Skip to contents

Glasdegib (Lin 2020)

Source: vignettes/articles/Lin_2020_glasdegib.Rmd
Lin_2020_glasdegib.Rmd

Models and source

Lin 2020 developed a population PK model for oral glasdegib in patients with cancer. Table 3 of the paper also reports, for comparison, a separately fitted model from healthy volunteers. The two models were fitted to different cohorts, so the paper contributes two model files.

mod_pat <- rxode2::rxode(readModelDb("Lin_2020_glasdegib"))
#> ℹ parameter labels from comments will be replaced by 'label()'
mod_hv <- rxode2::rxode(readModelDb("Lin_2020_glasdegib_healthy"))
#> ℹ parameter labels from comments will be replaced by 'label()'

tibble::tibble(
  Model = c("Lin_2020_glasdegib", "Lin_2020_glasdegib_healthy"),
  Cohort = c(
    "269 patients with hematologic malignancies or solid tumors (final model)",
    "49 healthy volunteers (exploratory comparison model, Table 3 footnote b)"
  )
) |>
  knitr::kable()
Model Cohort
Lin_2020_glasdegib 269 patients with hematologic malignancies or solid tumors (final model)
Lin_2020_glasdegib_healthy 49 healthy volunteers (exploratory comparison model, Table 3 footnote b)
  • Citation: Lin S, Shaik N, Martinelli G, Wagner AJ, Cortes J, Ruiz-Garcia A. Population Pharmacokinetics of Glasdegib in Patients With Advanced Hematologic Malignancies and Solid Tumors. J Clin Pharmacol. 2020;60(5):605-616. doi:10.1002/jcph.1556. Studies B1371001 (NCT00953758), B1371002 (NCT01286467) and B1371003 (NCT01546038).
  • Article: https://doi.org/10.1002/jcph.1556 (open access, PMC7187372)
  • Supplement (Supplemental Tables S1-S5, Figures S1-S2): https://europepmc.org/article/PMC/PMC7187372

Both models are two-compartment models with first-order absorption. The patient model’s covariates are baseline body weight (allometric), the weight-standardized creatinine clearance, the baseline percentage of bone marrow blasts and concomitant CYP3A inhibitors on CL/F, and tumor type (solid vs hematologic) on Vp/F and Q/F.

The steady-state exposures from this model were used in the exposure-response analyses of the same group; the overall-survival models from that work are packaged as Lin_2020_glasdegib_treatment, Lin_2020_glasdegib_exposure and Lin_2020_glasdegib_decitabine.

Population

The analysis pooled three studies: B1371001 (47 patients with hematologic malignancies, 5-600 mg QD), B1371002 (23 patients with solid tumors, 80-640 mg QD) and B1371003 (202 patients with AML or high-risk MDS, 100 or 200 mg QD combined with low-dose cytarabine, decitabine or 7 + 3 chemotherapy). Of the 272 enrolled patients, 3 had no PK samples, which left 269 patients (246 hematologic, 23 solid tumor) and 3616 concentrations. Table 2 gives the baseline characteristics of all 272: median age 69 years (25-92), median weight 78.6 kg (43.5-145.6), 181 male and 91 female, 86% White. Median Cockcroft-Gault creatinine clearance was 80.9 mL/min (31.4-238.4), and 62 and 56 patients took a moderate or a strong CYP3A inhibitor. 187 patients (69%) received the clinical dose of 100 mg QD.

The healthy-volunteer model used 1937 concentrations from 49 noncancer subjects of studies B1371010 (a ketoconazole interaction study with a 200 mg single dose of glasdegib) and B1371014. The paper does not report their demographics.

str(readModelDb("Lin_2020_glasdegib")()$population[c("n_subjects", "age_range", "weight_range")])
#> List of 3
#>  $ n_subjects  : int 269
#>  $ age_range   : chr "25-92 years (median 69) (Table 2, all 272 enrolled patients)"
#>  $ weight_range: chr "43.5-145.6 kg (median 78.6) (Table 2)"

Source trace

Every ini() value has a comment in the model file giving its source. The covariate coefficients are taken from the final-model equations in Results (“Covariate Analyses and Final Model”), which print more digits than Table 3.

