Teicoplanin (Zhou 2025)

Model and source

mod <- readModelDb("Zhou_2025_teicoplanin")
ui <- rxode2::rxode(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'

• Citation: Zhou Y, Peng J, Xu P, Wang F, Xi J, Zhang H, Hu S, Yan H, Tan L, Cai H, Zhang B, Lan G. Population pharmacokinetics and dosing optimization of teicoplanin in renal transplant patients. Antimicrob Agents Chemother. 2025;69(6):e01568-24. doi:10.1128/aac.01568-24
• Article: https://doi.org/10.1128/aac.01568-24 (PMC12135518, open access)
• Supplement: aac.01568-24-s0001.docx (Tables S1-S2, Figures S1-S2), retrieved from the EuropePMC supplementary-files endpoint for PMC12135518.

Teicoplanin disposition in 79 adult renal transplant recipients was described by a two-compartment model with first-order elimination and a zero-order (1 h infusion) input. Real-time Cockcroft-Gault creatinine clearance on clearance was the only covariate retained.

Population

The development cohort was 79 adults (56 male, 70.9%) enrolled prospectively at the Second Xiangya Hospital, Changsha, between January 2022 and December 2023, contributing 306 plasma concentrations (about four samples each). Median age was 42.5 years (IQR 36-52.8), median weight 64.8 kg (IQR 54-74.8) and median height 168 cm (Zhou 2025 Table 1). Every patient was on tacrolimus and mycophenolate mofetil; hypertension (81.0%) and diabetes (25.3%) were the commonest comorbidities. Sixty-five of the 79 were within one month of transplantation.

The defining feature of the cohort is renal: median baseline Cockcroft-Gault creatinine clearance was 9 mL/min (IQR 6.6-13.8) and median blood urea nitrogen 18.8 mmol/L. No patient received renal replacement therapy or mechanical ventilation. Because renal function recovers rapidly after transplantation, the authors modelled the real-time rather than the baseline creatinine clearance; the 17 mL/min normalising constant in the clearance equation sits well above the baseline median for exactly that reason.

All patients received the same empirical regimen: three 400 mg loading doses every 12 h, then 400 mg once daily for at least 3 days, each as a 1 h infusion. Eighty of the 99 enrolled patients received teicoplanin as perioperative prophylaxis and 19 for a confirmed or probable Gram-positive infection (Table S1: pneumonia 14, urinary tract infection 5; MRSA 9 and Enterococcus faecium 5, teicoplanin MIC 0.5-1 mg/L).

A separate cohort of 20 patients / 80 samples, recruited November 2024 to March 2025, was used for external validation (RMSE 8.49%, mean prediction error 0.704%).

The same information is available programmatically via readModelDb("Zhou_2025_teicoplanin")()$population.

Source trace

Every ini() entry in inst/modeldb/specificDrugs/Zhou_2025_teicoplanin.R carries an in-file comment naming its source. They are collected here for review.

Equation / parameter	Value	Source location
lcl (CL at CrCL = 17 mL/min)	0.711 L/h	Table 2, CL; Results equation block; Abstract
lvc (Vc)	11.3 L	Table 2, Vc; Results equation block
lq (CLd)	4.22 L/h	Table 2, CLd; Results equation block
lvp (Vp)	35.3 L	Table 2, Vp; Results equation block (the Abstract and the sentence below Table 2 round to 35.2)
e_crcl_cl (power exponent of CrCL on CL)	0.198	Table 2, CrCLCL, “the influence coefficient of CrCL on CL”
etalcl	omega = 0.401 (variance 0.1608)	Table 2, omega CL = 40.1%
etalvc	omega = 0.373 (variance 0.1391)	Table 2, omega Vc = 37.3%
propSd	0.146	Table 2, “Proportional residual (%)” = 14.6
cl <- exp(lcl + etalcl) * (CRCL / 17)^e_crcl_cl	n/a	Results, “Population pharmacokinetic modeling”: CL (L/h) = 0.711 x (CrCL / 17)^0.198 x e^eta1
vc <- exp(lvc + etalvc)	n/a	same equation block: Vc (L) = 11.3 x e^eta2
q <- exp(lq), vp <- exp(lvp)	n/a	same equation block: CLd (L/h) = 4.22, Vp (L) = 35.3 (no e^eta factor)
two-compartment ODEs, zero-order input	n/a	Results, “Population pharmacokinetic modeling”; Table S2 base model, “two-compartment model with a zero-order input rate”; Methods, 1 h infusion
Cc ~ prop(propSd)	n/a	Results, “proportional residual variability”; Methods lists the three candidate log-scale error forms

The equation block is typeset as a display equation and is lost by a naive text extraction of the PDF (it appears as formula-not-decoded in the preprocessed markdown). It was recovered with pdftotext -layout, which preserves it verbatim.

Two source-internal disagreements are recorded in Assumptions, deviations and errata below: the Results prose calls the retained covariate relationship “a proportional model” while the printed equation is a power model, and Vp is printed as both 35.3 and 35.2.

ui$iniDf |>
  dplyr::select(name, est, fix, label) |>
  dplyr::rename(
    "Parameter" = name, "Estimate" = est,
    "Fixed" = fix, "Label" = label
  ) |>
  knitr::kable(digits = 6, caption = "Packaged `ini()` block.")

Packaged ini() block.

Parameter	Estimate	Fixed	Label
lcl	-0.341083	FALSE	Clearance at CrCL = 17 mL/min (L/h)
lvc	2.424803	FALSE	Central volume Vc (L)
lq	1.439835	FALSE	Inter-compartmental clearance CLd (L/h)
lvp	3.563883	FALSE	Peripheral volume Vp (L)
e_crcl_cl	0.198000	FALSE	Power exponent of real-time CrCL on CL (unitless)
propSd	0.146000	FALSE	Proportional residual error (fraction)
etalcl	0.160801	FALSE	Zhou 2025 Table 2: omega_CL = 40.1% (SE 12%; bootstrap median 39.6%, 90% CI 30.6-49.2) -> omega^2 = 0.401^2
etalvc	0.139129	FALSE	Zhou 2025 Table 2: omega_Vc = 37.3% (SE 21%; bootstrap median 36.2%, 90% CI 16.7-53.0) -> omega^2 = 0.373^2

Structural check: ODE solution against the closed form

The two-compartment model with a constant-rate infusion has an exact bi-exponential solution. Both sides use the same parameter values, so the only difference between them is numerical integration error and the bound below is correspondingly tight.

