Skip to contents

Model and source

  • Citation: Jeong SH, Jang JH, Lee YB. Is Gender an Important Factor in the Precision Medicine Approach to Levocetirizine? Pharmaceutics. 2024;16(1):146. doi:10.3390/pharmaceutics16010146. The wheal and flare observations used to fit the pharmacodynamic sub-model were taken from Simons FER, Simons KJ. Levocetirizine: pharmacokinetics and pharmacodynamics in children age 6 to 11 years. J Allergy Clin Immunol. 2005;116(2):355-361. doi:10.1016/j.jaci.2005.04.010.
  • Description: Co-linked population PK-PD model for a single oral 5 mg levocetirizine immediate-release tablet in 40 healthy Korean adults (24 men, 16 women) from a two-way crossover bioequivalence study (Jeong 2024). PK is a two-compartment disposition model (central + peripheral1) fed by a chain of five sequential first-order absorption compartments (depot -> transit1 -> transit2 -> transit3 -> transit4 -> central, with distinct rate constants Ka1 to Ka5) preceded by an absorption lag time Tlag; volumes and clearances are apparent (Vc/F, CLc/F, Vp/F, CLp/F). Retained covariates: body surface area (power), serum albumin (power) and female sex (linear fractional decrease) on Vp/F, body surface area (power) on CLp/F, and body weight (power) on the fifth absorption rate constant Ka5, each continuous covariate centred on the cohort value. Residual error is log-additive, i.e. log-normal on the linear concentration scale. PD is two parallel indirect-response (input-inhibition) models for the histamine-induced skin wheal and flare sizes, expressed as a percentage of the pre-dose response (baseline kin / kout = 100), in which plasma levocetirizine inhibits the zero-order production rate through a sigmoid Imax function. The PD parameters were estimated from published mean data in children aged 6 to 11 years (Simons 2005) and carry no inter-individual variability.
  • Article: https://doi.org/10.3390/pharmaceutics16010146
  • Supplement (Supplementary Information 1-8, Tables S1-S8, Figures S1-S6): https://www.mdpi.com/article/10.3390/pharmaceutics16010146/s1

Jeong 2024 asked whether men and women handle a 5 mg levocetirizine tablet differently, and whether any difference matters for the drug’s antihistamine effect. Women reached a higher peak and cleared the drug from plasma faster than men, which the final model attributes to a smaller peripheral volume (Vp/F) in women, even after body surface area and serum albumin are accounted for. Total exposure (AUC) was the same in both sexes, and the simulated wheal and flare suppression differed by only a few percentage points.

The pharmacokinetic model and its covariates come from the authors’ own bioequivalence cohort. The pharmacodynamic model, two parallel indirect-response models for the histamine-induced skin wheal and flare, was fitted to published mean data from children aged 6 to 11 years (Simons 2005) and then linked to the adult PK model. Both layers are in a single model file because the paper’s simulations (Figures 4 and 5, Table 1) drive both PD endpoints from the one PK model.

Population

Forty healthy Korean adults (24 men, 16 women) received a single oral 5 mg levocetirizine tablet (UCB Pharma reference formulation) with 150 mL of water after a fast of more than 10 h. This was the reference period of a randomised, open-label, two-way crossover bioequivalence study with a 7-day washout (Supplementary Information 3). Plasma was sampled at 0, 0.17, 0.33, 0.5, 0.75, 1, 1.25, 1.5, 2, 3, 4, 6, 8, 12, 24 and 36 h post-dose (LLOQ 0.5 ng/mL).

Baseline demographics (Table S1, mean (SD)): age 28.38 (8.30) years; body weight 66.97 (12.59) kg (men 71.58 (12.81), women 60.05 (8.68)); height 168.33 (9.10) cm; body surface area 1.76 (0.20) m^2 (men 1.85 (0.19), women 1.63 (0.13)); serum albumin 4.70 (0.24) g/dL (men 4.79 (0.23), women 4.57 (0.17)). Renal and hepatic function were normal; none of the renal indices survived covariate selection.

The same information is available programmatically via readModelDb("Jeong_2024_levocetirizine")()$population.

Source trace

Every ini() entry in inst/modeldb/specificDrugs/Jeong_2024_levocetirizine.R carries an in-file comment naming its source. They are collected here for review.

