Skip to contents

Model and source

Kaikousidis et al. (2023) modelled serum and brain donepezil concentrations in mice after an oral solution and after a nasal film, in four stages. Each stage was fitted twice, once to the per-time-point mean concentrations and once to the raw concentrations of every animal pooled as one typical subject. The library carries the five models the paper reports with full parameter tables:

stems <- c(
  oral_mean = "Kaikousidis_2023_donepezil_oral_meandata_mouse",
  oral_raw = "Kaikousidis_2023_donepezil_oral_rawdata_mouse",
  in_mean = "Kaikousidis_2023_donepezil_intranasal_meandata_mouse",
  in_raw = "Kaikousidis_2023_donepezil_intranasal_rawdata_mouse",
  combined = "Kaikousidis_2023_donepezil_combined_mouse"
)
mods <- lapply(stems, function(s) rxode2::rxode2(readModelDb(s)))
tibble::tibble(
  Model = unname(stems),
  Source = c(
    "Table 2, Per Os - Mean Data", "Table 2, Per Os - Raw Data",
    "Table 3, IN - Mean Data", "Table 3, IN - Raw Data",
    "Table 4, Total Fit (Per Os + IN raw data)"
  )
) |>
  knitr::kable()
Model Source
Kaikousidis_2023_donepezil_oral_meandata_mouse Table 2, Per Os - Mean Data
Kaikousidis_2023_donepezil_oral_rawdata_mouse Table 2, Per Os - Raw Data
Kaikousidis_2023_donepezil_intranasal_meandata_mouse Table 3, IN - Mean Data
Kaikousidis_2023_donepezil_intranasal_rawdata_mouse Table 3, IN - Raw Data
Kaikousidis_2023_donepezil_combined_mouse Table 4, Total Fit (Per Os + IN raw data)
  • Oral models (Figure 2): first-order absorption from a gut depot into blood (central), a blood-peripheral exchange, a one-way blood-to-brain transfer k_central_brain (paper k23) that depletes blood, and a lumped brain elimination keff_brain (paper k30). The blood volume is fixed to 0.279 L from the paper’s preliminary blood-only fit (Table 1), which makes the brain volume identifiable.
  • Intranasal models (Figure 4): the nasal-film dose in depot_nasal is split into a fraction P1 absorbed into blood, a fraction (1 - P1) * P2 absorbed directly into brain (both zero-order over Tk0 after a lag Tlag), and the remainder (1 - P1) * (1 - P2), which reaches brain after a Savic transit chain (mean transit time Mtt, rate ktr) and a first-order absorption step out of depot_brain (ka). There is no peripheral compartment. k23 and k30 are fixed to the oral values of the matching data representation.
  • Combined model (Section 3.3): both sub-models fitted jointly to the raw data, sharing the blood elimination, blood-to-brain and brain elimination rate constants, with route-specific apparent volumes and residual errors selected by ROUTE_ORAL (1 = oral, 0 = nasal film). The authors present it as a check on the robustness of the staged fits and did not refine it further.

None of the models estimates between-animal variability: each is a typical-value model with residual error only.

Population

Eight-week-old C57BL/6J mice (Section 2.1) received either 0.2 mL of a 1.25 mg/mL donepezil solution by gastric gavage (0.25 mg, 10 mg/kg) or one nasal film containing 0.1 mg donepezil (4 mg/kg), which implies a body weight of about 25 g. Serum and perfused brain were sampled destructively at 5-360 min after the oral dose (n >= 5 per time point) and at 5-120 min after the nasal film (n = 3-5 per time point). Sex and the total number of animals are not reported. Concentrations below the LOQ (serum 0.043 ug/mL, brain 0.038 ug/mL) were set to LOQ/2 for fitting.

The same information is available programmatically, e.g. readModelDb("Kaikousidis_2023_donepezil_combined_mouse")()$population.