crcl_ref <- 30
p <- list(cl = 0.711 * (crcl_ref / 17)^0.198, vc = 11.3, q = 4.22, vp = 35.3)
kel <- p$cl / p$vc
k12 <- p$q / p$vc
k21 <- p$q / p$vp
bsum <- kel + k12 + k21
disc <- sqrt(bsum^2 - 4 * kel * k21)
alpha <- (bsum + disc) / 2
beta <- (bsum - disc) / 2

dose_mg <- 400
t_inf <- 1
coef_a <- (k21 - alpha) / (p$vc * (beta - alpha))
coef_b <- (k21 - beta) / (p$vc * (alpha - beta))

closed_form <- function(t) {
  arm <- function(t, k, coef) {
    ifelse(
      t <= t_inf,
      coef * (dose_mg / t_inf) / k * (1 - exp(-k * t)),
      coef * (dose_mg / t_inf) / k * (1 - exp(-k * t_inf)) * exp(-k * (t - t_inf))
    )
  }
  arm(t, alpha, coef_a) + arm(t, beta, coef_b)
}

cf_times <- c(0.25, 0.5, 1, 2, 4, 8, 12, 24, 48, 96, 168)
cf_ev <- data.frame(
  id = 1L,
  time = c(0, cf_times),
  amt = c(dose_mg, rep(NA_real_, length(cf_times))),
  evid = c(1L, rep(0L, length(cf_times))),
  dur = c(t_inf, rep(NA_real_, length(cf_times))),
  cmt = "central",
  CRCL = crcl_ref
)
cf_sim <- rxode2::rxSolve(
  rxode2::zeroRe(mod), cf_ev, returnType = "data.frame"
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
cf_cmp <- data.frame(
  time = cf_sim$time,
  ode = cf_sim$Cc,
  closed = closed_form(cf_sim$time)
) |>
  dplyr::mutate(pct_diff = 100 * (ode - closed) / closed)

cf_cmp |>
  dplyr::mutate(pct_diff = sprintf("%.1e", pct_diff)) |>
  dplyr::rename(
    "Time (h)" = time, "ODE solve (mg/L)" = ode,
    "Closed form (mg/L)" = closed, "% difference" = pct_diff
  ) |>
  knitr::kable(digits = c(2, 4, 4, NA),
               caption = "400 mg over 1 h, CrCL 30 mL/min, typical subject.")

400 mg over 1 h, CrCL 30 mL/min, typical subject.

Time (h)	ODE solve (mg/L)	Closed form (mg/L)	% difference
0.25	8.3801	8.3801	4.2e-13
0.50	15.9018	15.9018	4.1e-13
1.00	28.7923	28.7923	3.8e-13
2.00	19.4383	19.4383	2.0e-13
4.00	10.7960	10.7960	-6.6e-14
8.00	6.6432	6.6432	-2.0e-13
12.00	5.8554	5.8554	-1.5e-13
24.00	4.8260	4.8260	-1.3e-13
48.00	3.3381	3.3381	-2.5e-13
96.00	1.5971	1.5971	-4.6e-13
168.00	0.5285	0.5285	-8.0e-13

# Both sides use the SAME parameters, so this is pure integration error, not a
# cohort statistic; a tight all() bound is the correct assertion here.
stopifnot(max(abs(cf_cmp$pct_diff)) < 1e-6)

The terminal (beta) half-life implied by these parameters is 45.1 h at CrCL 30 mL/min. The authors are explicit that this is not teicoplanin’s true terminal half-life of 83-163 h: the sampling window was too short to resolve the third exponential phase, so the model describes the alpha and beta phases only (Discussion, first paragraph, and “limitations”).

Typical-value profiles and PKNCA validation

Non-compartmental analysis of a single 400 mg infusion in a typical subject recovers two quantities the paper prints independently of the packaged parameters:

• Clearance. The printed equation CL = 0.711 * (CrCL / 17)^0.198 gives a reference value at every renal stratum. Recovering it from Dose / AUC0-inf after solving the ODE system checks that the micro-constants, the infusion input and the Cc <- central / vc observation are wired correctly.
• Steady-state volume. The Discussion states the model’s volume of distribution as 46.6 L (“The distribution in the present model is smaller than previous models (46.6 L VS 60.5-121.1 L)”). That is Vc + Vp, and it adjudicates the Vp rounding: 11.3 + 35.3 = 46.6, whereas 11.3 + 35.2 = 46.5.

strata_crcl <- c(5, 17, 20, 45, 75, 105)
nca_grid <- sort(unique(c(seq(0, 4, by = 0.25), seq(4, 24, by = 1),
                          seq(24, 336, by = 6))))

typ_ev <- do.call(rbind, lapply(seq_along(strata_crcl), function(i) {
  rbind(
    data.frame(id = i, time = 0, amt = dose_mg, evid = 1L, dur = t_inf,
               cmt = "central", CRCL = strata_crcl[i]),
    data.frame(id = i, time = nca_grid, amt = NA_real_, evid = 0L,
               dur = NA_real_, cmt = "central", CRCL = strata_crcl[i])
  )
}))
typ_ev$stratum <- sprintf("CrCL %g mL/min", typ_ev$CRCL)

typ_sim <- rxode2::rxSolve(
  rxode2::zeroRe(mod), typ_ev, keep = c("stratum"), returnType = "data.frame"
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc'
#> Warning: multi-subject simulation without without 'omega'
stopifnot(all(typ_sim$Cc >= 0))

typ_sim |>
  dplyr::filter(time > 0) |>
  ggplot(aes(time, Cc, colour = stratum)) +
  geom_line(linewidth = 0.7) +
  scale_x_continuous(limits = c(0, 240)) +
  scale_y_log10() +
  labs(x = "Time after a single 400 mg 1 h infusion (h)",
       y = "Teicoplanin concentration (mg/L)",
       colour = "Renal stratum",
       title = "Typical-value disposition by renal function",
       caption = paste("Only clearance carries the CrCL covariate, so the",
                       "curves separate in the terminal phase only.")) +
  theme_bw()
#> Warning: Removed 96 rows containing missing values or values outside the scale range
#> (`geom_line()`).

nca_conc <- typ_sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::select(id, time, Cc, stratum)

