Skip to contents

Model and source

  • Citation: Avachat C, Hung CC, Birnbaum AK, Healy DP, Sherwin CMT, Lin YC. Pharmacokinetic effect of disease severity and use of extracorporeal membrane oxygenation in critically ill Asian patients receiving vancomycin. Front Pharmacol. 2025;16:1506793. doi:10.3389/fphar.2025.1506793. PMID 40078297; PMCID PMC11897554.
  • Description: One-compartment IV population PK model for vancomycin in critically ill, predominantly paediatric Asian patients (Taiwan; age 0.1-36.6 years, weight 0.7-93 kg) receiving therapeutic drug monitoring (Avachat 2025). Clearance scales allometrically with body weight (exponent fixed at 0.75) and volume linearly (exponent fixed at 1), both normalised to the 12.6 kg cohort median; clearance is further reduced by exp(-0.81) (about 56%) in patients on extracorporeal membrane oxygenation (ECMO) and by exp(-0.49) (about 39%) in patients whose vancomycin indication was a milder infection (cellulitis, surgical prophylaxis, neutropenic fever or other) rather than a severe one (pneumonia, bacteremia, osteomyelitis, meningitis or deep tissue infection). Exponential between-subject variability on CL and V; proportional residual error.
  • Article: https://doi.org/10.3389/fphar.2025.1506793
  • Open-access copy: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11897554/

Population

Avachat 2025 is a retrospective analysis of therapeutic drug monitoring (TDM) data from 90 critically ill patients treated with intravenous vancomycin at Kaohsiung Veterans General Hospital, Taiwan, between 2014 and 2016 (Table 1). Ninety percent were children: median age 3.7 years (range 0.1 to 36.6) and median weight 12.6 kg (range 0.7 to 93). There were 47 males and 43 females. Median eGFR (bedside Schwartz) was 70.4 (range 2.4 to 150.6); one patient was on regular haemodialysis and one on CRRT. Twelve patients (13.3%) were on extracorporeal membrane oxygenation (ECMO) and 33 (36.7%) had surgery.

The authors grouped the nine admission diagnoses by the severity of the infection for which vancomycin was given: severe = pneumonia (20), bacteremia (6), osteomyelitis (7), meningitis (4) and deep tissue infection (1), 38 patients; mild = cellulitis (6), surgical prevention (35), neutropenic fever (4) and others (7), 52 patients. The 263 concentrations were mostly troughs (drawn within 30 min before a dose), with some peaks (> 1 h after the end of infusion). The model was fitted in Phoenix NLME 8.2 with FOCE-ELS.

The same information is available programmatically via readModelDb("Avachat_2025_vancomycin")()$population.

Source trace

Equation / parameter Value Source location
lcl = log(1.03) 1.03 L/h Table 4, tvCL (L/h) (SE 0.099, 95% CI 0.84-1.23)
lvc = log(11.61) 11.61 L Table 4, tvV (L) (SE 1.22, 95% CI 9.2-14.03)
e_wt_cl 0.75 (fixed) Table 4, dCldWT (fixed); Section 2.2
e_wt_vc 1 (fixed) Table 4, dVdWT (fixed); Section 2.2
e_ecmo_status_cl -0.81 Table 4, dCldECMO (SE 0.17, 95% CI -1.14 to -0.47)
e_infect_mild_cl -0.49 Table 4, dCldDiag (SE 0.18, 95% CI -0.84 to -0.14)
etalcl 0.26 (variance) Table 4, eta Cl (SE 0.08, shrinkage 0.27)
etalvc 0.24 (variance) Table 4, eta V (SE 0.07, shrinkage 0.52)
propSd 0.44 Table 4, Stdev0 (SE 0.03); Section 3.2 proportional error
Reference weight 12.6 kg n/a Section 3.3 equations (WT/median(WT)) and the following paragraph; Table 1 median
cl equation n/a Section 3.3, first final-model equation
vc equation n/a Section 3.3, second final-model equation
One compartment, first-order elimination n/a Section 3.2

Verification against the published worked example

Table 3 of the paper substitutes the final estimates into the clearance equation for a 12.6 kg patient with no random effect. The packaged model, solved with the random effects removed, must reproduce it.

