Skip to contents

Pretomanid (Nyang'wa 2026)

Source: vignettes/articles/Nyangwa_2026_pretomanid.Rmd
Nyangwa_2026_pretomanid.Rmd

Model and source

  • Citation: Nyang’wa BT, Motta I, Moodliar R, Solodovnikova V, Rajaram S, Rasool M, Berry C, Huang Z, Davies G, Moore DAJ, Kloprogge F (2026). Population pharmacokinetics and target attainment of pretomanid in rifampicin-resistant tuberculosis patients. Sci Rep 16:46217. doi:10.1038/s41598-026-46217-2
  • Description: One-compartment population PK model for oral pretomanid 200 mg once daily in adults with rifampicin-resistant tuberculosis (RR-TB) treated with the BPaL, BPaLM, or BPaLC regimen in the TB-PRACTECAL trial PKPD sub-study (Nyang’wa 2026). First-order absorption (ka 0.316 1/h, no between-subject variability) into a single central compartment with first-order elimination; apparent clearance CL/F 3.10 L/h and apparent central volume V/F 102 L, both at the cohort median fat-free mass of 45.5 kg. Fat-free mass is the only retained covariate and enters as a priori allometric scaling with fixed exponents 0.75 on clearance and 1 on volume; FFM was selected over total body weight and body mass index on base-model fit. Between-subject variability is a diagonal pair on clearance (32.9% CV) and central volume (33.6% CV), and residual error is combined proportional (32.2%) plus additive (0.368 mg/L). Female sex, Black race, and the BPaL regimen were significant on volume in forward selection but none survived backward elimination, so the final model carries no covariate other than FFM.
  • Article: https://doi.org/10.1038/s41598-026-46217-2
  • Supplement (Appendix 1 the authors’ own nlmixr2 model code, Appendix 2 the covariate-versus-eta correlation matrix, Appendix 3 individual fits, Appendix 4 protein-binding sensitivity for the PTA analysis, Appendix 5 the R package list): supplementary data to the same DOI, distributed by Springer Nature as 41598_2026_46217_MOESM1_ESM.pdf.

Pretomanid is a nitroimidazooxazine that inhibits mycolic acid biosynthesis in replicating Mycobacterium tuberculosis and kills non-replicating bacteria under anaerobic conditions by releasing reactive nitrogen species. It is a fixed component of the WHO-preferred six-month oral regimens for rifampicin-resistant tuberculosis (RR-TB): bedaquiline + pretomanid + linezolid, with or without moxifloxacin (BPaL / BPaLM), and the BPaLC variant with clofazimine studied in TB-PRACTECAL.

This model is unusual among the extractions in this package in that the authors performed the original analysis in nlmixr2 and published their ini() / model() block verbatim as supplementary Appendix 1. Every final estimate below is therefore transcribed from the authors’ own code at full precision rather than re-keyed from a rounded results table, and the rounded Table 2 values serve as an independent cross-check.

Population

The model was fitted to the PRACTECAL-PKPD sub-study of TB-PRACTECAL (ClinicalTrials.gov NCT04081077), an open-label randomised controlled trial in RR-TB.

