Skip to contents

Model and source

Kong et al. 2025 pooled individual piperacillin/tazobactam data from 13 published studies – premature neonates to very elderly adults, critically ill, obese and underweight patients, and febrile neutropenic patients with haematological malignancy – into one general-purpose population PK model intended as an a priori model for Bayesian dose individualisation.

The final combined model has four compartments: a two-compartment disposition model for each drug, with linear elimination from each central compartment.

  • Volumes scale linearly with total body weight and elimination clearance with exponent 0.75 (70 kg reference). The inter-compartmental clearances use compartmental allometry, scaling with the individual peripheral volume.
  • Elimination clearance of both drugs shares a sigmoidal postmenstrual-age (PMA) maturation function (MAT50 = 54.2 weeks PMA), declines with age through a drug-specific sigmoidal function (50% at 89.1 years for piperacillin and 61.6 years for tazobactam, shared shape factor), and carries a shared exponential serum-creatinine (SCR) effect centred on a PMA-standardised SCR.
  • The random effects on the central volume, peripheral volume and inter-compartmental clearance are each shared by the two drugs; clearance has drug-specific random effects.
  • Three observables: total piperacillin, unbound piperacillin (fraction unbound 0.645) and total tazobactam, each with its own residual error. Dried blood spot concentrations are 0.368 (piperacillin) and 0.448 (tazobactam) times the plasma concentration.
  • Two study-specific effects: in the Sime 2014 haematological-malignancy study, piperacillin clearance is 1.73-fold and its peripheral volume 0.512-fold; in the Sukarnjanaset 2019 study the piperacillin unbound fraction is 1.

The authors’ companion paper on patients on intermittent haemodialysis extends this model and is packaged separately as Kong_2025_piperacillin_tazobactam.

  • Citation: Kong D, Koomen JV, Colin PJ, et al. A Pooled Pharmacokinetic Analysis for Piperacillin/Tazobactam Across Different Patient Populations: From Premature Infants to the Elderly. Clin Pharmacokinet. 2025;64(1):107-126.
  • Article: https://doi.org/10.1007/s40262-024-01460-6
mod <- readModelDb("Kong_2025b_piperacillin_tazobactam")
ui <- rxode2::rxode(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'
ui$state
#> [1] "central"         "peripheral1"     "central_taz"     "peripheral1_taz"

Population

str(ui$population)
#> List of 10
#>  $ species       : chr "human"
#>  $ n_subjects    : num 415
#>  $ n_studies     : num 13
#>  $ age_range     : chr "Postnatal age 1 day to 94 years; postmenstrual age 0.5-94.8 years"
#>  $ weight_range  : chr "0.5-200 kg"
#>  $ sex_female_pct: num 41
#>  $ disease_state : chr "Pooled individual data from 13 published piperacillin/tazobactam PK studies: critically ill adults with sepsis "| __truncated__
#>  $ dose_range    : chr "Intravenous piperacillin/tazobactam (8:1) by intermittent, extended or continuous infusion as per the component studies"
#>  $ regions       : chr "Multinational; pooled from 13 published studies (Table 1), countries not tabulated by Kong 2025"
#>  $ notes         : chr "415 patients by the Results text: 32 newborns, 127 children and adolescents, 163 adults (42 obese, 12 underweig"| __truncated__

The pooled dataset (Kong 2025 Table 1 and Results, “Data”) holds 415 patients: 32 newborns (PMA < 0.87 years), 127 children and adolescents, 163 adults (42 obese, 12 underweight), 74 elderly and 19 very elderly patients. Postnatal age ran from 1 day to 94 years, body weight from 0.5 to 200 kg and SCR from 0.17 to 5.82 mg/dL. There were 3798 piperacillin observations (2855 total plasma, 888 unbound plasma and 55 dried blood spot) and 1948 tazobactam observations (1893 total plasma and 55 dried blood spot). Patients on renal replacement therapy or extracorporeal membrane oxygenation were excluded.

Source trace

The Electronic Supplementary Material 3 (ESM3) of Kong 2025 is the complete final NONMEM control stream with its output, so every equation can be checked against the implemented code as well as against the printed Eqs. 6-25.