mod <- readModelDb("Avachat_2025_vancomycin")

tab3 <- tibble::tibble(
  id              = 1:4,
  group           = c("No ECMO, severe infection", "ECMO, severe infection",
                      "No ECMO, mild infection", "ECMO, mild infection"),
  ECMO_STATUS     = c(0, 1, 0, 1),
  DIS_INFECT_MILD = c(0, 0, 1, 1),
  # Table 3 'Clearance value (L/h)'; the paper does not tabulate the
  # combined ECMO + mild group.
  cl_paper        = c(1.03, 0.48, 0.63, NA),
  pct_paper       = c(NA, 56, 39, NA)
)

typ_ev <- tab3 |>
  select(id, ECMO_STATUS, DIS_INFECT_MILD) |>
  mutate(WT = 12.6) |>
  tidyr::crossing(time = c(0, 1)) |>
  mutate(evid = 0L, cmt = "central", amt = NA_real_)

typ <- rxode2::rxSolve(rxode2::zeroRe(mod), events = typ_ev) |>
  as.data.frame() |>
  group_by(id) |>
  summarise(cl = first(cl), vc = first(vc), .groups = "drop") |>
  left_join(tab3, by = "id") |>
  mutate(pct_reduction = 100 * (1 - cl / cl[1]))
#> 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'

typ |>
  select(group, cl, cl_paper, pct_reduction, pct_paper, vc) |>
  rename(
    "Group"                       = group,
    "CL, model (L/h)"             = cl,
    "CL, Table 3 (L/h)"           = cl_paper,
    "CL reduction, model (%)"     = pct_reduction,
    "CL reduction, paper (%)"     = pct_paper,
    "V, model (L)"                = vc
  ) |>
  knitr::kable(
    digits  = 3,
    caption = "Typical clearance at the 12.6 kg reference weight, against Avachat 2025 Table 3."
  )
Typical clearance at the 12.6 kg reference weight, against Avachat 2025 Table 3.
Group CL, model (L/h) CL, Table 3 (L/h) CL reduction, model (%) CL reduction, paper (%) V, model (L)
No ECMO, severe infection 1.030 1.03 0.000 NA 11.61
ECMO, severe infection 0.458 0.48 55.514 56 11.61
No ECMO, mild infection 0.631 0.63 38.737 39 11.61
ECMO, mild infection 0.281 NA 72.747 NA 11.61

# Deterministic arithmetic: no random effects on either side.
stopifnot(
  abs(typ$cl[1] - 1.03) < 1e-8,
  abs(typ$vc[1] - 11.61) < 1e-8,
  abs(typ$cl[3] - 0.63) < 0.005,
  abs(typ$pct_reduction[2] - 56) < 1,
  abs(typ$pct_reduction[3] - 39) < 1
)

The model reproduces the reference clearance, the mild-infection clearance (0.631 L/h against the printed 0.63) and both percentage reductions (55.5% and 38.7%, printed as about 56% and 39%). Table 3 and the Results text print the ECMO clearance as 0.48 L/h, but the substitution the table itself writes out, 1.03 * 1^0.75 * exp(-0.81), evaluates to 0.458 L/h; the printed 0.48 does not follow from the printed estimates, while the printed 56% reduction does. The model uses the Table 4 estimate of -0.81.

Typical elimination half-life for a 12.6 kg patient with a severe infection and no ECMO is log(2) * 11.61 / 1.03 = 7.8 h.

Virtual cohort

Individual data are not published. The cohort below draws body weight from a log-normal distribution with the cohort median of 12.6 kg, truncated to the observed 0.7 to 93 kg range. The paper reports no dosing regimen (only that infusion rates ranged from 8 to 1500 mg/h), so every subject receives 15 mg/kg every 6 h as a 1-h infusion (60 mg/kg/day), a common paediatric starting regimen. This is an assumption of the vignette.

Each of the four covariate groups (ECMO yes/no by severe/mild infection) gets 100 subjects with identical weight draws, so the groups differ only by the covariate effects and the random effects. Concentrations are simulated at steady state with ss = 1, over one dosing interval.