Parameter Value Source
lka log(0.06) Table 3, ka = 0.06 1/h
lcl log(6.27) Table 3 and CL/F equation
lvc log(3.32) Table 3 and Vc/F equation
lvp log(279.21) Table 3 and Vp/F equation
lq log(1.29) Table 3 and Q/F equation
e_wt_cl_q, e_wt_vc_vp 0.75, 1 (fixed) Results “Base Model”; final-model equations
e_crcl_cl 0.406 CL/F equation, (WNCL/71.23)^0.406
e_bmblast_pct_cl -0.004 CL/F equation, (1 - 0.004 (BPBL - 38.20))
e_cyp3a4_inh_mod_cl -0.173 CL/F equation, (1 - 0.173 CYPmoderate)
e_cyp3a4_inh_strong_cl -0.303 CL/F equation, (1 - 0.303 CYPstrong)
e_tumtp_solid_vp -0.825 Vp/F equation, (1 - 0.825 Solid)
e_tumtp_solid_q -0.653 Q/F equation, (1 - 0.653 Solid)
IIV CL/F, Vc/F, Vp/F, Q/F, ka 43.0, 215.3, 112.8, 65.8, 13.5 %CV Table 3; omega^2 = log(1 + CV^2)
expSdHeme, expSdSolid 0.658, 0.595 Table 3 residual error; Methods (log-transformed data, thetarized sigma)
WNCL = CRCL x 70 / WT n/a Table 1 footnote a
Two-compartment ODEs, first-order absorption n/a Methods; Figure 1
Healthy-volunteer model: CL/F, Vc/F, Vp/F, Q/F, ka 10.1, 112, 53.5, 1.92, 0.79 Table 3, “Healthy Volunteer Model” column
Healthy-volunteer IIV CL/F, Vc/F, Vp/F, Q/F, ka 19.6, 11.0, 3.2 (fixed), 3.2 (fixed), 53.4 %CV Table 3
Healthy-volunteer proportional residual error 58.6% Table 3, footnote b

Typical-value covariate effects (Table 4)

Table 4 lists the typical CL/F, Vc/F, Vp/F and Q/F for a 70 kg hematologic patient with 38.2% marrow blasts and no CYP3A inhibitor, then changes one covariate at a time. The chunk below computes the same quantities from the packaged model: it solves the model with the random effects set to zero and reads the individual parameters from the output.

mod_pat_typ <- rxode2::zeroRe(mod_pat)
#> Warning: No sigma parameters in the model

# Reference: WT 70 kg and WNCL = 71.23 mL/min, i.e. raw CRCL = 71.23 at 70 kg.
t4 <- tibble::tribble(
  ~scenario, ~WT, ~wncl, ~BMBLAST_PCT, ~CONMED_CYP3A4_INH_MOD, ~CONMED_CYP3A4_INH_STRONG, ~TUMTP_SOLID,
  "Typical patient", 70, 71.23, 38.2, 0, 0, 0,
  "Body weight 61.24 kg", 61.24, 71.23, 38.2, 0, 0, 0,
  "Body weight 102.1 kg", 102.1, 71.23, 38.2, 0, 0, 0,
  "Marrow blasts 15%", 70, 71.23, 15, 0, 0, 0,
  "Marrow blasts 83%", 70, 71.23, 83, 0, 0, 0,
  "Moderate CYP3A inhibitor", 70, 71.23, 38.2, 1, 0, 0,
  "Strong CYP3A inhibitor", 70, 71.23, 38.2, 0, 1, 0,
  "Solid tumor", 70, 71.23, 38.2, 0, 0, 1
) |>
  mutate(id = row_number(), CRCL = wncl * WT / 70)

ev_t4 <- t4 |>
  select(id, WT, CRCL, BMBLAST_PCT, CONMED_CYP3A4_INH_MOD, CONMED_CYP3A4_INH_STRONG, TUMTP_SOLID) |>
  mutate(time = 0, evid = 0, amt = 0, cmt = "central")

par_t4 <- rxode2::rxSolve(mod_pat_typ, events = ev_t4, returnType = "data.frame") |>
  select(id, cl, vc, vp, q) |>
  left_join(t4 |> select(id, scenario), by = "id")
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etalq', 'etalka'
#> Warning: multi-subject simulation without without 'omega'