# Guarantee a time = 0 record so PKNCA anchors AUC0-* without warning. For an
# IV infusion the pre-dose concentration is 0.
nca_conc <- dplyr::bind_rows(
  nca_conc,
  nca_conc |> dplyr::distinct(id, stratum) |> dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::distinct(id, stratum, time, .keep_all = TRUE) |>
  dplyr::arrange(id, stratum, time)

nca_dose <- typ_ev |>
  dplyr::filter(evid == 1) |>
  dplyr::select(id, time, amt, dur, stratum)

conc_obj <- PKNCA::PKNCAconc(nca_conc, Cc ~ time | stratum + id,
                             concu = "mg/L", timeu = "h")
dose_obj <- PKNCA::PKNCAdose(nca_dose, amt ~ time | stratum + id,
                             doseu = "mg", duration = "dur")

nca_intervals <- data.frame(
  start = 0, end = Inf,
  cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE, aucpext.obs = TRUE,
  half.life = TRUE, cl.obs = TRUE, vss.iv.obs = TRUE
)
nca_res <- PKNCA::pk.nca(
  PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = nca_intervals)
)

# The 336 h window must capture essentially all of AUC0-inf for cl.obs and
# vss.iv.obs to be meaningful.
pext <- as.data.frame(nca_res$result)
stopifnot(max(pext$PPORRES[pext$PPTESTCD == "aucpext.obs"]) < 5)

published_nca <- tibble::tibble(
  stratum = sprintf("CrCL %g mL/min", strata_crcl),
  # Zhou 2025 Results equation block: CL = 0.711 * (CrCL / 17)^0.198.
  cl.obs = 0.711 * (strata_crcl / 17)^0.198,
  # Zhou 2025 Discussion: "The distribution in the present model is smaller
  # than previous models (46.6 L VS 60.5-121.1 L)". No covariate acts on either
  # volume, so the value is the same in every stratum.
  vss.iv.obs = 46.6
)

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = nca_res,
  reference = published_nca,
  by = "stratum",
  units = c(cl.obs = "L/h", vss.iv.obs = "L"),
  tolerance_pct = 20
)

knitr::kable(
  cmp,
  caption = paste("Simulated vs published typical-value NCA.",
                  "* differs from the reference by >20%."),
  align = c("l", "l", "r", "r", "r")
)

Simulated vs published typical-value NCA. * differs from the reference by >20%.

NCA parameter	stratum	Reference	Simulated	% diff
CL/F (L/h)	CrCL 5 mL/min	0.558	0.558	-0.0%
CL/F (L/h)	CrCL 17 mL/min	0.711	0.711	-0.0%
CL/F (L/h)	CrCL 20 mL/min	0.734	0.734	-0.0%
CL/F (L/h)	CrCL 45 mL/min	0.862	0.862	-0.0%
CL/F (L/h)	CrCL 75 mL/min	0.954	0.954	-0.0%
CL/F (L/h)	CrCL 105 mL/min	1.02	1.02	-0.0%
Vss (IV) (L)	CrCL 5 mL/min	46.6	46.6	-0.1%
Vss (IV) (L)	CrCL 17 mL/min	46.6	46.6	-0.0%
Vss (IV) (L)	CrCL 20 mL/min	46.6	46.6	-0.0%
Vss (IV) (L)	CrCL 45 mL/min	46.6	46.6	-0.0%
Vss (IV) (L)	CrCL 75 mL/min	46.6	46.6	-0.0%
Vss (IV) (L)	CrCL 105 mL/min	46.6	46.6	-0.0%

cmp_num <- as.data.frame(nca_res$result) |>
  dplyr::filter(PPTESTCD %in% c("cl.obs", "vss.iv.obs")) |>
  dplyr::left_join(
    tidyr::pivot_longer(published_nca, -stratum,
                        names_to = "PPTESTCD", values_to = "ref"),
    by = c("stratum", "PPTESTCD")
  ) |>
  dplyr::mutate(pct = 100 * (PPORRES - ref) / ref)

stopifnot(nrow(cmp_num) == 2L * length(strata_crcl), !anyNA(cmp_num$pct))
# Deterministic (typical-value) quantities: no cohort draw is involved, so the
# bound is tight. Realised max |pct| is well under 0.5% -- 1% leaves room for
# the AUC extrapolation and quadrature while still going red on any parameter,
# unit or wiring error.
stopifnot(max(abs(cmp_num$pct)) < 1)

ncaComparisonTable() prints PKNCA’s cl.obs under its generic label “CL/F”; teicoplanin is given intravenously here, so there is no bioavailability term and the quantity is plain clearance.

The model half-lives recovered by PKNCA (36, 38, 42, 48, 50, 62 h across the strata) are the beta half-lives discussed above, not the 83-163 h terminal values quoted from the literature in the paper’s Introduction.

Virtual cohort on the study regimen

The trial regimen was 400 mg q12h x 3 followed by 400 mg q24h, each infused over 1 h. Creatinine clearance is drawn to match the observed baseline distribution (median 9 mL/min, IQR 6.6-13.8) and held constant per subject; see Assumptions for why that is a simplification of the paper’s time-varying covariate.

set.seed(20250423)
rxode2::rxSetSeed(20250423)

n_sub <- 200L
# Log-normal matched to the Table 1 development-set baseline CrCL: median 9,
# IQR 6.6-13.8 mL/min -> sdlog = log(13.8 / 6.6) / (2 * qnorm(0.75)).
crcl_sdlog <- log(13.8 / 6.6) / (2 * stats::qnorm(0.75))
cohort_crcl <- stats::rlnorm(n_sub, meanlog = log(9), sdlog = crcl_sdlog)

study_dose_t <- c(0, 12, 24, seq(48, 336, by = 24))
study_obs_t <- sort(unique(c(seq(0, 60, by = 1), seq(60, 336, by = 3),
                             study_dose_t - 1e-6)))
study_obs_t <- study_obs_t[study_obs_t >= 0]

cohort_ev <- do.call(rbind, lapply(seq_len(n_sub), function(i) {
  rbind(
    data.frame(id = i, time = study_dose_t, amt = 400, evid = 1L, dur = 1,
               cmt = "central", CRCL = cohort_crcl[i]),
    data.frame(id = i, time = study_obs_t, amt = NA_real_, evid = 0L,
               dur = NA_real_, cmt = "central", CRCL = cohort_crcl[i])
  )
}))
stopifnot(!anyDuplicated(unique(cohort_ev[, c("id", "time", "evid")])))

cohort_sim <- rxode2::rxSolve(mod, cohort_ev, keep = c("CRCL"),
                              returnType = "data.frame")
#> ℹ parameter labels from comments will be replaced by 'label()'