Equation / parameter Value Source location
Two-compartment disposition + five sequential first-order absorption compartments + Tlag on the dosing depot n/a Results 3.2; Table S5 model 02-05 (selected); Figure S2 (topology)
d/dt(depot) … d/dt(transit4), d/dt(central), d/dt(peripheral1) n/a Supplementary Information 7 (Ka1 = dosing depot to depot 1 … Ka5 = depot 4 to central); Figure S2
lka1 9.87 1/h Table S7, tv Ka1 (SE 1.53; bootstrap 9.86 [7.07-12.74])
lka2 28.41 1/h Table S7, tv Ka2 (SE 5.39; bootstrap 28.60 [20.87-40.82])
lka3 11.53 1/h Table S7, tv Ka3 (SE 1.87; bootstrap 11.45 [7.91-15.35])
lka4 11.29 1/h Table S7, tv Ka4 (SE 1.95; bootstrap 10.71 [8.29-15.67])
lka5 0.90 1/h Table S7, tv Ka5 (SE 0.15; bootstrap 0.90 [0.62-1.18])
ltlag 0.08 h Table S7, tv Tlag (SE 0.01; bootstrap 0.08 [0.07-0.10])
lvc 1.07 L Table S7, tv Vc/F (SE 0.28; bootstrap 1.03 [0.63-1.44])
lcl 2.94 L/h Table S7, tv CLc/F (SE 0.07; bootstrap 2.95 [2.79-3.07])
lvp 28.12 L Table S7, tv Vp/F (SE 1.07; bootstrap 28.02 [26.44-30.70])
lq 13.80 L/h Table S7, tv CLp/F (SE 2.24; bootstrap 13.88 [9.57-18.16])
e_wt_ka5, applied as Ka5 = tvKa5 * (WT / 66.97)^0.69 0.69 Table S7, dKa5dWeight; form from Supplementary Information 7 Eq. 5
e_bsa_vp, applied as (BSA / 1.76)^1.20 on Vp/F 1.20 Table S7, dVp/FdBSA; form from Eq. 9
e_alb_vp, applied as (ALB / 47.0 g/L)^(-0.79) on Vp/F -0.79 Table S7, dVp/FdAlbumin; form from Eq. 9
e_sexf_vp, applied as (1 - 0.18 * SEXF) on Vp/F -0.18 Table S7, dVp/FdGender; linear form from Eq. 9
e_bsa_q, applied as (BSA / 1.76)^1.94 on CLp/F 1.94 Table S7, dCLp/FdBSA; form from Eq. 10
etalvc, etalcl, etaltlag 0.89, 0.03, 0.37 Table S7, omega^2 Vc/F, CLc/F, Tlag
etalka1 … etalka5 2.09, 4.14, 2.36, 2.11, 0.02 Table S7, omega^2 Ka1 … Ka5
etalvp, etalq fixed(0) Table S7, omega^2 Vp/F and CLp/F printed as 0.00 (bootstrap 0.00-0.00); retained per Table S5 models 02-05-02-03 and -04
expSd (log-additive residual) 0.11 Table S7, epsilon; model form from Table S5 model 02-05-02
dR/dt = Kin * (1 - Imax * C^gamma / (IC50^gamma + C^gamma)) - Kout * R n/a Supplementary Information 8 Eq. 11; Figure S2
lkin_wheal, lkout_wheal, imax_wheal, lic50_wheal, lhill_wheal 87.9 %/h, 0.879 1/h, 0.92, 45.5 ng/mL, 1.11 Table S8, wheal column
lkin_flare, lkout_flare, imax_flare, lic50_flare, lhill_flare 99.9 %/h, 0.999 1/h, 0.999, 10.5 ng/mL, 2.51 Table S8, flare column
addSd_wheal, addSd_flare fixed(0) Table S8 reports no residual error for the PD model
centring values 66.97 kg, 1.76 m^2, 4.70 g/dL n/a Table S1 cohort means (the medians named in Supplementary Information 7 are not reported)

Virtual cohort

The individual data are not published, so the checks below use a virtual cohort of 100 men and 100 women, sampled on the paper’s 16-point schedule. Height, weight and albumin are drawn per sex from the Table S1 means and SDs; BSA is then computed with the Mosteller formula the paper used (Supplementary Information 1).

