Skip to contents

Decitabine (Han 2015)

Source: vignettes/articles/Han_2015_decitabine.Rmd
Han_2015_decitabine.Rmd

Model and source

  • Citation: Han S, Kim Y-J, Lee J, Jeon S, Hong T, Park G-J, Yoon J-H, Yahng S-A, Shin S-H, Lee S-E, Eom K-S, Kim H-J, Min C-K, Lee S, Yim D-S. (2015). Model-based adaptive phase I trial design of post-transplant decitabine maintenance in myelodysplastic syndrome. Journal of Hematology & Oncology 8:118. doi:10.1186/s13045-015-0208-3 (PMID 26482429). ClinicalTrials.gov NCT01277484.
  • Description: Two-compartment IV population pharmacokinetic model coupled with two parallel Friberg-style myelosuppression PD chains (absolute neutrophil count, ANC, and platelet count, PC) for decitabine post-transplant maintenance in adult Korean patients with higher-risk myelodysplastic syndrome or secondary acute myeloid leukemia (Han 2015). The platelet feedback baseline rises asymptotically over cycles per the paper’s IMP / IMK extension (BASE_P_t = BASE_P + IMP * (1 - exp(-IMK * t))); the neutrophil chain uses a time-invariant baseline. PK parameters are body-surface-area-normalized (per m^2): doses must be supplied in mg/m^2 and central-compartment concentrations are returned in mg/L (= ug/mL = 1000 ng/mL). PD outputs ANC and PLT are in 10^9 cells/L.
  • Article (open access): https://doi.org/10.1186/s13045-015-0208-3
  • PMID: 26482429
  • ClinicalTrials.gov: NCT01277484

Decitabine is a hypomethylating cytidine analogue. Han et al. (2015) report an adaptive phase I trial of post-allogeneic-hematopoietic-stem-cell- transplantation (allo-HSCT) decitabine maintenance in 16 Korean adults with higher-risk myelodysplastic syndrome (MDS) or secondary acute myeloid leukemia (sAML) evolving from MDS. The PK-PD model was used to individualize the Cycle 2-4 doses for each patient and to estimate the initial doses for cohorts 2-5. The popPK-PD analysis excluded one female patient with immune thrombocytopenia (managed with steroids), leaving 15 patients in the mixed-effect analysis (95 PK observations + 311 ANC + 311 PC observations).

The structural model has three layers (Han 2015 Results, “Overall mixed-effect PK-PD analysis”):

  • PK – two-compartment IV-infusion model parameterised on a body-surface-area-normalized (per m^2) scale: CL (L/h per m^2), Vc (L per m^2), Vp (L per m^2), Q (L/h per m^2). The trial administered decitabine as a 60-minute IV infusion at 4-12 mg/m^2/day for five consecutive days, repeated every ~4 weeks (mean actual interval 34.5 days, SD 8.7).
  • Absolute neutrophil count (ANC) – Friberg / Wallin transit compartment model with feedback (1 proliferating + 3 transit + 1 circulating; uniform ktr_N applied across proliferation, transit, and circulating-pool clearance). Linear drug effect on the proliferation rate (1 - SLOPE_N * Cc) and feedback (BASE_N / circ_anc)^GAMMA_N. Baseline is time-invariant: the paper reports “baseline cell count increase was not meaningful” for neutrophils.
  • Platelet count (PC) – Friberg / Wallin transit compartment model with feedback, identical structure to the ANC chain but with a time-dependent baseline that increases asymptotically over treatment cycles: BASE_P_t = BASE_P + IMP * (1 - exp(-IMK * TIME)). The time-dependent BASE_P_t substitutes for BASE_P inside the feedback term only; the transit kinetics use the structural ktr_P unchanged.

Population 

mod_meta$meta$population
#> $species
#> [1] "human"
#> 
#> $n_subjects
#> [1] 15
#> 
#> $n_studies
#> [1] 1
#> 
#> $age_range
#> [1] "19-64 years"
#> 
#> $weight_range
#> [1] NA
#> 
#> $sex_female_pct
#> [1] 40
#> 
#> $race_ethnicity
#> Asian 
#>   100 
#> 
#> $disease_state
#> [1] "Adult Korean patients with higher-risk myelodysplastic syndrome (MDS; intermediate-2 or high IPSS risk) or secondary acute myeloid leukemia (sAML) evolving from MDS, receiving decitabine maintenance after allogeneic hematopoietic stem cell transplantation (allo-HSCT). Eligibility: age <= 65; PC > 30,000/mm^3 and ANC > 1000/mm^3 maintained > 7 days without transfusion or growth factors; no grade III/IV acute GVHD; ECOG 0-2; no renal or hepatic impairment. Decitabine started on days 42-90 post-transplant (median 86 days)."
#> 
#> $dose_range
#> [1] "Decitabine 4-12 mg/m^2/day (Cycle 2-4 range 1.5-12 mg/m^2/day from Table 2) as a 60-minute IV infusion for 5 consecutive days, cycles repeated every 4 weeks up to Cycle 12. Initial Cycle-1 dose 5 mg/m^2/day for cohort 1; subsequent cohort initial doses estimated by cohort dose estimation (CDE) at 4, 5, 5.5, and 5 mg/m^2/day for cohorts 2-5. Doses for Cycles 2-4 individualized via PK-PD adaptive titration (IDT); the Cycle-4 dose was maintained for all subsequent cycles. The actual dosing interval was 34.5 +/- 8.7 days (mean +/- SD)."
#> 
#> $regions
#> [1] "Republic of Korea (Seoul St. Mary's Hospital, The Catholic University of Korea)."
#> 
#> $co_medication
#> [1] "GVHD prophylaxis: calcineurin inhibitor (cyclosporine for related donors, tacrolimus for unrelated donors) plus short-course methotrexate. Antithymocyte globulin given to all patients prior to transplant. At decitabine initiation: acute GVHD grade 0-2 in all patients, mild chronic GVHD in 1 patient."
#> 
#> $notes
#> [1] "Baseline demographics from Han 2015 Tables 1 and 2. 16 patients were enrolled (9 male / 7 female); the popPK-PD analysis excluded 1 female patient whose dose-limiting toxicity factor was platelet count due to immune thrombocytopenia after transplantation (managed with steroids), leaving 15 patients in the mixed-effect analysis (8 male / 6 female assumed; sex_female_pct = 40 reflects 6/15). Donor type for the 16 enrolled: matched sibling donor (n=7), matched unrelated donor (n=8), partially matched unrelated donor (n=1). WHO diagnoses: RAEB-1 (n=1), RAEB-2 (n=10), AML evolving from MDS (n=5). Final analysis dataset: 95 PK observations and 622 PD observations (311 ANC + 311 PC) across the 15 subjects (Han 2015 Results, 'Overall mixed-effect PK-PD analysis'). PK proportional residual error variance sigma^2_PK = 0.441 (= 66% CV); PD additive residual error variances sigma^2_PD,P = 25000 /mm^3 squared and sigma^2_PD,N = 754 /mm^3 squared (Table 3). Successful bootstrap convergence proportion was 78.8% (PK) and 78.0% (PD)."