Source trace

Every ini() value carries an in-file comment naming its table row. The table below collects the typical values of all five models, back-transformed from the packaged files, next to the paper symbol and source table.

back_transform <- function(name, est) {
  dplyr::case_when(
    grepl("^logit", name) ~ 1 / (1 + exp(-est)),
    grepl("Sd", name) ~ est^2,
    grepl("^l", name) ~ exp(est),
    TRUE ~ est
  )
}
symbols <- tibble::tribble(
  ~name, ~symbol,
  "lka", "k12 (1/min)",
  "lkel", "k20 (1/min)",
  "lk_central_brain", "k23 (1/min)",
  "lk12", "k24 (1/min)",
  "lk21", "k42 (1/min)",
  "lkeff_brain", "k30 (1/min)",
  "lvc", "Vblood / V1 (L)",
  "lvc_oral", "Vblood, oral (L)",
  "lvc_intranasal", "Vblood_IN (L)",
  "lv_brain", "Vbrain / V2 (L)",
  "lv_brain_oral", "Vbrain_PO (L)",
  "lv_brain_intranasal", "Vbrain_IN (L)",
  "ltlag", "Tlag (min)",
  "ld1", "Tk0 (min)",
  "lmtt", "Mtt (min)",
  "lktr", "ktr (1/min)",
  "logitfrel", "P1",
  "logitfrel2", "P2",
  "lka_brain", "ka (1/min)",
  "propSd", "b1 (variance)",
  "propSd_Cbrain", "b2 (variance)",
  "addSd", "a1 (variance)",
  "addSd_Cbrain", "a2 (variance)",
  "addSd_oral", "a1 (variance)",
  "addSd_Cbrain_oral", "a2 (variance)",
  "addSd_intranasal", "a3 (variance)",
  "propSd_Cbrain_intranasal", "b1 (variance)"
)
trace <- dplyr::bind_rows(lapply(names(mods), function(m) {
  mods[[m]]$iniDf |>
    dplyr::filter(!is.na(ntheta)) |>
    dplyr::transmute(model = m, name, value = signif(back_transform(name, est), 3),
                     fixed = fix)
})) |>
  dplyr::mutate(value = ifelse(fixed, paste0(value, " (fixed)"), as.character(value))) |>
  dplyr::select(-fixed) |>
  tidyr::pivot_wider(names_from = model, values_from = value, values_fill = "") |>
  dplyr::left_join(symbols, by = "name") |>
  dplyr::relocate(symbol, .after = name)
stopifnot(!anyNA(trace$symbol))
trace |>
  dplyr::rename(
    "Parameter" = name, "Paper symbol" = symbol,
    "Oral, mean (Table 2)" = oral_mean, "Oral, raw (Table 2)" = oral_raw,
    "IN, mean (Table 3)" = in_mean, "IN, raw (Table 3)" = in_raw,
    "Combined (Table 4)" = combined
  ) |>
  knitr::kable(caption = "Typical values on the paper's scale. Residual-error rows show the tabulated variance (the model stores its square root).")
Typical values on the paper’s scale. Residual-error rows show the tabulated variance (the model stores its square root).
Parameter Paper symbol Oral, mean (Table 2) Oral, raw (Table 2) IN, mean (Table 3) IN, raw (Table 3) Combined (Table 4)
lka k12 (1/min) 0.059 0.065 0.065
lkel k20 (1/min) 0.012 0.017 0.046 0.05 0.039
lk_central_brain k23 (1/min) 0.014 0.008 0.014 (fixed) 0.008 (fixed) 0.011
lk12 k24 (1/min) 0.034 0.036 0.018
lk21 k42 (1/min) 0.011 0.01 0.031
lkeff_brain k30 (1/min) 0.325 0.369 0.325 (fixed) 0.369 (fixed) 0.276
lvc Vblood / V1 (L) 0.279 (fixed) 0.279 (fixed) 0.021 0.027
lv_brain Vbrain / V2 (L) 0.018 0.009 0.005 0.003
propSd b1 (variance) 0.012
propSd_Cbrain b2 (variance) 0.062 0.043 0.238
addSd a1 (variance) 0.016 0.016 0.082
addSd_Cbrain a2 (variance) 0.004
ltlag Tlag (min) 4.06 3.72 3.59
ld1 Tk0 (min) 11.3 11.6 11.4
lmtt Mtt (min) 56 60.9 53.8
lktr ktr (1/min) 0.171 0.183 0.205
logitfrel P1 0.543 0.683 0.557
logitfrel2 P2 0.3 0.276 0.306
lka_brain ka (1/min) 0.173 0.566 0.127
lvc_oral Vblood, oral (L) 0.279 (fixed)
lv_brain_oral Vbrain_PO (L) 0.015
lvc_intranasal Vblood_IN (L) 0.024
lv_brain_intranasal Vbrain_IN (L) 0.005
addSd_oral a1 (variance) 0.017
addSd_Cbrain_oral a2 (variance) 0.004
addSd_intranasal a3 (variance) 0.084
propSd_Cbrain_intranasal b1 (variance) 0.236
Equation Source location
Oral structure: depot -> blood (k12), blood <-> peripheral (k24, k42), blood -> brain (k23), elimination from blood (k20) and brain (k30) Figure 2; Section 3.1
Intranasal structure: depot -> blood and depot -> brain with Tk0 and Tlag; depot -> transit chain -> Abs -> brain (ktr, ka); blood -> brain (k23); elimination k20, k30 Figure 4; Section 3.2
Dose split F2 = P2, F3 = (1 - P2) * P3, F4 = (1 - P2) * (1 - P3) Section 3.2 (printed as P1, P2 in Tables 3-4)
Transit count N from ktr = (N + 1) / Mtt (8.58 and 10.14 transit compartments) Section 3.2
Residual error: proportional / additive per matrix and fit Sections 3.1-3.3; Tables 2-4