Model element Source location Value / form
lvc, lcl, lvp, lq Table 3 piperacillin; ESM3 TH1-TH4 10.4 L, 10.6 L/h, 11.6 L, 15.2 L/h (70 kg)
lvc_taz, lcl_taz, lvp_taz, lq_taz Table 3 tazobactam; ESM3 TH5-TH8 10.5 L, 9.58 L/h, 13.7 L, 16.8 L/h (70 kg)
e_wt_cl, e_wt_vc, e_wt_q Methods, “Weight-Based and Compartmental Allometry”; Eqs. 1, 6-9 0.75, 1, 0.75 (not estimated)
lhill_mat, ltmat50 Table 3; ESM3 TH9-TH10 gamma1 = 3.35; MAT50 = 54.2 weeks
lhill_dec, ltdec50, ltdec50_taz Table 3; ESM3 TH11-TH13 gamma2 = 1.92; DEC50 = 89.1 / 61.6 years
e_creat_cl Table 3; ESM3 TH14 0.346 dL/mg
lfu, lfu_sukarnjanaset Table 3; Eq. 23; ESM3 TH15-TH16 0.645; 1 (held constant)
lfdbs, lfdbs_taz Table 3; Eqs. 24-25; ESM3 TH17-TH18 0.368; 0.448
e_sime2014_cl, e_sime2014_vp Table 3; Eqs. 21-22; ESM3 TH19-TH20 1.73; 0.512
etalvc, etalcl, etalcl_taz, etalvp, etalq ESM3 final OMEGA; Table 3 IIV (CV% = sqrt(exp(omega) - 1)) 0.167, 0.171, 0.159, 0.548, 0.358
propSd, addSd Table 3 sigma TOT&DBS piperacillin; ESM3 SIGMA 1-2 30.2%, 0.147 mg/L
propSd_Cu, addSd_Cu Table 3 sigma UNB; ESM3 SIGMA 3-4 36.5%, 0.747 mg/L
propSd_taz Table 3 sigma TOT&DBS tazobactam; ESM3 SIGMA 5 28.5% (additive 0, held constant)
fsize Eq. 16 WT / 70
fmat, fmat_ref Eq. 17; ESM3 FMAT/RFMAT sigmoid in PMA, normalised to PMA = 35 + 40/52 years
fdec, fdec_taz and references Eqs. 18-19; ESM3 FDEC/RFDEC 1 - sigmoid in PMA, normalised to the same reference
creat_std Eq. 20; ESM3 SCRstd exp[1.42 - (1.17 + 0.203 ln(PMA/100)) / sqrt(PMA/100)]
f_scr Eq. 4 exp(-theta_SCR (SCR - SCR_std))
cl, vc, vp, q Eqs. 6-9 as printed, with the ESM3 reference normalisation
cl_taz, vc_taz, vp_taz, q_taz Eqs. 11-14 as printed, with the ESM3 reference normalisation
ODEs ESM3 $DES two independent two-compartment systems
Cc, Cu, Cc_taz Eqs. 10, 15, 23-25; ESM3 $ERROR A1/V1 times fUNB and fDBS
Residual error form ESM3 $ERROR Y = IPRED*(1+PROP) + ADD combined proportional + additive

Published estimates reproduce from the packaged ini()

th <- ui$theta
om <- diag(ui$omega)

back <- tibble::tribble(
  ~Parameter, ~Published, ~Recovered,
  "theta_V1 pip (L)", 10.4, exp(th[["lvc"]]),
  "theta_CL pip (L/h)", 10.6, exp(th[["lcl"]]),
  "theta_V2 pip (L)", 11.6, exp(th[["lvp"]]),
  "theta_Q2 pip (L/h)", 15.2, exp(th[["lq"]]),
  "theta_V1 taz (L)", 10.5, exp(th[["lvc_taz"]]),
  "theta_CL taz (L/h)", 9.58, exp(th[["lcl_taz"]]),
  "theta_V2 taz (L)", 13.7, exp(th[["lvp_taz"]]),
  "theta_Q2 taz (L/h)", 16.8, exp(th[["lq_taz"]]),
  "MAT50 (weeks)", 54.2, 52 * exp(th[["ltmat50"]]),
  "gamma1", 3.35, exp(th[["lhill_mat"]]),
  "DEC50 pip (years)", 89.1, exp(th[["ltdec50"]]),
  "DEC50 taz (years)", 61.6, exp(th[["ltdec50_taz"]]),
  "gamma2", 1.92, exp(th[["lhill_dec"]]),
  "theta_SCR (dL/mg)", 0.346, th[["e_creat_cl"]],
  "fUNB", 0.645, exp(th[["lfu"]]),
  "fUNB Sukarnjanaset", 1, exp(th[["lfu_sukarnjanaset"]]),
  "fDBS pip", 0.368, exp(th[["lfdbs"]]),
  "fDBS taz", 0.448, exp(th[["lfdbs_taz"]]),
  "theta_CL_Sime", 1.73, th[["e_sime2014_cl"]],
  "theta_V2_Sime", 0.512, th[["e_sime2014_vp"]],
  "IIV V1 (%)", 42.6, 100 * sqrt(exp(om[["etalvc"]]) - 1),
  "IIV CL pip (%)", 43.2, 100 * sqrt(exp(om[["etalcl"]]) - 1),
  "IIV CL taz (%)", 41.5, 100 * sqrt(exp(om[["etalcl_taz"]]) - 1),
  "IIV V2 (%)", 85.4, 100 * sqrt(exp(om[["etalvp"]]) - 1),
  "IIV Q2 (%)", 65.6, 100 * sqrt(exp(om[["etalq"]]) - 1),
  "sigma prop pip (%)", 30.2, 100 * th[["propSd"]],
  "sigma add pip (mg/L)", 0.147, th[["addSd"]],
  "sigma prop unbound (%)", 36.5, 100 * th[["propSd_Cu"]],
  "sigma add unbound (mg/L)", 0.747, th[["addSd_Cu"]],
  "sigma prop taz (%)", 28.5, 100 * th[["propSd_taz"]]
)
back$pct_diff <- round(100 * (back$Recovered / back$Published - 1), 2)
back |>
  mutate(Recovered = signif(Recovered, 4)) |>
  rename("% difference" = pct_diff) |>
  knitr::kable(caption = "Kong 2025 Table 3 recovered from the packaged ini() block.")
Kong 2025 Table 3 recovered from the packaged ini() block.
Parameter Published Recovered % difference
theta_V1 pip (L) 10.400 10.400 0.00
theta_CL pip (L/h) 10.600 10.600 0.00
theta_V2 pip (L) 11.600 11.600 0.00
theta_Q2 pip (L/h) 15.200 15.200 0.00
theta_V1 taz (L) 10.500 10.500 0.00
theta_CL taz (L/h) 9.580 9.580 0.00
theta_V2 taz (L) 13.700 13.700 0.00
theta_Q2 taz (L/h) 16.800 16.800 0.00
MAT50 (weeks) 54.200 54.200 0.00
gamma1 3.350 3.350 0.00
DEC50 pip (years) 89.100 89.100 0.00
DEC50 taz (years) 61.600 61.600 0.00
gamma2 1.920 1.920 0.00
theta_SCR (dL/mg) 0.346 0.346 0.00
fUNB 0.645 0.645 0.00
fUNB Sukarnjanaset 1.000 1.000 0.00
fDBS pip 0.368 0.368 0.00
fDBS taz 0.448 0.448 0.00
theta_CL_Sime 1.730 1.730 0.00
theta_V2_Sime 0.512 0.512 0.00
IIV V1 (%) 42.600 42.630 0.08
IIV CL pip (%) 43.200 43.180 -0.04
IIV CL taz (%) 41.500 41.510 0.03
IIV V2 (%) 85.400 85.430 0.03
IIV Q2 (%) 65.600 65.610 0.02
sigma prop pip (%) 30.200 30.200 0.00
sigma add pip (mg/L) 0.147 0.147 0.00
sigma prop unbound (%) 36.500 36.500 0.00
sigma add unbound (mg/L) 0.747 0.747 0.00
sigma prop taz (%) 28.500 28.500 0.00

