Skip to contents
library(nlmixr2lib)
library(PKNCA)
#> 
#> Attaching package: 'PKNCA'
#> The following object is masked from 'package:stats':
#> 
#>     filter
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(ggplot2)

The paper

Merali et al. (2023) report the phase I study NCT01707394 (CV185118), in which 49 children from birth to <18 years at risk of venous or arterial thrombosis received a single oral dose of apixaban (0.1 mg sprinkle capsule in the neonate; 0.4 mg/mL oral solution otherwise; 1.08 to 2.43 mg/m^2). A population PK model was fitted to these data pooled with eight adult phase I studies and a second paediatric phase I study (Table S2): 254 subjects and 5,220 concentrations.

The final model (Table 2 and the displayed equations of the Methods) is two-compartment with first-order absorption and elimination. CL/F and Vc/F scale with baseline body weight (power exponents 0.71 and 0.879, reference 67.4 kg), CL/F carries a fixed CYP3A4 maturation function of age taken from Bouillon-Pichault 2011, Q/F is a power function of age (reference 25 years), ka is 142% higher between 9 months and 18 years, relative bioavailability falls with dose above 2.5 mg, and the neonatal capsule has a fixed relative bioavailability of 1.1. The paper also reports a linear relationship between plasma anti-factor Xa activity and apixaban concentration (Figure 4), which the model file carries as the algebraic output antiFXa.

mod <- rxode2::rxode2(nlmixr2lib::modellib("Merali_2023_apixaban"))
modTyp <- rxode2::zeroRe(mod)
mod$state
#> [1] "depot"       "central"     "peripheral1"

Population

From Table 1 (the 49 dosed paediatric subjects of CV185118; mean (SD)):

Age group n Age Weight (kg) BSA (m^2) Dose
Birth to 27 days 1 9 days at enrolment, 15 days at dosing 3.1 0.2 0.1 mg capsule
28 days to <9 months 11 126.9 (67.2) days 4.9 (1.7) 0.3 (0.1) 1.08 mg/m^2
9 months to <2 years 9 13.3 (4.2) months 8.3 (2.4) 0.4 (0.1) 1.08 (n = 6) or 2.43 (n = 3) mg/m^2
2 to <6 years 8 3.6 (1.2) years 15.9 (4.7) 0.7 (0.1) 1.17 mg/m^2
6 to <12 years 10 7.4 (1.2) years 29.8 (11.1) 1.0 (0.2) 1.80 mg/m^2
12 to <18 years 10 13.7 (1.4) years 49.9 (17.1) 1.5 (0.3) 2.19 mg/m^2

57% were female, 86% White, 10% Black and 4% other race. One 6-to-<12-year-old who vomited within 45 minutes of dosing was excluded from the PK analysis (n = 48). The pooled adult studies enrolled healthy volunteers (mean ages 26.7 to 42.0 years, mean weights 63.1 to 95.6 kg) dosed 2.5 to 50 mg (Table S2).

Source trace

Quantity Source
Two-compartment, first-order absorption and elimination Methods “Population PK analysis”; Results “PK results”
Covariate equations for ka, CL/F, Vc/F, Vp/F, Q/F, TVF1 and F1 Methods, displayed equations
PED9M definition (1 for 9 months to <18 years, else 0) Methods, text after the equations
lka, lcl, lvc, lq, lvp Table 2, theta1 to theta5
lgamma (theta6), logitifdepot50 (theta7) Table 2
e_wt_vc (theta8), e_wt_cl (theta9), e_age_q (theta10), e_ped9m_ka (theta11) Table 2
cyp3a4_hill (theta12), cyp3a4_t50 (theta13), cyp3a4_birth (theta15), all FIXED Table 2; maturation from Bouillon-Pichault 2011 (reference 10)
fdepot_capsule (theta14, FIXED) Table 2; Methods text “equal to 1.1 for subjects receiving neonatal formulation”
etalka, block etalcl / etalvc / etalvp / etalq Table 2 random effects (variances and covariances)
expSd Table 2 residual ‘Exponential’ sigma = 0.0844 (variance); SD = sqrt(0.0844)
lslope_antiFXa Discussion: “overall slope estimate of 0.0155 IU/ng”
lrbase_antiFXa Digitised from the regression line of Figure 4 (see below)

Deterministic checks

These checks use typical values only (zeroRe()), so both sides of every comparison share the same parameters and no cohort is involved.