# Published Table 4 values (rounded to 0.1 in the paper); NA where Table 4 has
# no entry for that parameter in that scenario.
pub_t4 <- tibble::tribble(
  ~scenario, ~cl_pub, ~vc_pub, ~vp_pub, ~q_pub,
  "Typical patient", 6.3, 3.3, 279.2, 1.3,
  "Body weight 61.24 kg", 5.7, 2.9, NA, NA,
  "Body weight 102.1 kg", 8.3, 4.8, NA, NA,
  "Marrow blasts 15%", 6.9, NA, NA, NA,
  "Marrow blasts 83%", 5.1, NA, NA, NA,
  "Moderate CYP3A inhibitor", 5.2, NA, NA, NA,
  "Strong CYP3A inhibitor", 4.4, NA, NA, NA,
  "Solid tumor", NA, NA, 48.9, 0.4
)

t4_cmp <- par_t4 |> left_join(pub_t4, by = "scenario")

t4_cmp |>
  select(scenario, cl, cl_pub, vc, vc_pub, vp, vp_pub, q, q_pub) |>
  dplyr::rename(
    Scenario = scenario,
    "CL/F model" = cl, "CL/F Table 4" = cl_pub,
    "Vc/F model" = vc, "Vc/F Table 4" = vc_pub,
    "Vp/F model" = vp, "Vp/F Table 4" = vp_pub,
    "Q/F model" = q, "Q/F Table 4" = q_pub
  ) |>
  knitr::kable(digits = 2, caption = "Replicates Table 4 of Lin 2020 (L/h and L).")
Replicates Table 4 of Lin 2020 (L/h and L).
Scenario CL/F model CL/F Table 4 Vc/F model Vc/F Table 4 Vp/F model Vp/F Table 4 Q/F model Q/F Table 4
Typical patient 6.27 6.3 3.32 3.3 279.21 279.2 1.29 1.3
Body weight 61.24 kg 5.67 5.7 2.90 2.9 244.27 NA 1.17 NA
Body weight 102.1 kg 8.32 8.3 4.84 4.8 407.25 NA 1.71 NA
Marrow blasts 15% 6.85 6.9 3.32 NA 279.21 NA 1.29 NA
Marrow blasts 83% 5.15 5.1 3.32 NA 279.21 NA 1.29 NA
Moderate CYP3A inhibitor 5.19 5.2 3.32 NA 279.21 NA 1.29 NA
Strong CYP3A inhibitor 4.37 4.4 3.32 NA 279.21 NA 1.29 NA
Solid tumor 6.27 NA 3.32 NA 48.86 48.9 0.45 0.4 

# Deterministic check: the model values must round to the published values.
# Table 4 prints one decimal place, so the tolerance is half a unit in that
# place (0.05) plus a little slack for the paper's own rounding of inputs.
long_cmp <- bind_rows(
  t4_cmp |> transmute(scenario, model = cl, pub = cl_pub),
  t4_cmp |> transmute(scenario, model = vc, pub = vc_pub),
  t4_cmp |> transmute(scenario, model = vp, pub = vp_pub),
  t4_cmp |> transmute(scenario, model = q, pub = q_pub)
) |>
  filter(!is.na(pub))
stopifnot(nrow(long_cmp) == 14, all(abs(long_cmp$model - long_cmp$pub) <= 0.06))

All fourteen Table 4 entries are reproduced to the printed precision. Table 4 also lists CL/F by renal-function group (6.5, 6.2 and 4.8 L/h at median raw CRCL 110.2, 75.5 and 51.1 mL/min). Its footnote and the Discussion describe these as body-weight-normalized post hoc estimates, not typical values from the equation, so they are not a check on the equation and are left out of the comparison above.

