Skip to contents

Avatrombopag (Liu 2025)

Source: vignettes/articles/Liu_2025_avatrombopag.Rmd
Liu_2025_avatrombopag.Rmd

Model and source

  • Citation: Liu X, Chen L, Ju G, Li C, Liu B, Fei Y, Wang X, Gao Y, He Q, Zhu X, Ouyang D. (2025). Investigation of the ABCB1 Gene Polymorphism and Food Effects on the Avatrombopag Pharmacokinetics in Chinese Individuals: A Population Pharmacokinetic/ Pharmacodynamic Analysis. Pharmaceuticals 18(6):903. doi:10.3390/ph18060903
  • Description: Population PK model for avatrombopag in 92 healthy Chinese adults given a single 20 mg oral dose under fasting or high-fat-fed conditions in a two-sequence, four-period replicate bioequivalence trial. One-compartment disposition with first-order elimination preceded by a nine-transit-compartment absorption chain (dose -> depot -> transit1 .. transit9, transfer rate ktr = 9 / MTT) followed by first-order absorption (ka) from transit9 into the central compartment. A food-by-genotype co-effect acts on the absorption fraction Fa (relative bioavailability): Fa = 1 + COV1 * FED_HIGHFAT + COV2 * (1 - ABCB1_C1236T_HET) + COV3 * FED_HIGHFAT * (1 - ABCB1_C1236T_HET), so the reference Fa = 1 is a fasting ABCB1 (C1236T) TC heterozygote. Between-subject variability is estimated on CL/F, Vd/F, ka and MTT; inter-occasion variability is estimated on ka across the four study periods. The source paper additionally simulated platelet counts by linking this PK model to a previously published chronic-liver-disease platelet-count PK/PD life-span model (Nomoto 2018, J Clin Pharmacol 58:1629-1638); no PD parameter of that model is reported in Liu 2025, so only the PK layer is implemented here.
  • Article (open access): https://doi.org/10.3390/ph18060903

Avatrombopag is an orally active, second-generation thrombopoietin receptor agonist (TPO-RA) approved for thrombocytopenia in chronic liver disease and in chronic immune thrombocytopenia. It is a BCS Class II compound whose absorption is solubility-limited, so its exposure is unusually sensitive to food, and it is a substrate of both CYP2C9 and the P-glycoprotein efflux transporter encoded by ABCB1. Liu 2025 asks whether that combination of food and pharmacogenetic sensitivity is large enough to warrant dose individualisation in Chinese patients.

What this paper contains, and what is packaged here

Liu 2025 has three analytical layers:

  1. A non-compartmental analysis of a four-period bioequivalence trial, comparing exposure across food conditions, sexes, and CYP2C9 / ABCB1 genotypes (Tables 1, 2, and S1-S7).
  2. An original population PK model built in NONMEM 7.5 from all 5923 avatrombopag concentrations. Its structure and every parameter estimate are reported in Table 3. This is the model packaged as Liu_2025_avatrombopag and validated below.
  3. A PK/PD simulation of platelet-count trajectories. The PD layer is not this paper’s model: Methods 4.5 states that “a previously published PD model developed in chronic liver disease (CLD) patients [10] was adopted and linked to our PK model”, with a 32% slope reduction applied for East Asians. Reference [10] is Nomoto, Ferry and Hussein (2018), J Clin Pharmacol 58:1629-1638, doi:10.1002/jcph.1267. Liu 2025 reports no PD parameter of that model at all – no baseline, no lifespan, no slope, no turnover rate, no residual error – so the PD layer cannot be source-traced from anything on disk and is deliberately not implemented here. See “Assumptions and deviations” for the details, including which of the paper’s own PD numbers can be back-solved and which cannot.

Population

Liu 2025 pooled 92 healthy Chinese volunteers enrolled at the Clinical Trials Unit of Hunan Province People’s Hospital (Changsha, Hunan) into a single-centre, open-label, two-sequence, four-period replicate bioequivalence trial with a 14-day washout between periods. Each subject received a single 20 mg avatrombopag tablet with 200 mL of water in each period. Forty-seven subjects were dosed after an overnight fast and 45 were dosed roughly 30 min after an FDA-standard high-fat, high-calorie breakfast (800-1000 kcal, of which 500-600 kcal from fat); the two arms are disjoint subject groups with separately tabulated demographics (Table 1).