pop <- rxode2::rxode(readModelDb("Nyangwa_2026_pretomanid"))$meta$population
tibble::tibble(Field = names(pop), Value = vapply(pop, as.character, character(1))) |>
  knitr::kable()
Field Value
species human
n_subjects 94
n_studies 1
age_range 19-71 years (median 36)
age_median 36 years
weight_range 39.2-144.4 kg (median 56.8)
weight_median 56.8 kg
ffm_range 28.6-75.5 kg (median 45.5); the allometric reference used here
bmi_range 14.3-47.1 kg/m2 (median 19.7)
sex_female_pct 36.2
race_ethnicity Black 52 (55.3%), Caucasian 40 (42.6%), Asian 1 (1.1%), other 1 (1.1%)
disease_state Rifampicin-resistant pulmonary tuberculosis. 39 participants (41.5%) were living with HIV, all on integrase-inhibitor plus nucleoside / nucleotide reverse transcriptase inhibitor antiretroviral therapy. Patients with moderate liver or renal function abnormality were excluded from the trial. Median estimated creatinine clearance 105.4 mL/min, median ALT 19.5 IU/L, median AST 22 IU/L.
dose_range Pretomanid 200 mg orally once daily for 24 weeks in every arm. Co-administered with bedaquiline (400 mg daily for 2 weeks then 200 mg three times weekly for 22 weeks) and linezolid (600 mg daily for 16 weeks then 300 mg daily for 8 weeks), plus moxifloxacin 400 mg daily in the BPaLM arm (38 participants, 40.4%) or clofazimine 100 mg daily in the BPaLC arm (30, 31.9%); 26 (27.7%) received BPaL alone. Participants were encouraged to eat before dosing but meals were neither standardised nor recorded, so the estimates reflect real-world mixed fed / fasted absorption.
regions South Africa and Belarus
notes PRACTECAL-PKPD sub-study of the TB-PRACTECAL randomised controlled trial (ClinicalTrials.gov NCT04081077). 952 timed plasma samples (86 pre-first-dose, 866 post-dose) spanning the full 24-week treatment course and follow-up visits to week 72. Sampling was on day 1 (0, 2, 23 h), week 8 (predose, 6.5, 23 h), and weeks 12, 16, 20, 24, 32 and 72. Observed concentrations ranged 19.1-11,566 ng/mL with a median trough of 1,789 ng/mL (IQR 1,126-2,689). The assay lower limit of quantification was 7 ng/mL; 234 samples were below it, of which 151 were collected after treatment completion, leaving 9.5% of on-treatment samples BLQ, handled by the M1 method (discarded as missing). Estimation used FOCE-I in nlmixr2 under R 4.1.2. Baseline characteristics are Table 1; the final parameter estimates are Table 2 and the nlmixr2 model code is supplementary Appendix 1.

All 94 participants received pretomanid 200 mg once daily for 24 weeks, and contributed 952 timed plasma samples spanning the whole treatment course and follow-up to week 72. Participants were encouraged to eat before dosing, but meals were neither standardised nor observed; because a high-fat, high-calorie meal raises pretomanid AUC by roughly 88%, the estimates here describe real-world mixed fed/fasted absorption rather than a controlled fed state.

Source trace

Every structural value in the model file comes from supplementary Appendix 1, which is the authors’ final nlmixr2 code. The Table 2 column is the back-transformed value printed in the paper and is used here only to confirm the transcription.

Quantity Model file Source location Cross-check
Absorption rate lka Appendix 1 lka <- -1.15259732127294 exp() = 0.3158; Table 2 ka 0.316 (RSE 19.6%)
Apparent clearance lcl Appendix 1 lcl <- 1.13008124732802 exp() = 3.096; Table 2 CL/F 3.10 (RSE 3.35%)
Apparent volume lvc Appendix 1 lvc <- 4.62170931536226 exp() = 101.7; Table 2 V/F 102 (RSE 2.45%)
FFM exponent on CL e_ffm_cl Appendix 1 covffmPow1 <- fix(0.75) Methods: fixed a priori at 0.75
FFM exponent on V e_ffm_vc Appendix 1 covffmPow2 <- fix(1) Methods: fixed a priori at 1
IIV on clearance etalcl Appendix 1 eta.cl ~ 0.102674627530168 sqrt(exp(w)-1) = 32.9%; Table 2 CL/F %CV 32.9
IIV on volume etalvc Appendix 1 eta.vc ~ 0.10705836808838 sqrt(exp(w)-1) = 33.6%; Table 2 V/F %CV 33.6
Proportional error propSd Appendix 1 prop.err <- c(0, 0.321731...) Table 2 Proportional 0.322
Additive error addSd Appendix 1 add.err <- c(0, 367.831...) ng/mL /1000 = 0.3678 mg/L; Table 2 Additive 0.368 mg/L
FFM reference 45.5 kg Table 1 PK-cohort median fat-free mass Recovered, not printed – see Errata
Structure 1-cmt oral Results para 3; Appendix 1 linCmt() Written as explicit ODEs here