# One subject per row of `cov`; dose into depot at time 0 and observations of
# the central compartment at `obsTimes`.
mkEvents <- function(cov, amt, obsTimes) {
  do.call(rbind, lapply(seq_len(nrow(cov)), function(i) {
    covRow <- cov[i, , drop = FALSE]
    dose <- data.frame(id = i, time = 0, amt = amt[i], evid = 1, cmt = "depot")
    obs <- data.frame(
      id = i, time = obsTimes, amt = 0, evid = 0, cmt = "central"
    )
    ev <- rbind(dose, obs)
    cbind(ev, covRow[rep(1, nrow(ev)), , drop = FALSE], row.names = NULL)
  }))
}

Relative bioavailability and maturation

grid <- data.frame(
  WT = 67.4,
  AGE = c(0, 0.244, 25, 25, 25, 25),
  DOSE_APIXABAN_MG = c(2.5, 2.5, 2.5, 50, 10, 2.5),
  FORM_CAPSULE = c(0, 0, 0, 0, 0, 1)
)
gridSim <- rxode2::rxSolve(
  modTyp, mkEvents(grid, amt = grid$DOSE_APIXABAN_MG, obsTimes = 1),
  returnType = "data.frame"
)
#> ℹ omega/sigma items treated as zero: 'etalka', 'etalcl', 'etalvc', 'etalvp', 'etalq'
#> Warning: multi-subject simulation without without 'omega'
gridRes <- cbind(grid, gridSim[, c("fmat", "frel", "fform")])
knitr::kable(
  gridRes |>
    dplyr::rename(
      "Weight (kg)" = WT, "Age (y)" = AGE, "Dose (mg)" = DOSE_APIXABAN_MG,
      "Neonatal capsule" = FORM_CAPSULE, "Maturation fraction" = fmat,
      "Dose F1" = frel, "Formulation F1" = fform
    ),
  digits = 4
)
Weight (kg) Age (y) Dose (mg) Neonatal capsule Maturation fraction Dose F1 Formulation F1
67.4 0.000 2.5 0 0.0500 1.0000 1.0
67.4 0.244 2.5 0 0.5250 1.0000 1.0
67.4 25.000 2.5 0 0.9801 1.0000 1.0
67.4 25.000 50.0 0 0.9801 0.6142 1.0
67.4 25.000 10.0 0 0.9801 0.9245 1.0
67.4 25.000 2.5 1 0.9801 1.0000 1.1

# Exact identities of the printed equations:
#   maturation fraction = FBIRTH (0.05) at birth and (1 + 0.05) / 2 at the
#   age of half-maturation, 0.244 y;
#   TVF1 = 1 at 2.5 mg and 1 - expit(-0.465) at the 50 mg maximum dose;
#   F1 for the neonatal formulation = 1.1.
stopifnot(
  abs(gridRes$fmat[1] - 0.05) < 1e-10,
  abs(gridRes$fmat[2] - 0.525) < 1e-10,
  abs(gridRes$frel[3] - 1) < 1e-10,
  abs(gridRes$frel[4] - (1 - 1 / (1 + exp(0.465)))) < 1e-10,
  abs(gridRes$fform[6] - 1.1) < 1e-10
)

The maturation function is still below 1 in adults (0.980 at 25 years), so exp(lcl) = 4.7 L/h is the CL/F of a hypothetical fully mature 67.4-kg subject; the typical 25-year-old 67.4-kg adult has CL/F 4.6 L/h. At the paediatric doses (0.1 to about 3.2 mg) the dose term is inert below 2.5 mg and reduces F1 by under 2% at the adolescent mean dose of 3.22 mg.