pop <- rxode2::rxode(readModelDb("Liu_2025_avatrombopag"))$population
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: some etas defaulted to non-mu referenced, possible parsing error: etaiov_ka_1, etaiov_ka_2, etaiov_ka_3, etaiov_ka_4
#> as a work-around try putting the mu-referenced expression on a simple line
tibble::tibble(Field = names(pop), Value = vapply(pop, paste, character(1), collapse = "; ")) |>
  knitr::kable(caption = "Population metadata carried in the model file (Liu 2025 Table 1, Results 2.1 and Methods 4.1-4.4).")
Population metadata carried in the model file (Liu 2025 Table 1, Results 2.1 and Methods 4.1-4.4).
Field Value
species human
n_subjects 92
n_studies 1
n_pk_observations 5923
age_range 18-45 years
age_median 28 years
weight_range mean 63.22 +/- 7.88 kg (fasting arm) and 60.95 +/- 6.32 kg (fed arm)
sex_female_pct 9.8
race_ethnicity Chinese (single-centre Chinese cohort; no further ethnic breakdown reported)
disease_state healthy volunteers
dose_range single 20 mg oral avatrombopag tablet with 200 mL water, repeated over four periods separated by a 14-day washout; 47 subjects dosed under an overnight fast and 45 dosed about 30 min after a high-fat, high-calorie meal
regions China (Clinical Trials Unit of Hunan Province People’s Hospital, Changsha, Hunan)
genotypes CYP2C9: 90% 1/1 normal metabolizer, 10% 1/3 intermediate metabolizer, no 3/3. ABCB1 (C1236T): 14% CC, 51% TC, 35% TT. ABCB1 (C3435T): 40% CC, 47% TC, 13% TT. ABCB1 (G2677T/A): 20% GG, 67% heterozygous, 13% homozygous variant. All variants were in Hardy-Weinberg equilibrium (Results 2.1, 2.2.2 and Table 1).
notes Open-label, single-centre, two-sequence, four-period replicate bioequivalence trial (IRB approval [2023]-32.1, Hunan Province People’s Hospital). Sampling at predose and 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 16, 24, 48, 72 and 96 h post dose; plasma avatrombopag by UPLC-MS/MS with an AVA-D8 internal standard, linear range 1.000-400.000 ng/mL. Estimation in NONMEM 7.5 with FOCE-I; stepwise covariate modelling (forward p < 0.05, backward p < 0.01); evaluated by GOF plots, VPC and non-parametric bootstrap (Table 3). Baseline demographics are Liu 2025 Table 1.

Plasma avatrombopag was measured by UPLC-MS/MS against an AVA-D8 internal standard over a 1.000-400.000 ng/mL calibration range, with samples at predose and 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 16, 24, 48, 72 and 96 h after each dose.

Source trace

Every structural element and every ini() value traces to Liu 2025 Table 3 or to the Methods text. Table 3’s covariate rows are labelled only COV1-COV3; their meaning comes from the table footnote, reproduced verbatim in the last three rows below.