# Every entry agrees with the published value to its printed precision.
stopifnot(all(abs(back$pct_diff) < 0.4))

Typical-value checks against the paper’s worked numbers

Kong 2025 quotes typical clearances for several example patients in the Abstract, Results and Discussion, and the per-kg clearance of each simulated age group in the Discussion. Each is reproduced below from a typical-value (zeroRe()) solve, so every factor of Eqs. 6-20 is exercised: size, maturation, decline and the SCR effect.

mod_typ <- rxode2::zeroRe(ui)

# Eq. 20, used to build the virtual patients.
creat_std_fn <- function(pma) exp(1.42 - (1.17 + 0.203 * log(pma / 100)) / sqrt(pma / 100))

# zeroRe() solves of several subjects warn that no omega is present, which is
# the intent here.
muffle_no_omega <- function(w) {
  if (grepl("without 'omega'", conditionMessage(w), fixed = TRUE)) invokeRestart("muffleWarning")
}

# One dosing record and one observation per patient; the parameter columns
# (cl, cl_taz, ...) are read from the observation row.
typical_params <- function(pts) {
  ev <- bind_rows(lapply(seq_len(nrow(pts)), function(i) {
    data.frame(
      id = i, time = c(0, 1), amt = c(1000, 0), evid = c(1L, 0L),
      cmt = "central", dvid = c(NA_integer_, 1L),
      PAGE = pts$pma_y[i] * 12, WT = pts$WT[i], CREAT = pts$CREAT[i],
      SAMPLE_DBS = 0, STUDY_SIME2014 = pts$sime[i],
      STUDY_SUKARNJANASET2019 = 0
    )
  }))
  sim <- withCallingHandlers(
    as.data.frame(rxode2::rxSolve(mod_typ, ev, returnType = "data.frame")),
    warning = muffle_no_omega
  )
  sim <- sim[sim$time == 1, ]
  cbind(pts, sim[, c("cl", "cl_taz", "vc", "vp", "q", "vc_taz", "vp_taz", "creat_std")])
}

The neonate PMA is taken as the value whose standardised SCR equals the 0.764 mg/dL printed in ESM4 Table S3 (the virtual patients’ SCR was fixed to SCR_std), i.e. 7.17 months rather than the rounded 7.3 months of the same table. Because maturation is steep in this range (gamma1 = 3.35), the rounding alone moves the neonatal clearance by about 5%; see Assumptions.