Steady-state exposure by covariate scenario (Figure 4)

Figure 4 shows the median steady-state Cmax and AUCtau for each covariate scenario (500 simulated patients each). The caption does not state the dose; the clinical dose of 100 mg QD is assumed, and the agreement below supports it. The caption describes the reference as “a 70-kg male hematology patient” but also as having “the median values for baseline weight”. The printed numbers fit the median, 78.6 kg: at 70 kg every non-weight row would be about 9% above the printed value. Sex is not a covariate, so the “Female” row is the reference patient: 78.6 kg, median WNCL and median marrow blasts, hematologic malignancy and no CYP3A inhibitor. The typical-value model below uses these reference covariates and changes one of them per row. PKNCA computes Cmax and AUC over the dosing interval on day 60.

fig4 <- tibble::tribble(
  ~scenario, ~WT, ~wncl, ~BMBLAST_PCT, ~CONMED_CYP3A4_INH_MOD, ~CONMED_CYP3A4_INH_STRONG, ~TUMTP_SOLID, ~cmax_pub, ~auc_pub,
  "Female (reference)", 78.6, 71.23, 38.2, 0, 0, 0, 995.5, 14980.2,
  "Solid tumor", 78.6, 71.23, 38.2, 0, 0, 1, 1021.4, 15269.9,
  "Strong CYP inhibitor", 78.6, 71.23, 38.2, 0, 1, 0, 1327.1, 20317.5,
  "Moderate CYP inhibitor", 78.6, 71.23, 38.2, 1, 0, 0, 1103.0, 16829.0,
  "BPBL 80.0", 78.6, 71.23, 80.0, 0, 0, 0, 1099.1, 16759.5,
  "BPBL 60.0", 78.6, 71.23, 60.0, 0, 0, 0, 1043.3, 15749.1,
  "BPBL 25.0", 78.6, 71.23, 25.0, 0, 0, 0, 922.4, 13988.1,
  "BPBL 17.8", 78.6, 71.23, 17.8, 0, 0, 0, 904.9, 13512.3,
  "WNCL 109.59", 78.6, 109.59, 38.2, 0, 0, 0, 821.2, 12250.7,
  "WNCL 88.48", 78.6, 88.48, 38.2, 0, 0, 0, 899.6, 13538.4,
  "WNCL 57.45", 78.6, 57.45, 38.2, 0, 0, 0, 1047.4, 15800.2,
  "WNCL 46.95", 78.6, 46.95, 38.2, 0, 0, 0, 1126.3, 17102.6,
  "BWT 102.28", 102.28, 71.23, 38.2, 0, 0, 0, 748.0, 11371.7,
  "BWT 89.00", 89.00, 71.23, 38.2, 0, 0, 0, 853.2, 12874.9,
  "BWT 78.60", 78.60, 71.23, 38.2, 0, 0, 0, 871.8, 13285.4,
  "BWT 68.00", 68.00, 71.23, 38.2, 0, 0, 0, 1017.1, 15327.3,
  "BWT 61.26", 61.26, 71.23, 38.2, 0, 0, 0, 1099.0, 16480.8
) |>
  mutate(id = row_number(), CRCL = wncl * WT / 70)

tau <- 24
ss_start <- 59 * tau
obs_times <- ss_start + seq(0, tau, by = 0.25)

ev_fig4 <- bind_rows(
  fig4 |> select(id) |>
    tidyr::crossing(time = seq(0, ss_start, by = tau)) |>
    mutate(evid = 1, amt = 100, cmt = "depot"),
  fig4 |> select(id) |>
    tidyr::crossing(time = obs_times) |>
    mutate(evid = 0, amt = 0, cmt = "central")
) |>
  left_join(fig4 |> select(id, scenario, WT, CRCL, BMBLAST_PCT, CONMED_CYP3A4_INH_MOD,
                           CONMED_CYP3A4_INH_STRONG, TUMTP_SOLID), by = "id") |>
  arrange(id, time, desc(evid))