Source trace for every structural element and ini() value in Liu_2025_avatrombopag.
Element Value Source location
One-compartment disposition, first-order elimination structure Results 2.3: ‘A one-compartment model with nine transit compartments, along with first-order absorption and elimination, was selected as the final model.’
Nine transit absorption compartments, ktr = 9 / MTT structure Results 2.3 (same sentence); ktr convention per nlmixr2lib lktr = n_transit / MTT
Exponential IIV, combined proportional + additive residual error structure Methods 4.4: ‘IIV was modeled using an exponential approach, while residual variability was tested using proportional, additive, and combined error models’
lcl = log(7.85) CL/F = 7.85 L/h (RSE 14%) Table 3, row ‘CL/F (L/h)’
lvc = log(199) Vd/F = 199 L (RSE 13%) Table 3, row ‘Vd/F (L)’
lka = log(0.582) ka = 0.582 /h (RSE 13%) Table 3, row ‘ka (/h)’
lmtt = log(1.33) MTT = 1.33 h (RSE 5%) Table 3, row ‘MTT (h)’
lfdepot = fixed(log(1)) reference Fa = 1 Table 3 footnote: ‘Fa = 1 x (1 + COV1 x FOOD + COV2 x ABCB1 + COV3 x FOOD x ABCB1)’
e_fed_highfat_fdepot = -0.0653 COV1 (RSE 111%) Table 3, row ‘COV1’; footnote ‘the effect of fed state and TC genotype on Fa’
e_abcb1hom_fdepot = -0.272 COV2 (RSE 15%) Table 3, row ‘COV2’; footnote ‘the effect of fast state and non-TC genotype on Fa’
e_fed_abcb1hom_fdepot = 0.177 COV3 (RSE 56%) Table 3, row ‘COV3’; footnote ‘the effect of fed state and non-TC genotype on Fa’
etalcl ~ 0.182 BSV on CL/F (RSE 35%); 44.7% CV Table 3, ‘Between-subject variability’ block, row ‘CL/F’
etalvc ~ 0.159 BSV on Vd/F (RSE 32%); 41.6% CV Table 3, ‘Between-subject variability’ block, row ‘Vd/F’
etalka ~ 0.504 BSV on ka (RSE 35%); 81.6% CV Table 3, ‘Between-subject variability’ block, row ‘KA’
etalmtt ~ 0.131 BSV on MTT (RSE 23%); 37.5% CV Table 3, ‘Between-subject variability’ block, row ‘MTT’
etaiov_ka_1..4 ~ 0.913 IOV on ka (RSE 12%); 122% CV Table 3, ‘Between-subject variability’ block, row ‘OCC [KA]’
propSd = 0.239 proportional residual error 23.90% (RSE 4%) Table 3, ‘Residual unexplained variability’ block, row ‘sigma prop (%)’
addSd = 1.18 additive residual error 1.18 ng/mL (RSE 3%) Table 3, ‘Residual unexplained variability’ block, row ‘sigma add (ng/mL)’
FOOD = 1 fed / 0 fasting covariate coding Table 3 footnote: ‘If in the fed state, FOOD is 1; fasting state is 0.’
ABCB1 = 1 homozygote / 0 heterozygote covariate coding Table 3 footnote: ‘If the genotype is heterozygote, ABCB1 is 0; homozygote is 1.’
Four occasions per subject OCC = 1..4 Methods 4.2: ‘This open-label, crossover study consisted of four periods with two sequences, separated by a 14-day washout.’

The absorption-fraction covariate model

Table 3’s footnote is the only place the covariate model appears, and its two indicator conventions are both non-obvious: FOOD = 1 marks the fed state, and ABCB1 = 1 marks a homozygote (either CC wild-type or TT variant) while ABCB1 = 0 marks the TC heterozygote. The canonical nlmixr2lib column ABCB1_C1236T_HET runs the other way (1 = heterozygote), so the model file writes (1 - ABCB1_C1236T_HET) wherever the paper writes ABCB1. The four strata evaluate to:

Absorption fraction Fa by stratum, from the Liu 2025 Table 3 footnote equation.
Stratum FED_HIGHFAT ABCB1_C1236T_HET Fa
Fasting, ABCB1 (C1236T) TC 0 1 1.0000
Fed, TC 1 1 0.9347
Fasting, non-TC (CC or TT) 0 0 0.7280
Fed, non-TC 1 0 0.8397

The reference stratum (Fa = 1) is therefore a fasting TC heterozygote, and the paper’s headline finding – “a 1.37-fold increase in Cmax and AUC … the comparison between the TC and non-TC under fasting state” (Results 2.4) – is exactly 1 / 0.728 = 1.374.

Virtual cohort

The single-dose cohort replicates the trial as designed: 47 fasting and 45 fed subjects, each contributing four 20 mg occasions, with ABCB1 (C1236T) genotypes assigned at the per-arm counts of Table 1 (fasting 7 CC / 28 TC / 12 TT; fed 6 CC / 19 TC / 20 TT). Both arms are well under the 200-per-arm cap.

set.seed(20250616)

mod <- rxode2::rxode(readModelDb("Liu_2025_avatrombopag"))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: some etas defaulted to non-mu referenced, possible parsing error: etaiov_ka_1, etaiov_ka_2, etaiov_ka_3, etaiov_ka_4
#> as a work-around try putting the mu-referenced expression on a simple line

# Table 1 ABCB1 (C1236T) counts, per arm. TC = heterozygote = 1.
het_fasting <- sample(c(rep(0L, 7 + 12), rep(1L, 28)))   # 47 subjects
het_fed     <- sample(c(rep(0L, 6 + 20), rep(1L, 19)))   # 45 subjects

subjects <- bind_rows(
  tibble::tibble(subject = 1:47,  arm = "Fasting", FED_HIGHFAT = 0, ABCB1_C1236T_HET = het_fasting),
  tibble::tibble(subject = 48:92, arm = "Fed",     FED_HIGHFAT = 1, ABCB1_C1236T_HET = het_fed)
)