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry. Every numeric value below was taken from Table 3 of Han et al. 2015 (“Final parameter estimates and bootstrap outcomes”). The BSV columns (reported as %CV) were converted to log-scale variances via omega^2 = log(1 + CV^2); residual error variances were converted to SDs via sqrt(.) with a /mm^3 -> 10^9/L unit conversion (divide by 1000).

Equation / parameter Value Source location (Han 2015)
2-cmt IV infusion ODE n/a Methods + Results “A two-compartment model was found to best describe the PK data”
Friberg / Wallin chain n/a Results, formula dA(N)/dt = ktr*A(N-1) - ktr*A(N) + feedback; cites Wallin 2009 [33]
lcl = log(87.8) (L/h per m^2) 87.8 Table 3, PK parameters row CL
lvc = log(18.5) (L per m^2) 18.5 Table 3, PK parameters row Vc
lvp = log(22.9) (L per m^2) 22.9 Table 3, PK parameters row Vp
lq = log(13.1) (L/h per m^2) 13.1 Table 3, PK parameters row Q
lktr_anc = log(0.0132) (1/h) 0.0132 Table 3, neutrophil row ktr,N
lslope_anc = log(0.263) (L/mg) 0.263 Table 3, neutrophil row SLOPE_N
lbase_anc = log(3.24) (10^9/L) 3240/mm^3 Table 3, neutrophil row BASE_N
lgamma_anc = log(0.193) 0.193 Table 3, neutrophil row GAMMA_N
lktr_plt = log(0.0244) (1/h) 0.0244 Table 3, platelet row ktr,P
lslope_plt = log(0.0656) (L/mg) 0.0656 Table 3, platelet row SLOPE_P
lbase_plt = log(49.2) (10^9/L) 49200/mm^3 Table 3, platelet row BASE_P
lgamma_plt = log(0.304) 0.304 Table 3, platelet row GAMMA_P
limp = log(55) (10^9/L) 55000/mm^3 Table 3, platelet row IMP
limk = log(5.3e-4) (1/h) 0.000530 Table 3, platelet row IMK
etalcl ~ log(1+0.214^2) 0.0448 Table 3, BSV CL = 21.4% CV
etalslope_anc ~ log(1+0.575^2) 0.286 Table 3, BSV SLOPE_N = 57.5% CV
etalbase_anc ~ log(1+0.435^2) 0.173 Table 3, BSV BASE_N = 43.5% CV
etalgamma_anc ~ log(1+0.394^2) 0.144 Table 3, BSV GAMMA_N = 39.4% CV
etalslope_plt ~ log(1+0.257^2) 0.0640 Table 3, BSV SLOPE_P = 25.7% CV
etalbase_plt ~ log(1+1.050^2) 0.743 Table 3, BSV BASE_P = 105% CV
etalimp ~ log(1+0.783^2) 0.478 Table 3, BSV IMP = 78.3% CV
propSd = sqrt(0.441) 0.664 Table 3, sigma^2_PK = 0.441
addSd_ANC = sqrt(754)/1000 0.0275 Table 3, sigma^2_PD,N = 754 /mm32; /1000 -> 10^9/L
addSd_PLT = sqrt(25000)/1000 0.158 Table 3, sigma^2_PD,P = 25000 /mm32; /1000 -> 10^9/L

Table 3 reports “NE” (not estimated) BSV for Vc, Vp, Q, ktr_N, ktr_P, GAMMA_P, and IMK; these are carried as fixed structural parameters in the model (no corresponding eta). The paper’s covariate analysis (Results, “No meaningful covariate was found in either the patient demographic or clinical variables”) found no retained covariates for sex, age, baseline body weight, body-surface-area, or any clinical-lab variable, so the model has covariateData = list().

Virtual cohort

The trial schedule for the 15 popPK-PD subjects was decitabine 4-12 mg/m^2/day as a 60-minute IV infusion for 5 consecutive days, repeated every 4 weeks (28-day protocol) up to Cycle 12; the median maintenance dose was 7 mg/m^2/day. The first reported recommendation is 5 mg/m^2/day for five consecutive days as the appropriate starting dose.

