Skip to contents

Perioperative factor VIII (Preijers 2021)

Source: vignettes/articles/Preijers_2021_factor_viii.Rmd
Preijers_2021_factor_viii.Rmd

Model and source

  • Citation: Preijers T, Liesner R, Hazendonk HCAM, Chowdary P, Driessens MHE, Hart DP, Laros-van Gorkom BAP, van der Meer FJM, Meijer K, Fijnvandraat K, Leebeek FWG, Mathot RAA, Cnossen MH; OPTI-CLOT study group. Validation of a perioperative population factor VIII pharmacokinetic model with a large cohort of pediatric hemophilia A patients. Br J Clin Pharmacol. 2021;87(11):4408-4420. doi:[10.1111/bcp.14864](https://doi.org/10.1111/bcp.14864). PMID:33884664.
  • Description: two-compartment population PK model for perioperative factor VIII (FVIII) concentrates in children and adults with hemophilia A. The paper first externally validated the Hazendonk 2016 perioperative model (Hazendonk_2016_factor_viii in this package) on 87 children from Great Ormond Street Hospital (GOSH), found that it under-predicted their FVIII levels, and then re-estimated the model on the pooled original and new data. The re-estimated final model is packaged here.
  • Modality: recombinant (88% of procedures) and plasma-derived FVIII concentrates given by intravenous bolus or continuous infusion; FVIII activity measured by one-stage clotting assay in IU/mL.

The final model (Preijers 2021 Table 2 footnote) is

CL(mL/h)=171⋅(BW/68)0.75⋅(AGE/40)−0.12⋅1.14BGV1(mL)=2930⋅(BW/68)1.0⋅(AGE/40)−0.09Q2(mL/h)=172⋅(BW/68)0.75,V2(mL)=1810⋅(BW/68) \begin{aligned} \mathrm{CL}\,(\mathrm{mL/h}) &= 171 \cdot (\mathrm{BW}/68)^{0.75} \cdot (\mathrm{AGE}/40)^{-0.12} \cdot 1.14^{\mathrm{BG}} \\ V_{1}\,(\mathrm{mL}) &= 2930 \cdot (\mathrm{BW}/68)^{1.0} \cdot (\mathrm{AGE}/40)^{-0.09} \\ Q_{2}\,(\mathrm{mL/h}) &= 172 \cdot (\mathrm{BW}/68)^{0.75}, \qquad V_{2}\,(\mathrm{mL}) = 1810 \cdot (\mathrm{BW}/68) \end{aligned}

with BG = 1 for blood group O. The measured FVIII level is the model prediction plus the patient’s endogenous baseline level, reduced by 30% for samples from patients on the B-domain-deleted product Refacto AF (Methods Eq. 1; see Assumptions and deviations for how that equation is read).

Population

The final model was fitted to 206 patients undergoing 342 surgical procedures (Preijers 2021 Table 1):

  • the original cohort of Hazendonk 2016: 119 patients (children and adults) from five Dutch hemophilia treatment centres, median age 39.6 years (0.24-77.6), median weight 75 kg (5-111), 70% severe hemophilia A, 197 procedures of which 49% were major surgery and 59% used continuous infusion;
  • the new cohort: 87 children from GOSH, London, median age 2.57 years (0.03-15.2), median weight 14 kg (4-57), all with severe hemophilia A, 145 minor procedures to insert, replace or remove a central venous access device, all dosed by bolus.

The total cohort had a median age of 7.79 years (0.03-77.6) and median weight of 30 kg (4-111); 83% had severe hemophilia A and 39% blood group O (of 175 with a recorded blood group). Recombinant FVIII was used in 88% of procedures. There were 2092 FVIII measurements (one-stage clotting assay), none below the 0.01 IU/mL quantification limit.

The same information is available programmatically via readModelDb("Preijers_2021_factor_viii")()$population.