# Liu 2025 Methods 4.2 sampling schedule.
sample_times <- c(0, 1:12, 16, 24, 48, 72, 96)

# One record per subject x occasion x nominal time. The four periods are
# modelled as four independent single-dose profiles (14-day washout and a
# 17.6 h terminal half-life, so no carry-over), which is what makes the
# OCC[KA] inter-occasion variability identifiable. Each subject-occasion gets
# its own rxode2 `id` so that occasions are solved independently; the
# subject-level BSV is therefore shared only through the covariate values,
# which is the conservative reading for a between-subject food design.
profiles <- tidyr::expand_grid(subjects, OCC = 1:4) |>
  mutate(id = dplyr::row_number())

ev <- profiles |>
  tidyr::expand_grid(time = sample_times) |>
  mutate(
    evid = ifelse(time == 0, 1L, 0L),
    amt  = ifelse(time == 0, 20, NA_real_),
    cmt  = ifelse(time == 0, "depot", "central")
  ) |>
  select(id, time, evid, amt, cmt, FED_HIGHFAT, ABCB1_C1236T_HET, OCC) |>
  arrange(id, time, desc(evid))

c(subjects = nrow(subjects), profiles = nrow(profiles), records = nrow(ev))
#> subjects profiles  records 
#>       92      368     6624

Each of the 92 subjects contributes four occasions, so the simulation carries 368 single-dose profiles – the same 5923-observation scale as the analysis dataset.

sim <- rxode2::rxSolve(
  mod, ev,
  returnType = "data.frame",
  addDosing  = FALSE,
  atol = 1e-10, rtol = 1e-10
) |>
  left_join(profiles |> select(id, arm, subject), by = "id")

Replicating the observed concentration-time profile

Liu 2025 Figure 3 is the visual predictive check of the final PK model against all 5923 observations. The simulated median and 5th/95th percentiles below are the model’s own prediction interval for the same 20 mg single-dose design.

sim |>
  filter(!is.na(Cc), time > 0) |>
  group_by(arm, time) |>
  summarise(
    lo = quantile(Cc, 0.05), md = median(Cc), hi = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, md, ymin = lo, ymax = hi, colour = arm, fill = arm)) +
  geom_ribbon(alpha = 0.2, colour = NA) +
  geom_line(linewidth = 0.8) +
  scale_y_log10() +
  labs(x = "Time after dose (h)", y = "Avatrombopag (ng/mL)",
       colour = "Arm", fill = "Arm") +
  theme_bw()
Simulated 20 mg single-dose avatrombopag concentrations by food arm (median with 5th-95th percentile band). Compare with the VPC of Figure 3 of Liu 2025.

Simulated 20 mg single-dose avatrombopag concentrations by food arm (median with 5th-95th percentile band). Compare with the VPC of Figure 3 of Liu 2025.

PKNCA validation 

sim_nca <- sim |>
  filter(!is.na(Cc)) |>
  transmute(profile = id, arm, time, Cc)

# Guarantee a time-zero row per profile. Filtering on time > 0 or Cc > 0 would
# drop it and trigger PKNCA's "AUC range starting before the first measurement"
# warning on every profile.
sim_nca <- bind_rows(
  sim_nca,
  sim_nca |> distinct(profile, arm) |> mutate(time = 0, Cc = 0)
) |>
  distinct(profile, time, .keep_all = TRUE) |>
  arrange(profile, time)

conc_obj <- PKNCA::PKNCAconc(sim_nca, Cc ~ time | arm + profile)

dose_df <- sim_nca |> distinct(profile, arm) |> mutate(time = 0, amt = 20)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | arm + profile)

intervals <- data.frame(
  start = 0, end = Inf,
  cmax = TRUE, tmax = TRUE, auclast = TRUE, aucinf.obs = TRUE, half.life = TRUE
)

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

Comparison against the published non-compartmental analysis

Liu 2025 Table 2 reports the observed NCA parameters for the whole fasting (n = 47) and fed (n = 45) arms, as mean +/- SD except for tmax, which is a median. The simulated arm means below are computed over the same design and the same 18-point sampling schedule.

published_nca <- tibble::tribble(
  ~arm,      ~cmax, ~tmax, ~auclast, ~aucinf.obs, ~half.life,
  "Fasting", 80.44, 7.02,  2335.67,  2428.42,     18.92,
  "Fed",     85.06, 6.00,  2389.39,  2478.66,     18.18
)

sim_summary <- nca_res |>
  as.data.frame() |>
  filter(PPTESTCD %in% c("cmax", "tmax", "auclast", "aucinf.obs", "half.life")) |>
  group_by(arm, PPTESTCD) |>
  summarise(
    PPORRES = if (PPTESTCD[1] == "tmax") median(PPORRES) else mean(PPORRES),
    .groups = "drop"
  )