The simulation below builds a 4-cycle (16-week) virtual cohort at the recommended initial dose of 5 mg/m^2/day for 5 consecutive days per cycle, with cycles repeated every 28 days (the protocol interval). PK parameters are on a per-m^2 scale and the simulated doses are entered in mg/m^2 directly, so the simulated central concentration is in mg/L (= ug/mL = 1000 ng/mL).

set.seed(20151028)

n_subjects <- 200L
mg_per_m2  <- 5            # initial dose
dose_h     <- 1            # 60-minute infusion
inf_rate   <- mg_per_m2 / dose_h  # mg/m^2/h
days_per_cycle <- 5
cycle_h       <- 28 * 24   # 4-week protocol interval (hours)
n_cycles      <- 4
total_h       <- n_cycles * cycle_h

# Dense PK sampling on Cycle 1 Day 1 + sparse PD sampling weekly thereafter,
# matching the paper's protocol (PK on Cycle 1 Day 1; PD weekly through
# Cycle 4).
pk_obs_times <- c(0, 1/3, 2/3, 1, 1.5, 2, 3)  # h after first dose
pd_obs_times <- seq(0, total_h, by = 24 * 7)  # weekly through 16 weeks

# Five daily 1-h infusions per cycle, starting at the cycle start time
dose_starts <- as.vector(outer(
  seq(0, by = 24, length.out = days_per_cycle),         # 5 daily doses
  seq(0, by = cycle_h, length.out = n_cycles),          # cycle starts
  FUN = "+"
))

make_events <- function(id) {
  ev_dose <- data.frame(
    id   = id,
    time = dose_starts,
    evid = 1L,
    amt  = mg_per_m2,
    cmt  = "central",
    rate = inf_rate,
    dur  = NA_real_
  )
  ev_obs_Cc  <- data.frame(id = id, time = pk_obs_times,
                           evid = 0L, amt = 0, cmt = "Cc",
                           rate = NA_real_, dur = NA_real_)
  ev_obs_ANC <- data.frame(id = id, time = pd_obs_times,
                           evid = 0L, amt = 0, cmt = "ANC",
                           rate = NA_real_, dur = NA_real_)
  ev_obs_PLT <- data.frame(id = id, time = pd_obs_times,
                           evid = 0L, amt = 0, cmt = "PLT",
                           rate = NA_real_, dur = NA_real_)
  dplyr::bind_rows(ev_dose, ev_obs_Cc, ev_obs_ANC, ev_obs_PLT)
}

events <- dplyr::bind_rows(lapply(seq_len(n_subjects), make_events))
nrow(events); length(unique(events$id))
#> [1] 12200
#> [1] 200

Simulation 

mod <- mod_meta

sim <- rxode2::rxSolve(
  mod,
  events = events
) |>
  as.data.frame()

mod_typical <- mod |> rxode2::zeroRe()
sim_typical <- rxode2::rxSolve(
  mod_typical,
  events = events |> dplyr::filter(id == 1L)
) |>
  as.data.frame()
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalslope_anc', 'etalbase_anc', 'etalgamma_anc', 'etalslope_plt', 'etalbase_plt', 'etalimp'

Replicate published figures

Cmax cross-check (Discussion, page 4)

The paper’s Discussion states: “the predicted C_max after 1-h infusion of 5 mg/m^2 was 66.0 ng/mL.” Reproduce by reading off the typical-value trajectory at the end of the first infusion (t = 1 h).

Paper-internal discrepancy. The published Table 3 parameters imply an asymptotic infusion concentration k0/CL = 5/87.8 = 57 ng/mL as a ceiling (i.e. the highest Cc reachable for an infinite-duration infusion at this dose rate); the actual 1-h end-of-infusion typical value computed analytically from the 2-cmt parameters in Table 3 is ~51 ng/mL. The paper’s quoted 66 ng/mL therefore cannot come from the Table 3 typical-value PK; it may reflect an alternative dose or rate than the text states, an inclusion of IIV (a higher percentile of the predicted-Cmax distribution), or a transcription error in the paper. The model file in this package faithfully reproduces Table 3, so the simulated Cmax below matches the parameters as published (~51 ng/mL). This is flagged as a > 20% discrepancy with the paper’s narrative; see the verification-checklist policy of “flag, do not tune.”

cmax_check <- sim_typical |>
  dplyr::filter(time >= 0, time <= 3) |>
  dplyr::select(time, Cc) |>
  dplyr::mutate(Cc_ngmL = Cc * 1000)

ggplot(cmax_check, aes(time, Cc_ngmL)) +
  geom_line(linewidth = 0.9) +
  geom_vline(xintercept = 1, linetype = 2) +
  labs(x = "Time after first dose (h)",
       y = "Decitabine plasma concentration (ng/mL)",
       title = "Cycle 1 Day 1 typical-value profile after 1-h infusion of 5 mg/m^2",
       caption = "Vertical dashed line marks end of infusion.") +
  theme_minimal()


cmax_pred <- cmax_check |>
  dplyr::filter(time <= 1) |>
  dplyr::pull(Cc_ngmL) |>
  max()

knitr::kable(data.frame(
  metric   = c("Predicted Cmax (ng/mL) at end of 1-h infusion of 5 mg/m^2"),
  this_sim = round(cmax_pred, 1),
  paper    = "~66.0"
))
metric this_sim paper
Predicted Cmax (ng/mL) at end of 1-h infusion of 5 mg/m^2 51.3 ~66.0