sim_fig4 <- rxode2::rxSolve(mod_pat_typ, events = ev_fig4, keep = "scenario",
                            returnType = "data.frame")
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etalq', 'etalka'
#> Warning: multi-subject simulation without without 'omega'
conc4 <- sim_fig4 |>
  dplyr::filter(!is.na(Cc)) |>
  select(id, time, Cc, scenario)
dose4 <- ev_fig4 |>
  dplyr::filter(evid == 1) |>
  select(id, time, amt, scenario)

nca4 <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(conc4, Cc ~ time | scenario + id),
  PKNCA::PKNCAdose(dose4, amt ~ time | scenario + id),
  intervals = data.frame(start = ss_start, end = ss_start + tau, cmax = TRUE, auclast = TRUE)
))

res4 <- as.data.frame(nca4) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "auclast")) |>
  select(scenario, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES) |>
  left_join(fig4 |> select(scenario, cmax_pub, auc_pub), by = "scenario") |>
  mutate(
    cmax_pct = 100 * (cmax / cmax_pub - 1),
    auc_pct = 100 * (auclast / auc_pub - 1)
  )

res4 |>
  select(scenario, cmax, cmax_pub, cmax_pct, auclast, auc_pub, auc_pct) |>
  dplyr::rename(
    Scenario = scenario,
    "Cmax,ss model (ng/mL)" = cmax, "Cmax,ss Figure 4" = cmax_pub, "Cmax diff (%)" = cmax_pct,
    "AUCtau,ss model (ng*h/mL)" = auclast, "AUCtau,ss Figure 4" = auc_pub, "AUC diff (%)" = auc_pct
  ) |>
  knitr::kable(digits = 1, caption = "Replicates the printed medians of Figure 4 of Lin 2020 (100 mg QD, day 60).")
Replicates the printed medians of Figure 4 of Lin 2020 (100 mg QD, day 60).
Scenario Cmax,ss model (ng/mL) Cmax,ss Figure 4 Cmax diff (%) AUCtau,ss model (ng*h/mL) AUCtau,ss Figure 4 AUC diff (%)
BPBL 17.8 906.1 904.9 0.1 13499.7 13512.3 -0.1
BPBL 25.0 928.1 922.4 0.6 13868.7 13988.1 -0.9
BPBL 60.0 1052.1 1043.3 0.8 15993.5 15749.1 1.6
BPBL 80.0 1140.9 1099.1 3.8 17527.5 16759.5 4.6
BWT 102.28 793.3 748.0 6.1 11979.9 11371.7 5.3
BWT 61.26 1175.5 1099.0 7.0 17605.1 16480.8 6.8
BWT 68.00 1085.3 1017.1 6.7 16278.2 15327.3 6.2
BWT 78.60 971.2 871.8 11.4 14600.3 13285.4 9.9
BWT 89.00 882.8 853.2 3.5 13299.3 12874.9 3.3
Female (reference) 971.2 995.5 -2.4 14600.3 14980.2 -2.5
Moderate CYP inhibitor 1147.9 1103.0 4.1 17650.2 16829.0 4.9
Solid tumor 1009.3 1021.4 -1.2 14612.8 15269.9 -4.3
Strong CYP inhibitor 1331.6 1327.1 0.3 20934.3 20317.5 3.0
WNCL 109.59 830.8 821.2 1.2 12258.7 12250.7 0.1
WNCL 46.95 1127.3 1126.3 0.1 17288.9 17102.6 1.1
WNCL 57.45 1048.4 1047.4 0.1 15930.6 15800.2 0.8
WNCL 88.48 898.4 899.6 -0.1 13370.7 13538.4 -1.2