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated     = sim_summary,
  reference     = published_nca,
  by            = "arm",
  params        = c("cmax", "tmax", "auclast", "aucinf.obs", "half.life"),
  units         = c(cmax = "ng/mL", tmax = "h", auclast = "ng*h/mL",
                    aucinf.obs = "ng*h/mL", half.life = "h"),
  tolerance_pct = 20
)

knitr::kable(
  cmp,
  digits  = 2,
  caption = paste(
    "Simulated versus published non-compartmental parameters after a single",
    "20 mg oral dose (Liu 2025 Table 2). tmax is compared as a median, the",
    "rest as arithmetic means. * marks a discrepancy greater than 20%."
  )
)
Simulated versus published non-compartmental parameters after a single 20 mg oral dose (Liu 2025 Table 2). tmax is compared as a median, the rest as arithmetic means. * marks a discrepancy greater than 20%.
NCA parameter arm Reference Simulated % diff
Cmax (ng/mL) Fasting 80.4 73.1 -9.1%
Cmax (ng/mL) Fed 85.1 75.3 -11.4%
Tmax (h) Fasting 7.02 6 -14.5%
Tmax (h) Fed 6 7 +16.7%
AUC0-∞ (obs) (ng*h/mL) Fasting 2430 2490 +2.6%
AUC0-∞ (obs) (ng*h/mL) Fed 2480 2570 +3.6%
AUClast (ng*h/mL) Fasting 2340 2270 -2.8%
AUClast (ng*h/mL) Fed 2390 2300 -3.7%
t½ (h) Fasting 18.9 21.4 +13.3%
t½ (h) Fed 18.2 22.1 +21.3%*

The model was fitted to these data, so agreement is expected – but it is not guaranteed by construction, because the food effect enters only through the absorption fraction while Table 2’s arm difference also reflects the different ABCB1 genotype mix of the two arms and the very large inter-occasion variability on ka. Exposure agrees to within 4% on both AUC metrics and both arms; Cmax runs about 10% low and tmax lands within one nominal sampling interval.

The only starred row is the terminal half-life in the fed arm (+21.3%, with +13.3% in the fasting arm). This is a distributional artefact of comparing arithmetic means of a right-skewed NCA statistic, not a structural error: the model’s typical-value terminal half-life is exactly ln(2) / (CL/F / Vd/F).

cl <- 7.85; vc <- 199
tibble::tibble(
  Quantity = c("Typical-value t1/2 from Table 3 (ln2 / (CL/F / Vd/F))",
               "Published mean t1/2, fasting arm (Table 2)",
               "Published mean t1/2, fed arm (Table 2)",
               "Simulated median t1/2 across all profiles"),
  `Value (h)` = c(
    log(2) / (cl / vc),
    18.92, 18.18,
    nca_res |> as.data.frame() |> filter(PPTESTCD == "half.life") |>
      pull(PPORRES) |> median()
  )
) |>
  knitr::kable(digits = 2,
               caption = "Terminal half-life: the typical value and the simulated median both sit within the published means; only the arithmetic mean over a 44.7%-CV clearance distribution is inflated.")
Terminal half-life: the typical value and the simulated median both sit within the published means; only the arithmetic mean over a 44.7%-CV clearance distribution is inflated.
Quantity Value (h)
Typical-value t1/2 from Table 3 (ln2 / (CL/F / Vd/F)) 17.57
Published mean t1/2, fasting arm (Table 2) 18.92
Published mean t1/2, fed arm (Table 2) 18.18
Simulated median t1/2 across all profiles 18.94

Replicating Table 4: the simulated dosing regimens

Liu 2025 Results 2.4 and Table 4 simulate the two label regimens – 60 mg once daily for five days for patients with a baseline platelet count below 40 x 10^9/L, and 40 mg once daily for five days for 40-50 x 10^9/L – for each combination of ABCB1 (C1236T) stratum and food state, “using typical individual parameters”. This is the sharpest available test of the covariate model, because all eight cells share one set of structural parameters and differ only through Fa.