Figure 2 – ANC trajectory at 5 mg/m^2/day maintenance over 4 cycles

Han 2015 Figure 2 displays simulated time courses of ANC under the maintenance dosage of 5 mg/m^2/day for four treatment cycles, including the median and prediction intervals. The two qualitative features the text calls out are (i) the lowest ANC value appears in the second cycle (6-7 weeks after treatment initiation), and (ii) the time to recovery to ANC > 1000/mm^3 is approximately 5 weeks for the ANC.

sim_anc <- sim |>
  dplyr::filter(!is.na(ANC)) |>
  dplyr::mutate(week = time / 168)

anc_quants <- sim_anc |>
  dplyr::group_by(week) |>
  dplyr::summarise(
    p10 = stats::quantile(ANC, 0.10, na.rm = TRUE),
    p25 = stats::quantile(ANC, 0.25, na.rm = TRUE),
    p50 = stats::quantile(ANC, 0.50, na.rm = TRUE),
    p75 = stats::quantile(ANC, 0.75, na.rm = TRUE),
    p90 = stats::quantile(ANC, 0.90, na.rm = TRUE),
    .groups = "drop"
  )

ggplot(anc_quants, aes(week, p50)) +
  geom_ribbon(aes(ymin = p10, ymax = p90), fill = "steelblue", alpha = 0.15) +
  geom_ribbon(aes(ymin = p25, ymax = p75), fill = "steelblue", alpha = 0.25) +
  geom_line(linewidth = 0.9, colour = "steelblue") +
  geom_hline(yintercept = 1.0, linetype = 2, colour = "darkred") +   # 1000/mm^3 = 1.0 * 10^9/L
  geom_hline(yintercept = 0.5, linetype = 2, colour = "indianred") + # 500/mm^3 = 0.5 * 10^9/L grade 4 toxicity
  labs(x = "Time since treatment initiation (weeks)",
       y = "Absolute neutrophil count (10^9 cells/L)",
       title = "Replicates Han 2015 Figure 2 (ANC time course at 5 mg/m^2/day)",
       caption = paste0(
         "Median, 50% (dark) and 80% (light) prediction intervals from ",
         n_subjects, " subjects. ",
         "Dashed lines: ANC = 1.0 (recovery threshold) and ANC = 0.5 ",
         "(grade 4 toxicity) x 10^9/L."
       )) +
  theme_minimal()


# Time to nadir
nadir_week <- anc_quants$week[which.min(anc_quants$p50)]
recovery_week <- anc_quants |>
  dplyr::filter(week > nadir_week, p50 >= 1.0) |>
  dplyr::slice_head(n = 1) |>
  dplyr::pull(week)

knitr::kable(data.frame(
  metric        = c("Week of median-ANC nadir (Cycle 2 region per paper text)",
                    "Week of median ANC recovery to >= 1.0 x 10^9/L"),
  this_sim      = c(round(nadir_week, 1),
                    if (length(recovery_week)) round(recovery_week, 1) else NA),
  paper_text    = c("3.5 weeks (time-to-nadir), nadir achieved in 2nd cycle (6-7 weeks)",
                    "~5 weeks from last dose")
))
metric this_sim paper_text
Week of median-ANC nadir (Cycle 2 region per paper text) 7 3.5 weeks (time-to-nadir), nadir achieved in 2nd cycle (6-7 weeks)
Week of median ANC recovery to >= 1.0 x 10^9/L 8 ~5 weeks from last dose

Platelet trajectory with cycle-asymptotic baseline rise

The paper’s PD model adds an asymptotic baseline-rise structure for platelets (BASE_P_t = BASE_P + IMP * (1 - exp(-IMK * TIME))) to capture the gradual platelet-count increase over cycles – a contribution of decitabine to cell proliferation reported in previous studies.

sim_plt <- sim |>
  dplyr::filter(!is.na(PLT)) |>
  dplyr::mutate(week = time / 168)

plt_quants <- sim_plt |>
  dplyr::group_by(week) |>
  dplyr::summarise(
    p10 = stats::quantile(PLT, 0.10, na.rm = TRUE),
    p25 = stats::quantile(PLT, 0.25, na.rm = TRUE),
    p50 = stats::quantile(PLT, 0.50, na.rm = TRUE),
    p75 = stats::quantile(PLT, 0.75, na.rm = TRUE),
    p90 = stats::quantile(PLT, 0.90, na.rm = TRUE),
    .groups = "drop"
  )

ggplot(plt_quants, aes(week, p50)) +
  geom_ribbon(aes(ymin = p10, ymax = p90), fill = "darkgreen", alpha = 0.15) +
  geom_ribbon(aes(ymin = p25, ymax = p75), fill = "darkgreen", alpha = 0.25) +
  geom_line(linewidth = 0.9, colour = "darkgreen") +
  geom_hline(yintercept = 30, linetype = 2, colour = "darkred") +  # 30,000/mm^3 holding threshold
  geom_hline(yintercept = 25, linetype = 3, colour = "indianred") +# 25,000/mm^3 grade 4 toxicity
  labs(x = "Time since treatment initiation (weeks)",
       y = "Platelet count (10^9 cells/L)",
       title = "Platelet time course at 5 mg/m^2/day maintenance",
       caption = paste(
         "Median + 50% (dark) and 80% (light) prediction intervals.",
         "Dashed line: PLT = 30 x 10^9/L (dose-hold);",
         "dotted line: PLT = 25 x 10^9/L (grade 4 toxicity)."
       )) +
  theme_minimal()