Source trace

Model element Value Source location
lcl (CL at 68 kg, 40 y, non-O) log(0.171 L/h) Table 2 final model: 171 mL/h/68 kg (RSE 7%)
lvc (V1 at 68 kg, 40 y) log(2.930 L) Table 2 final model: 2930 mL/68 kg (RSE 4%)
lq (Q2 at 68 kg) log(0.172 L/h) Table 2 final model: 172 mL/h/68 kg (RSE 19%)
lvp (V2 at 68 kg) log(1.810 L) Table 2 final model: 1810 mL/68 kg (RSE 10%)
theta_bdp (Refacto AF fraction) 0.30 Table 2 final model: ‘B-domain deleted recombinant factor VIII’ 0.30 (RSE 14%)
e_age_cl -0.12 Table 2 final model ‘CL - Age’ (RSE 26%); footnote (AGE/40)^-0.12
e_age_vc -0.09 Table 2 final model ‘V1 - Age’ (RSE 24%); footnote (AGE/40)^-0.09
e_blood_cl 0.14 Table 2 final model ‘CL - Blood group O’ 14% (RSE 6%); footnote 1.14^BG
e_wt_cl, e_wt_vc (fixed) 0.75, 1.0 Methods after Eq. 2: exponents fixed to 0.75 (CL, Q2) and 1 (V1, V2)
etalcl, etalvc block 0.14568, 0.05833, 0.07290 Table 2 final model: IIV CL 39.6 %CV, IIV V1 27.5 %CV, correlation 56.6%; omega^2 = log(1 + CV^2), cov = r sqrt(omega^2_CL omega^2_V1)
addSd_center123 / _center45 / _center6 0.12 / 0.06 / 0.17 IU/mL Table 2 final model: additive SD by centre
propSd_center123 / _center45 / _center6 0.197 / 0.21 / 0.22 Table 2 final model: proportional %CV by centre
Two-compartment structure, 68 kg normalisation – Results 3.3; Methods Eq. 2
Age power model – Methods Eq. 5; Results 3.3 (power form performed best)
Blood group O on CL – Methods Eq. 3; Table 2 footnote
Baseline + Refacto AF observation equation Cc = (central/vc/1000 + FVIIIRECENT) * (1 - 0.30 * FORM_FVIII_BDD) Methods Eq. 1

Covariates tested but not retained: major surgery on CL (significant in the multivariate step, dropped after the bootstrap because its confidence interval contained 1; Results 3.4) and severe hemophilia (dropped in the multivariate analysis; Results 3.3). Inter-occasion variability on CL and V1 was tested and omitted because the model became unstable (Results 3.3).

Typical-value checks against the paper’s numbers

The Table 2 footnote equations can be evaluated directly. The Results also quote typical values for a one-year-old 10 kg non-O child under both the Hazendonk 2016 model and the present one.

mod <- readModelDb("Preijers_2021_factor_viii")
mod_hz <- readModelDb("Hazendonk_2016_factor_viii")

typ_cl <- function(wt, age, bg = 0) 171 * (wt / 68)^0.75 * (age / 40)^-0.12 * 1.14^bg
typ_v1 <- function(wt, age) 2930 * (wt / 68) * (age / 40)^-0.09
hz_cl <- function(wt, age) 150 * (wt / 68)^0.75 * (age / 40)^-0.17
hz_v1 <- function(wt, age) 2810 * (wt / 68) * (age / 40)^-0.09

# The packaged model must reproduce the footnote equations exactly.
chk_cov <- data.frame(
  id = 1:4,
  WT = c(68, 68, 10, 30),
  AGE = c(40, 40, 1, 8),
  BLOOD_GROUP_O = c(0, 1, 0, 0),
  FORM_FVIII_BDD = 0, FVIIIRECENT = 0, STUDY_OPTICLOT_CTR45 = 0, STUDY_GOSH = 0
)
ev_chk <- merge(chk_cov, data.frame(time = c(0, 1)))
ev_chk$evid <- ifelse(ev_chk$time == 0, 1L, 0L)
ev_chk$amt <- ifelse(ev_chk$evid == 1L, 1000, 0)
ev_chk$cmt <- "central"
ev_chk <- ev_chk[order(ev_chk$id, ev_chk$time), ]
sim_chk <- as.data.frame(rxSolve(zeroRe(mod), ev_chk, keep = c("WT", "AGE", "BLOOD_GROUP_O")))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
#> Warning: multi-subject simulation without without 'omega'
sim_chk <- sim_chk[!duplicated(sim_chk$id), ]
ref_cl <- typ_cl(sim_chk$WT, sim_chk$AGE, sim_chk$BLOOD_GROUP_O)
ref_v1 <- typ_v1(sim_chk$WT, sim_chk$AGE)
stopifnot(
  all(abs(sim_chk$cl * 1000 / ref_cl - 1) < 1e-8),
  all(abs(sim_chk$vc * 1000 / ref_v1 - 1) < 1e-8)
)