# Comparable to Figure 2 of Zhou 2025 (prediction-corrected VPC of the final
# model): the 5th, 50th and 95th percentiles of the model's own predictions
# over the observed sampling window.
cohort_sim |>
  dplyr::filter(time <= 240) |>
  dplyr::group_by(time) |>
  dplyr::summarise(
    Q05 = quantile(Cc, 0.05),
    Q50 = quantile(Cc, 0.50),
    Q95 = quantile(Cc, 0.95),
    .groups = "drop"
  ) |>
  ggplot(aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line(linewidth = 0.7) +
  geom_hline(yintercept = c(15, 40), linetype = c("dashed", "dotted")) +
  labs(x = "Time after first dose (h)", y = "Teicoplanin concentration (mg/L)",
       title = "Study regimen: 400 mg q12h x 3, then 400 mg q24h",
       caption = paste("Median with 5th-95th percentile band, 200 virtual",
                       "subjects. Dashed line: 15 mg/L efficacy target.",
                       "Dotted line: 40 mg/L toxicity threshold.")) +
  theme_bw()

The paper’s central clinical message is visible directly: on the empirical 400 mg regimen, the median trough in this severely renally impaired cohort sits close to the 15 mg/L target only after several days, which is why the authors conclude that higher-than-standard loading and maintenance doses are needed.

Reproducing the published Monte Carlo table

Table 3 of Zhou 2025 reports the probability of target attainment for 40 dose regimens across five renal strata, against a 15 mg/L efficacy trough and a 40 mg/L toxicity threshold, at 72 h and 168 h after the first dose. It is the paper’s only large block of printed numbers derived from the model, and it therefore exercises every packaged parameter at once: both structural clearances, both volumes, the CrCL power term and both between-subject variances.

Two reconstruction choices are needed, and the table itself pins both.

• When maintenance dosing starts. Taking it to begin 24 h after the final loading dose puts the doses at 0, 12, 24 then 48, 72, … for a three-dose load and at 0, 12, 24, 36, 48 then 72, 96, … for a five-dose load. The table confirms this: with a five-dose load the first maintenance dose lands at 72 h, so it cannot influence Cmin,72h – and indeed the paper reports identical Cmin,72h attainment for 400 mg x 5 with 200 mg QD and with 400 mg QD (both 91.8%), for 600 mg x 5 with 200 and 400 mg QD (both 99.3%), and for 800 mg x 5 with 200 and 400 mg QD (99 and 99.0%). With a three-dose load the maintenance dose at 48 h does fall inside the window, and there the same comparison separates (600 mg x 3 gives 84.1, 92.2 and 96.4% with 200, 400 and 600 mg QD).
• Which creatinine clearance represents a stratum. The paper resampled its enrolled subjects, so the within-stratum distribution is not recoverable. The midpoint of each stratum is used here (5, 20, 45, 75 and 105 mL/min). The exponent 0.198 is weak enough that this matters little: across the 90-120 stratum the clearance multiplier moves by only 6%.

table3 <- tibble::tribble(
  ~stratum,      ~crcl, ~load, ~n_load, ~maint, ~p72_15, ~p168_15, ~p72_40, ~p168_40,
  "CrCL <10",        5,   400,      5L,    200,    91.8,     56.7,     0.0,      0.0,
  "CrCL <10",        5,   400,      3L,    400,    70.7,     83.9,     0.0,      1.6,
  "CrCL <10",        5,   400,      5L,    400,    91.8,     86.8,     0.0,      4.8,
  "CrCL <10",        5,   600,      3L,    200,    84.1,     50.5,     0.0,      0.0,
  "CrCL <10",        5,   600,      5L,    200,    99.3,     71.9,    18.5,      2.4,
  "CrCL <10",        5,   600,      3L,    400,    92.2,     87.1,     0.0,      5.5,
  "CrCL <10",        5,   600,      5L,    400,    99.3,     90.2,    18.5,     15.2,
  "CrCL <10",        5,   600,      3L,    600,    96.4,     97.2,     0.6,     28.5,
  "CrCL 10-30",     20,   600,      3L,    400,    76.5,     66.5,     0.0,      0.8,
  "CrCL 10-30",     20,   600,      5L,    400,    96.2,     75.3,     5.2,      4.0,
  "CrCL 10-30",     20,   600,      3L,    600,    86.8,     89.0,     0.0,      9.4,
  "CrCL 10-30",     20,   600,      5L,    600,    95.8,     89.8,     5.2,     15.2,
  "CrCL 10-30",     20,   800,      5L,    200,    99.0,     57.6,    40.1,      1.9,
  "CrCL 10-30",     20,   800,      3L,    400,    89.2,     71.6,     1.0,      2.0,
  "CrCL 10-30",     20,   800,      5L,    400,    99.0,     79.6,    40.1,      9.2,
  "CrCL 10-30",     20,   800,      3L,    600,    92.9,     88.8,     3.0,     13.2,
  "CrCL 30-60",     45,   600,      3L,    600,    73.5,     76.6,     0.0,      3.2,
  "CrCL 30-60",     45,   600,      5L,    600,    89.4,     78.9,     1.7,      6.6,
  "CrCL 30-60",     45,   800,      5L,    400,    97.9,     65.4,    23.1,      3.5,
  "CrCL 30-60",     45,   800,      3L,    600,    84.8,     77.4,     0.7,      4.7,
  "CrCL 30-60",     45,   800,      5L,    600,    96.8,     80.1,    22.5,     11.7,
  "CrCL 30-60",     45,   800,      3L,    800,    90.6,     90.0,     2.3,     18.0,
  "CrCL 30-60",     45,  1000,      3L,    600,    91.7,     80.4,     6.7,      8.0,
  "CrCL 30-60",     45,  1000,      3L,    800,    94.6,     89.9,    11.3,     20.4,
  "CrCL 60-90",     75,   600,      3L,    600,    64.3,     67.8,     0.0,      2.3,
  "CrCL 60-90",     75,   600,      5L,    600,    86.8,     72.4,     0.9,      4.1,
  "CrCL 60-90",     75,   800,      5L,    600,    94.7,     73.2,    15.1,      7.2,
  "CrCL 60-90",     75,   800,      3L,    800,    82.9,     82.3,     1.3,     10.7,
  "CrCL 60-90",     75,   800,      5L,    800,    94.9,     84.5,    16.6,     17.7,
  "CrCL 60-90",     75,  1000,      3L,    600,    86.2,     70.3,     4.1,      5.0,
  "CrCL 60-90",     75,  1000,      3L,    800,    90.5,     85.0,     6.3,     12.7,
  "CrCL 60-90",     75,  1000,      3L,   1000,    93.7,     92.6,    12.7,     27.9,
  "CrCL 90-120",   105,   800,      5L,    600,    91.6,     64.6,    10.0,      3.8,
  "CrCL 90-120",   105,   800,      3L,    800,    79.6,     79.0,     0.7,      7.6,
  "CrCL 90-120",   105,   800,      5L,    800,    91.8,     79.8,     9.5,     10.2,
  "CrCL 90-120",   105,  1000,      3L,    600,    81.9,     61.8,     2.0,      2.5,
  "CrCL 90-120",   105,  1000,      3L,    800,    85.7,     79.2,     4.5,      9.7,
  "CrCL 90-120",   105,  1000,      5L,    800,    97.6,     83.2,    31.2,     13.8,
  "CrCL 90-120",   105,  1000,      3L,   1000,    90.3,     88.8,     9.2,     21.2,
  "CrCL 90-120",   105,  1000,      5L,   1000,    97.1,     89.0,    32.4,     24.6
)
stopifnot(nrow(table3) == 40L)