Typical parameters by age group against Table 3

Table 3 gives the geometric means of the individual (post hoc) ka, CL/F and Vc/F by age group. The model’s typical values at each group’s mean age and weight (Table 1) should land near them. This is a structural check: the individual estimates carry random effects and the covariates vary within a group, so agreement is expected to tens of percent, not exactly, but a mis-transcribed exponent, reference weight or maturation constant moves the young groups by a factor of two or more.

groups <- tibble::tribble(
  ~group, ~AGE, ~WT, ~ka_pub, ~cl_pub, ~vc_pub,
  "Birth to 27 days", 15 / 365.25, 3.1, 0.55, 0.10, 1.82,
  "28 days to <9 months", 126.9 / 365.25, 4.9, 0.72, 0.35, 2.69,
  "9 months to <2 years", 13.3 / 12, 8.3, 1.34, 0.76, 4.61,
  "2 to <6 years", 3.6, 15.9, 1.03, 1.62, 8.97,
  "6 to <12 years", 7.4, 29.8, 1.15, 2.69, 15.4,
  "12 to <18 years", 13.7, 49.9, 1.14, 3.85, 23.6
)
groupCov <- groups |>
  dplyr::transmute(
    AGE, WT,
    DOSE_APIXABAN_MG = 1,
    FORM_CAPSULE = as.integer(group == "Birth to 27 days")
  ) |>
  as.data.frame()
typSim <- rxode2::rxSolve(
  modTyp, mkEvents(groupCov, amt = rep(1, nrow(groupCov)), obsTimes = 1),
  returnType = "data.frame"
)
#> ℹ omega/sigma items treated as zero: 'etalka', 'etalcl', 'etalvc', 'etalvp', 'etalq'
#> Warning: multi-subject simulation without without 'omega'
typRes <- groups |>
  dplyr::mutate(
    ka_mod = typSim$ka, cl_mod = typSim$cl, vc_mod = typSim$vc,
    ka_pct = 100 * (ka_mod / ka_pub - 1),
    cl_pct = 100 * (cl_mod / cl_pub - 1),
    vc_pct = 100 * (vc_mod / vc_pub - 1)
  )
knitr::kable(
  typRes |>
    dplyr::select(group, ka_mod, ka_pub, cl_mod, cl_pub, cl_pct, vc_mod, vc_pub, vc_pct) |>
    dplyr::rename(
      "Age group" = group,
      "ka typical (1/h)" = ka_mod, "ka Table 3" = ka_pub,
      "CL/F typical (L/h)" = cl_mod, "CL/F Table 3" = cl_pub, "CL/F %" = cl_pct,
      "Vc/F typical (L)" = vc_mod, "Vc/F Table 3" = vc_pub, "Vc/F %" = vc_pct
    ),
  digits = 2,
  caption = "Typical ka, CL/F and Vc/F at each age group's mean age and weight against the Table 3 geometric means of the individual estimates."
)
Typical ka, CL/F and Vc/F at each age group’s mean age and weight against the Table 3 geometric means of the individual estimates.
Age group ka typical (1/h) ka Table 3 CL/F typical (L/h) CL/F Table 3 CL/F % Vc/F typical (L) Vc/F Table 3 Vc/F %
Birth to 27 days 0.50 0.55 0.12 0.10 19.48 2.09 1.82 14.81
28 days to <9 months 0.50 0.72 0.43 0.35 24.06 3.12 2.69 16.17
9 months to <2 years 1.21 1.34 0.84 0.76 10.36 4.97 4.61 7.73
2 to <6 years 1.21 1.03 1.53 1.62 -5.52 8.79 8.97 -1.96
6 to <12 years 1.21 1.15 2.49 2.69 -7.29 15.28 15.40 -0.81
12 to <18 years 1.21 1.14 3.67 3.85 -4.58 24.03 23.60 1.83
# Deterministic, but the published side is a geometric mean of individual
# estimates over a covariate range, not a typical value at the mean. Realised
# CL/F differences run from -7% to +24% and Vc/F from -2% to +16%. A wrong
# weight exponent, reference weight or maturation constant moves the youngest
# groups by 2- to 10-fold, so a 35% envelope still goes red for them.
stopifnot(
  nrow(typRes) == 6L,
  max(abs(typRes$cl_pct)) < 35,
  max(abs(typRes$vc_pct)) < 35,
  # ka is a step at 9 months: 0.502 below it, 0.502 * 2.42 above it.
  all(abs(typRes$ka_mod[1:2] - 0.502) < 1e-8),
  all(abs(typRes$ka_mod[3:6] - 0.502 * 2.42) < 1e-8)
)

