Skip to contents

Axitinib (Garrett 2014)

Source: vignettes/articles/Garrett_2014_axitinib.Rmd
Garrett_2014_axitinib.Rmd

Model and source

  • Citation: Garrett M, Poland B, Brennan M, Hee B, Pithavala YK, Amantea MA. Population pharmacokinetic analysis of axitinib in healthy volunteers. Br J Clin Pharmacol. 2014;77(3):480-492. doi:10.1111/bcp.12206
  • Description: Two-compartment population PK model for axitinib in healthy volunteers (Garrett 2014). First-order absorption with fixed lag time, allometric power-form effect of body weight on the central volume of distribution (reference 75 kg), linear-proportional fasting effects on the first-order absorption rate constant ka and on bioavailability F, and a linear-proportional reduction in F for the marketed crystal polymorph Form XLI relative to the earlier Form IV reference. Pooled data from 337 healthy subjects across ten Pfizer Phase I studies.
  • Article: https://doi.org/10.1111/bcp.12206

Garrett 2014 pooled plasma axitinib concentrations from 337 healthy volunteers across ten Pfizer Phase I studies (Table 1) and fit a single two-compartment population PK model with first-order absorption and a fixed lag time. The food / formulation effects of clinical interest – fasting on first-order absorption rate ka and on absolute oral bioavailability F, plus the F reduction for the marketed crystal polymorph Form XLI relative to the earlier Form IV polymorph – were included in the base model and retained through the formal covariate search (Methods ‘Model development’ and ‘Statistical analysis’). Among the eleven additional covariates tested (body weight, age, sex, race, smoking status, creatinine clearance, ALT, AST, bilirubin, the UGT1A1*28 genotype, and the CYP2C19 inferred phenotype), only the power-form effect of body weight on the central volume Vc reached formal-selection significance.

Population

The pooled dataset (Garrett 2014 Table 2; reproduced programmatically via readModelDb("Garrett_2014_axitinib")$population) is dominated by young (median 31 years), White (62%), male (93%), non-smoking (88%) healthy volunteers with normal kidney and liver function as evidenced by group-mean creatinine clearance 117.2 mL/min, AST 23.9 U/L, ALT 24.1 U/L, and total bilirubin 0.8 mg/dL. Median total body weight is 75.0 kg. Race distribution is White 62%, Black 8%, Asian 18%, Japanese 6%, Hispanic 3%, Other 3%. Genotyping for UGT1A1*28 and the CYP2C19 *2 / *3 / *17 alleles was performed on the majority of subjects; the bulk of the cohort is UGT1A1 TA6/TA6 (51%) and CYP2C19 extensive metaboliser (93%). All studies enrolled in the United States, Belgium, or Singapore; ten clinical sites are listed in the Methods ‘Study design’ section. Pre-dose to 48 h dense sampling was performed in most studies (per-study schedules in Table 1).

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/Garrett_2014_axitinib.R. The table below collects them in one place for review.

Equation / parameter Value Source location
Structural model – two-compartment with first-order absorption and lag time n/a Methods ‘Model development’; Results ‘PK base model’
Reference body weight for allometric Vc 75 kg Results ‘Full model and final model’; Table 3 footnote (population median)
lka (Form IV, fed reference) log(0.523) Table 3 ‘k a (h-1) Form IV, Fed’ = 0.523
lcl log(17.0) Table 3 ‘CL (L/h)’ final = 17.0
lvc (at 75 kg) log(45.3) Table 3 ‘Vc (L)’ final = 45.3
lq log(1.74) Table 3 ‘Q (L/h)’ final = 1.74
lvp log(45.9) Table 3 ‘Vp (L)’ final = 45.9
lfdepot (Form IV, fed reference) log(0.465) Table 3 ‘F, Form IV, Fed’ = 0.465
ltlag (FIXED) log(0.457) Table 3 ‘tlag (h)’ = 0.457 (Table 3 footnote: fixed at 0.457 in the final run)
e_wt_vc (allometric Vc exponent) 0.758 Table 3 ‘Weight effect on Vc’ = 0.758
e_fast_ka (fasting on ka) 2.07 Table 3 ‘ka Fasting’ = 2.07
e_fast_f (fasting on F) 0.338 Table 3 ‘F Fasting’ = 0.338
e_xli_f (Form XLI on F) -0.150 Table 3 ‘F Form XLI’ = -0.150
omega^2 CL, cov(CL,Vc), omega^2 Vc 0.272, 0.141, 0.0949 Table 3 (block)
omega^2 Q, cov(Q,Vp), omega^2 Vp 0.406, 0.619, 1.07 Table 3 (block)
omega^2 ka 0.506 Table 3 final
expSd (oral log-additive residual SD) 0.509 Table 3 ‘Oral, %’ final = 50.9%
IV log-additive residual SD (not encoded) 0.342 Table 3 ‘Intravenous, %’ final = 34.2% (see Assumptions below)