PKNCA validation

The paper’s Discussion provides two cross-check NCA values: predicted Cmax = 66.0 ng/mL after a 1-h infusion of 5 mg/m^2, and a typical half-life of 0.31 h. PKNCA below evaluates Cycle-1-Day-1 (single-dose) NCA on the dense PK sampling grid (0, 20, 40, 60, 90, 120, 180 min) and compares Cmax and half-life with the paper.

sim_nca <- sim |>
  dplyr::filter(time <= 3, !is.na(Cc)) |>
  dplyr::distinct(id, time, .keep_all = TRUE) |>
  dplyr::select(id, time, Cc) |>
  dplyr::mutate(Cc_ngmL = Cc * 1000)

dose_df <- events |>
  dplyr::filter(evid == 1L, time == 0) |>
  dplyr::distinct(id, time, .keep_all = TRUE) |>
  dplyr::transmute(id, time, amt, treatment = "5 mg/m^2 1-h IV")

conc_df <- sim_nca |> dplyr::mutate(treatment = "5 mg/m^2 1-h IV")

conc_obj <- PKNCA::PKNCAconc(
  conc_df,
  Cc_ngmL ~ time | treatment / id,
  concu = "ng/mL",
  timeu = "h"
)
dose_obj <- PKNCA::PKNCAdose(
  dose_df,
  amt ~ time | treatment + id,
  doseu = "mg/m^2"
)

intervals <- data.frame(
  start    = 0,
  end      = 3,
  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)
)