pma_neon <- uniroot(function(p) creat_std_fn(p) - 0.764, c(0.3, 0.9))$root

pts <- tibble::tribble(
  ~patient, ~pma_y, ~WT, ~CREAT, ~sime, ~pub_cl, ~pub_cl_taz, ~pub_clkg, ~source,
  "Reference (70 kg, 35 y)", 35 + 40 / 52, 70, NA, 0, 10.6, 9.58, NA, "Abstract; Table 3",
  "30 y, 70 kg, SCR 0.773", 30 + 40 / 52, 70, 0.773, 0, 11.0, 10.3, NA, "Discussion",
  "70 y, 70 kg, SCR 1.12", 70 + 40 / 52, 70, 1.12, 0, 7.33, 5.37, NA, "Discussion",
  "Infant 44.3 wk PMA, 4.5 kg, SCR 0.416", 44.3 / 52, 4.5, 0.416, 0, 0.543, 0.565, NA, "Discussion",
  "Neonate (ESM4 Table S3)", pma_neon, 1.30, 0.764, 0, NA, NA, 0.0648, "Discussion; Table S3",
  "Infant (ESM4 Table S3)", 18.0 / 12, 7.74, 0.312, 0, NA, NA, 0.235, "Discussion; Table S3",
  "Child (ESM4 Table S3)", 5.0, 16.4, 0.336, 0, NA, NA, 0.253, "Discussion; Table S3",
  "Adolescent (ESM4 Table S3)", 13.4, 37.0, 0.517, 0, NA, NA, 0.203, "Discussion; Table S3",
  "Adult (ESM4 Table S3)", 48.8, 75.0, 0.954, 0, NA, NA, 0.132, "Discussion; Table S3",
  "Elderly (ESM4 Table S3)", 74.8, 75.0, 1.145, 0, NA, NA, 0.102, "Discussion; Table S3",
  "Reference, Sime 2014 study", 35 + 40 / 52, 70, NA, 1, 10.6 * 1.73, 9.58, NA, "Eq. 21"
)
pts$CREAT[is.na(pts$CREAT)] <- creat_std_fn(pts$pma_y[is.na(pts$CREAT)])
typ <- typical_params(pts)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etalcl_taz', 'etalvp', 'etalq'

typ_tab <- typ |>
  mutate(
    sim_clkg = cl / WT,
    pct_cl = 100 * (cl / pub_cl - 1),
    pct_cl_taz = 100 * (cl_taz / pub_cl_taz - 1),
    pct_clkg = 100 * (sim_clkg / pub_clkg - 1)
  )
typ_tab |>
  transmute(
    patient, source,
    "CL pip published" = pub_cl, "CL pip simulated" = signif(cl, 4),
    "CL taz published" = pub_cl_taz, "CL taz simulated" = signif(cl_taz, 4),
    "CL pip/kg published" = pub_clkg, "CL pip/kg simulated" = signif(sim_clkg, 3)
  ) |>
  knitr::kable(caption = "Typical clearances (L/h; L/h/kg) quoted by Kong 2025 versus the packaged model.")
Typical clearances (L/h; L/h/kg) quoted by Kong 2025 versus the packaged model.
patient source CL pip published CL pip simulated CL taz published CL taz simulated CL pip/kg published CL pip/kg simulated
Reference (70 kg, 35 y) Abstract; Table 3 10.600 10.60000 9.580 9.5800 NA 0.1510
30 y, 70 kg, SCR 0.773 Discussion 11.000 11.01000 10.300 10.2500 NA 0.1570
70 y, 70 kg, SCR 1.12 Discussion 7.330 7.57000 5.370 5.6180 NA 0.1080
Infant 44.3 wk PMA, 4.5 kg, SCR 0.416 Discussion 0.543 0.54370 0.565 0.5662 NA 0.1210
Neonate (ESM4 Table S3) Discussion; Table S3 NA 0.08392 NA 0.0874 0.0648 0.0646
Infant (ESM4 Table S3) Discussion; Table S3 NA 1.83900 NA 1.9140 0.2350 0.2380
Child (ESM4 Table S3) Discussion; Table S3 NA 4.14900 NA 4.3040 0.2530 0.2530
Adolescent (ESM4 Table S3) Discussion; Table S3 NA 7.50900 NA 7.6190 0.2030 0.2030
Adult (ESM4 Table S3) Discussion; Table S3 NA 9.96500 NA 8.3230 0.1320 0.1330
Elderly (ESM4 Table S3) Discussion; Table S3 NA 7.63900 NA 5.5640 0.1020 0.1020
Reference, Sime 2014 study Eq. 21 18.338 18.34000 9.580 9.5800 NA 0.2620

# The SCR standardisation at the reference age reproduces the Abstract's 0.83 mg/dL.
stopifnot(abs(typ$creat_std[1] - 0.83) < 0.005)

# Every quoted clearance except the 70-year-old example reproduces within 1.5%.
not70 <- typ_tab$patient != "70 y, 70 kg, SCR 1.12"
stopifnot(
  all(abs(typ_tab$pct_cl[not70]) < 1.5, na.rm = TRUE),
  all(abs(typ_tab$pct_cl_taz[not70]) < 1.5, na.rm = TRUE),
  all(abs(typ_tab$pct_clkg) < 1.5, na.rm = TRUE)
)
# The 70-year-old example sits 3-5% above the quoted values for both drugs;
# see Assumptions.
stopifnot(all(abs(c(typ_tab$pct_cl[!not70], typ_tab$pct_cl_taz[!not70])) < 6))

# Discussion: volumes per kg for the 70 kg reference.
vkg <- c(typ$vc[1], typ$vp[1], typ$vc_taz[1], typ$vp_taz[1]) / 70
stopifnot(all(abs(vkg / c(0.149, 0.166, 0.150, 0.196) - 1) < 0.005))

# Results: every 0.20 mg/dL rise in SCR lowers both clearances by 6.7%.
stopifnot(abs(100 * (1 - exp(-th[["e_creat_cl"]] * 0.2)) - 6.7) < 0.05)