# set.seed() fixes the weight draws. rxode2's own random-effect draws are not
# reproducible across rxode2 builds or thread counts, so the assertions below
# compare each subject with itself or with closed forms.
set.seed(20251306)

n_per_group <- 100L
tau         <- 6
infusion_h  <- 1

draw_trunc_lnorm <- function(n, median_value, sdlog, lower, upper) {
  out <- numeric(0)
  while (length(out) < n) {
    cand <- stats::rlnorm(n * 3L, log(median_value), sdlog)
    out <- c(out, cand[cand >= lower & cand <= upper])
  }
  out[seq_len(n)]
}

weights <- draw_trunc_lnorm(n_per_group, 12.6, 1.1, 0.7, 93)

groups <- tab3 |>
  select(group, ECMO_STATUS, DIS_INFECT_MILD) |>
  mutate(group_index = seq_len(n()))

subjects <- groups |>
  tidyr::expand_grid(subject = seq_len(n_per_group)) |>
  mutate(
    id  = (group_index - 1L) * n_per_group + subject,
    WT  = weights[subject],
    amt = 15 * WT
  )

doses <- subjects |>
  mutate(
    time = 0, evid = 1L, cmt = "central",
    rate = amt / infusion_h, ss = 1L, ii = tau
  )

# Observation rows reference the ODE state, never the observable Cc.
obs <- subjects |>
  tidyr::expand_grid(time = seq(0, tau, by = 0.05)) |>
  mutate(
    evid = 0L, cmt = "central", amt = NA_real_,
    rate = NA_real_, ss = NA_integer_, ii = NA_real_
  )

events <- bind_rows(doses, obs) |>
  arrange(id, time, desc(evid)) |>
  select(id, time, amt, evid, cmt, rate, ss, ii, WT, ECMO_STATUS,
         DIS_INFECT_MILD, group)

stopifnot(length(unique(events$id)) == 4L * n_per_group)

cat(sprintf(
  "Weights: median %.1f kg (range %.1f-%.1f)\n",
  median(weights), min(weights), max(weights)
))
#> Weights: median 10.7 kg (range 1.2-91.7)

Simulation

sim <- rxode2::rxSolve(mod, events = events, keep = c("group", "WT")) |>
  as.data.frame()
#> ℹ parameter labels from comments will be replaced by 'label()'

stopifnot(all(c("Cc", "cl", "vc") %in% names(sim)))

Steady-state profiles by covariate group

Figure 3 of the paper is a visual predictive check against time since the first dose (up to about 8000 h) for the observed regimens, which are not published, so it cannot be redrawn. The plot below shows the steady-state interval for the assumed regimen in each covariate group, on the same logarithmic concentration axis.

sim |>
  filter(!is.na(Cc)) |>
  group_by(group, time) |>
  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() +
  facet_wrap(~group) +
  scale_y_log10() +
  labs(
    x = "Time after dose at steady state (h)",
    y = "Vancomycin concentration (mg/L)",
    caption = "15 mg/kg q6h, 1-h infusion. Median and 5th-95th percentile of individual predictions, 100 subjects per group."
  )

PKNCA validation

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

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

dose_df <- events |>
  filter(evid == 1) |>
  select(id, time, amt, group)

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

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

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

nca_wide <- as.data.frame(nca_res) |>
  select(group, id, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES)

nca_wide |>
  group_by(group) |>
  summarise(
    cmax    = median(cmax),
    cmin    = median(cmin),
    auclast = median(auclast),
    .groups = "drop"
  ) |>
  rename(
    "Group"                      = group,
    "Median Cmax,ss (mg/L)"      = cmax,
    "Median Cmin,ss (mg/L)"      = cmin,
    "Median AUC0-6,ss (mg*h/L)"  = auclast
  ) |>
  knitr::kable(digits = 1, caption = "Steady-state NCA by covariate group, 15 mg/kg q6h.")
Steady-state NCA by covariate group, 15 mg/kg q6h.
Group Median Cmax,ss (mg/L) Median Cmin,ss (mg/L) Median AUC0-6,ss (mg*h/L)
ECMO, mild infection 107.2 95.2 609.0
ECMO, severe infection 76.2 60.0 401.1
No ECMO, mild infection 62.1 45.4 301.3
No ECMO, severe infection 39.7 24.2 188.7

Mass balance: steady-state AUC over one interval equals Dose/CL

For a linear model the steady-state AUC over one dosing interval equals Dose / CL for each subject. Both sides use the same simulated individual clearance, so the only difference is trapezoidal error on the 0.05 h grid and a tight bound is right.