Virtual cohort

Original observed data are not publicly available. The simulation cohorts below approximate the published Phase I trial demographics (Garrett 2014 Table 2): a healthy-adult cohort with body weight distributed around the population median 75 kg.

set.seed(20140311)

n_per_arm <- 100L

make_cohort <- function(n, fed_value, form_xli_value, label, id_offset = 0L) {
  wt <- pmax(50, rnorm(n, mean = 75, sd = 11.6))
  dosing <- tibble::tibble(
    id   = id_offset + seq_len(n),
    time = 0,
    evid = 1L,
    amt  = 5,                                # 5 mg oral axitinib (Garrett 2014 Methods Table 1)
    cmt  = "depot",
    WT   = wt,
    FED  = fed_value,
    FORM_AXI_XLI = form_xli_value,
    treatment    = label
  )
  obs_times <- c(0.5, 1, 1.5, 2, 3, 4, 6, 8, 12, 16, 24, 32, 36, 48)
  obs <- tidyr::expand_grid(
    id   = id_offset + seq_len(n),
    time = obs_times
  ) |>
    dplyr::left_join(
      dosing |> dplyr::select(id, WT, FED, FORM_AXI_XLI, treatment),
      by = "id"
    ) |>
    dplyr::mutate(evid = 0L, amt = 0, cmt = "central")
  dplyr::bind_rows(dosing, obs) |> dplyr::arrange(id, time, dplyr::desc(evid))
}

events <- dplyr::bind_rows(
  make_cohort(n_per_arm, fed_value = 1L, form_xli_value = 0L, label = "Form IV fed",     id_offset = 0L),
  make_cohort(n_per_arm, fed_value = 0L, form_xli_value = 0L, label = "Form IV fasted",  id_offset =  n_per_arm),
  make_cohort(n_per_arm, fed_value = 1L, form_xli_value = 1L, label = "Form XLI fed",    id_offset = 2L * n_per_arm)
)

stopifnot(!anyDuplicated(unique(events[, c("id", "time", "evid")])))

Simulation 

mod <- readModelDb("Garrett_2014_axitinib")

sim <- rxode2::rxSolve(
  mod,
  events = as.data.frame(events),
  keep   = c("treatment", "WT", "FED", "FORM_AXI_XLI"),
  nStud  = 1L
) |>
  as.data.frame() |>
  dplyr::mutate(treatment = factor(treatment, levels = c("Form IV fed", "Form IV fasted", "Form XLI fed")))
#> ℹ parameter labels from comments will be replaced by 'label()'

Replicate published figures

Garrett 2014 Figure 4 shows visual predictive checks of the final model against pooled observed data from four of the ten studies (study 1 Form IV fasted; study 2 Form IV fed and fasted; study 6 Form IV fasted; study 7 Form IV vs Form XLI both fed). Observed data are not publicly distributed, so the panel below reproduces the simulated median and 90% prediction interval from the packaged model under the three formulation x food conditions Garrett 2014 emphasises.

sim |>
  dplyr::filter(time > 0) |>
  dplyr::group_by(treatment, time) |>
  dplyr::summarise(
    Q05 = stats::quantile(Cc, 0.05, na.rm = TRUE),
    Q50 = stats::quantile(Cc, 0.50, na.rm = TRUE),
    Q95 = stats::quantile(Cc, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  facet_wrap(~ treatment) +
  scale_y_log10() +
  labs(
    x = "Time (h)",
    y = "Plasma axitinib (ng/mL)",
    title = "Simulated VPC: 5 mg single oral dose of axitinib",
    caption = "Replicates the Form IV fed / Form IV fasted / Form XLI fed panels of Garrett 2014 Figure 4."
  )

PKNCA validation

Single-dose NCA over the 0-48 h sampling window, stratified by formulation x food state. The grouping variable treatment matches the cohort labels carried through rxSolve(..., keep = ).