# set.seed() fixes the covariate draws. rxSetSeed() fixes rxode2's random
# effects for a given solver thread count but not across thread counts, so
# every assertion on this stochastic cohort below uses a robust centre with a
# generous envelope; the tight assertions run on typical-value solves.
set.seed(20240121)
rxode2::rxSetSeed(20240121)

sampling_times <- c(0, 0.17, 0.33, 0.5, 0.75, 1, 1.25, 1.5, 2, 3, 4, 6, 8, 12, 24, 36)

make_cohort <- function(n, sexf, ht, ht_sd, wt, wt_sd, wt_min, alb, alb_sd,
                        label, id_offset = 0L) {
  subjects <- tibble(
    id = id_offset + seq_len(n),
    SEXF = sexf,
    HT = stats::rnorm(n, ht, ht_sd),
    WT = pmax(stats::rnorm(n, wt, wt_sd), wt_min),
    # Table S1 albumin is in g/dL; the model's ALB column is g/L.
    ALB = 10 * stats::rnorm(n, alb, alb_sd),
    treatment = label
  ) |>
    mutate(BSA = sqrt(HT * WT / 3600))
  doses <- subjects |>
    mutate(time = 0, amt = 5, evid = 1L, cmt = "depot", dvid = NA_integer_)
  obs <- subjects |>
    tidyr::expand_grid(time = sampling_times) |>
    mutate(amt = NA_real_, evid = 0L, cmt = "central", dvid = 1L)
  bind_rows(doses, obs) |>
    arrange(id, time, desc(evid))
}

events <- bind_rows(
  make_cohort(100, 0, 173.41, 7.50, 71.58, 12.81, 50, 4.79, 0.23, "Male", 0L),
  make_cohort(100, 1, 160.70, 5.08, 60.05, 8.68, 45, 4.57, 0.17, "Female", 100L)
)
stopifnot(!anyDuplicated(events[, c("id", "time", "evid")]))

events |>
  distinct(id, treatment, WT, BSA, ALB) |>
  group_by(treatment) |>
  summarise(across(c(WT, BSA, ALB), \(x) round(mean(x), 2)), .groups = "drop") |>
  dplyr::rename(
    Sex = treatment, "Mean WT (kg)" = WT, "Mean BSA (m^2)" = BSA,
    "Mean ALB (g/L)" = ALB
  ) |>
  knitr::kable(caption = "Virtual-cohort covariate means (Table S1: men 71.58 kg, 1.85 m^2, 47.9 g/L; women 60.05 kg, 1.63 m^2, 45.7 g/L).")
Virtual-cohort covariate means (Table S1: men 71.58 kg, 1.85 m^2, 47.9 g/L; women 60.05 kg, 1.63 m^2, 45.7 g/L).
Sex Mean WT (kg) Mean BSA (m^2) Mean ALB (g/L)
Female 59.99 1.63 45.83
Male 70.51 1.84 48.12

Simulation

The model has three endpoints (plasma Cc, wheal and flare), so observation rows carry dvid = 1L to tie them to the first endpoint; the wheal and flare states are returned as columns at the same rows.

mod <- readModelDb("Jeong_2024_levocetirizine")

sim <- rxode2::rxSolve(
  mod,
  events = events,
  keep = c("treatment"),
  useLinCmt = FALSE
) |>
  as.data.frame()
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalvp', 'etalq'

Typical-value solves use the per-sex mean covariates from Table S1, as the paper did for Table 1 and Figures 4D-4F and 5D-5F (“applying the mean biochemical parameter values in each gender”).

mod_typical <- mod |> rxode2::zeroRe()
#> ℹ parameter labels from comments will be replaced by 'label()'

sex_covariates <- tibble(
  treatment = c("Male", "Female"),
  SEXF = c(0, 1),
  WT = c(71.58, 60.05),
  BSA = c(1.85, 1.63),
  ALB = c(47.9, 45.7)
)

