Skip to contents

Isavuconazole (Desai 2016)

Source: vignettes/articles/Desai_2016_isavuconazole.Rmd
Desai_2016_isavuconazole.Rmd

Model and source

  • Citation: Desai A, Kovanda L, Kowalski D, Lu Q, Townsend R, Bonate PL. Population Pharmacokinetics of Isavuconazole from Phase 1 and Phase 3 (SECURE) Trials in Adults and Target Attainment in Patients with Invasive Infections Due to Aspergillus and Other Filamentous Fungi. Antimicrob Agents Chemother. 2016;60(9):5483-5491. doi:10.1128/AAC.02819-15
  • Description: Two-compartment population PK model with a Weibull-function first-order absorption stage for isavuconazole administered as the prodrug isavuconazonium sulfate (p.o. or i.v.) to healthy adults and patients with invasive fungal infections (Desai 2016 SECURE pooled phase 1 / phase 3 popPK)
  • Article: https://doi.org/10.1128/AAC.02819-15 (open access, CC-BY 4.0)

Isavuconazole is the active moiety of the water-soluble triazole-antifungal prodrug isavuconazonium sulfate (372 mg prodrug = 200 mg isavuconazole; 1 h IV infusion or oral). Desai 2016 pooled data from nine phase 1 studies in healthy subjects with the phase 3 SECURE clinical trial in patients with invasive aspergillosis or other filamentous-fungi IFIs to develop a population-PK model in NONMEM 7.2 (PsN 3.7.6 for stepwise covariate modeling). A two-compartment model with a Weibull-function first-order absorption stage and linear elimination was retained as the best model; race (Asian vs Caucasian) modified clearance, and BMI plus subject-vs-patient status (SP) modified the peripheral volume of distribution.

Population

The pooled analysis dataset comprised 6,363 isavuconazole plasma concentrations from 421 individuals: 189 healthy subjects (5,828 records) across nine phase 1 studies and 232 patients (535 records) from the SECURE phase 3 trial. Age ranged from 17 to 85 years (median 43 in healthy subjects, 54 in patients); weight ranged from 41.0 to 127.7 kg; BMI ranged from 13.9 to 41.1 kg/m^2 (43 subjects had BMI > 30 and were classified obese). Sex was 64.6% male overall. Race was 87.4% predominantly Caucasian (which included 1 African American and 5 in other-race categories pooled into the reference category per Desai 2016 Table 3 footnote) and 12.6% Asian. SECURE dosing was 200 mg isavuconazole-equivalent (372 mg isavuconazonium sulfate) q8h for 6 loading doses on days 1-2 and 200 mg q24h thereafter, p.o. or i.v. Demographics are summarised in Desai 2016 Table 3; data sources are listed in Table 1. The same information is available programmatically:

str(rxode2::rxode(readModelDb("Desai_2016_isavuconazole"))$population)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> List of 14
#>  $ species       : chr "human"
#>  $ n_subjects    : int 421
#>  $ n_studies     : int 10
#>  $ n_observations: int 6363
#>  $ age_range     : chr "17-85 years"
#>  $ age_median    : chr "43 years (healthy subjects) / 54 years (patients)"
#>  $ weight_range  : chr "41.0-127.7 kg"
#>  $ weight_median : chr "77.8 kg (healthy subjects) / 67.0 kg (patients)"
#>  $ sex_female_pct: num 35.4
#>  $ race_ethnicity: Named num [1:2] 87.4 12.6
#>   ..- attr(*, "names")= chr [1:2] "predominantly_Caucasian" "Asian"
#>  $ disease_state : chr "Pooled cohort: 189 healthy subjects from 9 phase 1 studies (single- and multiple-ascending dose, hepatic impair"| __truncated__
#>  $ dose_range    : chr "Isavuconazole 40-400 mg, p.o. or i.v. (1-h infusion), single or multiple doses across phase 1; SECURE: 200 mg i"| __truncated__
#>  $ regions       : chr "International (phase 1 sites plus the global SECURE phase 3 trial including Asian regions)"
#>  $ notes         : chr "Dosed as the water-soluble prodrug isavuconazonium sulfate; the model represents the active moiety isavuconazol"| __truncated__

Source trace

Each ini() parameter in inst/modeldb/specificDrugs/Desai_2016_isavuconazole.R carries an in-file comment pointing to its source. The table below collects the provenance in one place.