regimen_label <- function(load, n_load, maint) {
  sprintf("%s mg q12h x %d, %s mg QD",
          format(load, big.mark = ",", trim = TRUE), n_load,
          format(maint, big.mark = ",", trim = TRUE))
}

build_pta_events <- function(load, n_load, maint, crcl, n, id_offset) {
  load_t <- seq(0, by = 12, length.out = n_load)
  maint_t <- seq(load_t[n_load] + 24, 192, by = 24)
  dose_t <- c(load_t, maint_t)
  dose_a <- c(rep(load, n_load), rep(maint, length(maint_t)))
  do.call(rbind, lapply(seq_len(n), function(i) {
    rbind(
      data.frame(id = id_offset + i, time = dose_t, amt = dose_a, evid = 1L,
                 dur = 1, cmt = "central", CRCL = crcl),
      data.frame(id = id_offset + i, time = c(72, 168), amt = NA_real_,
                 evid = 0L, dur = NA_real_, cmt = "central", CRCL = crcl)
    )
  }))
}

n_pta <- 200L
rxode2::rxSetSeed(97531)
pta_sim <- lapply(seq_len(nrow(table3)), function(k) {
  ev <- build_pta_events(table3$load[k], table3$n_load[k], table3$maint[k],
                         table3$crcl[k], n_pta, (k - 1L) * n_pta)
  s <- rxode2::rxSolve(mod, ev, returnType = "data.frame")
  c72 <- s$Cc[s$time == 72]
  c168 <- s$Cc[s$time == 168]
  stopifnot(length(c72) == n_pta, length(c168) == n_pta)
  data.frame(
    row = k,
    s72_15 = 100 * mean(c72 > 15), s168_15 = 100 * mean(c168 > 15),
    s72_40 = 100 * mean(c72 > 40), s168_40 = 100 * mean(c168 > 40)
  )
}) |>
  dplyr::bind_rows()

pta <- dplyr::bind_cols(table3, dplyr::select(pta_sim, -row)) |>
  dplyr::mutate(regimen = regimen_label(load, n_load, maint))

pta |>
  dplyr::mutate(
    `Cmin,72h > 15` = sprintf("%.1f / %.1f", p72_15, s72_15),
    `Cmin,168h > 15` = sprintf("%.1f / %.1f", p168_15, s168_15),
    `Cmin,72h > 40` = sprintf("%.1f / %.1f", p72_40, s72_40),
    `Cmin,168h > 40` = sprintf("%.1f / %.1f", p168_40, s168_40)
  ) |>
  dplyr::select(stratum, regimen, `Cmin,72h > 15`, `Cmin,168h > 15`,
                `Cmin,72h > 40`, `Cmin,168h > 40`) |>
  dplyr::rename("Stratum (mL/min)" = stratum, "Dose regimen" = regimen) |>
  knitr::kable(
    caption = paste("Zhou 2025 Table 3 reproduced. Each cell is",
                    "published / simulated PTA in percent."),
    align = c("l", "l", "r", "r", "r", "r")
  )

Zhou 2025 Table 3 reproduced. Each cell is published / simulated PTA in percent.