# Eqs. 9 and 22: in the Sime 2014 study V2 is 0.512-fold, and the
# compartmental allometry carries that into Q2 with exponent 0.75.
sime <- typ[typ$patient == "Reference, Sime 2014 study", ]
stopifnot(
  abs(sime$vp / typ$vp[1] - 0.512) < 1e-6,
  abs(sime$q / typ$q[1] - 0.512^0.75) < 1e-6
)

Observables

On a single subject, the unbound piperacillin observable is exactly 0.645 times the total concentration (1 in the Sukarnjanaset 2019 study), and dried blood spot records are 0.368 (piperacillin) and 0.448 (tazobactam) times plasma.

obs_ev <- bind_rows(lapply(1:3, function(i) {
  data.frame(
    id = i,
    time = c(0, 0, 1, 2, 4),
    amt = c(4000, 500, 0, 0, 0),
    rate = c(8000, 1000, 0, 0, 0),
    evid = c(1L, 1L, 0L, 0L, 0L),
    cmt = c("central", "central_taz", "central", "central", "central"),
    dvid = c(NA, NA, 1L, 1L, 1L),
    PAGE = (35 + 40 / 52) * 12, WT = 70, CREAT = 0.829,
    SAMPLE_DBS = as.numeric(i == 2), STUDY_SIME2014 = 0,
    STUDY_SUKARNJANASET2019 = as.numeric(i == 3)
  )
}))
obs <- withCallingHandlers(
  as.data.frame(rxode2::rxSolve(mod_typ, obs_ev, returnType = "data.frame")),
  warning = muffle_no_omega
)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etalcl_taz', 'etalvp', 'etalq'
obs$plasma <- obs$central / obs$vc
obs$plasma_taz <- obs$central_taz / obs$vc_taz
stopifnot(
  all(abs(obs$Cu[obs$id == 1] / obs$plasma[obs$id == 1] - 0.645) < 1e-9),
  all(abs(obs$Cu[obs$id == 3] / obs$plasma[obs$id == 3] - 1) < 1e-9),
  all(abs(obs$Cc[obs$id == 2] / obs$plasma[obs$id == 2] - 0.368) < 1e-9),
  all(abs(obs$Cc_taz[obs$id == 2] / obs$plasma_taz[obs$id == 2] - 0.448) < 1e-9),
  all(abs(obs$Cc[obs$id == 1] / obs$plasma[obs$id == 1] - 1) < 1e-9)
)

Replicate Kong 2025 Figure 1

Typical standardised clearance (L/h per 70 kg, SCR equal to SCR_std) over the lifespan. The post hoc individual values of the figure are not reproducible without the data.

grid <- data.frame(pma_y = exp(seq(log(0.45), log(100), length.out = 120)))
grid$WT <- 70
grid$CREAT <- creat_std_fn(grid$pma_y)
grid$sime <- 0
fig1 <- typical_params(grid)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etalcl_taz', 'etalvp', 'etalq'

bind_rows(
  data.frame(pma = fig1$pma_y, cl = fig1$cl, drug = "Piperacillin"),
  data.frame(pma = fig1$pma_y, cl = fig1$cl_taz, drug = "Tazobactam")
) |>
  ggplot(aes(pma, cl)) +
  geom_line(linewidth = 1) +
  facet_wrap(~drug) +
  scale_x_log10() +
  labs(
    x = "Postmenstrual age (years)", y = "CLstd (L/h/70 kg)",
    title = "Replicates Figure 1 of Kong 2025 (typical lines)"
  ) +
  theme_bw()

Virtual cohort and Figure 5 simulation

Figure 5 of Kong 2025 shows the steady-state probability of target attainment (PTA) of unbound piperacillin under the highest dose of the summary of product characteristics (SmPC) for six age groups, by infusion mode. The virtual patients of each group are male, with the median PMA, weight and SCR of ESM4 Table S3 (SCR at SCR_std), and differ only through their random effects. The paper simulated 1000 patients per group and mode; 200 are used here.

groups <- tibble::tribble(
  ~group, ~pma_y, ~WT, ~CREAT, ~dose_pip, ~dose_taz, ~tau,
  "Neonates", pma_neon, 1.30, 0.764, 70 * 1.30, 8.75 * 1.30, 8,
  "Infants", 18.0 / 12, 7.74, 0.312, 70 * 7.74, 8.75 * 7.74, 8,
  "Children", 5.0, 16.4, 0.336, 80 * 16.4, 10 * 16.4, 6,
  "Adolescents", 13.4, 37.0, 0.517, 4000, 500, 6,
  "Adults", 48.8, 75.0, 0.954, 4000, 500, 6,
  "Elderly", 74.8, 75.0, 1.145, 4000, 500, 6
)
modes <- c("Intermittent", "Extended", "Continuous")
n_per_arm <- 200
t_end <- 72

arms <- tidyr::expand_grid(groups, mode = modes) |>
  mutate(arm = row_number())