strata <- tibble::tribble(
  ~group,           ~FED_HIGHFAT, ~ABCB1_C1236T_HET,
  "Non-TC Fasting", 0,            0,
  "Non-TC Fed",     1,            0,
  "TC Fasting",     0,            1,
  "TC Fed",         1,            1
)

tv <- rxode2::zeroRe(mod)
#> Warning: some etas defaulted to non-mu referenced, possible parsing error: etaiov_ka_1, etaiov_ka_2, etaiov_ka_3, etaiov_ka_4
#> as a work-around try putting the mu-referenced expression on a simple line

simulate_regimen <- function(dose, fed, het) {
  e <- rxode2::et(amt = dose, ii = 24, until = 96, cmt = "depot") |>
    rxode2::et(seq(96, 120, by = 0.05), cmt = "central")
  d <- as.data.frame(e)
  d$FED_HIGHFAT <- fed
  d$ABCB1_C1236T_HET <- het
  d$OCC <- 1
  s <- rxode2::rxSolve(tv, d, returnType = "data.frame", atol = 1e-10, rtol = 1e-10)
  s <- s[!is.na(s$Cc) & s$time >= 96, ]
  list(
    peak   = max(s$Cc),
    c24    = s$Cc[which.min(abs(s$time - 120))],
    auctau = sum(diff(s$time) * (head(s$Cc, -1) + tail(s$Cc, -1)) / 2)
  )
}

published_t4 <- tibble::tribble(
  ~dose, ~group,           ~cmax_pub, ~auc_pub,
  60,    "Non-TC Fasting", 160.92,    5480.21,
  60,    "Non-TC Fed",     185.54,    6321.18,
  60,    "TC Fasting",     220.88,    7527.41,
  60,    "TC Fed",         206.50,    7035.84,
  40,    "Non-TC Fasting", 107.24,    3653.12,
  40,    "Non-TC Fed",     123.76,    4214.01,
  40,    "TC Fasting",     147.41,    5018.61,
  40,    "TC Fed",         137.80,    4691.04
)

t4 <- published_t4 |>
  left_join(strata, by = "group") |>
  rowwise() |>
  mutate(res = list(simulate_regimen(dose, FED_HIGHFAT, ABCB1_C1236T_HET))) |>
  ungroup() |>
  mutate(
    peak   = vapply(res, `[[`, numeric(1), "peak"),
    c24    = vapply(res, `[[`, numeric(1), "c24"),
    auctau = vapply(res, `[[`, numeric(1), "auctau"),
    pct_auc  = 100 * (auctau - auc_pub) / auc_pub,
    pct_c24  = 100 * (c24 - cmax_pub) / cmax_pub,
    pct_peak = 100 * (peak - cmax_pub) / cmax_pub
  )
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'