The Table 3 ka geometric mean of the 28-day-to-<9-month group (0.72 1/h) lies between the two typical values because its upper age bound sits at the 9-month step and the individual estimates carry 54% CV of IIV on ka.

Anti-factor Xa activity

The Discussion gives the slope (0.0155 IU/ng, i.e. IU/mL per ng/mL) but no intercept. The maintainers digitised the regression line drawn in Figure 4 (23 points along the line, axes calibrated from the tick marks) and fitted a straight line to them: slope 0.01549 IU/mL per ng/mL, which reproduces the printed slope and confirms the line is the reported regression, and intercept 0.074 IU/mL, which is used in the model.

pdCov <- data.frame(WT = 67.4, AGE = 25, DOSE_APIXABAN_MG = 2.5, FORM_CAPSULE = 0)
pdSim <- rxode2::rxSolve(
  modTyp, mkEvents(pdCov, amt = 2.5, obsTimes = seq(0, 24, by = 0.5)),
  returnType = "data.frame"
)
#> ℹ omega/sigma items treated as zero: 'etalka', 'etalcl', 'etalvc', 'etalvp', 'etalq'
fit <- stats::lm(antiFXa ~ Cc, data = pdSim)
stats::coef(fit)
#> (Intercept)          Cc 
#>      0.0740      0.0155
stopifnot(
  abs(stats::coef(fit)[["Cc"]] - 0.0155) < 1e-10,
  abs(stats::coef(fit)[["(Intercept)"]] - 0.074) < 1e-10
)

Virtual paediatric cohort

One arm per age group, with the 9-month-to-<2-year group split by its two dose levels as in Table 3. Ages and weights are drawn independently from normal distributions with the Table 1 mean and SD of each group, truncated to the group’s age bounds and to a plausible weight range; the single neonate is represented by subjects at the neonate’s age (15 days) and weight (3.1 kg). Every subject in an arm receives the arm’s mean absolute dose: the per-group means printed in the Methods (0.1, 0.29, 0.71, 1.82 and 3.22 mg), and for the two 9-month-to-<2-year dose levels the dose per m^2 times the group’s mean BSA of 0.4 m^2 (0.43 and 0.97 mg).

rxode2::rxSetSeed(20230403)
set.seed(20230403)
nPerArm <- 100L

arms <- tibble::tribble(
  ~arm, ~ageMean, ~ageSd, ~ageLo, ~ageHi, ~wtMean, ~wtSd, ~dose, ~capsule,
  "Birth to 27 days", 15 / 365.25, 0, 15 / 365.25, 15 / 365.25, 3.1, 0, 0.1, 1L,
  "28 days to <9 months", 126.9 / 365.25, 67.2 / 365.25, 28 / 365.25, 0.749, 4.9, 1.7, 0.29, 0L,
  "9 months to <2 years (1.08 mg/m2)", 13.3 / 12, 4.2 / 12, 0.75, 1.999, 8.3, 2.4, 1.08 * 0.4, 0L,
  "9 months to <2 years (2.43 mg/m2)", 13.3 / 12, 4.2 / 12, 0.75, 1.999, 8.3, 2.4, 2.43 * 0.4, 0L,
  "2 to <6 years", 3.6, 1.2, 2, 5.999, 15.9, 4.7, 0.71, 0L,
  "6 to <12 years", 7.4, 1.2, 6, 11.999, 29.8, 11.1, 1.82, 0L,
  "12 to <18 years", 13.7, 1.4, 12, 17.999, 49.9, 17.1, 3.22, 0L
)

