veil_reduction(veil)) |>
  ggplot(aes(veil, reduction)) +
  geom_line() +
  geom_point(data = claims, aes(veil, paper), colour = "firebrick") +
  labs(
    x = "Veillonella relative abundance (%)",
    y = "Reduction in pitavastatin CL/F vs HV (%)",
    caption = "ESRD patients; points are the values quoted in the Kong 2025 Discussion."
  )

Assumptions and deviations

  • IIV scale. Table 3 reports IIV as a percentage, and Eqs. 1-5 print the random effect as e^0.096, e^0.225 and so on. The maintainers read the percentage as a CV and convert it with omega^2 = log(1 + CV^2); for the large values (atorvastatin Vc/F 99.3%, dabigatran Vp/F 86.0%) this gives smaller variances than reading the percentage as omega directly.
  • Residual error. The paper does not report the residual error model or its magnitude. propSd is fixed at 0 in all five models, so simulations are individual predictions; add a residual error before using the models for VPCs against observed data.
  • Final estimates vs Eqs. 1-5. The equations print rounded bootstrap medians (e.g. CL/F 17 and exponent -0.68 for pitavastatin); the models use the Table 3 final estimates (16.5 and -0.677). The 1.08 normaliser of the microbiome terms appears only in the equations and is used as printed for both Veillonella and Clostridium XIVb.
  • Albumin on rosuvastatin Vc/F. Table 3 gives the coefficient -6.42 without a functional form. The maintainers encode it as the power form (ALB / 44)^-6.42, with 44 g/L the rounded pooled Table 1 mean (43.7 g/L); an exponential or linear form with that coefficient would be implausible over the observed 40-48 g/L range.
  • Fixed absorption rates. Pitavastatin ka (18 1/h) and midazolam ka (13 1/h) have no RSE and a degenerate bootstrap interval, so they are encoded with fixed(). Pitavastatin Q/F is printed with RSE 0 but has a non-degenerate bootstrap interval (6.6-9.4 L/h), so it is treated as estimated. Atorvastatin ka has a bootstrap median of 22 1/h against the estimate of 7.53 1/h; the estimate is used.
  • Dabigatran dose. The model was fitted to dabigatran concentrations with the dose expressed as the administered dabigatran etexilate (375 ug); the molecular-weight ratio and prodrug conversion are part of CL/F. Enter doses as dabigatran etexilate. The typical-value check above (dose/CL = 2568 vs Table S7 2533 h*pg/mL) confirms this.
  • ESRD and dialysis. Only pre-dialysis samples were modelled, so the models describe the interdialytic period. Table S8 shows that dialysis removes dabigatran substantially, which the models do not include.
  • Microbiome covariates. The per-subject abundances are not reported; the virtual cohort holds them at the reference value of 0.077%, which understates the ESRD exposure of pitavastatin and, plausibly, of rosuvastatin.
  • Errata. No correction notice was found for Kong 2025 as of 2026-11-09.