typical_solve <- function(times, n_doses = 1L, tau = 24) {
  rows <- bind_rows(
    tibble(time = 0, amt = 5, evid = 1L, cmt = "depot",
           ii = if (n_doses > 1L) tau else 0, addl = n_doses - 1L,
           dvid = NA_integer_),
    tibble(time = times, amt = NA_real_, evid = 0L, cmt = "central",
           ii = 0, addl = 0L, dvid = 1L)
  )
  ev <- dplyr::cross_join(sex_covariates |> mutate(id = row_number()), rows) |>
    arrange(id, time, desc(evid))
  rxode2::rxSolve(mod_typical, ev, keep = "treatment", useLinCmt = FALSE) |>
    as.data.frame()
}

typical_sd <- typical_solve(seq(0, 36, by = 0.01))
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etaltlag', 'etalvp', 'etalq', 'etalka1', 'etalka2', 'etalka3', 'etalka4', 'etalka5'
#> Warning: multi-subject simulation without without 'omega'
typical_md <- typical_solve(seq(0, 240, by = 0.01), n_doses = 10L)
#> ℹ omega/sigma items treated as zero: 'etalvc', 'etalcl', 'etaltlag', 'etalvp', 'etalq', 'etalka1', 'etalka2', 'etalka3', 'etalka4', 'etalka5'
#> Warning: multi-subject simulation without without 'omega'

Replicate published figures

# Replicates Figure 1A of Jeong 2024: mean (SD) plasma concentration by sex.
sim |>
  group_by(treatment, time) |>
  summarise(mean = mean(Cc), sd = sd(Cc), .groups = "drop") |>
  ggplot(aes(time, mean, colour = treatment)) +
  geom_line(linewidth = 0.8) +
  geom_errorbar(aes(ymin = mean, ymax = mean + sd), width = 0.4) +
  labs(
    x = "Time after dose (h)", y = "Levocetirizine plasma concentration (ng/mL)",
    colour = NULL,
    title = "Figure 1A - simulated mean (SD) concentration by sex",
    caption = "100 virtual subjects per sex. Replicates Figure 1A of Jeong 2024."
  )

# Replicates Figure 2B-2C of Jeong 2024: prediction percentiles by sex on a
# log scale (individual predictions; residual error not added).
sim |>
  filter(time > 0) |>
  group_by(treatment, time) |>
  summarise(
    Q025 = quantile(Cc, 0.025), Q50 = quantile(Cc, 0.5),
    Q975 = quantile(Cc, 0.975), .groups = "drop"
  ) |>
  ggplot(aes(time, Q50, colour = treatment, fill = treatment)) +
  geom_ribbon(aes(ymin = Q025, ymax = Q975), alpha = 0.2, colour = NA) +
  geom_line(linewidth = 0.8) +
  scale_y_log10() +
  labs(
    x = "Time after dose (h)", y = "Levocetirizine plasma concentration (ng/mL)",
    colour = NULL, fill = NULL,
    title = "Figure 2 - 2.5th, 50th and 97.5th percentiles by sex",
    caption = "Replicates the VPC panels of Figure 2B-2C of Jeong 2024."
  )
#> Warning in scale_y_log10(): log-10 transformation introduced infinite values.

# Replicates Figure 4D-4F of Jeong 2024 as time courses: typical-value plasma
# concentration, wheal and flare after a single 5 mg dose.
typical_sd |>
  select(time, treatment, Cc, wheal, flare) |>
  pivot_longer(c(Cc, wheal, flare), names_to = "output", values_to = "value") |>
  mutate(output = factor(
    output, c("Cc", "wheal", "flare"),
    c("Plasma levocetirizine (ng/mL)", "Wheal size (% of baseline)",
      "Flare size (% of baseline)")
  )) |>
  ggplot(aes(time, value, colour = treatment)) +
  geom_line(linewidth = 0.8) +
  facet_wrap(~output, scales = "free_y") +
  labs(
    x = "Time after dose (h)", y = NULL, colour = NULL,
    title = "Figure 4 - single 5 mg dose, typical male and female",
    caption = "Per-sex mean covariates from Table S1. Replicates Figure 4A-4C of Jeong 2024 (mean-covariate curves)."
  )