cohort <- do.call(rbind, lapply(seq_len(nrow(arms)), function(i) {
  a <- arms[i, ]
  data.frame(
    arm = a$arm,
    AGE = pmin(pmax(rnorm(nPerArm, a$ageMean, a$ageSd), a$ageLo), a$ageHi),
    WT = pmin(pmax(rnorm(nPerArm, a$wtMean, a$wtSd), 0.5 * a$wtMean), 2 * a$wtMean),
    DOSE_APIXABAN_MG = a$dose,
    FORM_CAPSULE = a$capsule
  )
}))
cohort$id <- seq_len(nrow(cohort))
stopifnot(nrow(cohort) == nrow(arms) * nPerArm, max(table(cohort$arm)) <= 200)

obsTimes <- c(0, 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, 3, 4, 5, 6, 8, 10, 12, 16, 24, 36, 48, 72)
events <- mkEvents(
  cohort[, c("AGE", "WT", "DOSE_APIXABAN_MG", "FORM_CAPSULE")],
  amt = cohort$DOSE_APIXABAN_MG, obsTimes = obsTimes
)
sim <- rxode2::rxSolve(mod, events, returnType = "data.frame") |>
  dplyr::mutate(id = as.integer(as.character(id))) |>
  dplyr::left_join(cohort[, c("id", "arm")], by = "id")
stopifnot(nrow(sim) == nrow(cohort) * length(obsTimes), all(is.finite(sim$Cc)))
# Replicates the layout of Figure 2 (concentration-time by age group, semilog).
ggplot(dplyr::filter(sim, time > 0, time <= 26), aes(time, Cc, group = id)) +
  geom_line(alpha = 0.1) +
  stat_summary(aes(group = arm), fun = median, geom = "line", colour = "firebrick") +
  facet_wrap(~arm, ncol = 2) +
  scale_y_log10() +
  labs(
    x = "Time after dose (h)", y = "Apixaban concentration (ng/mL)",
    title = "Simulated single-dose apixaban profiles by age group",
    subtitle = "Grey: individual subjects; red: median. Compare Figure 2 of Merali 2023."
  ) +
  theme_bw()

# Replicates Figure 3: individual CL/F versus age, with the Table 3 geometric
# means at each group's mean age for comparison.
indiv <- sim |>
  dplyr::filter(time == 1) |>
  dplyr::select(id, arm, AGE, cl)
ggplot(indiv, aes(AGE, cl)) +
  geom_point(alpha = 0.3) +
  geom_point(
    data = groups, aes(AGE, cl_pub), colour = "firebrick", size = 3, shape = 17
  ) +
  scale_y_log10() +
  labs(
    x = "Age (years)", y = "CL/F (L/h)",
    title = "Individual CL/F versus age",
    subtitle = "Black: simulated subjects; red: Table 3 geometric means. Compare Figure 3."
  ) +
  theme_bw()

PKNCA validation

concData <- sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::select(id, time, Cc, arm)
stopifnot(sum(concData$time == 0) == nrow(cohort))

doseData <- cohort |>
  dplyr::transmute(id, time = 0, amt = DOSE_APIXABAN_MG, arm)

concObj <- PKNCA::PKNCAconc(concData, Cc ~ time | arm + id, concu = "ng/mL", timeu = "h")
doseObj <- PKNCA::PKNCAdose(doseData, amt ~ time | arm + id, route = "extravascular", doseu = "mg")
intervals <- data.frame(start = 0, end = Inf, cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE)
ncaRes <- PKNCA::pk.nca(PKNCA::PKNCAdata(concObj, doseObj, intervals = intervals))

Comparison against Table 3

Table 3 reports geometric means of the post hoc Cmax and AUCinf and the median Tmax per age group. ncaComparisonTable() summarises the simulated subjects by their median, which for these log-normally distributed exposures estimates the same quantity as a geometric mean.