make_arm <- function(a) {
  ids <- (a$arm - 1) * n_per_arm + seq_len(n_per_arm)
  if (a$mode == "Continuous") {
    daily <- 24 / a$tau
    dose <- data.frame(
      time = 0, amt = c(a$dose_pip, a$dose_taz) * daily * t_end / 24,
      dur = t_end, cmt = c("central", "central_taz")
    )
  } else {
    dur <- if (a$mode == "Intermittent") 0.5 else a$tau / 2
    times <- seq(0, t_end - a$tau, by = a$tau)
    dose <- rbind(
      data.frame(time = times, amt = a$dose_pip, dur = dur, cmt = "central"),
      data.frame(time = times, amt = a$dose_taz, dur = dur, cmt = "central_taz")
    )
  }
  dose$evid <- 1L
  dose$dvid <- NA_integer_
  obs <- data.frame(
    time = seq(t_end - a$tau, t_end, by = 0.1), amt = 0, dur = 0,
    cmt = "central", evid = 0L, dvid = 1L
  )
  one <- rbind(dose, obs)
  ev <- one[rep(seq_len(nrow(one)), n_per_arm), ]
  ev$id <- rep(ids, each = nrow(one))
  ev$rate <- ifelse(ev$evid == 1L, ev$amt / ev$dur, 0)
  ev$dur <- NULL
  ev$PAGE <- a$pma_y * 12
  ev$WT <- a$WT
  ev$CREAT <- a$CREAT
  ev$SAMPLE_DBS <- 0
  ev$STUDY_SIME2014 <- 0
  ev$STUDY_SUKARNJANASET2019 <- 0
  ev
}
events <- bind_rows(lapply(seq_len(nrow(arms)), function(i) make_arm(arms[i, ]))) |>
  arrange(id, time, desc(evid))
rxode2::rxSetSeed(20250101)
sim <- as.data.frame(rxode2::rxSolve(ui, events, returnType = "data.frame"))
sim$arm <- (sim$id - 1) %/% n_per_arm + 1
sim <- left_join(sim, arms[, c("arm", "group", "mode", "tau")], by = "arm")
# fT>MIC 100% at steady state: the unbound concentration stays above the MIC
# for the whole last dosing interval, i.e. the interval minimum exceeds it.
cmin_u <- sim |>
  group_by(group, mode, id) |>
  summarise(cmin_u = min(Cu), cl = first(cl), .groups = "drop")

mics <- 2^(0:7)
pta <- bind_rows(lapply(mics, function(m) {
  cmin_u |>
    group_by(group, mode) |>
    summarise(
      pta_mic = 100 * mean(cmin_u > m), pta_4mic = 100 * mean(cmin_u > 4 * m),
      .groups = "drop"
    ) |>
    mutate(mic = m)
}))

pta |>
  tidyr::pivot_longer(c(pta_mic, pta_4mic), names_to = "target", values_to = "pta") |>
  mutate(
    target = ifelse(target == "pta_mic", "fT>MIC 100%", "fT>4xMIC 100%"),
    mode = factor(mode, levels = modes)
  ) |>
  ggplot(aes(mic, pta, colour = group)) +
  geom_line() +
  geom_point() +
  geom_hline(yintercept = 90, linetype = "dashed") +
  scale_x_log10(breaks = mics) +
  facet_grid(target ~ mode) +
  labs(
    x = "MIC (mg/L)", y = "PTA (%)", colour = NULL,
    title = "Replicates Figure 5 of Kong 2025 (SmPC highest dose)"
  ) +
  theme_bw() +
  theme(legend.position = "bottom")

Continuous infusion: a closed-form check

Under continuous infusion the steady-state unbound concentration of each subject is fu x R / CL_i, using that subject’s own drawn clearance. The simulated interval minimum must equal it.

cont <- cmin_u |>
  filter(mode == "Continuous") |>
  left_join(groups[, c("group", "dose_pip", "tau")], by = "group") |>
  mutate(
    css_u = 0.645 * (dose_pip / tau) / cl,
    rel = cmin_u / css_u - 1
  )
stopifnot(
  abs(median(cont$rel)) < 0.01,
  quantile(abs(cont$rel), 0.9) < 0.03
)

Comparison with the published PTA curves

The PTA values below were digitised by the maintainers from the fT>MIC 100% row of Figure 5 (intermittent and continuous panels, readable to about 3 percentage points).