# Replicates Figure 5A-5C of Jeong 2024: 5 mg once daily for 10 days.
typical_md |>
  select(time, treatment, Cc, wheal, flare) |>
  pivot_longer(c(Cc, wheal, flare), names_to = "output", values_to = "value") |>
  mutate(output = factor(
    output, c("Cc", "wheal", "flare"),
    c("Plasma levocetirizine (ng/mL)", "Wheal size (% of baseline)",
      "Flare size (% of baseline)")
  )) |>
  ggplot(aes(time, value, colour = treatment)) +
  geom_line(linewidth = 0.6) +
  facet_wrap(~output, scales = "free_y", ncol = 1) +
  labs(
    x = "Time after first dose (h)", y = NULL, colour = NULL,
    title = "Figure 5 - 5 mg every 24 h, typical male and female",
    caption = "Replicates Figure 5A-5C of Jeong 2024 (mean-covariate curves)."
  )

Figure 3 of the paper (the PD fit to the paediatric data of Simons 2005) is not replicated: it was produced with the children’s concentrations, which are not part of this model.

Table 1 of the paper

Table 1 reports typical-value simulations at the per-sex mean covariates, for a single dose (0-36 h) and at steady state (216-240 h, once daily). The rows are recomputed below with the paper’s definitions: the minimum is taken after the peak (footnote a), the wheal and flare maxima after the maximum effect (footnote b), the fluctuation is maximum / minimum (footnote d), and the single-dose AUEC is the area between the 100% baseline and the response curve.

window_metrics <- function(d) {
  d <- arrange(d, time)
  trap <- function(y) sum(diff(d$time) * (head(y, -1) + tail(y, -1)) / 2)
  dur <- diff(range(d$time))
  after <- function(y, i) y[i:length(y)]
  i_c <- which.max(d$Cc)
  i_w <- which.min(d$wheal)
  i_f <- which.min(d$flare)
  tibble(
    metric = c(
      "Cc maximum", "Cc minimum", "Cc mean", "Cc AUCall",
      "Wheal maximum", "Wheal minimum", "Wheal mean", "Wheal AUEC",
      "Flare maximum", "Flare minimum", "Flare mean", "Flare AUEC"
    ),
    value = c(
      max(d$Cc), min(after(d$Cc, i_c)), trap(d$Cc) / dur, trap(d$Cc),
      max(after(d$wheal, i_w)), min(d$wheal), trap(d$wheal) / dur,
      trap(100 - d$wheal),
      max(after(d$flare, i_f)), min(d$flare), trap(d$flare) / dur,
      trap(100 - d$flare)
    )
  )
}

simulated_t1 <- bind_rows(
  typical_sd |> group_by(treatment) |>
    reframe(window_metrics(pick(everything()))) |> mutate(period = "Single dose"),
  typical_md |> filter(time >= 216) |> group_by(treatment) |>
    reframe(window_metrics(pick(everything()))) |> mutate(period = "Steady state")
)

# Jeong 2024 Table 1. The steady-state AUEC rows are omitted: their values
# (2431.14 and 3517.14 h*% in men) are inconsistent with a 24 h window around
# the reported steady-state means, so the integration window is unclear.
published_t1 <- tibble::tribble(
  ~period, ~metric, ~Male, ~Female,
  "Single dose", "Cc maximum", 212.00, 220.52,
  "Single dose", "Cc minimum", 6.53, 4.39,
  "Single dose", "Cc mean", 45.15, 46.09,
  "Single dose", "Cc AUCall", 1624.90, 1659.09,
  "Single dose", "Wheal maximum", 89.45, 92.77,
  "Single dose", "Wheal minimum", 33.81, 31.87,
  "Single dose", "Wheal mean", 63.02, 64.30,
  "Single dose", "Wheal AUEC", 1330.75, 1284.66,
  "Single dose", "Flare maximum", 72.56, 87.14,
  "Single dose", "Flare minimum", 0.77, 0.65,
  "Single dose", "Flare mean", 20.31, 25.93,
  "Single dose", "Flare AUEC", 2867.43, 2665.73,
  "Steady state", "Cc maximum", 231.37, 235.37,
  "Steady state", "Cc minimum", 20.69, 16.19,
  "Steady state", "Cc mean", 70.73, 70.71,
  "Steady state", "Cc AUCall", 1702.49, 1702.49,
  "Steady state", "Wheal maximum", 70.81, 75.54,
  "Steady state", "Wheal minimum", 30.17, 29.39,
  "Steady state", "Wheal mean", 48.79, 50.22,
  "Steady state", "Flare maximum", 13.11, 21.10,
  "Steady state", "Flare minimum", 0.44, 0.43,
  "Steady state", "Flare mean", 3.49, 5.05
) |>
  pivot_longer(c(Male, Female), names_to = "treatment", values_to = "published")