published <- tibble::tribble(
  ~arm, ~cmax, ~tmax, ~aucinf.obs,
  "Birth to 27 days", 45.9, 4.54, 1120,
  "28 days to <9 months", 64, 2.32, 806,
  "9 months to <2 years (1.08 mg/m2)", 59.1, 1.60, 501,
  "9 months to <2 years (2.43 mg/m2)", 148, 1.67, 1260,
  "2 to <6 years", 49.9, 1.91, 423,
  "6 to <12 years", 80, 1.71, 662,
  "12 to <18 years", 96.5, 1.95, 815
)

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = ncaRes,
  reference = published,
  by = "arm",
  units = c(cmax = "ng/mL", tmax = "h", aucinf.obs = "ng*h/mL"),
  tolerance_pct = 30
)
knitr::kable(
  cmp,
  caption = "Simulated (median) vs. Table 3 (geometric mean; Tmax median). * differs by more than 30%."
)
Simulated (median) vs. Table 3 (geometric mean; Tmax median). * differs by more than 30%.
NCA parameter arm Reference Simulated % diff
Cmax (ng/mL) Birth to 27 days 45.9 38.5 -16.1%
Cmax (ng/mL) 28 days to <9 months 64 55.3 -13.6%
Cmax (ng/mL) 9 months to <2 years (1.08 mg/m2) 59.1 59.3 +0.4%
Cmax (ng/mL) 9 months to <2 years (2.43 mg/m2) 148 134 -9.7%
Cmax (ng/mL) 2 to <6 years 49.9 54.7 +9.7%
Cmax (ng/mL) 6 to <12 years 80 83.4 +4.2%
Cmax (ng/mL) 12 to <18 years 96.5 95.3 -1.3%
Tmax (h) Birth to 27 days 4.54 5 +10.1%
Tmax (h) 28 days to <9 months 2.32 4 +72.4%*
Tmax (h) 9 months to <2 years (1.08 mg/m2) 1.6 1.5 -6.3%
Tmax (h) 9 months to <2 years (2.43 mg/m2) 1.67 2 +19.8%
Tmax (h) 2 to <6 years 1.91 1.5 -21.5%
Tmax (h) 6 to <12 years 1.71 2 +17.0%
Tmax (h) 12 to <18 years 1.95 2 +2.6%
AUC0-∞ (obs) (ng*h/mL) Birth to 27 days 1120 834 -25.5%
AUC0-∞ (obs) (ng*h/mL) 28 days to <9 months 806 652 -19.1%
AUC0-∞ (obs) (ng*h/mL) 9 months to <2 years (1.08 mg/m2) 501 493 -1.6%
AUC0-∞ (obs) (ng*h/mL) 9 months to <2 years (2.43 mg/m2) 1260 1160 -8.1%
AUC0-∞ (obs) (ng*h/mL) 2 to <6 years 423 497 +17.4%
AUC0-∞ (obs) (ng*h/mL) 6 to <12 years 662 726 +9.6%
AUC0-∞ (obs) (ng*h/mL) 12 to <18 years 815 899 +10.3%
simGm <- as.data.frame(ncaRes) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "aucinf.obs")) |>
  dplyr::group_by(arm, PPTESTCD) |>
  dplyr::summarise(sim = exp(mean(log(PPORRES))), .groups = "drop") |>
  dplyr::left_join(
    tidyr::pivot_longer(published, c(cmax, aucinf.obs), names_to = "PPTESTCD", values_to = "pub") |>
      dplyr::select(arm, PPTESTCD, pub),
    by = c("arm", "PPTESTCD")
  ) |>
  dplyr::mutate(pct = 100 * (sim / pub - 1))
simGm
#> # A tibble: 14 × 5
#>    arm                               PPTESTCD      sim    pub    pct
#>    <chr>                             <chr>       <dbl>  <dbl>  <dbl>
#>  1 12 to <18 years                   aucinf.obs  864.   815     6.07
#>  2 12 to <18 years                   cmax         91.0   96.5  -5.65
#>  3 2 to <6 years                     aucinf.obs  491.   423    16.0 
#>  4 2 to <6 years                     cmax         56.6   49.9  13.5 
#>  5 28 days to <9 months              aucinf.obs  662.   806   -17.9 
#>  6 28 days to <9 months              cmax         55.1   64   -13.9 
#>  7 6 to <12 years                    aucinf.obs  723.   662     9.19
#>  8 6 to <12 years                    cmax         81.3   80     1.67
#>  9 9 months to <2 years (1.08 mg/m2) aucinf.obs  489.   501    -2.40
#> 10 9 months to <2 years (1.08 mg/m2) cmax         58.2   59.1  -1.52
#> 11 9 months to <2 years (2.43 mg/m2) aucinf.obs 1152.  1260    -8.59
#> 12 9 months to <2 years (2.43 mg/m2) cmax        135.   148    -8.90
#> 13 Birth to 27 days                  aucinf.obs  790.  1120   -29.5 
#> 14 Birth to 27 days                  cmax         37.8   45.9 -17.7