The transit count is not a parameter of the files: it follows from Mtt and ktr. The paper reports N = 8.58 (mean data) and 10.14 (raw data). These equal ktr * Mtt - 1 for both fits, i.e. the Savic relation ktr = (N + 1) / Mtt; the formula as printed, (N - 1) / Mtt, would give 10.58 and 12.14.

ntr <- function(m) {
  th <- mods[[m]]$theta
  exp(th[["lmtt"]]) * exp(th[["lktr"]]) - 1
}
n_check <- c(in_mean = ntr("in_mean"), in_raw = ntr("in_raw"), combined = ntr("combined"))
round(n_check, 2)
#>  in_mean   in_raw combined 
#>     8.58    10.14    10.03
stopifnot(abs(n_check[["in_mean"]] - 8.58) < 0.01, abs(n_check[["in_raw"]] - 10.14) < 0.01)

Simulation

All five models are deterministic at the typical value, so one simulated mouse per route reproduces the paper’s fitted curves. Doses are 0.25 mg into depot (oral) and 0.1 mg into depot_nasal (nasal film). Each model has two observation endpoints (Cc for serum, Cbrain for brain), so the observation rows carry dvid = 1; rxode2 returns both concentrations on every row.

make_events <- function(route, tend, by = 0.5) {
  dose <- if (route == "oral") {
    data.frame(id = 1L, time = 0, amt = 0.25, evid = 1L, cmt = "depot", dvid = NA_integer_)
  } else {
    data.frame(id = 1L, time = 0, amt = 0.1, evid = 1L, cmt = "depot_nasal", dvid = NA_integer_)
  }
  obs <- data.frame(id = 1L, time = seq(0, tend, by = by), amt = 0, evid = 0L,
                    cmt = NA_character_, dvid = 1L)
  ev <- rbind(dose, obs)
  ev$ROUTE_ORAL <- as.integer(route == "oral")
  ev
}
runs <- tibble::tribble(
  ~fit, ~model, ~route, ~tend, ~figure,
  "Oral, mean data", "oral_mean", "oral", 360, "Figure 3A (upper)",
  "Oral, raw data", "oral_raw", "oral", 360, "Figure 3A (lower)",
  "IN, mean data", "in_mean", "intranasal", 120, "Figure 5A (upper)",
  "IN, raw data", "in_raw", "intranasal", 120, "Figure 5A (lower)",
  "Combined, oral", "combined", "oral", 360, "Figure 6 (upper)",
  "Combined, IN", "combined", "intranasal", 120, "Figure 6 (lower)"
)
sim <- dplyr::bind_rows(lapply(seq_len(nrow(runs)), function(i) {
  ev <- make_events(runs$route[i], runs$tend[i])
  s <- as.data.frame(rxode2::rxSolve(mods[[runs$model[i]]], events = ev))
  s$fit <- runs$fit[i]
  s$route <- runs$route[i]
  s
})) |>
  dplyr::mutate(fit = factor(fit, levels = runs$fit))
# Integration noise only: states never go meaningfully negative.
stopifnot(all(sim$Cc >= -1e-6), all(sim$Cbrain >= -1e-6))