digitised <- tibble::tribble(
  ~group, ~mode, ~mic, ~pub,
  "Neonates", "Intermittent", 1, 99, "Neonates", "Intermittent", 4, 96,
  "Neonates", "Intermittent", 16, 80, "Neonates", "Intermittent", 64, 25,
  "Infants", "Intermittent", 1, 52, "Infants", "Intermittent", 4, 24,
  "Infants", "Intermittent", 16, 4,
  "Children", "Intermittent", 1, 74, "Children", "Intermittent", 4, 46,
  "Children", "Intermittent", 16, 11,
  "Adolescents", "Intermittent", 1, 88, "Adolescents", "Intermittent", 4, 70,
  "Adolescents", "Intermittent", 16, 30,
  "Adults", "Intermittent", 1, 96, "Adults", "Intermittent", 4, 81,
  "Adults", "Intermittent", 16, 39,
  "Elderly", "Intermittent", 1, 99, "Elderly", "Intermittent", 4, 92,
  "Elderly", "Intermittent", 16, 61, "Elderly", "Intermittent", 64, 8,
  "Neonates", "Continuous", 64, 77, "Neonates", "Continuous", 128, 17,
  "Infants", "Continuous", 16, 84, "Infants", "Continuous", 32, 25,
  "Children", "Continuous", 32, 56, "Children", "Continuous", 64, 7,
  "Adolescents", "Continuous", 32, 92, "Adolescents", "Continuous", 64, 39,
  "Adults", "Continuous", 32, 77, "Adults", "Continuous", 64, 17,
  "Elderly", "Continuous", 32, 91, "Elderly", "Continuous", 64, 37
)
cmp <- digitised |>
  left_join(pta[, c("group", "mode", "mic", "pta_mic")], by = c("group", "mode", "mic")) |>
  mutate(diff = pta_mic - pub)
cmp |>
  rename(
    "Age group" = group, "Infusion" = mode, "MIC (mg/L)" = mic,
    "Published PTA (%)" = pub, "Simulated PTA (%)" = pta_mic,
    "Difference (points)" = diff
  ) |>
  knitr::kable(caption = "Simulated versus digitised Figure 5 PTA (fT>MIC 100%).")
Simulated versus digitised Figure 5 PTA (fT>MIC 100%).
Age group Infusion MIC (mg/L) Published PTA (%) Simulated PTA (%) Difference (points)
Neonates Intermittent 1 99 98.5 -0.5
Neonates Intermittent 4 96 93.5 -2.5
Neonates Intermittent 16 80 79.0 -1.0
Neonates Intermittent 64 25 28.5 3.5
Infants Intermittent 1 52 53.5 1.5
Infants Intermittent 4 24 25.0 1.0
Infants Intermittent 16 4 3.5 -0.5
Children Intermittent 1 74 72.0 -2.0
Children Intermittent 4 46 47.0 1.0
Children Intermittent 16 11 13.5 2.5
Adolescents Intermittent 1 88 87.5 -0.5
Adolescents Intermittent 4 70 66.0 -4.0
Adolescents Intermittent 16 30 32.5 2.5
Adults Intermittent 1 96 94.0 -2.0
Adults Intermittent 4 81 80.0 -1.0
Adults Intermittent 16 39 37.0 -2.0
Elderly Intermittent 1 99 99.0 0.0
Elderly Intermittent 4 92 94.0 2.0
Elderly Intermittent 16 61 60.0 -1.0
Elderly Intermittent 64 8 7.0 -1.0
Neonates Continuous 64 77 78.5 1.5
Neonates Continuous 128 17 18.5 1.5
Infants Continuous 16 84 84.0 0.0
Infants Continuous 32 25 22.5 -2.5
Children Continuous 32 56 59.5 3.5
Children Continuous 64 7 9.0 2.0
Adolescents Continuous 32 92 93.0 1.0
Adolescents Continuous 64 39 39.0 0.0
Adults Continuous 32 77 75.5 -1.5
Adults Continuous 64 17 15.5 -1.5
Elderly Continuous 32 91 89.5 -1.5
Elderly Continuous 64 37 35.0 -2.0

# Centre and envelope, not extremes: each simulated PTA carries a Monte Carlo
# standard error of up to 3.5 points at n = 200, on top of digitisation error.
stopifnot(
  abs(median(cmp$diff)) < 4,
  quantile(abs(cmp$diff), 0.9) < 12
)

The qualitative conclusions of the Results reproduce: neonates attain the targets most often and infants least often, the elderly more often than adults, adolescents more often than children, and longer infusions raise the PTA in every age group. Each comparison below is on the PTA averaged over the eight MICs, where the gaps are tens of points.

avg <- pta |>
  group_by(group, mode) |>
  summarise(avg_pta = mean(pta_mic), .groups = "drop")
int_avg <- with(filter(avg, mode == "Intermittent"), setNames(avg_pta, group))
stopifnot(
  names(which.max(int_avg)) == "Neonates",
  names(which.min(int_avg)) == "Infants",
  int_avg[["Elderly"]] > int_avg[["Adults"]] + 5,
  int_avg[["Adolescents"]] > int_avg[["Children"]] + 5
)
by_mode <- tidyr::pivot_wider(avg, names_from = mode, values_from = avg_pta)
stopifnot(
  all(by_mode$Extended > by_mode$Intermittent),
  all(by_mode$Continuous > by_mode$Extended)
)

PKNCA validation

Kong 2025 reports no NCA parameters. PKNCA is run on the last steady-state intermittent-infusion dosing interval of each age group, and the interval AUC is checked against the steady-state identity AUCtau = Dose / CL for each subject’s own clearance.

int_ids <- sim |>
  filter(mode == "Intermittent")
conc <- int_ids |>
  filter(!is.na(Cc)) |>
  select(id, time, Cc, group)
dose_last <- distinct(int_ids, id, group) |>
  left_join(groups[, c("group", "dose_pip", "tau")], by = "group") |>
  mutate(time = t_end - tau, amt = dose_pip)

conc_obj <- PKNCAconc(conc, Cc ~ time | group + id)
dose_obj <- PKNCAdose(dose_last, amt ~ time | group + id)
intervals <- groups |>
  transmute(
    group, start = t_end - tau, end = t_end,
    auclast = TRUE, cmax = TRUE, cmin = TRUE
  )
nca <- pk.nca(PKNCAdata(conc_obj, dose_obj, intervals = intervals))
nca_res <- as.data.frame(nca$result)