The CYP3A, marrow-blast, renal-function and tumor-type rows agree with the printed medians to within about 5%. The body-weight rows are 3-11% higher than printed. Figure 4 has its own internal noise of about that size. The “Female” row and the “BWT 78.60” row describe the same patient, since sex is not a covariate and 78.6 kg is the median weight, yet the figure prints AUCs of 14980 and 13285 ng*h/mL for them, 11% apart. Each scenario appears to be a separate 500-patient Monte Carlo simulation. The median AUC over log-normal clearance is the typical value, so the typical-value solve here has no such noise. The fold changes relative to the “Female” row are:

ref_row <- res4 |> dplyr::filter(scenario == "Female (reference)")
fold4 <- res4 |>
  mutate(
    fold_model = auclast / ref_row$auclast,
    fold_pub = auc_pub / ref_row$auc_pub,
    fold_pct = 100 * (fold_model / fold_pub - 1)
  )
fold4 |>
  select(scenario, fold_model, fold_pub, fold_pct) |>
  dplyr::rename(
    Scenario = scenario, "AUC fold, model" = fold_model,
    "AUC fold, Figure 4" = fold_pub, "Difference (%)" = fold_pct
  ) |>
  knitr::kable(digits = 3, caption = "AUCtau,ss fold change relative to the reference patient.")
AUCtau,ss fold change relative to the reference patient.
Scenario AUC fold, model AUC fold, Figure 4 Difference (%)
BPBL 17.8 0.925 0.902 2.506
BPBL 25.0 0.950 0.934 1.726
BPBL 60.0 1.095 1.051 4.194
BPBL 80.0 1.200 1.119 7.303
BWT 102.28 0.821 0.759 8.089
BWT 61.26 1.206 1.100 9.601
BWT 68.00 1.115 1.023 8.967
BWT 78.60 1.000 0.887 12.757
BWT 89.00 0.911 0.859 5.984
Female (reference) 1.000 1.000 0.000
Moderate CYP inhibitor 1.209 1.123 7.608
Solid tumor 1.001 1.019 -1.813
Strong CYP inhibitor 1.434 1.356 5.716
WNCL 109.59 0.840 0.818 2.669
WNCL 46.95 1.184 1.142 3.719
WNCL 57.45 1.091 1.055 3.448
WNCL 88.48 0.916 0.904 1.331 

# Typical-value solve, so the model side is deterministic (no rxode2 RNG);
# the tolerance covers the Monte Carlo noise in the published medians, which
# the Female / BWT 78.60 pair puts at about 11%. Realised on authoring: median
# AUC difference +1.6%, median Cmax difference +0.8%, 90th percentile of the
# absolute fold difference about 9.2%. A mis-transcribed clearance, dose or
# unit moves these by tens of percent.
stopifnot(
  abs(median(res4$auc_pct)) < 10,
  abs(median(res4$cmax_pct)) < 10,
  quantile(abs(fold4$fold_pct), 0.9) < 15
)

The fold changes all point in the published direction and have the published magnitudes. The largest differences are in the body-weight rows, which inherit the offset described above. The CYP3A rows follow the equation exactly: 1 / (1 - 0.303) = 1.43-fold and 1 / (1 - 0.173) = 1.21-fold. Figure 4 prints 1.36 and 1.12. The Results and Table 4 quote the same 17% and 30% CL/F reductions as the equation, so the equation is carried as published.

Concentration-time profiles (Figure 3)

Figure 3 is a prediction-corrected VPC of the observed data by study, which cannot be reproduced without the data. The chunk below shows the simulated 5th, 50th and 95th percentiles for 100 mg QD, the clinical dose, over the first day and at steady state, for 200 hematologic and 200 solid-tumor patients.

rxode2::rxSetSeed(20200501)
set.seed(20200501)

make_cohort <- function(n, solid, id_offset) {
  tibble(
    id = id_offset + seq_len(n),
    WT = pmin(pmax(exp(rnorm(n, log(78.6), 0.22)), 43.5), 145.6),
    CRCL = pmin(pmax(exp(rnorm(n, log(80.9), 0.35)), 31.4), 238.4),
    BMBLAST_PCT = if (solid) 38.2 else pmin(pmax(rnorm(n, 39.3, 25), 0), 100),
    CONMED_CYP3A4_INH_MOD = 0,
    CONMED_CYP3A4_INH_STRONG = 0,
    TUMTP_SOLID = as.integer(solid),
    group = if (solid) "Solid tumor" else "Hematologic"
  )
}
cohort <- bind_rows(
  make_cohort(200, solid = FALSE, id_offset = 0L),
  make_cohort(200, solid = TRUE, id_offset = 200L)
)
stopifnot(!anyDuplicated(cohort$id))