# Cohort-derived geometric means against geometric means of 1 to 11 real
# subjects per group, with ages, weights and doses only matched in their
# group means. Realised: median -5%, 75th percentile of the absolute
# difference 16%, largest -30% (the neonate AUC, discussed below). The centre
# of the seven-arm comparison is gated tightly and the envelope loosely: a
# mis-transcribed clearance, weight exponent, maturation constant or unit
# shifts every arm (or all the young arms) by tens of percent to several-fold.
stopifnot(
  nrow(simGm) == 14L,
  all(is.finite(simGm$pct)),
  abs(stats::median(simGm$pct)) < 20,
  stats::quantile(abs(simGm$pct), 0.75) < 35
)

Tmax is read off the simulation grid (1.5, 2, 2.5, 3, 4, 5 h), so its comparison is coarse. The 28-day-to-<9-month arm is the one starred row: below 9 months the typical ka is the adult 0.502 1/h, giving a simulated median Tmax of about 4 h against the paper’s 2.32 h (range 1.46 to 9.42 h), and the Table 3 ka geometric mean for that group (0.72 1/h) indicates that the individual estimates sat above the typical value.

The paper does not compute AUCinf by NCA: it defines it as Dose / (CL/F) with the individual post hoc CL/F. In the youngest children the unscaled peripheral volume (18 L) and the very small age-scaled Q/F produce a slow return phase that a 72-hour profile cannot extrapolate, so the PKNCA AUCinf above underestimates the neonate’s exposure. The paper’s own definition, evaluated with each simulated subject’s CL/F and relative bioavailability, reads it directly:

aucDef <- sim |>
  dplyr::filter(time == 1) |>
  dplyr::mutate(auc = 1000 * DOSE_APIXABAN_MG * frel * fform / cl) |>
  dplyr::group_by(arm) |>
  dplyr::summarise(auc_sim = exp(mean(log(auc))), .groups = "drop") |>
  dplyr::left_join(published[, c("arm", "aucinf.obs")], by = "arm") |>
  dplyr::mutate(pct = 100 * (auc_sim / aucinf.obs - 1))
knitr::kable(
  aucDef |>
    dplyr::rename(
      "Arm" = arm, "Dose / (CL/F), simulated (ng*h/mL)" = auc_sim,
      "Table 3 AUCinf (ng*h/mL)" = aucinf.obs, "% diff" = pct
    ),
  digits = 1,
  caption = "Geometric mean of Dose * F1 / (CL/F) per arm against Table 3 AUCinf."
)
Geometric mean of Dose * F1 / (CL/F) per arm against Table 3 AUCinf.
Arm Dose / (CL/F), simulated (ng*h/mL) Table 3 AUCinf (ng*h/mL) % diff
12 to <18 years 863.7 815 6.0
2 to <6 years 494.0 423 16.8
28 days to <9 months 748.1 806 -7.2
6 to <12 years 723.2 662 9.2
9 months to <2 years (1.08 mg/m2) 505.1 501 0.8
9 months to <2 years (2.43 mg/m2) 1192.7 1260 -5.3
Birth to 27 days 927.4 1120 -17.2
# Same cohort caveats as the NCA gate above. Realised -17% (the neonate) to
# +17%, median +1%.
stopifnot(
  nrow(aucDef) == 7L,
  abs(stats::median(aucDef$pct)) < 20,
  max(abs(aucDef$pct)) < 40
)