Equation / parameter Value Source location
lcl (Caucasian CL) log(2.36) Desai 2016 Table 5 theta_1
lvc (V_1) log(49.10) Desai 2016 Table 5 theta_2
lq (Q) log(26.60) Desai 2016 Table 5 theta_3
lvp (V_p, patient) log(417) Desai 2016 Table 5 theta_4 (assigned to patients per Discussion typical V_p ~390 L)
lkamax (KAMAX) log(1.08) Desai 2016 Table 5 theta_5
lra (RA) log(0.72) Desai 2016 Table 5 theta_6
lgam1 (GAM1) log(4.88) Desai 2016 Table 5 theta_7
lfdepot (F) fixed(0) Desai 2016 Methods (F fixed at 1 from prior NCA in healthy subjects)
e_race_asian_cl -0.3602 Derived from theta_9 (CL_Asian = 1.51) and theta_1 (CL_Caucasian = 2.36): (1.51-2.36)/2.36
e_dis_healthy_vp -0.3765 Derived from theta_11 (V_p_healthy = 260) and theta_4 (V_p_patient = 417): (260-417)/417
e_bmi_vp 0.060 Desai 2016 Table 5 theta_10
IIV etalcl (omega^2) 0.3293 Desai 2016 Table 5 CL (patients) CV 62.44%; omega^2 = log(0.6244^2 + 1)
IIV etalvp (omega^2) 0.0962 Desai 2016 Table 5 V_p CV 31.78%
IIV etalq (omega^2) 0.2159 Desai 2016 Table 5 Q CV 49.09%
IIV etalra (omega^2) 0.1500 Desai 2016 Table 5 RA CV 40.24%
IIV etalgam1 (omega^2) 0.1898 Desai 2016 Table 5 GAM1 CV 45.71%
propSd (residual) 0.4494 Desai 2016 Table 5 theta_8 W = 44.94%; Ln-Ln additive error mapped to linear-space proportional
d/dt(depot) -WB*depot Desai 2016 Eq for dA(1)/dt
WB (time-varying ka) Desai 2016 Eq: KAMAX * (1 - exp(-(RA*TAD)^GAM1))
d/dt(central) Desai 2016 Eq for dA(2)/dt
d/dt(peripheral1) Desai 2016 Eq for dA(3)/dt

Virtual cohort

The original observed data are not publicly available. The replications below use virtual patient cohorts whose covariate distributions approximate the SECURE-trial patients (BMI, race) and whose dosing follows the SECURE clinical regimen.

set.seed(2026L)

n_per_arm <- 250L

# BMI ranges per Desai 2016: 14-41 kg/m^2 for Caucasian patients (matching
# NHANES 2014 demographic distribution) and 17-28 kg/m^2 for Asian patients.
# Approximate the within-arm BMI distribution as truncated normal centered at
# the patient median (Desai 2016 Table 3: 23.6 kg/m^2 for patients).
make_arm <- function(n, race_asian, id_offset, bmi_lo, bmi_hi, bmi_mean, bmi_sd) {
  bmi <- pmin(bmi_hi, pmax(bmi_lo, rnorm(n, mean = bmi_mean, sd = bmi_sd)))
  tibble(
    id          = id_offset + seq_len(n),
    RACE_ASIAN  = race_asian,
    BMI         = bmi,
    DIS_HEALTHY = 0L,
    cohort      = if (race_asian == 1L) "Asian patient" else "Caucasian patient"
  )
}

cohort_def <- bind_rows(
  make_arm(n_per_arm, race_asian = 0L, id_offset =                0L,
           bmi_lo = 14, bmi_hi = 41, bmi_mean = 25.0, bmi_sd = 4.5),
  make_arm(n_per_arm, race_asian = 1L, id_offset =        n_per_arm,
           bmi_lo = 17, bmi_hi = 28, bmi_mean = 22.5, bmi_sd = 3.0)
)