sim_nca <- sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::select(id, time, Cc, treatment)

sim_nca <- dplyr::bind_rows(
  sim_nca,
  sim_nca |> dplyr::distinct(id, treatment) |>
    dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::distinct(id, treatment, time, .keep_all = TRUE) |>
  dplyr::arrange(id, treatment, time)

dose_df <- events |>
  dplyr::filter(evid == 1L) |>
  dplyr::select(id, time, amt, treatment) |>
  as.data.frame()

conc_obj <- PKNCA::PKNCAconc(
  as.data.frame(sim_nca),
  Cc ~ time | treatment + id,
  concu = "ng/mL",
  timeu = "h"
)

dose_obj <- PKNCA::PKNCAdose(
  dose_df,
  amt ~ time | treatment + id,
  doseu = "mg"
)

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

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

nca_summary <- as.data.frame(nca_res$result) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "tmax", "aucinf.obs", "half.life")) |>
  dplyr::group_by(treatment, PPTESTCD) |>
  dplyr::summarise(median_value = stats::median(PPORRES, na.rm = TRUE), .groups = "drop") |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = median_value)

knitr::kable(
  nca_summary,
  digits  = 3,
  caption = "Simulated single-dose median NCA parameters by formulation x food state."
)
Simulated single-dose median NCA parameters by formulation x food state.
treatment aucinf.obs cmax half.life tmax
Form IV fasted 174.644 41.883 21.714 1.5
Form IV fed 141.330 18.779 19.366 3.0
Form XLI fed 106.399 16.311 18.268 3.0

Comparison against published structural effects

Garrett 2014 does not tabulate single-dose NCA values directly, but the ratios between the three arms are completely determined by the Table 3 fasting and formulation effects on ka and F:

  • AUC(Form IV fasted) / AUC(Form IV fed) = 1 + 0.338 = 1.338 (fasting increases F by 33.8%; ka does not enter the AUC ratio because CL, Vc, Q, and Vp are unchanged).
  • AUC(Form XLI fed) / AUC(Form IV fed) = 1 - 0.150 = 0.850 (Form XLI reduces F by 15%).
  • Cmax(Form IV fasted) / Cmax(Form IV fed) increases with both the higher F and the 207%-higher ka (faster absorption shifts the peak earlier and higher).
ratio_table <- nca_summary |>
  dplyr::select(treatment, aucinf.obs) |>
  tibble::deframe()

paper_auc_ratios <- tibble::tibble(
  ratio                     = c("Form IV fasted / Form IV fed", "Form XLI fed / Form IV fed"),
  simulated_AUC_ratio       = c(
    unname(ratio_table["Form IV fasted"] / ratio_table["Form IV fed"]),
    unname(ratio_table["Form XLI fed"]   / ratio_table["Form IV fed"])
  ),
  published_AUC_ratio_from_Table3 = c(1 + 0.338, 1 - 0.150)
)

knitr::kable(
  paper_auc_ratios,
  digits = 3,
  caption = "Simulated AUC(0-Inf) ratios vs published F-effect ratios (Garrett 2014 Table 3)."
)
Simulated AUC(0-Inf) ratios vs published F-effect ratios (Garrett 2014 Table 3).
ratio simulated_AUC_ratio published_AUC_ratio_from_Table3
Form IV fasted / Form IV fed 1.236 1.338
Form XLI fed / Form IV fed 0.753 0.850

Body-weight body-Cmax simulation (Garrett 2014 Results section)

Garrett 2014 separately reports a typical-value simulation of steady-state Cmax for the marketed 5 mg BID Form XLI fed regimen at three body-weight strata (Results ‘Full model and final model’): the 10th-percentile 62 kg subject (20.6 ng/mL), the median 75 kg subject (19.3 ng/mL), and the 90th-percentile 93 kg subject (17.9 ng/mL). The following block reproduces those Cmax values with a deterministic typical-value simulation (random effects zeroed).

mod_typ <- rxode2::zeroRe(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'

ss_cohort <- function(weight_kg, label) {
  doses <- tibble::tibble(
    id   = 1L,
    time = seq(0, 24 * 14, by = 12),    # 5 mg BID for two weeks (more than enough to reach steady state)
    evid = 1L,
    amt  = 5,
    cmt  = "depot",
    WT   = weight_kg,
    FED  = 1L,
    FORM_AXI_XLI = 1L,
    treatment    = label
  )
  obs <- tibble::tibble(
    id   = 1L,
    time = seq(0, 24 * 14 + 12, by = 0.5),
    evid = 0L,
    amt  = 0,
    cmt  = "central",
    WT   = weight_kg,
    FED  = 1L,
    FORM_AXI_XLI = 1L,
    treatment    = label
  )
  dplyr::bind_rows(doses, obs) |> dplyr::arrange(time, dplyr::desc(evid))
}