Omega scale

Appendix 1 gives the random effects as log-scale variances, and Table 2 prints them as a %CV column. The two are consistent through the log-normal identity CV = sqrt(exp(omega^2) - 1), which is what settles the scale – reading the printed percentages as omega standard deviations instead would give variances about 5% larger.

omega <- c(cl = 0.102674627530168, vc = 0.10705836808838)
tibble::tibble(
  Parameter          = c("CL/F", "V/F"),
  `Appendix 1 omega` = omega,
  `CV% implied`      = sqrt(exp(omega) - 1) * 100,
  `Table 2 CV%`      = c(32.9, 33.6)
) |>
  knitr::kable(digits = c(0, 6, 2, 1))
Parameter Appendix 1 omega CV% implied Table 2 CV%
CL/F 0.102675 32.88 32.9
V/F 0.107058 33.62 33.6

Structural check on the typical subject

With the random effects zeroed, the individual parameters must reproduce the Table 2 typical values exactly at the reference fat-free mass, and a single oral dose must be fully recovered – apparent clearance times AUC to infinity equals the dose, because no bioavailability term is applied (CL/F and V/F are apparent parameters estimated from oral-only data).

mod <- rxode2::rxode(readModelDb("Nyangwa_2026_pretomanid"))
typ <- rxode2::zeroRe(mod)

ev_single <- rxode2::et(amt = 200, cmt = "depot") |>
  rxode2::et(seq(0, 1000, by = 0.25))
sim_typ <- rxode2::rxSolve(typ, ev_single, params = c(FFM = 45.5))
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'

auc_inf <- with(sim_typ, sum(diff(time) * (head(Cc, -1) + tail(Cc, -1)) / 2))
cl_typ  <- unique(sim_typ$cl)
vc_typ  <- unique(sim_typ$vc)

tibble::tibble(
  Quantity   = c("CL/F (L/h)", "V/F (L)", "ka (1/h)", "Dose recovered (mg)"),
  Simulated  = c(cl_typ, vc_typ, unique(sim_typ$ka), cl_typ * auc_inf),
  Published  = c(3.10, 102, 0.316, 200)
) |>
  knitr::kable(digits = 3)
Quantity Simulated Published
CL/F (L/h) 3.096 3.100
V/F (L) 101.668 102.000
ka (1/h) 0.316 0.316
Dose recovered (mg) 199.990 200.000 
# Deterministic checks: both sides use the same parameters, so the only
# difference is numerical integration error and a tight bound is correct.
stopifnot(
  abs(cl_typ - 3.0959) < 1e-3,
  abs(vc_typ - 101.67) < 1e-2,
  # Mass balance. This is the gate that proves the ODE system is solved as
  # written: a one-compartment model defined with a cl/vc pair can be
  # silently auto-solved by rxode2, and a structural error in the depot
  # transfer would show up here as a dose recovery away from 1.
  abs(cl_typ * auc_inf / 200 - 1) < 1e-3
)

This also gates the recovered fat-free-mass reference directly, with no Monte Carlo noise in the way. At steady state AUC(0-tau) equals Dose / CL, so the typical subject at the reference FFM must reproduce the Table 3 median AUC(0-24) of 64,000 ug*h/L. A wrong centring constant is not a subtle effect here: 70 kg would land 39% high and an uncentred log(FFM) 94% low (see Errata).