Stratum (mL/min)	Dose regimen	Cmin,72h > 15	Cmin,168h > 15	Cmin,72h > 40	Cmin,168h > 40
CrCL <10	400 mg q12h x 5, 200 mg QD	91.8 / 92.0	56.7 / 58.0	0.0 / 0.0	0.0 / 0.5
CrCL <10	400 mg q12h x 3, 400 mg QD	70.7 / 65.5	83.9 / 79.0	0.0 / 0.0	1.6 / 1.0
CrCL <10	400 mg q12h x 5, 400 mg QD	91.8 / 90.0	86.8 / 87.0	0.0 / 0.0	4.8 / 5.5
CrCL <10	600 mg q12h x 3, 200 mg QD	84.1 / 82.0	50.5 / 44.0	0.0 / 0.0	0.0 / 0.0
CrCL <10	600 mg q12h x 5, 200 mg QD	99.3 / 99.5	71.9 / 64.5	18.5 / 19.5	2.4 / 1.5
CrCL <10	600 mg q12h x 3, 400 mg QD	92.2 / 93.5	87.1 / 88.5	0.0 / 0.5	5.5 / 5.5
CrCL <10	600 mg q12h x 5, 400 mg QD	99.3 / 99.5	90.2 / 89.5	18.5 / 23.5	15.2 / 20.5
CrCL <10	600 mg q12h x 3, 600 mg QD	96.4 / 95.0	97.2 / 97.5	0.6 / 0.5	28.5 / 19.5
CrCL 10-30	600 mg q12h x 3, 400 mg QD	76.5 / 72.0	66.5 / 62.5	0.0 / 0.0	0.8 / 0.5
CrCL 10-30	600 mg q12h x 5, 400 mg QD	96.2 / 97.0	75.3 / 71.5	5.2 / 6.5	4.0 / 5.0
CrCL 10-30	600 mg q12h x 3, 600 mg QD	86.8 / 78.5	89.0 / 84.0	0.0 / 0.0	9.4 / 10.0
CrCL 10-30	600 mg q12h x 5, 600 mg QD	95.8 / 96.0	89.8 / 87.5	5.2 / 4.5	15.2 / 13.0
CrCL 10-30	800 mg q12h x 5, 200 mg QD	99.0 / 98.5	57.6 / 53.0	40.1 / 36.5	1.9 / 2.5
CrCL 10-30	800 mg q12h x 3, 400 mg QD	89.2 / 85.0	71.6 / 72.0	1.0 / 1.0	2.0 / 3.0
CrCL 10-30	800 mg q12h x 5, 400 mg QD	99.0 / 100.0	79.6 / 72.5	40.1 / 35.5	9.2 / 7.0
CrCL 10-30	800 mg q12h x 3, 600 mg QD	92.9 / 90.0	88.8 / 85.5	3.0 / 3.0	13.2 / 13.0
CrCL 30-60	600 mg q12h x 3, 600 mg QD	73.5 / 74.5	76.6 / 79.0	0.0 / 0.0	3.2 / 5.5
CrCL 30-60	600 mg q12h x 5, 600 mg QD	89.4 / 89.5	78.9 / 75.5	1.7 / 4.0	6.6 / 9.0
CrCL 30-60	800 mg q12h x 5, 400 mg QD	97.9 / 98.0	65.4 / 66.0	23.1 / 22.5	3.5 / 4.0
CrCL 30-60	800 mg q12h x 3, 600 mg QD	84.8 / 84.5	77.4 / 78.5	0.7 / 1.0	4.7 / 9.0
CrCL 30-60	800 mg q12h x 5, 600 mg QD	96.8 / 95.5	80.1 / 81.0	22.5 / 27.0	11.7 / 14.0
CrCL 30-60	800 mg q12h x 3, 800 mg QD	90.6 / 90.5	90.0 / 89.5	2.3 / 4.0	18.0 / 20.0
CrCL 30-60	1,000 mg q12h x 3, 600 mg QD	91.7 / 92.5	80.4 / 81.0	6.7 / 11.0	8.0 / 12.5
CrCL 30-60	1,000 mg q12h x 3, 800 mg QD	94.6 / 92.5	89.9 / 87.0	11.3 / 15.0	20.4 / 25.0
CrCL 60-90	600 mg q12h x 3, 600 mg QD	64.3 / 64.5	67.8 / 67.0	0.0 / 0.0	2.3 / 3.5
CrCL 60-90	600 mg q12h x 5, 600 mg QD	86.8 / 82.5	72.4 / 68.5	0.9 / 2.5	4.1 / 7.5
CrCL 60-90	800 mg q12h x 5, 600 mg QD	94.7 / 97.0	73.2 / 69.0	15.1 / 16.5	7.2 / 10.5
CrCL 60-90	800 mg q12h x 3, 800 mg QD	82.9 / 82.5	82.3 / 82.0	1.3 / 2.5	10.7 / 15.5
CrCL 60-90	800 mg q12h x 5, 800 mg QD	94.9 / 92.0	84.5 / 81.5	16.6 / 13.5	17.7 / 15.5
CrCL 60-90	1,000 mg q12h x 3, 600 mg QD	86.2 / 81.0	70.3 / 71.5	4.1 / 5.5	5.0 / 6.0
CrCL 60-90	1,000 mg q12h x 3, 800 mg QD	90.5 / 90.0	85.0 / 82.5	6.3 / 10.5	12.7 / 16.0
CrCL 60-90	1,000 mg q12h x 3, 1,000 mg QD	93.7 / 89.0	92.6 / 86.5	12.7 / 17.5	27.9 / 28.5
CrCL 90-120	800 mg q12h x 5, 600 mg QD	91.6 / 89.5	64.6 / 67.0	10.0 / 10.5	3.8 / 5.0
CrCL 90-120	800 mg q12h x 3, 800 mg QD	79.6 / 74.0	79.0 / 73.5	0.7 / 0.5	7.6 / 8.5
CrCL 90-120	800 mg q12h x 5, 800 mg QD	91.8 / 87.0	79.8 / 73.0	9.5 / 10.5	10.2 / 12.0
CrCL 90-120	1,000 mg q12h x 3, 600 mg QD	81.9 / 82.5	61.8 / 65.5	2.0 / 2.5	2.5 / 4.0
CrCL 90-120	1,000 mg q12h x 3, 800 mg QD	85.7 / 83.5	79.2 / 77.0	4.5 / 6.0	9.7 / 11.5
CrCL 90-120	1,000 mg q12h x 5, 800 mg QD	97.6 / 95.0	83.2 / 85.0	31.2 / 36.0	13.8 / 21.0
CrCL 90-120	1,000 mg q12h x 3, 1,000 mg QD	90.3 / 88.5	88.8 / 87.5	9.2 / 12.0	21.2 / 22.0
CrCL 90-120	1,000 mg q12h x 5, 1,000 mg QD	97.1 / 94.0	89.0 / 84.5	32.4 / 35.0	24.6 / 28.5

pta_diff <- c(
  pta$s72_15 - pta$p72_15, pta$s168_15 - pta$p168_15,
  pta$s72_40 - pta$p72_40, pta$s168_40 - pta$p168_40
)
stopifnot(length(pta_diff) == 160L, !anyNA(pta_diff))

pta_summary <- data.frame(
  Statistic = c("cells compared", "median |difference| (pp)",
                "90th percentile |difference| (pp)", "maximum |difference| (pp)"),
  Value = c(length(pta_diff), median(abs(pta_diff)),
            unname(quantile(abs(pta_diff), 0.9)), max(abs(pta_diff)))
)
knitr::kable(pta_summary, digits = 2,
             caption = "Agreement with the published Table 3.")

Agreement with the published Table 3.

Statistic	Value
cells compared	160.00
median |difference| (pp)	1.55
90th percentile |difference| (pp)	4.81
maximum |difference| (pp)	9.00

# These are cohort statistics, so the bounds are robust summaries, not
# per-cell limits. Each simulated cell is a binomial proportion from 200
# subjects (about 2.8 pp standard error near 80%) against a published cell from
# 1,000 (about 1.3 pp), so an individual cell can legitimately be several
# points off; assert the CENTRE and a robust upper quantile instead. At 1,000
# subjects per arm the realised values are 1.5 / 3.7 / 4.5 pp -- the bounds
# below sit outside the 200-subject sampling spread and still go red on any
# parameter that moves the distribution, which is what a mis-read variance,
# a dropped covariate term or a wrong reference constant would do.
stopifnot(
  median(abs(pta_diff)) < 6,
  quantile(abs(pta_diff), 0.9) < 12
)

The AUC24/MIC target

The paper’s second efficacy criterion is AUC24h/MIC >= 610.4, evaluated over two 24 h windows: 72-96 h, which the authors describe as loading-dose driven (Figure 6), and 168-192 h, which is maintenance driven (Figure 7). Two claims from the Results are checked below in both windows.