tibble::tibble(
  Quantity = c(
    "CL, 68 kg, 40 y, non-O (mL/h)",
    "V1, 68 kg, 40 y (mL)",
    "CL, 1 y, 10 kg, non-O: Hazendonk 2016 model (mL/h)",
    "CL, 1 y, 10 kg, non-O: Preijers 2021 model (mL/h)",
    "V1, 1 y, 10 kg: Hazendonk 2016 model (mL)",
    "V1, 1 y, 10 kg: Preijers 2021 model (mL)"
  ),
  Computed = round(c(
    typ_cl(68, 40), typ_v1(68, 40),
    hz_cl(10, 1), typ_cl(10, 1),
    hz_v1(10, 1), typ_v1(10, 1)
  ), 1),
  `Paper` = c(
    "171 (Table 2)", "2930 (Table 2)",
    "63.2 as 'from' value (Results 3.4)", "66.7 as 'to' value (Results 3.4)",
    "601 as 'from' value (Results 3.4)", "576 as 'to' value (Results 3.4)"
  )
) |>
  knitr::kable(caption = "Typical values from the packaged model versus the numbers quoted in Preijers 2021.")
Typical values from the packaged model versus the numbers quoted in Preijers 2021.
Quantity Computed Paper
CL, 68 kg, 40 y, non-O (mL/h) 171.0 171 (Table 2)
V1, 68 kg, 40 y (mL) 2930.0 2930 (Table 2)
CL, 1 y, 10 kg, non-O: Hazendonk 2016 model (mL/h) 66.7 63.2 as ‘from’ value (Results 3.4)
CL, 1 y, 10 kg, non-O: Preijers 2021 model (mL/h) 63.2 66.7 as ‘to’ value (Results 3.4)
V1, 1 y, 10 kg: Hazendonk 2016 model (mL) 575.9 601 as ‘from’ value (Results 3.4)
V1, 1 y, 10 kg: Preijers 2021 model (mL) 600.5 576 as ‘to’ value (Results 3.4)

The four one-year-old values are reproduced to the printed precision, but attached to the opposite model: the Hazendonk 2016 equations give 66.7 mL/h and 576 mL and the Preijers 2021 equations give 63.2 mL/h and 601 mL, whereas Results 3.4 says CL “increased from 63.2 to 66.7” and V1 was “reduced from 601 to 576”. The numbers themselves confirm the Table 2 footnote equations; the from/to wording in the text is swapped. The Table 2 equations are used here.

Virtual cohort

Observed data are not public. Two virtual cohorts of 100 patients each mimic Table 1: a pediatric arm resembling the GOSH cohort and an adult arm resembling the adults of the original cohort. Pediatric weight follows an approximate median weight-for-age curve with 12% log-normal scatter.

set.seed(2021)
rxode2::rxSetSeed(2021)
n_per_arm <- 100L

age_knots <- c(0, 0.5, 1, 2, 3, 5, 8, 10, 12, 15)
wt_knots <- c(3.5, 7.8, 10, 12.5, 14.5, 18.5, 26, 32, 40, 56)

children <- tibble::tibble(
  id = seq_len(n_per_arm),
  AGE = pmin(pmax(exp(rnorm(n_per_arm, log(2.6), 1.0)), 0.05), 15.2),
  cohort = "Pediatric"
) |>
  mutate(WT = pmin(pmax(approx(age_knots, wt_knots, AGE, rule = 2)$y * exp(rnorm(n(), 0, 0.12)), 4), 57))

adults <- tibble::tibble(
  id = n_per_arm + seq_len(n_per_arm),
  AGE = runif(n_per_arm, 19, 78),
  WT = pmin(pmax(rnorm(n_per_arm, 78, 13), 45), 111),
  cohort = "Adult"
)

cohort <- bind_rows(children, adults) |>
  mutate(
    BLOOD_GROUP_O = as.integer(runif(n()) < 0.39),
    FORM_FVIII_BDD = 0L,
    FVIIIRECENT = 0,
    STUDY_OPTICLOT_CTR45 = 0L,
    STUDY_GOSH = as.integer(cohort == "Pediatric")
  )

cohort |>
  group_by(cohort) |>
  summarise(
    n = n(),
    `median age (y)` = round(median(AGE), 2),
    `median weight (kg)` = round(median(WT), 1),
    `blood group O (%)` = round(100 * mean(BLOOD_GROUP_O)),
    .groups = "drop"
  ) |>
  knitr::kable(caption = "Virtual cohort summary.")
Virtual cohort summary.
cohort n median age (y) median weight (kg) blood group O (%)
Adult 100 47.61 77.8 30
Pediatric 100 2.27 13.1 42

FVIIIRECENT = 0 (severe hemophilia, drug-attributable FVIII only) and FORM_FVIII_BDD = 0 are used throughout so the NCA below reflects the PK parameters; their effect on the readout is shown separately at the end.