mb <- nca_wide |>
  left_join(
    sim |> group_by(id) |> summarise(cl = first(cl), WT = first(WT), .groups = "drop"),
    by = "id"
  ) |>
  mutate(ratio = auclast / (15 * WT / cl))

tibble(
  Subjects          = nrow(mb),
  `Max |ratio - 1|` = max(abs(mb$ratio - 1))
) |>
  knitr::kable(digits = 6, caption = "PKNCA AUC0-tau at steady state against Dose/CL.")
PKNCA AUC0-tau at steady state against Dose/CL.
Subjects Max |ratio - 1|
400 6e-05

stopifnot(nrow(mb) == 4L * n_per_group, max(abs(mb$ratio - 1)) < 1e-3)

Group contrasts

The median steady-state AUC ratio between groups is the clearance ratio: exp(-0.81) = 0.445 for ECMO and exp(-0.49) = 0.613 for a mild infection, so AUC rises 2.25-fold on ECMO and 1.63-fold for a mild infection. Because the groups share weight draws but not random effects, the comparison is on medians.

auc_med <- nca_wide |>
  group_by(group) |>
  summarise(auc = median(auclast), .groups = "drop")

ref_auc <- auc_med$auc[auc_med$group == "No ECMO, severe infection"]

contrast <- auc_med |>
  mutate(
    ratio_sim      = auc / ref_auc,
    ratio_expected = c(
      "No ECMO, severe infection" = 1,
      "ECMO, severe infection"    = exp(0.81),
      "No ECMO, mild infection"   = exp(0.49),
      "ECMO, mild infection"      = exp(0.81 + 0.49)
    )[group],
    pct_diff = 100 * (ratio_sim / ratio_expected - 1)
  )

contrast |>
  rename(
    "Group"                       = group,
    "Median AUC0-6,ss (mg*h/L)"   = auc,
    "AUC ratio, simulated"        = ratio_sim,
    "AUC ratio, 1/CL multiplier"  = ratio_expected,
    "% difference"                = pct_diff
  ) |>
  knitr::kable(digits = 2, caption = "Median steady-state AUC relative to the reference group.")
Median steady-state AUC relative to the reference group.
Group Median AUC0-6,ss (mg*h/L) AUC ratio, simulated AUC ratio, 1/CL multiplier % difference
ECMO, mild infection 609.00 3.23 3.67 -12.03
ECMO, severe infection 401.12 2.13 2.25 -5.43
No ECMO, mild infection 301.28 1.60 1.63 -2.18
No ECMO, severe infection 188.68 1.00 1.00 0.00

# Centre-based: the group medians are drawn from 100 subjects each with
# omega^2(CL) = 0.26, so the median ratio scatters by roughly 10-15%. A
# mis-signed or mis-transcribed coefficient moves a ratio by 50% or more.
stopifnot(all(abs(contrast$pct_diff) < 30))

Assumptions and deviations

  • Omega scale. Table 4 lists eta Cl = 0.26 and eta V = 0.24 without stating whether these are variances or standard deviations. Phoenix NLME reports the Omega matrix as variances, so they are used as variances (lognormal CV about 54% and 52%). The paper’s %CV column beside them is the relative standard error of the estimate (0.08 / 0.26 = 30.7%), not the between-subject CV.
  • Residual error. Section 3.2 selects a proportional error model and Section 5 describes “a relatively high residual error of 44%”; Table 4’s Stdev0 = 0.44 is therefore the proportional SD. Section 2.2 mentions natural-log transformed concentrations, which the text ties to the plotting of the data; the error model is taken as proportional, as the Results state.
  • ECMO clearance printed as 0.48 L/h. The printed estimates give 0.458 L/h (see the Table 3 check above); the model uses the Table 4 coefficient.
  • Reference weight. The equations print WT/median(WT); the 12.6 kg median is stated in the text and Table 1.
  • ECMO and infection category are subject-level. The paper gives no within-subject timing for ECMO cannulation; diagnosis is the admission diagnosis.
  • Covariates not retained. Surgery status (removed at backward elimination, Table 2) and eGFR (not significant) are not in the model.
  • Dosing regimen. The paper does not report individual regimens; the 15 mg/kg q6h, 1-h infusion regimen used here is illustrative.
  • No published NCA. The paper reports no Cmax, AUC or half-life values, so there is no NCA comparison table; the checks above are against the paper’s own Table 3 arithmetic and against closed forms.