1. “All of the selected regimens achieved the PTA target when MIC = 0.5 mg/L.”
2. At MIC = 1.0 mg/L the recommended regimens are 600 mg q12h x 5 then 400 mg QD for CrCL <10; 800 mg q12h x 3 then 600 mg QD for 10-30; 800 mg q12h x 3 then 800 mg QD for 30-60; 1,000 mg q12h x 3 then 800 mg QD for 60-90; and 1,000 mg q12h x 3 then 1,000 mg QD for 90-120.

Unlike Table 3, both of these are read off graphical figures, so only the 80% threshold line the authors drew on them is available as a reference value.

auc_reg <- tibble::tribble(
  ~stratum,      ~crcl, ~load, ~n_load, ~maint,
  "CrCL <10",        5,   600,      5L,    400,
  "CrCL 10-30",     20,   800,      3L,    600,
  "CrCL 30-60",     45,   800,      3L,    800,
  "CrCL 60-90",     75,  1000,      3L,    800,
  "CrCL 90-120",   105,  1000,      3L,   1000
)
auc_grid <- sort(unique(c(seq(72, 96, by = 0.5), seq(168, 192, by = 0.5))))

rxode2::rxSetSeed(24680)
auc_res <- lapply(seq_len(nrow(auc_reg)), function(k) {
  load_t <- seq(0, by = 12, length.out = auc_reg$n_load[k])
  maint_t <- seq(load_t[auc_reg$n_load[k]] + 24, 192, by = 24)
  dose_t <- c(load_t, maint_t)
  dose_a <- c(rep(auc_reg$load[k], auc_reg$n_load[k]),
              rep(auc_reg$maint[k], length(maint_t)))
  ev <- do.call(rbind, lapply(seq_len(n_pta), function(i) {
    rbind(
      data.frame(id = i, time = dose_t, amt = dose_a, evid = 1L, dur = 1,
                 cmt = "central", CRCL = auc_reg$crcl[k]),
      data.frame(id = i, time = auc_grid, amt = NA_real_, evid = 0L,
                 dur = NA_real_, cmt = "central", CRCL = auc_reg$crcl[k])
    )
  }))
  s <- rxode2::rxSolve(mod, ev, returnType = "data.frame")
  s <- s[!is.na(s$Cc), c("id", "time", "Cc")]
  s$stratum <- auc_reg$stratum[k]
  s
}) |>
  dplyr::bind_rows()

auc_conc <- PKNCA::PKNCAconc(auc_res, Cc ~ time | stratum + id,
                             concu = "mg/L", timeu = "h")
auc_out <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  auc_conc,
  intervals = data.frame(start = c(72, 168), end = c(96, 192), auclast = TRUE)
))
#> No dose information provided, calculations requiring dose will return NA.

auc24 <- as.data.frame(auc_out$result) |>
  dplyr::filter(PPTESTCD == "auclast") |>
  dplyr::mutate(window = ifelse(start == 72, "72-96 h", "168-192 h")) |>
  dplyr::group_by(stratum, window) |>
  dplyr::summarise(
    median_auc = median(PPORRES),
    pta_mic05 = 100 * mean(PPORRES / 0.5 >= 610.4),
    pta_mic10 = 100 * mean(PPORRES / 1.0 >= 610.4),
    n = dplyr::n(),
    .groups = "drop"
  )
stopifnot(all(auc24$n == n_pta), nrow(auc24) == 2L * nrow(auc_reg))

auc24 |>
  dplyr::left_join(
    auc_reg |> dplyr::mutate(regimen = regimen_label(load, n_load, maint)) |>
      dplyr::select(stratum, regimen),
    by = "stratum"
  ) |>
  dplyr::select(stratum, regimen, window, median_auc, pta_mic05, pta_mic10) |>
  dplyr::rename(
    "Stratum (mL/min)" = stratum, "Dose regimen" = regimen,
    "Window" = window, "Median AUC24 (mg*h/L)" = median_auc,
    "PTA (%) at MIC 0.5" = pta_mic05, "PTA (%) at MIC 1.0" = pta_mic10
  ) |>
  knitr::kable(digits = 1,
               caption = paste("Attainment of AUC24h/MIC >= 610.4 for the",
                               "regimens the paper recommends at",
                               "MIC = 1.0 mg/L."))

Attainment of AUC24h/MIC >= 610.4 for the regimens the paper recommends at MIC = 1.0 mg/L.

Stratum (mL/min)	Dose regimen	Window	Median AUC24 (mg*h/L)	PTA (%) at MIC 0.5	PTA (%) at MIC 1.0
CrCL 10-30	800 mg q12h x 3, 600 mg QD	168-192 h	828.9	100.0	80.0
CrCL 10-30	800 mg q12h x 3, 600 mg QD	72-96 h	820.2	100.0	85.0
CrCL 30-60	800 mg q12h x 3, 800 mg QD	168-192 h	912.5	100.0	88.0
CrCL 30-60	800 mg q12h x 3, 800 mg QD	72-96 h	859.4	100.0	88.5
CrCL 60-90	1,000 mg q12h x 3, 800 mg QD	168-192 h	820.1	99.5	80.5
CrCL 60-90	1,000 mg q12h x 3, 800 mg QD	72-96 h	847.7	99.5	84.5
CrCL 90-120	1,000 mg q12h x 3, 1,000 mg QD	168-192 h	1002.5	100.0	89.0
CrCL 90-120	1,000 mg q12h x 3, 1,000 mg QD	72-96 h	964.7	100.0	89.0
CrCL <10	600 mg q12h x 5, 400 mg QD	168-192 h	800.0	98.5	76.5
CrCL <10	600 mg q12h x 5, 400 mg QD	72-96 h	902.9	100.0	92.5

mic10_load <- auc24$pta_mic10[auc24$window == "72-96 h"]
mic10_maint <- auc24$pta_mic10[auc24$window == "168-192 h"]

# Claim 1 reproduces in both windows with a very wide margin: the realised
# minimum at 1,000 subjects per arm is 98.6%. The bound is one-sided and comes
# from the paper's own 80% criterion, not from this run.
stopifnot(all(auc24$pta_mic05 >= 90))

# Claim 2 reproduces in the loading-dose window. Realised at 1,000 subjects per
# arm: 81.1-90.8%. The 70 bound leaves room for the 200-subject binomial
# standard error (about 2.8 pp near 80%) while still going red on any change
# that moves the exposure distribution materially.
stopifnot(all(mic10_load >= 70))

# Claim 2 in the MAINTENANCE window is a DOCUMENTED DEVIATION -- three of five
# strata land a few points under 80% (see the prose below). It is deliberately
# NOT gated at 80; the assertion below only keeps it in a plausible band so a
# gross regression still fails.
stopifnot(median(mic10_maint) >= 65)

Claim 1 reproduces with a wide margin in both windows: every recommended regimen exceeds 98% attainment at MIC 0.5 mg/L.