t4 |>
  select(
    "Dose (mg)"                       = dose,
    "Group"                           = group,
    "AUCtau published (ng*h/mL)"      = auc_pub,
    "AUCtau simulated (ng*h/mL)"      = auctau,
    "AUC diff (%)"                    = pct_auc,
    "Cmax published (ng/mL)"          = cmax_pub,
    "C at 24 h post dose (ng/mL)"     = c24,
    "C24 diff (%)"                    = pct_c24,
    "True model peak (ng/mL)"         = peak,
    "Peak diff (%)"                   = pct_peak
  ) |>
  knitr::kable(
    digits  = 1,
    caption = paste(
      "Replication of Liu 2025 Table 4. The published AUCtau is reproduced to",
      "within 0.6% in every cell. The published 'Cmax' is NOT the model peak",
      "(uniformly about 1.88-fold lower); it matches the concentration 24 h",
      "after the last dose to within 3.2%, i.e. it is a value read off a",
      "daily output grid. See 'Assumptions and deviations'."
    )
  )
Replication of Liu 2025 Table 4. The published AUCtau is reproduced to within 0.6% in every cell. The published ‘Cmax’ is NOT the model peak (uniformly about 1.88-fold lower); it matches the concentration 24 h after the last dose to within 3.2%, i.e. it is a value read off a daily output grid. See ‘Assumptions and deviations’.
Dose (mg) Group AUCtau published (ng*h/mL) AUCtau simulated (ng*h/mL) AUC diff (%) Cmax published (ng/mL) C at 24 h post dose (ng/mL) C24 diff (%) True model peak (ng/mL) Peak diff (%)
60 Non-TC Fasting 5480.2 5509.0 0.5 160.9 156.0 -3.1 302.0 87.7
60 Non-TC Fed 6321.2 6354.3 0.5 185.5 179.9 -3.0 348.4 87.8
60 TC Fasting 7527.4 7567.3 0.5 220.9 214.2 -3.0 414.9 87.8
60 TC Fed 7035.8 7073.2 0.5 206.5 200.2 -3.0 387.8 87.8
40 Non-TC Fasting 3653.1 3672.7 0.5 107.2 104.0 -3.0 201.3 87.8
40 Non-TC Fed 4214.0 4236.2 0.5 123.8 119.9 -3.1 232.2 87.7
40 TC Fasting 5018.6 5044.9 0.5 147.4 142.8 -3.1 276.6 87.6
40 TC Fed 4691.0 4715.4 0.5 137.8 133.5 -3.1 258.5 87.6 
t4 |>
  mutate(ratio_auc = auc_pub / auctau, ratio_c24 = cmax_pub / c24,
         ratio_peak = cmax_pub / peak) |>
  summarise(
    `AUCtau published / simulated (min)`  = min(ratio_auc),
    `AUCtau published / simulated (max)`  = max(ratio_auc),
    `Cmax published / C(24 h) (min)`      = min(ratio_c24),
    `Cmax published / C(24 h) (max)`      = max(ratio_c24),
    `Cmax published / true peak (min)`    = min(ratio_peak),
    `Cmax published / true peak (max)`    = max(ratio_peak)
  ) |>
  tidyr::pivot_longer(everything(), names_to = "Quantity", values_to = "Ratio") |>
  knitr::kable(
    digits  = 4,
    caption = paste(
      "The published-to-simulated ratios are near-constant across all eight",
      "cells for both readings, which is what identifies the Table 4 'Cmax'",
      "column as a 24-hour-grid concentration rather than a modelling error."
    )
  )
The published-to-simulated ratios are near-constant across all eight cells for both readings, which is what identifies the Table 4 ‘Cmax’ column as a 24-hour-grid concentration rather than a modelling error.
Quantity Ratio
AUCtau published / simulated (min) 0.9947
AUCtau published / simulated (max) 0.9948
Cmax published / C(24 h) (min) 1.0310
Cmax published / C(24 h) (max) 1.0322
Cmax published / true peak (min) 0.5324
Cmax published / true peak (max) 0.5330

Reproducing the paper’s own PK figure

Liu 2025 Figure 4 plots median PK/PD profiles for the four strata at 40 and 60 mg. The PK panels are reproduced here from typical-value profiles.

profile_grid <- tidyr::expand_grid(dose = c(40, 60), strata) |>
  rowwise() |>
  mutate(prof = list({
    e <- rxode2::et(amt = dose, ii = 24, until = 96, cmt = "depot") |>
      rxode2::et(seq(0, 168, by = 0.5), cmt = "central")
    d <- as.data.frame(e)
    d$FED_HIGHFAT <- FED_HIGHFAT
    d$ABCB1_C1236T_HET <- ABCB1_C1236T_HET
    d$OCC <- 1
    s <- rxode2::rxSolve(tv, d, returnType = "data.frame")
    s[!is.na(s$Cc), c("time", "Cc")]
  })) |>
  ungroup() |>
  tidyr::unnest(prof)
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalka', 'etalmtt', 'etaiov_ka_1', 'etaiov_ka_2', 'etaiov_ka_3', 'etaiov_ka_4'

ggplot(profile_grid, aes(time, Cc, colour = group)) +
  geom_line(linewidth = 0.7) +
  facet_wrap(~ paste0(dose, " mg QD x 5 days")) +
  labs(x = "Time (h)", y = "Avatrombopag (ng/mL)", colour = "Stratum") +
  theme_bw()
Typical-value avatrombopag concentration-time profiles over the five-day 40 mg and 60 mg once-daily regimens, by ABCB1 (C1236T) stratum and food state. Replicates the PK panels of Figure 4 of Liu 2025.

Typical-value avatrombopag concentration-time profiles over the five-day 40 mg and 60 mg once-daily regimens, by ABCB1 (C1236T) stratum and food state. Replicates the PK panels of Figure 4 of Liu 2025.