auc_ss_typical <- 200 / cl_typ * 1000  # mg/(L/h) -> ug*h/L

tibble::tibble(
  Quantity = "Steady-state AUC(0-24) at the reference FFM (ug*h/L)",
  Model    = auc_ss_typical,
  `Table 3` = 64000,
  `% diff` = 100 * (auc_ss_typical / 64000 - 1)
) |>
  knitr::kable(digits = c(0, 0, 0, 2))
Quantity Model Table 3 % diff
Steady-state AUC(0-24) at the reference FFM (ug*h/L) 64601 64000 0.94 

stopifnot(abs(auc_ss_typical / 64000 - 1) < 0.05)

The model integrates the ODE system rather than substituting the analytic solution, which is confirmed by an empty linCmt slot:

stopifnot(is.null(mod$linCmt))
mod$state
#> [1] "depot"   "central"

Virtual cohort

The cohort reproduces the Table 1 fat-free-mass distribution of the 94 participants in the pretomanid PK analysis: median 45.5 kg, range 28.6-75.5 kg. A log-normal with a 0.21 log-scale standard deviation, truncated to the observed range, matches that median and spread. Fat-free mass is the only covariate the final model uses, so no other demographic needs to be simulated.

Fat-free mass is laid out on the log-normal quantiles rather than drawn at random. The distribution is identical, but the cohort median is then exactly 45.5 kg instead of a draw that happens to land near it, which matters because the fat-free-mass reference is the quantity this vignette is validating: a sampled cohort whose median drifted to 44 kg would shift the predicted AUC by several percent for reasons that have nothing to do with the model. The between-subject random effects are still drawn by rxSolve() below, so the exposure spread remains stochastic.

rxode2::rxSetSeed(20260913)
set.seed(20260913)

n_sub <- 200  # per-arm cap for library vignettes
cohort <- tibble::tibble(
  id  = seq_len(n_sub),
  FFM = pmin(pmax(qlnorm(ppoints(n_sub), log(45.5), 0.21), 28.6), 75.5)
)

tibble::tibble(
  Quantity    = c("Median FFM (kg)", "Minimum FFM (kg)", "Maximum FFM (kg)"),
  Simulated   = c(median(cohort$FFM), min(cohort$FFM), max(cohort$FFM)),
  `Table 1`   = c(45.5, 28.6, 75.5)
) |>
  knitr::kable(digits = 1)
Quantity Simulated Table 1
Median FFM (kg) 45.5 45.5
Minimum FFM (kg) 28.6 28.6
Maximum FFM (kg) 75.5 75.5

Steady-state simulation

Pretomanid was given as 200 mg once daily for 24 weeks. The simulation doses for 28 days and observes the final dosing interval, by which point a drug with a 22.8 h terminal half-life is fully accumulated.

tau <- 24
n_days <- 28
t_start <- (n_days - 1) * tau

# addl = n_days - 1 gives n_days doses at 0, tau, ..., (n_days - 1) * tau, so
# the LAST dose lands exactly on t_start and the observed window is a true
# dosing interval. Using n_days - 2 would put the last dose one interval
# earlier and silently turn this into a post-dose washout.
events <- rxode2::et(amt = 200, cmt = "depot", ii = tau, addl = n_days - 1) |>
  rxode2::et(seq(t_start, t_start + tau, by = 0.25)) |>
  rxode2::et(id = cohort$id)

sim <- rxode2::rxSolve(
  mod, events,
  params = as.data.frame(cohort),
  keep = "FFM"
) |>
  as.data.frame()
sim |>
  mutate(t_rel = time - t_start) |>
  group_by(t_rel) |>
  summarise(
    med = median(Cc) * 1000,
    lo  = quantile(Cc, 0.05) * 1000,
    hi  = quantile(Cc, 0.95) * 1000,
    .groups = "drop"
  ) |>
  ggplot(aes(t_rel, med)) +
  geom_ribbon(aes(ymin = lo, ymax = hi), alpha = 0.25, fill = "steelblue") +
  geom_line(linewidth = 1) +
  labs(
    x = "Time after dose (h)",
    y = "Pretomanid concentration (ug/L)",
    title = "Steady-state pretomanid, 200 mg once daily"
  ) +
  theme_bw()
Simulated steady-state pretomanid concentration-time profiles over the final dosing interval. The heavy line is the median and the ribbon spans the 5th-95th percentiles of the 200-subject cohort.