Anti-factor Xa versus concentration (Figure 4)

ggplot(dplyr::filter(sim, time %in% c(0.5, 1, 2, 4, 6, 12, 24)), aes(Cc, antiFXa)) +
  geom_point(aes(colour = arm), alpha = 0.3, size = 0.8) +
  coord_cartesian(xlim = c(0, 200), ylim = c(0, 3.1)) +
  labs(
    x = "Apixaban concentration (ng/mL)", y = "Anti-FXa (IU/mL)", colour = NULL,
    title = "Anti-factor Xa activity versus apixaban concentration",
    subtitle = "Typical-value regression evaluated at simulated concentrations. Compare Figure 4."
  ) +
  theme_bw() +
  theme(legend.position = "bottom", legend.text = element_text(size = 6))

Assumptions and deviations

  • Printed CL/F equation. In the published typesetting of the CL/F equation the bracket of the maturation term encloses 1 - theta15 * Age^theta12, which taken literally makes the maturation fraction negative above about 37 years (the adults in the dataset). The model uses the standard form of the cited Bouillon-Pichault 2011 function, theta15 + (1 - theta15) * Age^theta12 / (Age^theta12 + theta13^theta12), with theta15 named FBIRTH in Table 2. This form reproduces the Table 3 CL/F geometric means (see the typical-value check above).
  • Anti-factor Xa intercept digitised. The paper reports only the slope of the paediatric anti-FXa regression. The intercept, 0.074 IU/mL, was digitised by the maintainers from the regression line drawn in Figure 4; the slope digitised from the same line (0.01549) reproduces the printed 0.0155. The linear mixed-effects regression’s random effects and residual error are not reported, so antiFXa is a typical-value relationship with no error model. The assay was heparin-calibrated (STA Liquid Anti-Xa Heparin or Rotachrom), so antiFXa is in IU/mL, not drug-equivalent ng/mL.
  • IIV covariance block. The printed variances and covariances of the CL/F, Vc/F, Vp/F and Q/F block reproduce the r^2 values in the Table 2 footnote, but at their printed precision the 4 x 4 block is very slightly indefinite (smallest eigenvalue -2.4e-5), so it cannot be sampled. The off-diagonal covariances are scaled by 0.99 (each correlation reduced by 1%); the variances are as printed.
  • Residual error. Table 2 reports a single “Exponential” residual variance (0.0844, SD 0.29) and a negative minimum objective function value (-5805.653), consistent with log-transformed concentrations; the model uses lnorm(expSd) with expSd = sqrt(0.0844) = 0.2905.
  • PED9M. Derived from AGE inside the model (AGE >= 0.75 and AGE < 18, with 9 months taken as 0.75 years), so no separate column is needed. Adults and infants under 9 months share the reference ka.
  • Age and weight. AGE is postnatal age in years; the paper does not use gestational or postmenstrual age. WT is baseline weight. The Q/F age effect makes Q/F very small in infants (0.08 L/h at 4 months), so the disposition is effectively one-compartment in the youngest children.
  • Virtual cohort. Ages and weights are drawn independently within each age group from the Table 1 means and SDs, and each arm receives the group’s mean absolute dose rather than a per-subject BSA-based dose (subject heights are not reported). Table 3 is built from individual post hoc estimates in 1 to 11 subjects per group, so the comparison is a centre-of-distribution check, not a reproduction of each group’s individuals.
  • Not extracted. The adult phase I studies’ individual data and Figure S1 (VPC) are not reproduced; the paper’s push-pull sampling comparison is not a model.
  • No erratum found. No correction notice for doi:10.1002/psp4.12935 was found on Europe PMC as of 2026-10-26.