Assumptions and deviations

  • The platelet-count PD layer is not implemented. Liu 2025 Methods 4.5 adopts the chronic-liver-disease platelet PK/PD model of Nomoto 2018 (doi:10.1002/jcph.1267) wholesale and reports none of its parameters – no baseline, lifespan, transit rate, drug-effect slope, variance, or residual error appears anywhere in Liu 2025 or its supplement. Implementing it would mean transcribing another paper’s model from a one-sentence citation, which is exactly the secondary-source extraction this library forbids. Nomoto 2018 is closed access (Unpaywall is_oa: false, no PMC record) and has been registered for acquisition so the CLD platelet model can be extracted properly, from its own source, as its own model file.
  • What the PD gap does and does not cost. The PK layer is complete and self-contained, and Table 4’s PK columns are fully reproduced above. Table 4’s PD columns are not reproducible here, but they are also self-consistent with a purely additive drug-induced platelet increment on a fixed baseline: regressing the tabulated platelet Cmax on Fa across the four strata gives an intercept of 30.9 x 10^9/L at 60 mg and 42.1 x 10^9/L at 40 mg, matching the two label cohorts’ baseline platelet bands (< 40 x 10^9/L and 40-50 x 10^9/L). That is consistent with, but does not pin down, the Nomoto structure, which is why nothing was encoded from it.
  • Table 4’s “Cmax” column is not a peak concentration. The model reproduces Table 4’s AUCtau to within 0.6% in all eight cells – and that alone confirms CL/F, the Fa covariate model, and dose proportionality – but the “Cmax” column is uniformly 1.88-fold below the model’s true peak. It instead matches the concentration 24 h after the last dose to within 3.2% in all eight cells, with a near-constant ratio, which identifies it as a value read off the daily output grid used for the 960-hour platelet simulation rather than a true Cmax. The model’s true single-dose peak is independently corroborated by Table 2’s observed NCA (simulated 20 mg typical-value Cmax about 82 ng/mL versus observed 80.44 fasting / 85.06 fed ng/mL), so this is a reporting artefact in Table 4, not a model-implementation error. The 40 mg rows of Table 4 are exactly two-thirds of the 60 mg rows, confirming the simulation was linear in dose.
  • Transit-chain convention. Results 2.3 gives “nine transit compartments” and Table 3 gives MTT = 1.33 h, but does not state whether the dose compartment is counted in the mean transit time. The model uses the nlmixr2lib convention ktr = n_transit / MTT with n_transit = 9, i.e. the chain depot -> transit1 -> ... -> transit9 has nine ktr steps whose mean duration is exactly the reported 1.33 h, followed by first-order absorption at ka from transit9 into central. The Savic (n + 1) / MTT alternative changes only the shape of the Erlang transit-time distribution, not its mean, and shifts simulated tmax by under 0.15 h.
  • Inter-occasion variability encoding. Table 3 reports a single OCC [KA] variance shared across occasions – the NONMEM $OMEGA BLOCK(1) SAME idiom. nlmixr2 has no SAME shortcut, so four etas are declared and occasions 2-4 are fixed to the occasion-1 estimate. For single-occasion simulations pass OCC = 1.
  • ABCB1 indicator orientation. Liu 2025 codes its ABCB1 covariate as 1 = homozygote (CC or TT, pooled) and 0 = TC heterozygote. The canonical column ABCB1_C1236T_HET has the opposite orientation, so the model writes (1 - ABCB1_C1236T_HET). Pooling CC with TT is the paper’s own finding (Results 2.2.2), not an extraction convenience.
  • Between-subject variability scale. Table 3’s BSV rows are unlabelled as to scale. They are read as variances of an exponential IIV model, per Methods 4.4. Reading 0.182 as a variance gives 44.7% CV on CL/F, which sits between the 62.4% (fasting) and 36.6% (fed) CL/F variability the Discussion reports for the two arms; reading it as an SD would give 18.3% CV, far below the 62% coefficient of variation of the observed fasting AUC in Table 2.
  • Screened but unretained covariates. Age, body weight, sex, ALT, AST, CYP2C9 phenotype, ABCB1 (C3435T) and ABCB1 (G2677T/A) were all screened (Methods 4.4) and none entered the final model. They are recorded in the model file’s covariatesDataExcluded list rather than covariateData, so the provenance of the covariate screen survives without creating unused covariate declarations. Notably the 1.70-fold CYP2C9 intermediate-metabolizer exposure increase reported in the abstract is an NCA finding, not a model covariate.
  • The supplement was not obtainable. MDPI’s supplementary endpoint (https://www.mdpi.com/article/10.3390/ph18060903/s1) returns HTTP 403 to automated requests. Tables S1-S7 contain only the statistical comparisons behind Results 2.2; every model parameter is in the main-text Table 3, so nothing needed for this extraction is missing.
  • Study design in the cohort. The four crossover periods are simulated as four independent single-dose occasions. With a 14-day washout and a 17.6 h terminal half-life there is no measurable carry-over, and this is what makes the OCC [KA] inter-occasion variability identifiable.