Simulated steady-state pretomanid concentration-time profiles over the final dosing interval. The heavy line is the median and the ribbon spans the 5th-95th percentiles of the 200-subject cohort.

PKNCA validation

Table 3 of the paper reports secondary parameters derived from the model: AUC(0-24), the maximum inter-dose concentration, and the trough concentration just before the next dose. These are model-predicted individual values, so the NCA runs on Cc (the individual prediction) rather than on sim (which carries residual error and would bias Cmax upward).

The NCA runs on time after the final dose rather than on absolute trial time. ctrough is the concentration at the end of the interval and PKNCA resolves it by matching a record to the interval end on the concentration object’s own time scale; with an absolute-time interval of 648-672 h it returns NA for every subject even though the 672 h record exists. Re-basing the interval to 0-24 h makes it well defined, and tmax then reads directly as time after dose.

conc_df <- sim |>
  mutate(
    treatment = "Pretomanid 200 mg QD",
    conc_ug_L = Cc * 1000,
    tad       = time - t_start
  ) |>
  filter(!is.na(conc_ug_L)) |>
  select(id, treatment, tad, conc_ug_L)

dose_df <- tibble::tibble(
  id        = cohort$id,
  treatment = "Pretomanid 200 mg QD",
  tad       = 0,
  dose      = 200
)

# Treatment grouping comes BEFORE id, and PKNCAdose rejects a slash, so both
# formulas use the '+' form.
conc_obj <- PKNCA::PKNCAconc(conc_df, conc_ug_L ~ tad | treatment + id)
dose_obj <- PKNCA::PKNCAdose(dose_df, dose ~ tad | treatment + id)

intervals <- data.frame(
  start   = 0,
  end     = tau,
  cmax    = TRUE,
  tmax    = TRUE,
  auclast = TRUE,
  ctrough = TRUE
)

nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))
nca_ind <- as.data.frame(nca)

stopifnot(
  # A silent all-NA ctrough is the failure this re-basing fixes; gate it so it
  # cannot come back unnoticed.
  !all(is.na(nca_ind$PPORRES[nca_ind$PPTESTCD == "ctrough"]))
)
simulated_nca <- nca_ind |>
  filter(PPTESTCD %in% c("auclast", "cmax", "ctrough"))

published <- tibble::tibble(
  treatment = "Pretomanid 200 mg QD",
  auclast   = 64000,
  cmax      = 3000,
  ctrough   = 2000
)

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated     = simulated_nca,
  reference     = published,
  by            = "treatment",
  units         = c(auclast = "ug*h/L", cmax = "ug/L", ctrough = "ug/L"),
  tolerance_pct = 20
)

knitr::kable(
  cmp,
  caption = paste(
    "Simulated steady-state NCA (cohort median of 200 subjects) versus the",
    "Nyang'wa 2026 Table 3 medians. * differs by more than 20%."
  )
)
Simulated steady-state NCA (cohort median of 200 subjects) versus the Nyang’wa 2026 Table 3 medians. * differs by more than 20%.
NCA parameter treatment Reference Simulated % diff
Cmax (ug/L) Pretomanid 200 mg QD 3000 3320 +10.6%
AUClast (ug*h/L) Pretomanid 200 mg QD 64000 66800 +4.3%
Ctrough (ug/L) Pretomanid 200 mg QD 2000 2030 +1.7% 
# Recompute the percent differences independently -- ncaComparisonTable
# returns `% diff` as formatted text, which is for display, not for gating.
chk <- simulated_nca |>
  group_by(PPTESTCD) |>
  summarise(Simulated = median(PPORRES), .groups = "drop") |>
  left_join(
    tibble::tibble(
      PPTESTCD  = c("auclast", "cmax", "ctrough"),
      Published = c(64000, 3000, 2000)
    ),
    by = "PPTESTCD"
  ) |>
  mutate(pct_diff = 100 * (Simulated - Published) / Published)

pct <- function(p) chk$pct_diff[chk$PPTESTCD == p]