ss_events <- dplyr::bind_rows(
  ss_cohort(62, "62 kg (10th pctile)") |> dplyr::mutate(id = 1L),
  ss_cohort(75, "75 kg (median)")      |> dplyr::mutate(id = 2L),
  ss_cohort(93, "93 kg (90th pctile)") |> dplyr::mutate(id = 3L)
)

ss_sim <- rxode2::rxSolve(
  mod_typ,
  events = as.data.frame(ss_events),
  keep   = c("treatment", "WT")
) |>
  as.data.frame()
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalq', 'etalvp', 'etalka'
#> Warning: multi-subject simulation without without 'omega'

tau          <- 12
last_dose    <- max(ss_events$time[ss_events$evid == 1L])
ss_window_lo <- last_dose
ss_window_hi <- last_dose + tau

ss_cmax_sim <- ss_sim |>
  dplyr::filter(time >= ss_window_lo, time <= ss_window_hi) |>
  dplyr::group_by(treatment) |>
  dplyr::summarise(cmax_ss_sim = max(Cc), .groups = "drop")

ss_cmax_published <- tibble::tibble(
  treatment           = c("62 kg (10th pctile)", "75 kg (median)", "93 kg (90th pctile)"),
  cmax_ss_published   = c(20.6, 19.3, 17.9)
)

ss_cmax_compare <- dplyr::left_join(ss_cmax_sim, ss_cmax_published, by = "treatment") |>
  dplyr::mutate(pct_diff = 100 * (cmax_ss_sim - cmax_ss_published) / cmax_ss_published)

knitr::kable(
  ss_cmax_compare,
  digits  = 2,
  caption = "Simulated vs Garrett 2014 published typical-value steady-state Cmax for 5 mg BID Form XLI fed axitinib (ng/mL) at three body weights."
)
Simulated vs Garrett 2014 published typical-value steady-state Cmax for 5 mg BID Form XLI fed axitinib (ng/mL) at three body weights.
treatment cmax_ss_sim cmax_ss_published pct_diff
62 kg (10th pctile) 20.57 20.6 -0.14
75 kg (median) 19.30 19.3 0.01
93 kg (90th pctile) 17.91 17.9 0.08

Assumptions and deviations

  • Route-stratified residual error not encoded. Garrett 2014 fit two log-additive residual SDs – oral 50.9% and intravenous 34.2% (Table 3 final). The packaged model encodes a single expSd = 0.509 (the oral residual) because the model is intended for simulating oral dosing of the marketed axitinib tablet; the 16 IV-arm subjects from study 3 contributed primarily to the absolute-bioavailability anchor and are not the typical-use case. Users who need to simulate the IV arm at its reported residual variability should override expSd <- 0.342 after readModelDb().
  • Cohort body-weight distribution truncated at 50 kg. The Phase I virtual cohort draws weights from Normal(mean = 75, SD = 11.6) matched to Garrett 2014 Table 2 but truncated below at 50 kg so the simulated lighter-end weights remain consistent with the actual Phase I enrolment ceiling on lower body weight (no observed subjects below approximately 50 kg). The truncation has no effect on the Form XLI body-weight Cmax replication (62 / 75 / 93 kg are all well above the 50 kg floor).
  • Race / sex / smoking / genotype distributions are not simulated. Garrett 2014 found none of these covariates significant on CL or Vc in the final model (Results ‘Full model and final model’), so the packaged virtual cohort omits them. Users who want to simulate specific subgroups can pass additional columns through the event table, but the model itself ignores them.
  • tlag fixed at 0.457 h. The Table 3 footnote (double-dagger) notes that the final-model run reported in Table 3 was rerun with tlag fixed at 0.457 h to obtain a successful covariance step; estimates with fixed tlag matched the free-tlag run or differed by no more than 2 in the last significant digit. The packaged model encodes tlag as fixed accordingly.
  • FED covariate sense. The packaged model uses FED = 1 for the fed state (the reference category matching Garrett 2014 Table 3’s ‘Form IV, Fed’ typical values). The fasting effect is activated by FED = 0 and applied internally via the linear-proportional shift (1 + theta_fast * (1 - FED)) on both ka and F, reproducing the paper’s theta1 * (1 + theta2 * indicator) covariate form (Methods ‘Statistical analysis’).