Build a SECURE-style dosing event table: a 1-hour i.v. infusion of 200 mg loading dose q8h for 6 doses (days 1-2), then 200 mg p.o. q24h to t = 864 h (end of last steady-state dosing interval used for the AUC calculation in Desai 2016 PTA simulations). Observation times sample the post-loading-dose absorption phase densely and the steady-state interval q24h for AUC calculation.

loading_times     <- seq(0, by = 8, length.out = 6L)     # 0, 8, 16, 24, 32, 40 h
maintenance_times <- seq(48, 840, by = 24)               # day 3 onward
all_dose_times    <- c(loading_times, maintenance_times)

# Loading doses i.v. (compartment = central, 1-h infusion);
# maintenance doses p.o. (compartment = depot).
dose_rows <- bind_rows(
  tibble(
    time = loading_times,
    amt  = 200,
    cmt  = "central",
    evid = 1L,
    rate = 200             # 1-hour infusion -> rate = 200 mg/h
  ),
  tibble(
    time = maintenance_times,
    amt  = 200,
    cmt  = "depot",
    evid = 1L,
    rate = 0
  )
)

# Observation grid: dense in the early-absorption phase post-first-dose, then
# sparse to the steady-state interval, then dense again over the steady-state
# interval, and a terminal decay phase past the last dose for half-life.
obs_grid <- sort(unique(c(
  seq(0, 12, by = 0.25),
  seq(12, 48, by = 1),
  seq(48, 840, by = 12),
  seq(840, 864, by = 0.5),
  seq(864, 1500, by = 12)
)))

obs_rows <- tibble(
  time = obs_grid,
  amt  = NA_real_,
  cmt  = "central",
  evid = 0L,
  rate = 0
)

# Cross dose + obs records with each subject's covariates.
events <- cohort_def |>
  rowwise() |>
  do({
    sub <- .
    bind_rows(dose_rows, obs_rows) |>
      mutate(
        id          = sub$id,
        RACE_ASIAN  = sub$RACE_ASIAN,
        BMI         = sub$BMI,
        DIS_HEALTHY = sub$DIS_HEALTHY,
        cohort      = sub$cohort
      )
  }) |>
  ungroup() |>
  arrange(id, time, desc(evid))

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

Simulation 

mod <- rxode2::rxode(readModelDb("Desai_2016_isavuconazole"))
#> ℹ parameter labels from comments will be replaced by 'label()'

sim <- rxode2::rxSolve(
  mod,
  events = events,
  keep = c("cohort", "RACE_ASIAN", "BMI", "DIS_HEALTHY")
) |>
  as.data.frame()

Replicate published figures 

# Visual concentration-time profile by cohort, showing the SECURE loading +
# maintenance regimen approach to steady state and terminal decay after the
# last dose at 840 h. Median +/- 5th-95th percentiles across the simulated
# subjects.
sim |>
  dplyr::filter(time <= 240) |>     # focus on loading + first week
  dplyr::filter(time > 0) |>
  dplyr::group_by(time, cohort) |>
  dplyr::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"
  ) |>
  ggplot(aes(time, Q50, colour = cohort, fill = cohort)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.2, colour = NA) +
  geom_line(linewidth = 0.7) +
  scale_y_log10() +
  labs(
    x = "Time post first dose (h)",
    y = "Plasma isavuconazole (mg/L, log scale)",
    title = "SECURE-style dosing: loading q8h x 6 + maintenance q24h",
    subtitle = "Caucasian vs Asian patient cohorts (median, 5-95th percentile)"
  ) +
  theme_minimal()

PKNCA validation

Compute steady-state AUC0-24, Cmax, and Tmax over the last simulated dosing interval (840-864 h), which matches the Desai 2016 PTA simulation window.

sim_nca_ss <- sim |>
  dplyr::filter(!is.na(Cc), time >= 840, time <= 864) |>
  dplyr::select(id, time, Cc, cohort)

dose_for_nca <- events |>
  dplyr::filter(evid == 1L) |>
  dplyr::select(id, time, amt) |>
  dplyr::left_join(cohort_def |> dplyr::select(id, cohort), by = "id")

conc_obj <- PKNCA::PKNCAconc(
  sim_nca_ss,
  Cc ~ time | cohort + id,
  concu = "mg/L",
  timeu = "h"
)