Simulation 1: single 50 IU/kg bolus 

obs_grid <- sort(unique(c(seq(0, 12, 0.5), seq(13, 48, 1), seq(50, 120, 2), seq(126, 168, 6))))

ev_single <- bind_rows(
  cohort |> mutate(time = 0, evid = 1L, amt = 50 * WT, cmt = "central"),
  cohort |> tidyr::crossing(time = obs_grid) |> mutate(evid = 0L, amt = 0, cmt = "central")
) |>
  arrange(id, time, desc(evid)) |>
  as.data.frame()

sim_single <- rxSolve(mod, ev_single, keep = c("cohort", "WT", "AGE")) |>
  as.data.frame()
#> ℹ parameter labels from comments will be replaced by 'label()'

Simulation 2: perioperative regimen

A representative bolus regimen: 50 IU/kg before surgery, then 25 IU/kg every 12 h for 5 days (children every 8 h, reflecting their faster weight-normalised clearance). The dashed lines are the Dutch consensus targets used by the original cohort (0.8-1.0 IU/mL day 1, 0.5-0.8 IU/mL days 2-5).

make_regimen <- function(d) {
  tau <- ifelse(d$cohort[1] == "Pediatric", 8, 12)
  dose_times <- seq(0, 120 - tau, by = tau)
  bind_rows(
    d |> tidyr::crossing(time = dose_times) |>
      mutate(evid = 1L, amt = ifelse(time == 0, 50, 25) * WT, cmt = "central"),
    d |> tidyr::crossing(time = seq(0, 132, 1)) |>
      mutate(evid = 0L, amt = 0, cmt = "central")
  )
}
ev_periop <- cohort |>
  group_split(cohort) |>
  lapply(make_regimen) |>
  bind_rows() |>
  arrange(id, time, desc(evid)) |>
  as.data.frame()

sim_periop <- rxSolve(mod, ev_periop, keep = c("cohort")) |>
  as.data.frame()

sim_periop |>
  group_by(cohort, time) |>
  summarise(
    med = median(Cc), lo = quantile(Cc, 0.05), hi = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, med, colour = cohort, fill = cohort)) +
  geom_ribbon(aes(ymin = lo, ymax = hi), alpha = 0.2, colour = NA) +
  geom_line() +
  geom_hline(yintercept = c(0.5, 0.8, 1.0), linetype = "dashed", colour = "grey40") +
  labs(
    x = "Time from pre-operative dose (h)", y = "FVIII (IU/mL)",
    title = "Perioperative FVIII levels, median and 90% prediction interval",
    caption = "Illustrative regimen; compare with the pcVPC in Preijers 2021 Figure 6."
  ) +
  theme_bw()

Comparison with the Hazendonk 2016 model

Preijers 2021 Figures 1-2 show that the Hazendonk 2016 model under-predicted typical CL and V1 in young children. The typical 50 IU/kg profiles below compare both models for a one-year-old 10 kg child and a 40-year-old 68 kg adult (non-O, minor surgery, full-length product).

typ_subj <- data.frame(
  id = 1:2, WT = c(10, 68), AGE = c(1, 40), who = c("1 y, 10 kg", "40 y, 68 kg"),
  BLOOD_GROUP_O = 0, SURG_SEV_MAJOR = 0, FORM_FVIII_BDD = 0, FVIIIRECENT = 0,
  STUDY_OPTICLOT_CTR45 = 0, STUDY_GOSH = 0
)
ev_typ <- bind_rows(
  typ_subj |> mutate(time = 0, evid = 1L, amt = 50 * WT, cmt = "central"),
  typ_subj |> tidyr::crossing(time = seq(0, 72, 0.5)) |> mutate(evid = 0L, amt = 0, cmt = "central")
) |>
  arrange(id, time, desc(evid)) |>
  as.data.frame()

typ_both <- bind_rows(
  as.data.frame(rxSolve(zeroRe(mod), ev_typ, keep = "who")) |> mutate(model = "Preijers 2021"),
  as.data.frame(rxSolve(zeroRe(mod_hz), ev_typ, keep = "who")) |> mutate(model = "Hazendonk 2016")
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> 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'

ggplot(typ_both, aes(time, Cc, colour = model)) +
  geom_line() +
  facet_wrap(~who) +
  scale_y_log10() +
  labs(x = "Time after 50 IU/kg (h)", y = "Typical FVIII (IU/mL, log scale)") +
  theme_bw()