vpc_times <- c(seq(0, 24, by = 1), ss_start + seq(0, 24, by = 1))
ev_vpc <- bind_rows(
  cohort |> tidyr::crossing(time = seq(0, ss_start, by = tau)) |>
    mutate(evid = 1, amt = 100, cmt = "depot"),
  cohort |> tidyr::crossing(time = vpc_times) |>
    mutate(evid = 0, amt = 0, cmt = "central")
) |>
  arrange(id, time, desc(evid))

sim_vpc <- rxode2::rxSolve(mod_pat, events = ev_vpc, keep = "group", returnType = "data.frame")
sim_vpc |>
  mutate(day = ifelse(time < ss_start, "Day 1", "Day 60 (steady state)"),
         tad = ifelse(time < ss_start, time, time - ss_start)) |>
  group_by(group, day, tad) |>
  summarise(
    Q05 = quantile(Cc, 0.05, na.rm = TRUE),
    Q50 = quantile(Cc, 0.50, na.rm = TRUE),
    Q95 = quantile(Cc, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  dplyr::filter(tad > 0) |>
  ggplot(aes(tad, Q50, colour = group, fill = group)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.2, colour = NA) +
  geom_line() +
  facet_wrap(~day) +
  scale_y_log10() +
  labs(x = "Time after dose (h)", y = "Glasdegib Cc (ng/mL)", colour = NULL, fill = NULL,
       caption = "Simulated 5th-95th percentiles, 100 mg QD; compare Figure 3 of Lin 2020.")

Figure 3C, study B1371003 where most patients took 100 mg QD, has its prediction-corrected median at roughly 500-1000 ng/mL, with most observations between about 100 and 3000 ng/mL. The simulated profiles fall in the same range.

Healthy-volunteer model

The Discussion states that the healthy-volunteer CL/F is “over 1.6-fold higher” and ka “over 13-fold higher” than in patients. Both follow from the two parameter sets:

hv_ratio <- c(
  cl = exp(mod_hv$theta[["lcl"]] - mod_pat$theta[["lcl"]]),
  ka = exp(mod_hv$theta[["lka"]] - mod_pat$theta[["lka"]])
)
round(hv_ratio, 2)
#>    cl    ka 
#>  1.61 13.17
stopifnot(hv_ratio[["cl"]] > 1.6, hv_ratio[["ka"]] > 13)

For a 70 kg subject, a 200 mg single dose (the B1371010 regimen) is simulated with the typical healthy-volunteer model and with 200 simulated volunteers. PKNCA’s AUC0-inf from the typical-value profile must equal Dose / (CL/F).

mod_hv_typ <- rxode2::zeroRe(mod_hv)
hv_times <- c(0, 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 12, 24, 36, 48, 72, 96, 120, 168, 240, 336)
ev_hv <- bind_rows(
  tibble(id = 1L, time = 0, evid = 1, amt = 200, cmt = "depot"),
  tibble(id = 1L, time = hv_times, evid = 0, amt = 0, cmt = "central")
) |>
  mutate(WT = 70, arm = "200 mg single dose") |>
  arrange(id, time, desc(evid))
sim_hv_typ <- rxode2::rxSolve(mod_hv_typ, events = ev_hv, returnType = "data.frame") |>
  # a one-subject solve returns no id column
  mutate(id = 1L, arm = "200 mg single dose")
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalvp', 'etalq', 'etalka'