Checks of the intranasal input

The intranasal files write the lagged zero-order release and the transit input as explicit functions of the nasal-film dose (podo() / tad() of depot_nasal) rather than with f(), alag() and dur() on three dose records. With two endpoints, that form (a lagged modelled-duration dose alongside the f() = 0 bolus that anchors the transit input) does not solve in rxode2 5.1.8 (“Duration is zero/negative”). The check below solves the intranasal mean-data model against an independent single-endpoint formulation that uses rxode2’s own lag, duration and bioavailability on three dose records, and confirms mass balance with elimination switched off.

th <- mods$in_mean$theta
ref_mod <- rxode2::rxode2({
  ntr <- mtt * ktr - 1
  t_tr <- tad(depot_brain)
  transit_in <- 0
  if (t_tr > 0) transit_in <- (1 - p1) * (1 - p2) * exp(log(podo(depot_brain)) + log(ktr) + ntr * log(ktr * t_tr) - ktr * t_tr - lgamma(ntr + 1))
  d/dt(depot_brain) <- transit_in - ka_brain * depot_brain
  d/dt(central) <- -(kel + k_central_brain) * central
  d/dt(brain) <- ka_brain * depot_brain + k_central_brain * central - keff_brain * brain
  f(depot_brain) <- 0
  f(central) <- p1
  alag(central) <- tlag
  dur(central) <- d1
  f(brain) <- (1 - p1) * p2
  alag(brain) <- tlag
  dur(brain) <- d1
  Cc <- central / vc
  Cbrain <- brain / v_brain
})
ref_par <- c(
  mtt = exp(th[["lmtt"]]), ktr = exp(th[["lktr"]]), tlag = exp(th[["ltlag"]]),
  d1 = exp(th[["ld1"]]), p1 = plogis(th[["logitfrel"]]), p2 = plogis(th[["logitfrel2"]]),
  ka_brain = exp(th[["lka_brain"]]), kel = exp(th[["lkel"]]),
  k_central_brain = exp(th[["lk_central_brain"]]), keff_brain = exp(th[["lkeff_brain"]]),
  vc = exp(th[["lvc"]]), v_brain = exp(th[["lv_brain"]])
)
ref_ev <- rxode2::et(amt = 0.1, cmt = "central", rate = -2) |>
  rxode2::et(amt = 0.1, cmt = "brain", rate = -2) |>
  rxode2::et(amt = 0.1, cmt = "depot_brain") |>
  rxode2::et(seq(0, 120, by = 0.5))
ref <- as.data.frame(rxode2::rxSolve(ref_mod, params = ref_par, events = ref_ev))
pkg <- sim |> dplyr::filter(fit == "IN, mean data")
stopifnot(identical(ref$time, pkg$time))
rel_dev <- c(
  Cc = max(abs(pkg$Cc - ref$Cc)) / max(ref$Cc),
  Cbrain = max(abs(pkg$Cbrain - ref$Cbrain)) / max(ref$Cbrain)
)
signif(rel_dev, 2)
#>      Cc  Cbrain 
#> 2.3e-06 8.1e-06
# Two independent ODE solutions at the default tolerances (rtol 1e-6);
# the observed agreement is about 1e-6.
stopifnot(all(rel_dev < 1e-3))

# Mass balance: with blood and brain elimination switched off the whole
# 0.1 mg dose ends up in the model states.
mb <- as.data.frame(rxode2::rxSolve(
  mods$in_mean, events = make_events("intranasal", 600),
  params = c(lkel = -40, lkeff_brain = -40)
))
mass_end <- with(mb[nrow(mb), ], depot_nasal + depot_brain + central + brain)
mass_end
#> [1] 0.1
stopifnot(abs(mass_end / 0.1 - 1) < 1e-4)

The dose split implied by the fractions, compared with the Discussion’s summary (direct-to-blood about half the dose, delayed transit fraction 25-30%):

split <- dplyr::bind_rows(lapply(c("in_mean", "in_raw", "combined"), function(m) {
  th <- mods[[m]]$theta
  p1 <- plogis(th[["logitfrel"]])
  p2 <- plogis(th[["logitfrel2"]])
  tibble::tibble(model = m, blood = p1, direct_brain = (1 - p1) * p2,
                 transit_brain = (1 - p1) * (1 - p2))
}))
split |>
  dplyr::mutate(dplyr::across(-model, \(x) round(x, 3))) |>
  dplyr::rename("Model" = model, "Direct to blood" = blood,
                "Direct to brain" = direct_brain, "Transit to brain" = transit_brain) |>
  knitr::kable()
Model Direct to blood Direct to brain Transit to brain
in_mean 0.543 0.137 0.320
in_raw 0.683 0.087 0.230
combined 0.557 0.136 0.307
stopifnot(all(abs(rowSums(split[, -1]) - 1) < 1e-12))