t1 <- inner_join(published_t1, simulated_t1, by = c("period", "metric", "treatment")) |>
  mutate(pct_diff = 100 * (value - published) / published)
stopifnot(nrow(t1) == nrow(published_t1))

t1 |>
  mutate(across(c(published, value, pct_diff), \(x) round(x, 2))) |>
  select(period, metric, treatment, published, value, pct_diff) |>
  dplyr::rename(
    Period = period, Quantity = metric, Sex = treatment,
    "Jeong 2024 Table 1" = published, Simulated = value, "% difference" = pct_diff
  ) |>
  knitr::kable(caption = "Typical-value reproduction of Jeong 2024 Table 1 (concentrations in ng/mL, AUC in h*ng/mL, PD in % of baseline, AUEC in h*%).")
Typical-value reproduction of Jeong 2024 Table 1 (concentrations in ng/mL, AUC in hng/mL, PD in % of baseline, AUEC in h%).
Period Quantity Sex Jeong 2024 Table 1 Simulated % difference
Single dose Cc maximum Male 212.00 204.61 -3.49
Single dose Cc maximum Female 220.52 228.25 3.51
Single dose Cc minimum Male 6.53 7.02 7.54
Single dose Cc minimum Female 4.39 3.84 -12.57
Single dose Cc mean Male 45.15 44.85 -0.66
Single dose Cc mean Female 46.09 46.24 0.32
Single dose Cc AUCall Male 1624.90 1614.69 -0.63
Single dose Cc AUCall Female 1659.09 1664.54 0.33
Single dose Wheal maximum Male 89.45 88.71 -0.83
Single dose Wheal maximum Female 92.77 93.65 0.95
Single dose Wheal minimum Male 33.81 34.41 1.78
Single dose Wheal minimum Female 31.87 31.14 -2.28
Single dose Wheal mean Male 63.02 62.77 -0.40
Single dose Wheal mean Female 64.30 64.96 1.03
Single dose Wheal AUEC Male 1330.75 1340.30 0.72
Single dose Wheal AUEC Female 1284.66 1261.40 -1.81
Single dose Flare maximum Male 72.56 68.98 -4.94
Single dose Flare maximum Female 87.14 90.28 3.61
Single dose Flare minimum Male 0.77 0.79 2.65
Single dose Flare minimum Female 0.65 0.63 -2.71
Single dose Flare mean Male 20.31 19.40 -4.46
Single dose Flare mean Female 25.93 28.06 8.21
Single dose Flare AUEC Male 2867.43 2901.48 1.19
Single dose Flare AUEC Female 2665.73 2589.91 -2.84
Steady state Cc maximum Male 231.37 225.04 -2.74
Steady state Cc maximum Female 235.37 241.89 2.77
Steady state Cc minimum Male 20.69 21.78 5.26
Steady state Cc minimum Female 16.19 14.89 -8.01
Steady state Cc mean Male 70.73 70.86 0.19
Steady state Cc mean Female 70.71 70.86 0.21
Steady state Cc AUCall Male 1702.49 1700.68 -0.11
Steady state Cc AUCall Female 1702.49 1700.68 -0.11
Steady state Wheal maximum Male 70.81 69.74 -1.51
Steady state Wheal maximum Female 75.54 77.04 1.99
Steady state Wheal minimum Male 30.17 30.51 1.14
Steady state Wheal minimum Female 29.39 28.96 -1.47
Steady state Wheal mean Male 48.79 48.36 -0.89
Steady state Wheal mean Female 50.22 50.80 1.15
Steady state Flare maximum Male 13.11 11.79 -10.10
Steady state Flare maximum Female 21.10 24.56 16.42
Steady state Flare minimum Male 0.44 0.44 -0.42
Steady state Flare minimum Female 0.43 0.43 0.67
Steady state Flare mean Male 3.49 3.20 -8.19
Steady state Flare mean Female 5.05 5.76 14.14