PKNCA validation 

nca_in <- sim_single |>
  filter(!is.na(Cc)) |>
  select(id, cohort, time, Cc)
nca_in <- bind_rows(
  nca_in,
  nca_in |> distinct(id, cohort) |> mutate(time = 0, Cc = 0)
) |>
  distinct(id, cohort, time, .keep_all = TRUE) |>
  arrange(id, time)

dose_df <- ev_single |>
  filter(evid == 1) |>
  select(id, cohort, time, amt)

conc_obj <- PKNCAconc(nca_in, Cc ~ time | cohort + id, concu = "IU/mL", timeu = "h")
dose_obj <- PKNCAdose(dose_df, amt ~ time | cohort + id, doseu = "IU", route = "intravascular")
intervals <- data.frame(
  start = 0, end = Inf,
  cmax = TRUE, aucinf.obs = TRUE, half.life = TRUE, cl.obs = TRUE
)
nca_res <- pk.nca(PKNCAdata(conc_obj, dose_obj, intervals = intervals))
knitr::kable(summary(nca_res), caption = "PKNCA summary by cohort, single 50 IU/kg bolus.")
PKNCA summary by cohort, single 50 IU/kg bolus.
Interval Start Interval End cohort N Cmax (IU/mL) Half-life (h) AUCinf,obs (h*IU/mL) CL (based on AUCinf,obs) (IU/(h*IU/mL))
0 Inf Adult 100 1.24 [27.7] 24.4 [5.82] 21.4 [36.7] 177 [39.7]
0 Inf Pediatric 100 0.893 [25.8] 13.7 [4.75] 9.24 [50.6] 76.0 [54.5]

The paper reports no NCA, so the NCA is checked against the model itself: the NCA clearance Dose / AUCinf must recover each subject’s model clearance (same drawn parameters on both sides, so the difference is numerical error from the trapezoid rule and the extrapolated tail).

indiv <- sim_single |>
  distinct(id, cohort, cl, vc, q, vp)
cl_nca <- as.data.frame(nca_res$result) |>
  filter(PPTESTCD == "cl.obs") |>
  select(id, cohort, cl_nca = PPORRES)
chk <- inner_join(indiv, cl_nca, by = c("id", "cohort")) |>
  # cl.obs = IU / (IU/mL * h) is already in mL/h; model cl is in L/h.
  mutate(pct_diff = 100 * (cl_nca / (cl * 1000) - 1))
chk |>
  group_by(cohort) |>
  summarise(
    `median model CL (mL/h)` = round(median(cl * 1000), 1),
    `median NCA CL (mL/h)` = round(median(cl_nca), 1),
    `median % difference` = round(median(pct_diff), 2),
    .groups = "drop"
  ) |>
  knitr::kable(caption = "NCA clearance versus model clearance.")
NCA clearance versus model clearance.
cohort median model CL (mL/h) median NCA CL (mL/h) median % difference
Adult 169.2 169.2 0.00
Pediatric 73.6 73.6 -0.01 
stopifnot(
  abs(median(chk$pct_diff)) < 2,
  quantile(abs(chk$pct_diff), 0.9) < 5
)

For the typical 68 kg, 40-year-old adult, the terminal half-life follows from the two-compartment micro-constants:

k10 <- 0.171 / 2.930
k12 <- 0.172 / 2.930
k21 <- 0.172 / 1.810
s <- k10 + k12 + k21
beta <- (s - sqrt(s^2 - 4 * k10 * k21)) / 2
alpha <- (s + sqrt(s^2 - 4 * k10 * k21)) / 2
c(`distribution t1/2 (h)` = log(2) / alpha, `terminal t1/2 (h)` = log(2) / beta)
#> distribution t1/2 (h)     terminal t1/2 (h) 
#>               3.81808              22.68964