Replicate published figures

The points are values the maintainers read off the fitted curves in the paper’s figures (peaks, the intranasal brain trough between the two peaks, and selected late times); the lines are the packaged models.

fig_reads <- tibble::tribble(
  ~fit, ~matrix, ~time, ~value,
  "Oral, mean data", "Blood", 17, 0.355,
  "Oral, mean data", "Brain", 20, 0.215,
  "Oral, mean data", "Blood", 120, 0.065,
  "Oral, mean data", "Brain", 120, 0.045,
  "Oral, mean data", "Blood", 360, 0.022,
  "Oral, raw data", "Blood", 17, 0.355,
  "Oral, raw data", "Brain", 20, 0.215,
  "IN, mean data", "Blood", 15, 1.85,
  "IN, mean data", "Brain", 15, 1.03,
  "IN, mean data", "Brain", 30, 0.27,
  "IN, mean data", "Brain", 60, 0.49,
  "IN, raw data", "Blood", 15, 1.80,
  "IN, raw data", "Brain", 15, 0.97,
  "IN, raw data", "Brain", 30, 0.25,
  "IN, raw data", "Brain", 60, 0.50,
  "Combined, oral", "Blood", 17, 0.355,
  "Combined, oral", "Brain", 20, 0.23,
  "Combined, IN", "Blood", 15, 1.75,
  "Combined, IN", "Brain", 15, 1.00,
  "Combined, IN", "Brain", 60, 0.50
) |>
  dplyr::mutate(fit = factor(fit, levels = runs$fit))

sim_long <- sim |>
  dplyr::select(fit, time, Blood = Cc, Brain = Cbrain) |>
  tidyr::pivot_longer(c(Blood, Brain), names_to = "matrix", values_to = "conc")
ggplot(sim_long, aes(time, conc, colour = matrix)) +
  geom_line(linewidth = 0.8) +
  geom_point(data = fig_reads, aes(time, value), shape = 21, fill = "white", size = 2) +
  facet_wrap(~fit, ncol = 2, scales = "free") +
  scale_colour_manual(values = c(Blood = "#1f77b4", Brain = "#2ca02c")) +
  labs(x = "Time (min)", y = "Donepezil (ug/mL)", colour = NULL,
       caption = "Replicates Figures 3A, 5A and 6 of Kaikousidis 2023 (fitted curves).") +
  theme_bw() +
  theme(legend.position = "bottom")

fig_check <- fig_reads |>
  dplyr::left_join(sim_long, by = c("fit", "matrix", "time")) |>
  dplyr::mutate(pct_diff = 100 * (conc / value - 1))
fig_check |>
  dplyr::mutate(conc = signif(conc, 3), pct_diff = round(pct_diff, 1)) |>
  dplyr::rename("Fit" = fit, "Matrix" = matrix, "Time (min)" = time,
                "Read from figure (ug/mL)" = value, "Model (ug/mL)" = conc,
                "% diff" = pct_diff) |>
  knitr::kable()
Fit Matrix Time (min) Read from figure (ug/mL) Model (ug/mL) % diff
Oral, mean data Blood 17 0.355 0.3340 -5.9
Oral, mean data Brain 20 0.215 0.2190 1.7
Oral, mean data Blood 120 0.065 0.0651 0.2
Oral, mean data Brain 120 0.045 0.0444 -1.2
Oral, mean data Blood 360 0.022 0.0222 0.9
Oral, raw data Blood 17 0.355 0.3470 -2.3
Oral, raw data Brain 20 0.215 0.2300 6.8
IN, mean data Blood 15 1.850 1.8400 -0.8
IN, mean data Brain 15 1.030 0.9810 -4.7
IN, mean data Brain 30 0.270 0.2390 -11.4
IN, mean data Brain 60 0.490 0.4530 -7.6
IN, raw data Blood 15 1.800 1.8100 0.3
IN, raw data Brain 15 0.970 0.9540 -1.6
IN, raw data Brain 30 0.250 0.2330 -6.7
IN, raw data Brain 60 0.500 0.4950 -1.0
Combined, oral Blood 17 0.355 0.3300 -7.1
Combined, oral Brain 20 0.230 0.2370 3.2
Combined, IN Blood 15 1.750 1.7700 1.0
Combined, IN Brain 15 1.000 1.0700 7.1
Combined, IN Brain 60 0.500 0.5460 9.3
# Typical-value solves, no random draws: every model point sits within 12%
# of the curve read off the paper's figure (largest 11%, the shallow brain
# trough at 30 min). Fixing the raw-data intranasal fit's k23/k30 to the
# mean-data oral values instead moves its first brain peak by +34%.
stopifnot(!anyNA(fig_check$conc), all(abs(fig_check$pct_diff) < 15))