knitr::kable(summary(nca_res),
             caption = "Cycle-1 Day-1 single-dose NCA (200 subjects, 5 mg/m^2 1-h IV).")
Cycle-1 Day-1 single-dose NCA (200 subjects, 5 mg/m^2 1-h IV).
Interval Start Interval End id N AUClast (h*ng/mL) Cmax (ng/mL) Tmax (h) Half-life (h) AUCinf,obs (h*ng/mL)
0 3 1 1 70.0 65.5 1.00 0.696 72.4
0 3 2 1 55.8 53.7 1.00 0.733 57.3
0 3 3 1 66.5 62.7 1.00 0.701 68.7
0 3 4 1 42.5 42.0 1.00 0.821 43.4
0 3 5 1 52.4 50.8 1.00 0.749 53.7
0 3 6 1 56.4 54.3 1.00 0.730 57.9
0 3 7 1 53.7 52.0 1.00 0.742 55.2
0 3 8 1 57.7 55.4 1.00 0.725 59.3
0 3 9 1 60.9 58.1 1.00 0.715 62.7
0 3 10 1 63.2 59.9 1.00 0.709 65.1
0 3 11 1 55.1 53.2 1.00 0.736 56.6
0 3 12 1 61.4 58.5 1.00 0.713 63.2
0 3 13 1 71.5 66.6 1.00 0.694 74.0
0 3 14 1 62.0 59.0 1.00 0.712 63.9
0 3 15 1 74.0 68.6 1.00 0.691 76.6
0 3 16 1 47.5 46.6 1.00 0.778 48.7
0 3 17 1 52.4 50.8 1.00 0.749 53.8
0 3 18 1 73.4 68.1 1.00 0.692 76.0
0 3 19 1 46.6 45.7 1.00 0.785 47.7
0 3 20 1 66.6 62.8 1.00 0.701 68.8
0 3 21 1 58.3 55.9 1.00 0.723 59.9
0 3 22 1 50.7 49.4 1.00 0.758 52.0
0 3 23 1 64.6 61.1 1.00 0.705 66.6
0 3 24 1 36.7 36.8 1.00 0.888 37.5
0 3 25 1 41.6 41.2 1.00 0.830 42.5
0 3 26 1 61.1 58.2 1.00 0.714 62.9
0 3 27 1 67.8 63.7 1.00 0.699 70.0
0 3 28 1 62.3 59.2 1.00 0.711 64.2
0 3 29 1 57.0 54.8 1.00 0.728 58.6
0 3 30 1 62.0 59.0 1.00 0.712 63.9
0 3 31 1 42.8 42.3 1.00 0.818 43.8
0 3 32 1 45.1 44.4 1.00 0.798 46.1
0 3 33 1 43.3 42.8 1.00 0.813 44.3
0 3 34 1 36.5 36.6 1.00 0.891 37.3
0 3 35 1 52.6 51.0 1.00 0.747 54.0
0 3 36 1 46.4 45.6 1.00 0.787 47.5
0 3 37 1 47.2 46.3 1.00 0.781 48.3
0 3 38 1 48.1 47.0 1.00 0.775 49.3
0 3 39 1 39.2 39.0 1.00 0.857 40.0
0 3 40 1 57.5 55.2 1.00 0.726 59.2
0 3 41 1 72.3 67.3 1.00 0.693 74.8
0 3 42 1 46.7 45.9 1.00 0.784 47.9
0 3 43 1 37.5 37.5 1.00 0.878 38.3
0 3 44 1 47.2 46.2 1.00 0.781 48.3
0 3 45 1 55.9 53.9 1.00 0.732 57.4
0 3 46 1 40.9 40.6 1.00 0.837 41.8
0 3 47 1 35.4 35.5 1.00 0.908 36.1
0 3 48 1 68.3 64.1 1.00 0.698 70.5
0 3 49 1 71.0 66.2 1.00 0.695 73.4
0 3 50 1 47.3 46.4 1.00 0.780 48.5
0 3 51 1 57.3 55.1 1.00 0.727 58.9
0 3 52 1 35.7 35.8 1.00 0.903 36.4
0 3 53 1 58.6 56.2 1.00 0.722 60.3
0 3 54 1 54.7 52.8 1.00 0.738 56.1
0 3 55 1 52.2 50.6 1.00 0.750 53.5
0 3 56 1 73.5 68.2 1.00 0.692 76.1
0 3 57 1 66.8 62.9 1.00 0.701 69.0
0 3 58 1 68.5 64.3 1.00 0.698 70.7
0 3 59 1 55.3 53.3 1.00 0.735 56.8
0 3 60 1 66.2 62.4 1.00 0.702 68.3
0 3 61 1 54.6 52.8 1.00 0.738 56.1
0 3 62 1 57.7 55.4 1.00 0.725 59.3
0 3 63 1 52.3 50.8 1.00 0.749 53.7
0 3 64 1 47.9 46.9 1.00 0.776 49.1
0 3 65 1 46.1 45.3 1.00 0.789 47.2
0 3 66 1 48.3 47.2 1.00 0.773 49.4
0 3 67 1 46.5 45.6 1.00 0.786 47.6
0 3 68 1 59.3 56.7 1.00 0.720 61.0
0 3 69 1 29.2 29.6 1.00 1.02 29.7
0 3 70 1 49.7 48.5 1.00 0.764 50.9
0 3 71 1 50.5 49.2 1.00 0.759 51.8
0 3 72 1 46.2 45.4 1.00 0.788 47.3
0 3 73 1 44.5 43.8 1.00 0.802 45.5
0 3 74 1 47.9 46.9 1.00 0.776 49.1
0 3 75 1 48.7 47.6 1.00 0.770 49.9
0 3 76 1 52.0 50.5 1.00 0.751 53.4
0 3 77 1 45.2 44.5 1.00 0.796 46.3
0 3 78 1 79.8 73.0 1.00 0.689 82.9
0 3 79 1 77.7 71.4 1.00 0.689 80.6
0 3 80 1 41.3 41.0 1.00 0.833 42.2
0 3 81 1 51.0 49.6 1.00 0.756 52.3
0 3 82 1 41.7 41.3 1.00 0.829 42.6
0 3 83 1 61.6 58.6 1.00 0.713 63.4
0 3 84 1 62.3 59.2 1.00 0.711 64.2
0 3 85 1 22.7 23.4 1.00 1.15 23.1
0 3 86 1 51.4 49.9 1.00 0.754 52.7
0 3 87 1 32.9 33.2 1.00 0.948 33.6
0 3 88 1 35.2 35.3 1.00 0.912 35.9
0 3 89 1 56.5 54.4 1.00 0.730 58.1
0 3 90 1 53.6 51.8 1.00 0.743 55.0
0 3 91 1 60.7 57.9 1.00 0.715 62.5
0 3 92 1 62.8 59.6 1.00 0.710 64.7
0 3 93 1 47.3 46.3 1.00 0.781 48.4
0 3 94 1 45.1 44.4 1.00 0.797 46.1
0 3 95 1 60.1 57.4 1.00 0.717 61.8
0 3 96 1 48.4 47.4 1.00 0.772 49.6
0 3 97 1 53.8 52.1 1.00 0.742 55.2
0 3 98 1 65.3 61.7 1.00 0.704 67.4
0 3 99 1 75.6 69.8 1.00 0.690 78.4
0 3 100 1 63.6 60.3 1.00 0.708 65.5