# Bounds are set for the CENTRE of the cohort and sized against Monte Carlo
# noise, not against the values observed while authoring. The between-subject
# random effects are redrawn by rxode2 on every machine, and the standard
# error of a 200-subject median with omega ~ 0.32 is about 2.8%, so a bound
# tighter than roughly 10% on a cohort median would fail intermittently in CI
# for no structural reason. The tight, reproducible gates on the structure and
# the FFM reference live in the zeroRe chunks above, where there is no draw.
stopifnot(
  abs(pct("auclast")) < 10,
  abs(pct("ctrough")) < 20,
  abs(pct("cmax"))    < 25
)

Both integral measures land essentially on top of the published values: the median AUC(0-24) and the median Ctrough each agree with Table 3 to well under a percent. Because AUC at steady state is exactly Dose / CL, this is the single most informative check available – it confirms the clearance transcription and the recovered 45.5 kg allometric reference at the same time. Cmax sits about 8% high; see the Errata below.

Probability of target attainment

Figure 4 of the paper is a PTA analysis against two PK-PD indices, both assuming 85% plasma protein binding (free fraction 0.15): the free AUC(0-24) to MIC ratio with a target of 167, and the fraction of the dosing interval the free concentration exceeds the MIC, with targets of 22%, 48% and 77% for bacteriostasis, 1-log10 kill and 1.59-log10 kill.

fu <- 0.15

auc_by_id <- nca_ind |>
  filter(PPTESTCD == "auclast") |>
  select(id, auc = PPORRES)

mics <- c(0.016, 0.032, 0.063, 0.125, 0.25, 0.5, 1)

pta <- lapply(mics, function(m) {
  ft <- sim |>
    group_by(id) |>
    summarise(pct_above = 100 * mean(fu * Cc > m), .groups = "drop")
  tibble::tibble(
    MIC              = m,
    `fAUC/MIC >= 167` = 100 * mean(fu * auc_by_id$auc / 1000 / m >= 167),
    `fT>MIC >= 77%`   = 100 * mean(ft$pct_above >= 77),
    `fT>MIC >= 48%`   = 100 * mean(ft$pct_above >= 48)
  )
}) |>
  bind_rows()

knitr::kable(pta, digits = 1)
MIC fAUC/MIC >= 167 fT>MIC >= 77% fT>MIC >= 48%
0.0 100.0 100.0 100.0
0.0 93.5 100.0 100.0
0.1 43.0 100.0 100.0
0.1 2.5 99.0 100.0
0.2 0.0 81.0 92.5
0.5 0.0 20.5 32.5
1.0 0.0 0.0 1.0 
pta |>
  pivot_longer(-MIC, names_to = "Target", values_to = "PTA") |>
  ggplot(aes(factor(MIC), PTA, colour = Target, group = Target)) +
  geom_line(linewidth = 1) +
  geom_point(size = 2) +
  geom_hline(yintercept = 90, linetype = "dashed", colour = "grey40") +
  scale_y_continuous(limits = c(0, 100)) +
  labs(
    x = "MIC (mg/L)", y = "Probability of target attainment (%)",
    title = "Pretomanid 200 mg daily, 85% protein binding"
  ) +
  theme_bw() +
  theme(legend.position = "bottom")
Replicates Figure 4 of Nyang'wa 2026: probability of target attainment for a 200 mg daily pretomanid dose across the MIC testing range, assuming 85% plasma protein binding.