The intranasal brain curve shows the paper’s double peak: the direct nose-to-brain input peaks at the end of the zero-order window (Tlag + Tk0, about 15 min) and the transit-delayed input produces the second peak near 60 min.

second_peak <- sim |>
  dplyr::filter(route == "intranasal", time > 30) |>
  dplyr::group_by(fit) |>
  dplyr::summarise(t_second_peak = time[which.max(Cbrain)], .groups = "drop")
second_peak
#> # A tibble: 3 × 2
#>   fit           t_second_peak
#>   <fct>                 <dbl>
#> 1 IN, mean data          57.5
#> 2 IN, raw data           58.5
#> 3 Combined, IN           58.5
stopifnot(all(second_peak$t_second_peak > 45 & second_peak$t_second_peak < 75))

PKNCA validation

Non-compartmental analysis of the typical curves, one PKNCA block per matrix, grouped by fit. The oral curves end at the last oral sample (360 min) and the intranasal curves at the last intranasal sample (120 min).

run_nca <- function(conc_col) {
  conc <- sim |>
    dplyr::filter(!is.na(.data[[conc_col]])) |>
    dplyr::transmute(id = 1L, fit = as.character(fit), time, conc = pmax(.data[[conc_col]], 0))
  stopifnot(all(conc |> dplyr::group_by(fit) |> dplyr::summarise(t0 = min(time)) |> dplyr::pull(t0) == 0))
  dose <- runs |>
    dplyr::transmute(id = 1L, fit, time = 0, amt = ifelse(route == "oral", 0.25, 0.1))
  conc_obj <- PKNCA::PKNCAconc(conc, conc ~ time | fit + id)
  dose_obj <- PKNCA::PKNCAdose(dose, amt ~ time | fit + id)
  intervals <- data.frame(start = 0, end = Inf, cmax = TRUE, tmax = TRUE, auclast = TRUE)
  PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))
}
nca_blood <- run_nca("Cc")
nca_brain <- run_nca("Cbrain")
nca_tab <- dplyr::bind_rows(
  as.data.frame(nca_blood) |> dplyr::mutate(matrix = "Blood"),
  as.data.frame(nca_brain) |> dplyr::mutate(matrix = "Brain")
) |>
  dplyr::select(fit, matrix, PPTESTCD, PPORRES) |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = PPORRES) |>
  dplyr::mutate(fit = factor(fit, levels = runs$fit)) |>
  dplyr::arrange(matrix, fit)
nca_tab |>
  dplyr::mutate(cmax = signif(cmax, 3), auclast = signif(auclast, 3)) |>
  dplyr::rename("Fit" = fit, "Matrix" = matrix, "Cmax (ug/mL)" = cmax,
                "Tmax (min)" = tmax, "AUClast (ug*min/mL)" = auclast) |>
  knitr::kable()
Fit Matrix AUClast (ug*min/mL) Cmax (ug/mL) Tmax (min)
Oral, mean data Blood 29.3 0.334 17.5
Oral, raw data Blood 29.2 0.347 16.5
IN, mean data Blood 43.0 1.860 15.5
IN, raw data Blood 43.5 1.820 15.5
Combined, oral Blood 17.9 0.330 16.0
Combined, IN Blood 46.2 1.770 15.0
Oral, mean data Brain 19.5 0.219 21.5
Oral, raw data Brain 19.6 0.230 20.0
IN, mean data Brain 35.7 0.981 15.0
IN, raw data Brain 37.0 0.954 15.0
Combined, oral Brain 13.3 0.237 20.5
Combined, IN Brain 40.8 1.070 15.0