These are deterministic solves, so the gates below are tight. The structural quantities (peak, mean, AUC, the whole wheal response) agree within 5%. The remaining gap, largest for the post-peak minimum and the low flare values, comes from the unreported covariate medians (see Assumptions): the medians move the male and female Vp/F, and so the terminal phase, in opposite directions. A mis-transcribed volume, clearance or IC50 would move these rows by tens of percent.

gate <- t1 |>
  mutate(bound = case_when(
    metric == "Cc minimum" ~ 15,
    startsWith(metric, "Flare") ~ 20,
    TRUE ~ 5
  ))
stopifnot(all(abs(gate$pct_diff) < gate$bound))

# Steady-state AUC over a dosing interval equals Dose / (CL/F) exactly for a
# linear model: 5 mg / 2.94 L/h = 1700.68 h*ng/mL. Both sides use the same
# parameters, so the bound is numerical error only.
auc_ss <- t1 |> filter(period == "Steady state", metric == "Cc AUCall") |> pull(value)
stopifnot(all(abs(auc_ss / (5000 / 2.94) - 1) < 1e-3))

# Drug-free PD baseline is kin / kout = 100 for both endpoints.
stopifnot(
  isTRUE(all.equal(typical_sd$wheal[typical_sd$time == 0], c(100, 100))),
  isTRUE(all.equal(typical_sd$flare[typical_sd$time == 0], c(100, 100)))
)

PKNCA validation

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

conc_obj <- PKNCA::PKNCAconc(sim_nca, Cc ~ time | treatment + id)

dose_df <- events |>
  dplyr::filter(evid == 1) |>
  dplyr::select(id, time, amt, treatment)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | treatment + id)

intervals <- data.frame(
  start = 0,
  end = Inf,
  cmax = TRUE,
  tmax = TRUE,
  auclast = TRUE,
  aucinf.obs = TRUE,
  half.life = TRUE
)

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

Comparison against published NCA

# Jeong 2024 Table S2, observed mean by sex (AUCall over 0-36 h is auclast).
published <- tibble::tribble(
  ~treatment, ~cmax, ~tmax, ~auclast, ~aucinf.obs, ~half.life,
  "Male", 203.89, 1.11, 1647.94, 1751.08, 8.73,
  "Female", 230.84, 1.09, 1783.22, 1835.06, 7.05
)

cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = nca_res,
  reference = published,
  by = "treatment",
  units = c(
    cmax = "ng/mL", auclast = "ng*h/mL", aucinf.obs = "ng*h/mL",
    tmax = "h", half.life = "h"
  ),
  tolerance_pct = 20
)

knitr::kable(
  cmp,
  caption = "Simulated (median of 100 subjects per sex) vs. Jeong 2024 Table S2 (observed mean). * differs from reference by >20%.",
  align = c("l", "l", "r", "r", "r")
)
Simulated (median of 100 subjects per sex) vs. Jeong 2024 Table S2 (observed mean). * differs from reference by >20%.
NCA parameter treatment Reference Simulated % diff
Cmax (ng/mL) Male 204 161 -20.9%*
Cmax (ng/mL) Female 231 193 -16.4%
Tmax (h) Male 1.11 1.5 +35.1%*
Tmax (h) Female 1.09 1.25 +14.7%
AUC0-∞ (obs) (ng*h/mL) Male 1750 1720 -1.9%
AUC0-∞ (obs) (ng*h/mL) Female 1840 1730 -5.5%
AUClast (ng*h/mL) Male 1650 1620 -1.4%
AUClast (ng*h/mL) Female 1780 1680 -5.6%
t½ (h) Male 8.73 8.53 -2.3%
t½ (h) Female 7.05 6.87 -2.6%

AUC and half-life agree with the observed means, and the model reproduces the sex ordering the paper reports: women have the higher Cmax and the shorter half-life.

Cmax and Tmax in the stochastic cohort. The simulated Cmax is about 15-20% below the observed mean and Tmax is later, although the typical-value Cmax reproduces Table 1 within 4%. The cause is the published inter-individual variability on the four fast absorption rate constants (omega^2 2.09 to 4.14, i.e. SDs of 1.4 to 2.0 on the log scale) and on Vc/F (0.89). A sizeable minority of simulated subjects therefore absorb slowly or have a large central volume, which pulls the cohort peak down and spreads it out in time. The paper’s own visual predictive check shows the same thing: in Figure 2A the predicted 2.5th percentile over the first hours lies well below every observation. The simulated percentiles of the virtual cohort agree with the prediction lines read from Figure 2A (on the natural-log scale, a median of about 4.0 at 12 h and 1.8 at 36 h, and a 2.5th percentile of about 0.3 at 36 h). This is a property of the published model and is reproduced as published; no parameter was changed.