Endogenous baseline and Refacto AF readout 

ro <- data.frame(
  id = 1:3, WT = 68, AGE = 40, BLOOD_GROUP_O = 0,
  FORM_FVIII_BDD = c(0, 0, 1), FVIIIRECENT = c(0, 0.03, 0.03),
  STUDY_OPTICLOT_CTR45 = 0, STUDY_GOSH = 0
)
ev_ro <- bind_rows(
  ro |> mutate(time = 0, evid = 1L, amt = 50 * 68, cmt = "central"),
  ro |> tidyr::crossing(time = c(0, 1, 24)) |> mutate(evid = 0L, amt = 0, cmt = "central")
) |>
  arrange(id, time, desc(evid)) |>
  as.data.frame()
sim_ro <- as.data.frame(rxSolve(zeroRe(mod), ev_ro, keep = c("FVIIIRECENT", "FORM_FVIII_BDD")))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
#> Warning: multi-subject simulation without without 'omega'
sim_ro |>
  filter(time == 24) |>
  select(FVIIIRECENT, FORM_FVIII_BDD, Cc) |>
  knitr::kable(digits = 4, caption = "Typical FVIII 24 h after 50 IU/kg: baseline and BDD-product readout.")
Typical FVIII 24 h after 50 IU/kg: baseline and BDD-product readout.
FVIIIRECENT FORM_FVIII_BDD Cc
0.00 0 0.2466
0.03 0 0.2766
0.03 1 0.1936 
r <- sim_ro[sim_ro$time == 24, ]
stopifnot(
  abs(r$Cc[2] - r$Cc[1] - 0.03) < 1e-8,
  abs(r$Cc[3] / r$Cc[2] - 0.70) < 1e-8
)

Assumptions and deviations

  • Final model only. The Table 2 ‘Original model’ column is the Hazendonk 2016 model, already available as Hazendonk_2016_factor_viii; the ‘Structural model’ column is an intermediate step without covariates and is not packaged. The bootstrap medians are not used.
  • Methods Eq. 1 read as a multiplicative correction. As printed, Eq. 1 is C = (C_PRED + C_base) - theta_prod * (C_PRED + C_base)^theta_RefactoAF, which would subtract a constant 0.30 IU/mL from every non-Refacto sample (anything raised to the power 0 is 1). The text defines theta_prod as “the estimated effect fraction of a FVIII product on the measured FVIII level” and theta_RefactoAF as a 0/1 indicator, and the Hazendonk 2016 predecessor applied the same correction as C_pred * (1 - theta). The model therefore uses (C_PRED + C_base) * (1 - 0.30 * FORM_FVIII_BDD), which equals (C_PRED + C_base) * (1 - 0.30)^FORM_FVIII_BDD for a binary indicator.
  • Endogenous baseline. Eq. 1 adds the measured endogenous level C_base to the prediction; it is carried as the covariate FVIIIRECENT (IU/mL), the package’s canonical name for the patient’s most recently measured untreated FVIII:C (also used, as a power-scaled covariate, by Schutte_2018_desmopressin). Set it to 0 to simulate the drug-attributable FVIII only.
  • Pre-dose FVIII initialisation not encoded. When a pre-operative FVIII level was available without the preceding dose history, the authors initialised every compartment at that level times its volume (NONMEM A_0). This is a data-handling device, not a model parameter; users can reproduce it by setting central(0) and peripheral1(0) on the event table.
  • IIV variance scale. Table 2 reports IIV as %CV; the variances are log(1 + CV^2) (0.1457 and 0.0729). The Hazendonk 2016 model in this package used CV^2; the two differ by about 7% for CL and less than 1% for V1.
  • Centre-specific residual error. All three centre groups are encoded. STUDY_OPTICLOT_CTR45 = 1 selects centres 4-5 and STUDY_GOSH = 1 selects centre 6 (GOSH); both 0 selects centres 1-3. The paper does not say which Dutch centres are numbered 4 and 5. Table 2 prints the centre 1-3 proportional error as 19.7 (%CV) and the others as fractions (0.21, 0.22); all are used as fractions.
  • Narrative values not reproduced. Results 3.2 quotes typical CL of 68 mL/h and V1 of 930 mL for a one-year-old 10 kg child under the Hazendonk 2016 model; its equations give 66.7 mL/h and 576 mL. The 930 mL value cannot be recovered from any printed equation, so it is not used.
  • Virtual cohort. Weight-for-age knots are approximate medians and the dosing regimen is illustrative; the paper’s individual dosing histories are not published. Sex is taken as all-male (X-linked disease; not tabulated).
  • Errata. A Europe PMC search on 2026-09-29 found no correction notice for this article. The supplement (weight imputation model, model-building steps) contains no final parameter values and was not needed.