Comparison against published values

The Discussion reports that after the nasal film both serum and brain reach their maximum at the 15-min sampling time:

in_fits <- c("IN, mean data", "IN, raw data", "Combined, IN")
reference <- data.frame(fit = in_fits, tmax = 15)
cmp <- dplyr::bind_rows(
  nlmixr2lib::ncaComparisonTable(
    simulated = as.data.frame(nca_blood) |> dplyr::filter(fit %in% in_fits),
    reference = reference, by = "fit", params = "tmax", units = c(tmax = "min")
  ) |> dplyr::mutate(Matrix = "Blood", .before = 2),
  nlmixr2lib::ncaComparisonTable(
    simulated = as.data.frame(nca_brain) |> dplyr::filter(fit %in% in_fits),
    reference = reference, by = "fit", params = "tmax", units = c(tmax = "min")
  ) |> dplyr::mutate(Matrix = "Brain", .before = 2)
)
knitr::kable(cmp, caption = "Simulated vs. published Tmax after the nasal film. * differs from reference by >20%.")
Simulated vs. published Tmax after the nasal film. * differs from reference by >20%.
NCA parameter Matrix fit Reference Simulated % diff
Tmax (min) Blood IN, mean data 15 15.5 +3.3%
Tmax (min) Blood IN, raw data 15 15.5 +3.3%
Tmax (min) Blood Combined, IN 15 15 +0.0%
Tmax (min) Brain IN, mean data 15 15 +0.0%
Tmax (min) Brain IN, raw data 15 15 +0.0%
Tmax (min) Brain Combined, IN 15 15 +0.0%
tmax_in <- nca_tab |> dplyr::filter(fit %in% in_fits) |> dplyr::pull(tmax)
stopifnot(all(abs(tmax_in - 15) <= 1.5))

The Discussion also quotes intranasal-to-oral ratios of Cmax and AUC from the non-compartmental analysis of the observed data in the authors’ earlier study (Papakyriakopoulou et al. 2023): Cmax 3.9-fold (blood) and 5.7-fold (brain) higher, and AUC 1.4-fold (blood) and 3.1-fold (brain) higher after the nasal film, at a 2.5-fold lower dose. The model-based ratios are shown for comparison; they are not a gate, because that analysis is of the observed mean data rather than of these fitted curves, and the AUC interval of that analysis is not stated.

ratio <- function(oral_fit, in_fit) {
  nca_tab |>
    dplyr::filter(fit %in% c(oral_fit, in_fit)) |>
    dplyr::mutate(route = ifelse(fit == in_fit, "in", "oral")) |>
    dplyr::select(matrix, route, cmax, auclast) |>
    tidyr::pivot_wider(names_from = route, values_from = c(cmax, auclast)) |>
    dplyr::transmute(matrix, cmax_ratio = cmax_in / cmax_oral,
                     auclast_ratio = auclast_in / auclast_oral)
}
ratios <- dplyr::bind_rows(
  ratio("Oral, mean data", "IN, mean data") |> dplyr::mutate(source = "Mean-data models"),
  ratio("Oral, raw data", "IN, raw data") |> dplyr::mutate(source = "Raw-data models"),
  ratio("Combined, oral", "Combined, IN") |> dplyr::mutate(source = "Combined model"),
  tibble::tibble(matrix = c("Blood", "Brain"), cmax_ratio = c(3.9, 5.7),
                 auclast_ratio = c(1.4, 3.1), source = "Reported (observed NCA)")
)
ratios |>
  dplyr::mutate(dplyr::across(c(cmax_ratio, auclast_ratio), \(x) round(x, 1))) |>
  dplyr::arrange(matrix) |>
  dplyr::rename("Matrix" = matrix, "Cmax IN / oral" = cmax_ratio,
                "AUC IN / oral" = auclast_ratio, "Source" = source) |>
  knitr::kable()
Matrix Cmax IN / oral AUC IN / oral Source
Blood 5.6 1.5 Mean-data models
Blood 5.3 1.5 Raw-data models
Blood 5.3 2.6 Combined model
Blood 3.9 1.4 Reported (observed NCA)
Brain 4.5 1.8 Mean-data models
Brain 4.2 1.9 Raw-data models
Brain 4.5 3.1 Combined model
Brain 5.7 3.1 Reported (observed NCA)

The model-based brain Cmax ratio (4.2-4.5) is lower than the reported 5.7-fold, and the blood Cmax ratio (5.3-5.6) is higher than the reported 3.9-fold. The model-based AUC ratios over each route’s own sampling window match the reported blood ratio for the separate fits (1.5 vs 1.4) and the reported brain ratio for the combined fit (3.1 vs 3.1). The separate fits give 1.8-1.9 for brain, and the combined fit gives 2.6 for blood, because its faster blood elimination shortens the oral serum curve. The paper does not give the AUC interval or say whether the ratios are dose-normalised, so these differences cannot be attributed to the model.