0 3 101 1 53.2 51.6 1.00 0.744 54.7
0 3 102 1 59.7 57.0 1.00 0.719 61.4
0 3 103 1 52.4 50.9 1.00 0.749 53.8
0 3 104 1 56.7 54.5 1.00 0.729 58.3
0 3 105 1 54.9 53.0 1.00 0.737 56.4
0 3 106 1 46.0 45.2 1.00 0.790 47.1
0 3 107 1 50.5 49.2 1.00 0.759 51.8
0 3 108 1 63.0 59.8 1.00 0.709 64.9
0 3 109 1 56.2 54.1 1.00 0.731 57.7
0 3 110 1 70.6 65.9 1.00 0.695 73.0
0 3 111 1 38.2 38.1 1.00 0.868 39.1
0 3 112 1 65.7 62.0 1.00 0.703 67.8
0 3 113 1 56.3 54.2 1.00 0.731 57.9
0 3 114 1 61.5 58.6 1.00 0.713 63.3
0 3 115 1 53.6 51.9 1.00 0.743 55.0
0 3 116 1 31.2 31.5 1.00 0.980 31.8
0 3 117 1 47.0 46.1 1.00 0.782 48.2
0 3 118 1 45.4 44.6 1.00 0.795 46.4
0 3 119 1 42.0 41.6 1.00 0.826 42.9
0 3 120 1 53.2 51.6 1.00 0.744 54.6
0 3 121 1 59.1 56.6 1.00 0.720 60.8
0 3 122 1 69.3 64.9 1.00 0.697 71.6
0 3 123 1 64.7 61.2 1.00 0.705 66.7
0 3 124 1 57.1 54.9 1.00 0.727 58.7
0 3 125 1 51.2 49.8 1.00 0.755 52.5
0 3 126 1 41.3 40.9 1.00 0.833 42.2
0 3 127 1 54.7 52.8 1.00 0.738 56.2
0 3 128 1 36.6 36.6 1.00 0.891 37.4
0 3 129 1 64.2 60.8 1.00 0.706 66.2
0 3 130 1 60.2 57.5 1.00 0.717 62.0
0 3 131 1 39.9 39.6 1.00 0.849 40.7
0 3 132 1 69.0 64.7 1.00 0.697 71.3
0 3 133 1 41.9 41.5 1.00 0.827 42.8
0 3 134 1 48.9 47.8 1.00 0.769 50.1
0 3 135 1 52.9 51.2 1.00 0.746 54.2
0 3 136 1 41.4 41.0 1.00 0.832 42.3
0 3 137 1 72.9 67.7 1.00 0.692 75.4
0 3 138 1 48.3 47.3 1.00 0.773 49.5
0 3 139 1 46.0 45.2 1.00 0.790 47.1
0 3 140 1 38.0 37.9 1.00 0.872 38.8
0 3 141 1 48.0 47.0 1.00 0.775 49.2
0 3 142 1 30.8 31.2 1.00 0.986 31.4
0 3 143 1 67.7 63.6 1.00 0.699 69.9
0 3 144 1 67.4 63.4 1.00 0.700 69.6
0 3 145 1 54.4 52.6 1.00 0.739 55.9
0 3 146 1 52.0 50.5 1.00 0.751 53.3
0 3 147 1 70.0 65.4 1.00 0.696 72.3
0 3 148 1 36.4 36.4 1.00 0.893 37.2
0 3 149 1 29.7 30.1 1.00 1.01 30.3
0 3 150 1 73.2 68.0 1.00 0.692 75.8
0 3 151 1 69.6 65.2 1.00 0.696 72.0
0 3 152 1 57.5 55.2 1.00 0.726 59.1
0 3 153 1 45.4 44.7 1.00 0.795 46.5
0 3 154 1 41.8 41.4 1.00 0.828 42.7
0 3 155 1 43.2 42.6 1.00 0.814 44.1
0 3 156 1 71.4 66.6 1.00 0.694 73.9
0 3 157 1 58.0 55.7 1.00 0.724 59.7
0 3 158 1 52.8 51.1 1.00 0.747 54.1
0 3 159 1 52.2 50.6 1.00 0.750 53.5
0 3 160 1 73.9 68.5 1.00 0.692 76.5
0 3 161 1 48.4 47.3 1.00 0.773 49.6
0 3 162 1 50.7 49.4 1.00 0.758 52.0
0 3 163 1 76.4 70.5 1.00 0.690 79.3
0 3 164 1 40.4 40.1 1.00 0.843 41.3
0 3 165 1 64.3 60.8 1.00 0.706 66.3
0 3 166 1 63.8 60.5 1.00 0.707 65.8
0 3 167 1 49.2 48.0 1.00 0.768 50.4
0 3 168 1 63.0 59.8 1.00 0.709 65.0
0 3 169 1 65.5 61.9 1.00 0.703 67.6
0 3 170 1 47.1 46.2 1.00 0.782 48.3
0 3 171 1 56.4 54.2 1.00 0.730 57.9
0 3 172 1 38.7 38.6 1.00 0.862 39.6
0 3 173 1 71.1 66.3 1.00 0.694 73.5
0 3 174 1 57.1 54.9 1.00 0.728 58.7
0 3 175 1 66.2 62.4 1.00 0.702 68.3
0 3 176 1 71.2 66.4 1.00 0.694 73.7
0 3 177 1 79.2 72.6 1.00 0.689 82.3
0 3 178 1 64.1 60.7 1.00 0.706 66.1
0 3 179 1 69.7 65.2 1.00 0.696 72.0
0 3 180 1 65.9 62.2 1.00 0.703 68.0
0 3 181 1 54.4 52.6 1.00 0.739 55.9
0 3 182 1 65.3 61.7 1.00 0.704 67.3
0 3 183 1 72.6 67.5 1.00 0.693 75.1
0 3 184 1 38.9 38.8 1.00 0.860 39.8
0 3 185 1 55.9 53.9 1.00 0.732 57.5
0 3 186 1 56.7 54.5 1.00 0.729 58.2
0 3 187 1 48.8 47.7 1.00 0.770 50.1
0 3 188 1 51.5 50.0 1.00 0.754 52.8
0 3 189 1 46.2 45.4 1.00 0.788 47.3
0 3 190 1 45.4 44.6 1.00 0.795 46.4
0 3 191 1 48.6 47.5 1.00 0.771 49.8
0 3 192 1 46.9 46.0 1.00 0.783 48.0
0 3 193 1 68.8 64.5 1.00 0.698 71.0
0 3 194 1 53.1 51.4 1.00 0.745 54.5
0 3 195 1 44.5 43.8 1.00 0.802 45.5
0 3 196 1 39.8 39.6 1.00 0.850 40.7
0 3 197 1 45.0 44.3 1.00 0.798 46.1
0 3 198 1 66.5 62.7 1.00 0.701 68.7
0 3 199 1 54.0 52.2 1.00 0.741 55.4
0 3 200 1 45.6 44.9 1.00 0.793 46.7 
pp_res <- as.data.frame(nca_res$result)
cmax_median <- median(pp_res$PPORRES[pp_res$PPTESTCD == "cmax"], na.rm = TRUE)
hl_median   <- median(pp_res$PPORRES[pp_res$PPTESTCD == "half.life"], na.rm = TRUE)