# Stochastic cohort: assert the centre of the exposure and elimination
# quantities against the published means, with an envelope wide enough for any
# cohort the model can draw. A wrong clearance, volume or unit moves these by
# far more than 25%. Cmax and Tmax are not gated, for the reason given above.
nca_summary <- as.data.frame(nca_res$result) |>
  filter(PPTESTCD %in% c("aucinf.obs", "half.life")) |>
  group_by(treatment, PPTESTCD) |>
  summarise(median = median(PPORRES, na.rm = TRUE), .groups = "drop") |>
  left_join(
    published |>
      select(treatment, aucinf.obs, half.life) |>
      pivot_longer(-treatment, names_to = "PPTESTCD", values_to = "ref"),
    by = c("treatment", "PPTESTCD")
  )
stopifnot(nrow(nca_summary) == 4, all(abs(nca_summary$median / nca_summary$ref - 1) < 0.25))

Assumptions and deviations

  • Covariate centring values are the cohort means, not the medians. Supplementary Information 7 centres weight, BSA and albumin on “the median … values in the observed population”, but Table S1 reports only means and SDs. The model uses the Table S1 means (66.97 kg, 1.76 m^2, 4.70 g/dL). With these centres the typical-value reproduction of Table 1 is within 5% for Cmax, AUC, mean concentration and all wheal quantities. The post-peak minimum (up to 13% off) and the low-valued flare rows (up to 16% off) are the most sensitive to the true medians. No centre was tuned to match the table.
  • Albumin is supplied in g/L. The paper reports albumin in g/dL; the canonical ALB column is g/L, so the centre is 47.0 g/L. The ratio form makes the effect unit-invariant once the centre is converted as well.
  • The weight covariate acts on Ka5, not Ka4. Table S7 footnote c says “the quantitative correlation value between Ka4 and body weight”, but the printed equation (Supplementary Information 7 Eq. 5), the covariate table (Table S6 model 5), the Results and the Conclusions all place it on Ka5. The model follows the equation.
  • omega^2 for Vp/F and CLp/F is encoded as fixed(0). Table S7 prints both as 0.00 (bootstrap 0.00-0.00), while Table S5 shows that removing either worsened the fit by more than 400 units of -2LL. The variances are therefore non-zero but below the two-decimal reporting precision; they are carried as explicit zeros rather than invented.
  • The residual-error model is “log additive” in Phoenix NLME, i.e. additive on the natural-log scale, which is log-normal on the linear concentration scale and maps to ~ lnorm(expSd). Table S7 reports epsilon = 0.11; Phoenix reports this parameter as a standard deviation.
  • The PD sub-model has no IIV and no residual error. Table S8 gives point estimates only; addSd_wheal and addSd_flare are fixed(0).
  • The PD data are not from this cohort. The wheal and flare responses were taken from Simons 2005 (children aged 6 to 11 years given 5 mg levocetirizine) and fitted with the adult PK structure (Methods 2.4). The authors note that adult PD studies are needed to confirm the parameters (Discussion).
  • Response scale. Supplementary Information 8 calls R the “percent inhibition” of the wheal or flare, but Table 1 and Figures 3-5 plot it as the response size relative to a 100% baseline that falls towards 0 as the antihistamine effect grows, which is what kin / kout = 100 gives. The model follows the plotted scale.
  • The stochastic cohort’s Cmax is lower and its Tmax later than observed. This follows from the large published IIV on Ka1-Ka4 and Vc/F and matches the width of the paper’s own VPC (Figure 2); see the PKNCA section.
  • Steady-state AUEC is not compared. Table 1 reports 2431.14 and 3517.14 h*% (men, wheal and flare) at steady state, which cannot come from a 24 h window around steady-state mean responses of 48.79% and 3.49%; the integration window used is not stated.
  • Virtual-cohort covariates are assumed normal and independent within sex (height, weight, albumin), with BSA computed from height and weight. The paper publishes summary statistics only.