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

intervals_ss <- data.frame(
  start   = 840,
  end     = 864,
  cmax    = TRUE,
  tmax    = TRUE,
  auclast = TRUE,
  cav     = TRUE,
  cmin    = TRUE
)

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

ss_tbl <- as.data.frame(nca_ss$result) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "tmax", "auclast", "cav", "cmin")) |>
  dplyr::group_by(cohort, PPTESTCD) |>
  dplyr::summarise(
    median = median(PPORRES, na.rm = TRUE),
    q05    = quantile(PPORRES, 0.05, na.rm = TRUE),
    q95    = quantile(PPORRES, 0.95, na.rm = TRUE),
    .groups = "drop"
  )

knitr::kable(
  ss_tbl,
  caption = "Steady-state NCA over the dosing interval 840-864 h (median, 5-95th percentile across virtual patients).",
  digits = 2
)
Steady-state NCA over the dosing interval 840-864 h (median, 5-95th percentile across virtual patients).
cohort PPTESTCD median q05 q95
Asian patient auclast 128.02 55.36 253.30
Asian patient cav 5.33 2.31 10.55
Asian patient cmax 6.72 3.83 12.26
Asian patient cmin 4.99 1.87 9.93
Asian patient tmax 2.50 2.00 3.50
Caucasian patient auclast 80.64 35.75 179.88
Caucasian patient cav 3.36 1.49 7.49
Caucasian patient cmax 4.84 2.88 8.94
Caucasian patient cmin 2.88 1.10 7.01
Caucasian patient tmax 2.50 2.00 3.50

For the terminal half-life, compute lambda.z over the post-last-dose decay window (864-1500 h).

sim_nca_term <- sim |>
  dplyr::filter(!is.na(Cc), time >= 864, time <= 1500) |>
  dplyr::select(id, time, Cc, cohort)

conc_obj_term <- PKNCA::PKNCAconc(
  sim_nca_term,
  Cc ~ time | cohort + id,
  concu = "mg/L",
  timeu = "h"
)

intervals_term <- data.frame(
  start       = 864,
  end         = 1500,
  half.life   = TRUE,
  lambda.z    = TRUE,
  clast.obs   = TRUE
)

nca_term <- PKNCA::pk.nca(
  PKNCA::PKNCAdata(conc_obj_term, dose_obj, intervals = intervals_term)
)

hl_tbl <- as.data.frame(nca_term$result) |>
  dplyr::filter(PPTESTCD == "half.life") |>
  dplyr::group_by(cohort) |>
  dplyr::summarise(
    median_halflife = median(PPORRES, na.rm = TRUE),
    q05             = quantile(PPORRES, 0.05, na.rm = TRUE),
    q95             = quantile(PPORRES, 0.95, na.rm = TRUE),
    .groups = "drop"
  )

knitr::kable(
  hl_tbl,
  caption = "Terminal half-life (h) by cohort from the 864-1500 h decay window.",
  digits = 1
)
Terminal half-life (h) by cohort from the 864-1500 h decay window.
cohort median_halflife q05 q95
Asian patient 204.7 67.0 661.8
Caucasian patient 145.5 61.6 406.6

Comparison against published NCA

Desai 2016 reports the per-subject AUC0-24 at steady state computed as F * dose / CL = 200 / CL_indiv mg.h/L (Table 4): mean AUC0-24 was 92.0 mg.h/L for healthy subjects, 101.0 mg.h/L for patients, and 97.0 mg.h/L combined. The mean terminal half-life for the patient population was estimated to be 130 h (median 110 h, 5th-95th percentiles 53.5-248.1 h). The Discussion also reports complete absorption at 2-3 h postdosing and steady-state volume V_ss ~460 L.