Claim 2 reproduces in the loading-dose window (72-96 h), where all five recommended regimens clear 80%. In the maintenance window (168-192 h) three of the five land a few percentage points short. At 1,000 subjects per arm – enough to remove sampling noise as an explanation – the realised values are 73.8% (CrCL <10), 76.8% (10-30), 85.1% (30-60), 78.9% (60-90) and 88.2% (90-120).

This is recorded as a deviation rather than corrected. The most likely explanation is the stratum-representative creatinine clearance: this window’s attainment depends on maintenance-phase clearance alone, and a representative CrCL at the lower end of each band instead of its midpoint raises attainment in every stratum (at CrCL 2 rather than 5 mL/min, the <10 stratum’s median AUC168-192 rises by about a fifth and its attainment clears 80%). The paper resampled its enrolled subjects rather than using a single value per stratum, and it reports these results only as figures, so neither the distribution nor the exact plotted values can be recovered. Notably, the same representative values reproduce the printed Table 3 trough attainment closely (median absolute difference under 2 percentage points), so the disagreement is confined to the figure-derived claim.

Assumptions, deviations and errata

• “Proportional model” versus the printed power equation. The Results paragraph introducing the covariate says real-time CrCL was included in CL “using a proportional model”, but the equation block printed three lines later on the same page is unambiguously a power model, CL (L/h) = 0.711 x (CrCL / 17)^0.198 x e^eta1. The equation is used. Two further pieces of evidence support it: Table 2 labels the estimate CrCLCL, “the influence coefficient of CrCL on CL”, which is exponent-like language; and the linear reading is arithmetically impossible in this cohort. A centred proportional form 0.711 * (1 + 0.198 * (CrCL - 17)) turns negative below CrCL = 17 - 1/0.198 = 11.9 mL/min, and the development set’s median baseline CrCL is 9 mL/min – more than half the modelled subjects would carry a negative clearance, and the ODE system does not solve at all. The Table 3 reproduction above is the positive confirmation.
• Vp is printed as both 35.3 and 35.2 L. Table 2 and the Results equation block give 35.3; the Abstract and the sentence immediately below Table 2 give 35.2. 35.3 is used, because it is the value in both the parameter table and the equation, and because the Discussion’s total volume of 46.6 L equals 11.3 + 35.3 rather than 11.3 + 35.2.
• Between-subject variability scale. Table 2 reports the random effects under a “Random effects (%)” header as omega CL 40.1% and omega Vc 37.3%. They are taken as omega (the log-scale standard deviation) x 100, so the packaged variances are 0.401^2 and 0.373^2. The column cannot be variances: the residual row in the same block reads 14.6%, which as a variance would mean a 38% residual standard deviation – irreconcilable with the paper’s own external-validation RMSE of 8.49% and with an assay whose inter-day CV was at most 3.72%. Reading the same percentages as approximate CVs instead (omega = sqrt(log(1 + CV^2))) would give 0.386 and 0.361; the difference is immaterial for simulation and is inside the reproduction tolerance of the Table 3 check.
• Residual error form. Phoenix NLME writes its proportional error on the log scale as ln Y = ln F + eps1 (the second of the three candidate forms printed in Methods), which is formally exponential. The packaged model uses Cc ~ prop(propSd) with propSd = 0.146, matching the paper’s own “proportional residual” label; at that magnitude the two forms differ by well under 1% in the simulated concentration distribution.
• CrCL is held constant per subject in the simulations here. The published model uses real-time creatinine clearance, which rises substantially during a teicoplanin course as the graft recovers, but the paper publishes no trajectory for it and the Monte Carlo simulations in Table 3 are themselves run at fixed renal strata. Every simulation in this vignette therefore uses a time-constant CRCL. A user with longitudinal creatinine can supply a time-varying CRCL column directly; the model needs no change.
• Stratum representative CrCL. The paper resampled its enrolled subjects within each renal stratum, so the within-stratum covariate distribution is not recoverable. Stratum midpoints (5, 20, 45, 75, 105 mL/min) are used. The weak 0.198 exponent makes the Table 3 reproduction insensitive to this choice: at 1,000 subjects per arm, a uniform draw across the <10 stratum and a fixed 5 mL/min give median absolute Table 3 differences of 1.4 and 1.5 percentage points respectively.
• Known deviation: the figure-derived AUC24/MIC recommendations at MIC 1.0 mg/L in the 168-192 h window. Three of the five recommended regimens reach 74-79% rather than the 80% the authors’ Figure 7 threshold implies. The printed Table 3 trough attainment for the same strata reproduces to within about 2 percentage points, and the same regimens clear 80% in the paper’s other (72-96 h) window, so this is attributed to the unrecoverable within-stratum CrCL distribution rather than to a transcription error. It is reported in the table and excluded from the assertion rather than having the bound widened until it passes.
• Maintenance-dose start time. Taken as 24 h after the final loading dose. This is not stated in the Methods but is pinned by the internal structure of Table 3, as set out in the reproduction section above.
• Baseline CrCL distribution for the virtual cohort. Table 1 reports the development-set baseline CrCL as a median with an interquartile range only. A log-normal matched to that median and IQR is used; the paper reports no distributional form.
• No supplement gap. The article’s supplement (Tables S1-S2, Figures S1-S2) was retrieved and used. It contributed the covariate-screening delta-OFV values recorded in population$screened_covariates and the infection / pathogen detail in population$disease_state; it contains no final parameter estimates beyond those in Table 2.
• Model scope. The authors state that the short sampling window prevented the model from resolving teicoplanin’s true terminal phase (literature half-life 83-163 h). This two-compartment model describes the alpha and beta phases and should not be used to project exposure far beyond the 192 h horizon the paper itself simulates.