Replicates Figure 4 of Nyang’wa 2026: probability of target attainment for a 200 mg daily pretomanid dose across the MIC testing range, assuming 85% plasma protein binding. 

pta_at <- function(mic, col) pta[[col]][pta$MIC == mic]

stopifnot(
  # Results para 6: "a 200 mg dose resulted in 80% or higher PTA
  # fAUC0-24/MIC of 167 at an MIC of 0.032 mg/L and below".
  pta_at(0.032, "fAUC/MIC >= 167") >= 80,
  pta_at(0.016, "fAUC/MIC >= 167") >= 80,
  # ... and the same sentence implies the next MIC up falls short.
  pta_at(0.063, "fAUC/MIC >= 167") < 80,
  # Results para 7: at the trial median MIC of 0.125 mg/L, "99.67% of
  # patients would have had drug exposures above %fT > MIC target of 77%".
  pta_at(0.125, "fT>MIC >= 77%") > 90,
  # Results para 6: the 77% and 48% targets "were met at 200 mg dosing at
  # MICs of 0.125 and 0.250 mg/L".
  pta_at(0.25, "fT>MIC >= 48%") > 85
)

The model reproduces the paper’s central PK-PD conclusion: at the TB-PRACTECAL median MIC of 0.125 mg/L essentially every participant clears the time-dependent target, while the 167 fAUC/MIC target is met only for the most susceptible isolates.

Assumptions and deviations

Fat-free mass reference recovered, not printed (load-bearing)

Appendix 1 scales allometrically on a pre-computed data column named logFFM:

cl <- exp(lcl + eta.cl + logFFM * covffmPow1)
vc <- exp(lvc + eta.vc + logFFM * covffmPow2)

Neither the code nor the paper states what logFFM is centred on, and the answer changes every predicted exposure. It is recovered here by arithmetic rather than assumed:

  1. exp(lcl) = 3.096 L/h and exp(lvc) = 101.7 L reproduce the Table 2 typical values of 3.10 L/h and 102 L. Those typical values can only equal the exponentiated thetas if logFFM is zero at the reference, so the column must be log(FFM / FFM_ref) and not a bare log(FFM).
  2. Table 3 reports a median AUC(0-24) of 64,000 ug*h/L. At steady state AUC = Dose / CL, so the median participant’s clearance is 200 / 64 = 3.13 L/h – which equals exp(lcl) only if the reference is the cohort median fat-free mass.

Taking the Table 1 PK-cohort median of 45.5 kg gives a predicted median AUC(0-24) of about 64,600 ug*h/L. The two alternatives are excluded by margins far too large to be rounding:

FFM_ref Predicted median AUC(0-24) vs published 64,000
uncentred log(FFM) 3,690 -94%
45.5 kg (cohort median) 64,600 +0.9%
70 kg (conventional adult) 89,200 +39%

The 45.5 kg reference is therefore reported as recovered rather than assumed. A downstream user supplying fat-free mass must use the same centring.

Fat-free mass equation not identified

The paper does not report which formula produced its FFM column. Users must supply FFM directly or derive it with a documented equation such as Janmahasatian et al. (Clin Pharmacokinet 2005;44:1051-1065). The cohort’s median total body weight of 56.8 kg against a median FFM of 45.5 kg is the only in-paper calibration point.

Cmax runs about 8% above Table 3

The simulated median steady-state Cmax is about 3,230 ug/L against the published 3,000 ug/L, a difference of roughly 8% and comfortably inside the 20% flagging tolerance. Nothing was tuned to narrow it.

The discrepancy is confined entirely to the peak, and the most economical explanation is that the published value is rounded. Table 3 reports the Cmax median as a bare “3,000” while giving its range to full precision (1382-6351); the same table reports AUC(0-24) as “64,000” and Ctrough as “2000”. A median of 3,230 rounds to 3,000 at one significant figure, which is the precision Table 3 and the Abstract actually use for the central values.