published <- tibble::tribble(
  ~cohort,                ~param,            ~published_value,
  "Caucasian patient",    "AUC0-24 (mg.h/L)", 101.0,
  "Asian patient",        "AUC0-24 (mg.h/L)", NA,        # paper does not report Asian-only mean (PTA used pooled CL)
  "Caucasian patient",    "Half-life (h)",   130.0,
  "Asian patient",        "Half-life (h)",   130.0
)

ss_auc <- ss_tbl |>
  dplyr::filter(PPTESTCD == "auclast") |>
  dplyr::transmute(cohort, param = "AUC0-24 (mg.h/L)", simulated_median = median)

ss_hl <- hl_tbl |>
  dplyr::transmute(cohort, param = "Half-life (h)", simulated_median = median_halflife)

simulated <- bind_rows(ss_auc, ss_hl)

comparison <- published |>
  dplyr::left_join(simulated, by = c("cohort", "param")) |>
  dplyr::mutate(
    pct_diff = ifelse(
      is.na(published_value),
      NA_real_,
      100 * (simulated_median - published_value) / published_value
    )
  )

knitr::kable(
  comparison,
  caption = "Side-by-side comparison of published Desai 2016 NCA values with the simulated NCA values from this validation.",
  digits = 1
)
Side-by-side comparison of published Desai 2016 NCA values with the simulated NCA values from this validation.
cohort param published_value simulated_median pct_diff
Caucasian patient AUC0-24 (mg.h/L) 101 80.6 -20.2
Asian patient AUC0-24 (mg.h/L) NA 128.0 NA
Caucasian patient Half-life (h) 130 145.5 11.9
Asian patient Half-life (h) 130 204.7 57.5

The simulated Asian-cohort AUC0-24 is expected to be approximately 1.56-fold higher than the Caucasian cohort, matching the published ~36% lower CL in Asian subjects (AUC scales as 1/CL).

Assumptions and deviations

  • V_p baseline assignment (theta_4 = patient, theta_11 = healthy) is inferred from the Discussion typical values (V_p ~390 L in patients, ~292 L in healthy subjects) and the demographic BMI medians; the Table 5 column labels do not unambiguously identify which group goes with which theta. Back-calculating the typicals at the median BMIs gives V_p (patient) = 417 * (1 + 0.060 * (23.6 - 24.80)) = 387 L (matches ~390 L) and V_p (healthy) = 260 * (1 + 0.060 * (25.7 - 24.80)) = 274 L (close to ~292 L). The alternative assignment (theta_4 to healthy, theta_11 to patients) does not reconcile with the Discussion values.
  • Single omega for CL. Desai 2016 reports separate CL between-subject variability for healthy subjects (31.30% CV) and patients (62.44% CV), suggesting a categorical OMEGA structure in NONMEM. nlmixr2 supports a single omega per parameter, so the patient CV (62.44%) is used as the operational default. The model is intended primarily for clinical simulations of patients with invasive fungal infections, where the higher-variability estimate is the more representative choice. Downstream users who want to simulate the healthy subject population should set omega^2 = log(0.3130^2 + 1) = 0.0935 instead.
  • Residual error mapping. Desai 2016 used Ln-Ln transformation of observations and predictions with additive error on the log scale (W = 44.94%). In nlmixr2’s linear-concentration space this maps to a proportional error structure with propSd = 0.4494 (per references/verification-checklist.md); the approximation is exact in the small-sigma limit and gives a linear-space coefficient of variation of sqrt(exp(0.4494^2) - 1) = 47.3% at this value of sigma.
  • BMI distribution by race. The simulation cohort uses NHANES-style BMI ranges (Caucasian: 14-41 kg/m^2 with mean ~25.0 kg/m^2; Asian: 17-28 kg/m^2 with mean ~22.5 kg/m^2) approximating the Desai 2016 PTA Monte Carlo setup. The exact NHANES per-subject BMI distribution used by Desai 2016 is not reproducible here without the underlying NHANES sample.
  • Loading-dose route. SECURE protocol allowed either p.o. or i.v. loading; this vignette uses i.v. loading + p.o. maintenance to exercise both administration routes (and verify the Weibull absorption stage only acts when drug is in the depot compartment).
  • Time-after-dose (TAD) reset across the multi-dose regimen. The Weibull absorption function depends on rxode2’s tad() (time since the most recent dose). For the q24h maintenance regimen the absorption is essentially complete by 2-3 h, so residual depot at the next dose is near zero and the per-dose TAD reset has negligible impact on accumulation. The same dose-event semantics are used by the original NONMEM implementation.

Errata

No erratum is listed for Desai 2016 as of the model-extraction date (2026-06-02); a PubMed and Antimicrobial Agents and Chemotherapy notices-and-corrections search returned no related correction.