nca_hv <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(sim_hv_typ |> dplyr::filter(!is.na(Cc)) |> select(id, time, Cc, arm),
                   Cc ~ time | arm + id),
  PKNCA::PKNCAdose(ev_hv |> dplyr::filter(evid == 1) |> select(id, time, amt, arm),
                   amt ~ time | arm + id),
  intervals = data.frame(start = 0, end = Inf, cmax = TRUE, tmax = TRUE,
                         aucinf.obs = TRUE, half.life = TRUE)
))
hv_nca <- as.data.frame(nca_hv) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "tmax", "aucinf.obs", "half.life")) |>
  select(PPTESTCD, PPORRES)
knitr::kable(hv_nca, digits = 2, caption = "Typical healthy volunteer, 70 kg, 200 mg single dose.")
Typical healthy volunteer, 70 kg, 200 mg single dose.
PPTESTCD PPORRES
cmax 1306.36
tmax 3.00
half.life 24.61
aucinf.obs 19898.74 

auc_closed <- 200 / 10.1 * 1000
auc_nca <- hv_nca$PPORRES[hv_nca$PPTESTCD == "aucinf.obs"]
stopifnot(abs(auc_nca / auc_closed - 1) < 0.02)
set.seed(20200502)
hv_cohort <- tibble(id = 1:200, WT = pmin(pmax(rnorm(200, 75, 12), 50), 110))
ev_hv_vpc <- bind_rows(
  hv_cohort |> mutate(time = 0, evid = 1, amt = 200, cmt = "depot"),
  hv_cohort |> tidyr::crossing(time = hv_times) |> mutate(evid = 0, amt = 0, cmt = "central")
) |>
  arrange(id, time, desc(evid))
sim_hv <- rxode2::rxSolve(mod_hv, events = ev_hv_vpc, returnType = "data.frame")
sim_hv |>
  dplyr::filter(time > 0) |>
  group_by(time) |>
  summarise(
    Q05 = quantile(Cc, 0.05), Q50 = quantile(Cc, 0.5), Q95 = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  scale_y_log10() +
  labs(x = "Time (h)", y = "Glasdegib Cc (ng/mL)",
       caption = "Healthy-volunteer model, 200 mg single dose, simulated 5th-95th percentiles.")

Assumptions and deviations

  • IIV scale. Table 3 gives IIV as “CV (%)” without saying how it was computed. The maintainers converted with the log-normal relation omega^2 = log(1 + CV^2). For Vc/F (215.3%) this matters: reading the value as 100 x omega instead would give omega^2 = 4.64 rather than 1.73. The paper gives no second number that could tell the two readings apart.
  • Residual error. The Methods describe a proportional error on log-transformed concentrations with a thetarized sigma, so the patient model uses lnorm() with the Table 3 percentages as log-scale SDs (0.658 hematologic, 0.595 solid tumor).
  • Coefficient precision. The printed final-model equations give the covariate coefficients to three digits (0.173, 0.303, 0.406, 0.825, 0.653) where Table 3 rounds to two. The equation values are used.
  • Creatinine clearance input. The model takes the raw Cockcroft-Gault CRCL (mL/min) and forms the paper’s weight-standardized WNCL = CRCL x 70 / WT internally.
  • Marrow blasts in solid-tumor patients. Not collected (“Not applicable”, Table 2). The paper imputes missing continuous covariates at the population median, so the simulations here set 38.2% for solid-tumor patients, which makes the term neutral.
  • CYP3A inhibitor indicators. The paper does not say whether they were time-varying. They are simple 0/1 columns, so either use works. A subject with both flags gets both factors, as in the published equation.
  • Healthy-volunteer model. Table 3 footnote b states only “allometric body weight scaling”. The exponents and the 70 kg reference are assumed to be the same as in the patient model. The food effect on ka was “empirically added … by linear function”, but its coefficient and reference food state are not reported, so it is omitted; ka = 0.79 1/h is the typical value in the unstated reference state. The paper does not say how the ketoconazole period of B1371010 was handled; no CYP3A inhibitor covariate is reported. The residual error is taken as proportional on the linear scale (“proportional error model”, FOCEI).
  • Cohorts. The Figure 3 cohort draws log-normal weight and CRCL around the Table 2 medians, truncated to the Table 2 ranges, and excludes CYP3A inhibitors. The healthy-volunteer weights are an assumption, as the paper gives none.
  • No correction notice for this article was found in Europe PMC as of 2026-09-26.