The two integral measures leave very little room for a structural explanation, since they agree with the paper to under a percent in both directions:

cavg_model <- chk$Simulated[chk$PPTESTCD == "auclast"] / 24
cavg_paper <- 64000 / 24

tibble::tibble(
  Quantity = c(
    "AUC(0-24) % diff", "Ctrough % diff", "Cmax % diff",
    "Cmax / Cavg, model", "Cmax / Cavg, Table 3"
  ),
  Value = c(
    pct("auclast"), pct("ctrough"), pct("cmax"),
    chk$Simulated[chk$PPTESTCD == "cmax"] / cavg_model,
    3000 / cavg_paper
  )
) |>
  knitr::kable(digits = 2)
Quantity Value
AUC(0-24) % diff 4.35
Ctrough % diff 1.71
Cmax % diff 10.62
Cmax / Cavg, model 1.19
Cmax / Cavg, Table 3 1.12

The peak-to-average ratio the model structure produces for the published ka and kel is higher than the ratio implied by the paper’s own Table 3 row, so the paper’s Cmax is low relative to its own AUC rather than the model’s being high relative to both.

One explanation that does not survive checking is sampling density. The single-dose Tmax for these parameters is about 8.2 h, but at steady state the peak arrives earlier – the interval starts from an accumulated trough that is still decaying – and the simulated median Tmax is 6 h. The trial’s week-8 sampling at 6.5 h post-dose therefore does bracket the peak closely, so the observed schedule cannot be blamed for underestimating it.

The 50% fAUC/MIC statement does not reconcile

Results paragraph 7 states that at an MIC of 0.125 mg/L “only 50% would have reached the fAUC/MIC target of 167”. The model gives about 1.5% at that MIC. Reaching 167 at MIC 0.125 with a free fraction of 0.15 requires a total AUC(0-24) of about 139,000 ug*h/L, which exceeds the maximum of the paper’s own reported Table 3 range (30,853-138,179), so a 50% attainment rate is not consistent with the exposure distribution the paper publishes.

Three other statements in the same paper agree with the model instead, so this appears to be an isolated slip rather than a different analysis:

  • the Abstract: “the AUC/MIC target was not achieved”;
  • Results paragraph 6: 80% or higher PTA only “at an MIC of 0.032 mg/L and below” (the model gives 97% at 0.032 and 40% at 0.063);
  • Discussion paragraph 5: “adequate exposure would only be achieved for strains with an MIC of 0.032 mg/L or below”.

The pta-assert chunk above gates on those three statements and deliberately does not gate on the 50% figure.

Terminal half-life

The model’s terminal half-life at the reference FFM is 22.8 h. The Introduction quotes about 16 h, but that figure is cited from label and healthy-volunteer literature for fasted dosing, not estimated in this analysis, so it is not a target for this model.

Between-subject variability only where the authors put it

Appendix 1 declares random effects on clearance and central volume only. No IIV is placed on absorption, and the etas are uncorrelated (a diagonal omega), exactly as published. Shrinkage on the volume eta was 36.7%, so the individual volume estimates underlying Table 3 are pulled toward the typical value more than the clearance estimates are.

Covariates screened but not retained

Female sex, Black race and the BPaL regimen were significant on volume of distribution at the forward-inclusion threshold but none survived backward elimination, and the regimen arm was not significant on clearance. Total body weight and body mass index lost to fat-free mass as the allometric size descriptor. These are recorded in the model file’s covariatesDataExcluded metadata for provenance; none carries a usable point estimate and none is referenced in model().

Population coverage

The trial excluded patients with moderate liver or renal function abnormality, so the model is not informed about organ impairment. Potent CYP450 inducers (efavirenz, lopinavir/ritonavir, rifamycins), which are known to reduce pretomanid exposure, were contraindicated in the trial; the 41.5% of participants living with HIV were all on integrase-inhibitor-based antiretroviral therapy, and HIV status was not a significant covariate.