summary(nca)
#>  start end       group   N     auclast       cmax        cmin
#>     64  72    Neonates 200 1100 [42.7] 321 [30.5]  49.7 [155]
#>     64  72     Infants 200  298 [43.8] 216 [35.0] 1.27 [1030]
#>     66  72    Children 200  315 [45.2] 256 [28.6]  3.50 [730]
#>     66  72 Adolescents 200  507 [47.2] 371 [30.6]  9.04 [461]
#>     66  72      Adults 200  405 [46.4] 239 [33.1]  13.9 [218]
#>     66  72     Elderly 200  527 [41.9] 256 [29.1]  29.1 [120]
#> 
#> Caption: auclast, cmax, cmin: geometric mean and geometric coefficient of variation; N: number of subjects

auc_chk <- nca_res |>
  filter(PPTESTCD == "auclast") |>
  left_join(distinct(int_ids, id, cl), by = "id") |>
  left_join(groups[, c("group", "dose_pip")], by = "group") |>
  mutate(rel = PPORRES * cl / dose_pip - 1)

# Same drawn parameters on both sides; only trapezoid and solver error remain.
stopifnot(
  abs(median(auc_chk$rel)) < 0.01,
  quantile(abs(auc_chk$rel), 0.9) < 0.02
)

Tazobactam exposure

The Results report a median unbound tazobactam mean steady-state concentration (AUCss,24h / 24, unbound fraction 0.70) of 3.69 mg/L across the age groups (range 1.64-6.77 mg/L, lowest in infants). At steady state this equals 0.70 x daily dose / (24 x CL_taz); the typical values per group are below.

taz <- typical_params(transmute(groups, patient = group, pma_y, WT, CREAT, sime = 0)) |>
  left_join(groups[, c("group", "dose_taz", "tau")], by = c("patient" = "group")) |>
  mutate(cm_u = 0.70 * dose_taz * (24 / tau) / (24 * cl_taz))
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etalcl_taz', 'etalvp', 'etalq'
taz |>
  transmute("Age group" = patient, "CL taz (L/h)" = signif(cl_taz, 3), "Unbound Cm (mg/L)" = signif(cm_u, 3)) |>
  knitr::kable(caption = "Typical unbound tazobactam mean steady-state concentration by age group.")
Typical unbound tazobactam mean steady-state concentration by age group.
Age group CL taz (L/h) Unbound Cm (mg/L)
Neonates 0.0874 11.40
Infants 1.9100 3.10
Children 4.3000 4.45
Adolescents 7.6200 7.66
Adults 8.3200 7.01
Elderly 5.5600 10.50
stopifnot(taz$patient[which.min(taz$cm_u)] == "Infants")

Infants have the lowest exposure, as reported, but the packaged model gives higher values overall than the quoted 3.69 mg/L median (see Assumptions).

Assumptions and deviations

  • Reference normalisation of maturation and decline. The printed Eqs. 7, 12 and 17-19 omit it, but the ESM3 control stream divides the maturation and decline factors by their values at the reference PMA of 35 + 40/52 years, so theta_CL is the clearance of a 35-year-old. The model follows the control stream; the Abstract’s reference patient and the infant and 30-year-old examples of the Discussion reproduce only with it.
  • MAT50 units. Table 3 prints 54.2 weeks; the control stream estimates it on the log-year scale (0.0359, i.e. 1.037 years). The packaged value converts 54.2 weeks with the 52 weeks per year that the control stream uses for the reference PMA.
  • Missing serum creatinine. The source coded subjects without any SCR record as 0 and set F_SCR to 1 for them. The packaged model has no such sentinel; supply CREAT equal to SCR_std (model variable creat_std) to reproduce it.
  • Neonate PMA in the Table S3 simulations. The neonatal PMA was taken as 7.17 months, the value at which SCR_std equals the printed 0.764 mg/dL, instead of the rounded 7.3 months; with 7.3 months the neonatal clearance per kg is 0.068 instead of the quoted 0.0648 L/h/kg.
  • 70-year-old example. The Discussion quotes CL = 7.33 (piperacillin) and 5.37 L/h (tazobactam) for a 70-year-old, 70-kg patient with SCR 1.12 mg/dL; the model gives 3% and 5% more. Both drugs would match at an age of about 73.5 years, which points at the example’s inputs rather than the parameters. The same model reproduces the per-kg clearance of the 74.8-year-old elderly group to within 0.2%.
  • Tazobactam exposure. The packaged model does not reproduce the quoted median unbound tazobactam concentration of 3.69 mg/L (range 1.64-6.77). The paper does not state how the median was pooled across groups and scenarios, and the per-group typical values above are 1.7-2 times higher (median 7.3 mg/L, range 3.1-11.4 mg/L). The relative ordering (infants lowest) is reproduced. Piperacillin PTA, computed from the same clearances, matches Figure 5.
  • Residual error for tazobactam. The additive tazobactam term is 0 and held constant in the source and is omitted, leaving a proportional error.
  • Dried blood spots share the residual error of the total plasma observations (Table 3, sigma TOT&DBS), so SAMPLE_DBS changes only the prediction.
  • Population counts. The Results text gives 415 patients, while the per-study N column of Table 1 sums to 435. The female percentage (41%) is derived from the per-study percentage of males in Table 1.
  • PTA simulation size. 200 virtual patients per age group and infusion mode instead of 1000; PTA values carry Monte Carlo error of up to 3.5 points.