knitr::kable(data.frame(
  metric        = c("Cmax (ng/mL)", "Apparent half-life (h)"),
  this_sim_med  = c(round(cmax_median, 1), round(hl_median, 2)),
  paper_target  = c("~66 (Discussion, page 4)", "~0.31 (Discussion, page 4)")
))
metric this_sim_med paper_target
Cmax (ng/mL) 51.90 ~66 (Discussion, page 4)
Apparent half-life (h) 0.74 ~0.31 (Discussion, page 4)

Assumptions and deviations

  • Per-m^2 PK scale and dosing unit. Han 2015 reports PK parameters on a body-surface-area-normalized scale (L/h per m^2 and L per m^2); this is the convention for cytotoxic dosing in oncology. Doses are consequently entered in mg/m^2 (not mg), and the central compartment carries mg/m^2 – central / vc yields mg/L directly. Users who wish to dose in mg for a patient with known BSA must pre-multiply the absolute mg dose by 1 / BSA (m^2) before passing it to the model, or alternatively rescale the per-m^2 PK parameters by their patient’s BSA. The trial protocol always recorded doses in mg/m^2/day, so the per-m^2 model is the operational convention.
  • No covariates in the final model. Han 2015 reports “no meaningful covariate was found in either the patient demographic or clinical variables”, so covariateData = list(). Demographic facts (age range 19-64 years, 40% female, all Korean) are recorded in population metadata for downstream filtering but do not affect simulation.
  • Vc, Vp, Q, ktr_N, ktr_P, GAMMA_P, and IMK have no IIV. Table 3’s BSV column reports “NE” (not estimated) for these parameters; the Methods says “BSV … was the only random effect which could be estimated, except for the proportional residual error” for PK CL. The model carries these as typical-value-only parameters (no corresponding eta in ini()), consistent with the paper’s reporting.
  • Friberg / Wallin transit kinetics use a single ktr. Both ANC and PLT chains use one structural ktr (the rate constant of inter- compartmental movement, reported in 1/h directly in Table 3) applied uniformly to the proliferation, transit, and circulating compartments. This matches the Wallin 2009 form the paper cites (reference 33). The implied mean transit time MTT = 4 / ktr is 303 h (12.6 days) for ANC and 164 h (6.8 days) for PLT.
  • Time-varying platelet baseline enters via feedback only. The BASE_P_t = BASE_P + IMP * (1 - exp(-IMK * TIME)) structure substitutes for BASE_P inside the Friberg feedback term (BASE_P_t / circ_plt)^GAMMA_P only; the transit kinetics use the structural ktr_P unchanged, and the initial conditions precursor*_plt(0) = circ_plt(0) = BASE_P are correct because BASE_P_t(t = 0) = BASE_P. This faithfully reproduces the paper’s second formula in “Overall mixed-effect PK-PD analysis”.
  • Time origin for IMK. The paper writes the platelet asymptote in terms of TIME (time from initiation of decitabine treatment). The model uses rxode2’s reserved variable t, which is simulation time; for any rxode2 events table that places the first dose at t = 0 the two are identical. Users who define event tables with t < 0 for pre-dose observations will see BASE_P_t < BASE_P because 1 - exp(-IMK*t) < 0 for t < 0; the canonical usage is to start at t = 0.
  • Initial-condition simplification. Friberg chains are placed at steady-state at t = 0 (all five compartments equal to BASE, consistent with the Friberg derivation where the unperturbed system has precursor1 = ... = circ at the baseline value). The paper does not provide a separate ramp-in period; the trial enrolled patients whose blood counts had recovered to PC > 30,000/mm^3 and ANC > 1000/mm^3 prior to decitabine initiation.
  • Half-life cross-check. The paper’s Discussion reports an “average terminal half-life … of 0.31 h in this study”. With CL/Vc = 87.8 / 18.5 = 4.745 /h the rapid distribution half-life is ln(2)/4.745 = 0.146 h; the slower beta phase is around 1.4 h (per analytic 2-cmt eigenvalues from CL, Vc, Vp, Q). The simulated single-dose half-life PKNCA estimates (above) reflect the average behaviour across the observed sampling window (3 h post-dose) where the alpha and beta phases mix; the apparent half-life depends sensitively on which terminal-phase points the NCA algorithm uses. The published “0.31 h” is an empirical estimate from non-compartmental analysis on the actual clinical samples (Discussion, p. 4) – the PKNCA value here is on simulated typical-value data and is reported only as a sanity check, not as a faithful reproduction of the empirical estimate.
  • circ_anc, precursor*_anc, circ_plt, precursor*_plt compartment names deviate from the canonical compartment register (depot, central, peripheral<n>, effect, …). They are retained because they directly mirror the standard Friberg / Wallin notation (proliferation -> 3 transits -> circulating); the Friberg-precedent files (Friberg_2002_paclitaxel in DDMORE, Petrov_2024_romiplostim, Netterberg_2017_docetaxel, Guo_2022_PF_06939999) use the same naming. checkModelConventions() flags these as warnings of compartments severity; the deviation is intentional.
  • Residual error stored in ini() on the canonical 10^9/L PD unit. Han 2015 reports the PD additive residual variances as sigma^2_PD,N = 754 /mm^3^2 and sigma^2_PD,P = 25,000 /mm^3^2. The SDs in mm^-3 are sqrt(754) = 27.46 and sqrt(25000) = 158.1. To match addSd_ANC / addSd_PLT to the canonical PD observation unit of 10^9 cells/L (i.e. 1000 / mm^3), the SDs in ini() are divided by 1000 to give 0.0275 and 0.158.
  • Successful-bootstrap convergence not propagated. Han 2015 reports bootstrap convergence rates of 78.8% (PK) and 78.0% (PD) with median parameter estimates that closely match the point estimates (Table 3, Bootstrap median column). The model carries the point estimates rather than the bootstrap medians; the differences are < 5% for every reported parameter.