Assumptions and deviations

  • Residual-error scale. Tables 2-4 print the residual-error parameters (a1, a2, a3, b1, b2) without saying whether they are variances or standard deviations. The maintainers read them as NONMEM variances and store their square roots. Read as standard deviations, the oral raw-data additive error on serum (0.016 ug/mL) would put 90% of observations within +/-0.026 ug/mL of a 0.33 ug/mL peak. The raw serum data in Figure 3A (lower) span about 0.13-0.72 ug/mL around that peak, which matches the variance reading (SD 0.126 ug/mL, 90% interval about 0.12-0.54 ug/mL). The intranasal raw serum data (Figure 5A, lower: about 0.2-1.3 ug/mL at 30 min) likewise fit an SD of sqrt(0.082) = 0.29 ug/mL better than 0.082 ug/mL. The proportional rows are read on the same scale.
  • Mislabelled table blocks. Table 3 prints its residual-error block under “Per Os” headings, although the block sits in the intranasal table and has the intranasal structure the text describes (additive serum, proportional brain). Table 4’s footnote labels b1 / b2 as intranasal blood / brain, but the table rows are a1, a2, a3, b1, and the text says the intranasal blood error is additive. The rows are assigned by the text: a3 is the intranasal serum additive error and b1 the intranasal brain proportional error, matching the separate intranasal raw-data fit (0.084 vs 0.082 and 0.236 vs 0.238).
  • Fraction notation. Section 3.2 writes the dose split with P2 (blood) and P3 (direct brain), while Tables 3-4 print them as P1 and P2, and the Table 4 footnote calls P2 the “fraction of dose absorbed in brain”. The packaged models follow the printed equation (P2 is the share of the non-blood dose, not of the whole dose); where the paper’s equation and its labels disagree, the maintainers follow the equation. The Discussion’s figures of “25-30%” for the transit fraction and “20-25%” for the olfactory fraction match neither reading exactly.
  • Rate-constant labels in Section 3.2. The text says that “the blood-to-brain transfer (k23) and the brain elimination (k20) were fixed … while the blood elimination (k30) was taken as a prior”. Figure 4 and the Table 3 footnote name k20 as blood elimination (estimated, 0.046 / 0.05 1/min) and k30 as brain elimination, so k23 and k30 are the fixed pair. The text does not say which oral fit supplied them. Fixing them to the matching data representation (mean to mean, raw to raw) reproduces Figure 5. The other pairing moves the raw-data intranasal brain peak from 0.95 to 1.28 ug/mL (Figure 5 about 0.97).
  • Oral blood volume in the combined fit. Table 4 does not list the oral blood volume. It is taken as the fixed 0.279 L of the oral sub-model (Section 3.1), which reproduces the oral panels of Figure 6.
  • No peripheral compartment after intranasal dosing in the combined fit. Section 3.3 joins “the two previous individual models”, and the intranasal model (Figure 4) has no peripheral compartment. With the oral peripheral exchange also applied to intranasal dosing, the combined model’s intranasal serum peak would drop from 1.77 to 1.63 ug/mL and the curve would keep a slow tail (0.05 ug/mL at 120 min) that Figure 6 does not show.
  • Transit relation. The printed ktr = (N - 1) / Mtt is read as ktr = (N + 1) / Mtt, the only form that reproduces the reported N = 8.58 and 10.14 (checked above).
  • Lag time and duration in the combined fit. Table 4 reports Tlag and Tk0 with relative standard errors of 0.11% and 0.06%. They are estimates, not fixed values, and are packaged as estimated.
  • Single nasal-film dose. The explicit intranasal input is keyed on the most recent depot_nasal dose (podo() / tad()). The models are valid for the single doses the paper studied, or for doses spaced beyond the transit chain (about 3 x Mtt).
  • Preliminary blood-only fit not packaged. The two-compartment fit to the oral serum data (Table 1) is the first stage of the oral model and only serves to fix the blood volume. It is not packaged as a separate model.
  • No correction notice for the article was found (search on 2026-10-26).