Skip to contents

Model and source

Vellinga 2025 ran a three-period crossover trial in 24 healthy volunteers. In every period remimazolam was given by effect-site target-controlled infusion (TCI), stepped up and down through targets of 50 to 2,000 ng/mL, together with a constant remifentanil plasma TCI target of 0, 0.5, or 0.1 to 4.0 ng/mL. The paper fits one joint pharmacokinetic model and then, sequentially on its individual predictions, four separate pharmacodynamic models. The library mirrors that structure with five model files sharing one article:

Model Endpoint
Vellinga_2025_remimazolam Remimazolam, CNS 7054 and remifentanil plasma concentrations
Vellinga_2025_remimazolam_bis Bispectral Index (BIS, 0-100)
Vellinga_2025_remimazolam_moaas Modified Observer’s Assessment of Alertness and Sedation (MOAAS, 0-5)
Vellinga_2025_remimazolam_tol Tolerance to laryngoscopy (TOL, binary)
Vellinga_2025_remimazolam_tots Tolerance to tetanic stimulation (TOTS, binary)
  • Citation: Vellinga R, Koomen JV, Eleveld DJ, Stohr T, Pesic M, Struys MMRF, Colin PJ. Influence of Remifentanil on the Pharmacokinetics and Pharmacodynamics of Remimazolam in Healthy Volunteers. Anesthesiology. 2025;142(4):666-679. doi:10.1097/ALN.0000000000005348. PMC11892992. Remifentanil layer: Eleveld DJ, Proost JH, Vereecke H, Absalom AR, Olofsen E, Vuyk J, Struys MMRF. An Allometric Model of Remifentanil Pharmacokinetics and Pharmacodynamics. Anesthesiology. 2017;126(6):1005-1018.
  • Description (PK): Joint population pharmacokinetic model of remimazolam, its inactive metabolite CNS 7054 and co-administered remifentanil in 24 healthy adult volunteers given remimazolam by effect-site target-controlled infusion (TCI) alone and with a constant remifentanil plasma TCI target (Vellinga 2025, NCT04670471). Remimazolam: three-compartment disposition with fixed allometric weight scaling; 80% of its clearance forms CNS 7054 through one transit compartment into a one-compartment metabolite pool. Remifentanil: the Eleveld 2017 three-compartment model with all structural, covariate and variability terms held at their published values. Remifentanil inhibits the apparent CNS 7054 clearance through a sigmoid Imax function of its plasma concentration (IC50 8.0 ng/mL); remimazolam clearance is unaffected. The four pharmacodynamic endpoints fit on top of this model are in modellib(‘Vellinga_2025_remimazolam_bis’), ‘Vellinga_2025_remimazolam_moaas’, ‘Vellinga_2025_remimazolam_tol’ and ‘Vellinga_2025_remimazolam_tots’.
  • Article: https://doi.org/10.1097/ALN.0000000000005348 (open access, PMC11892992)
  • Supplemental Digital Content 5 (PK model code and estimates): https://links.lww.com/ALN/D790
  • Supplemental Digital Content 6 (PD model code): https://links.lww.com/ALN/D791

The structure, in brief:

  • Remimazolam: three compartments, weight-scaled allometrically to 70 kg with fixed exponents. Eighty percent of its clearance (fixed from the Byfavo EPAR) forms the inactive metabolite CNS 7054.
  • CNS 7054: formed through one transit compartment into a one-compartment pool. Remifentanil inhibits its apparent clearance by Cr^0.61 / (Cr^0.61 + 8.0^0.61), about 40% at 4 ng/mL (Discussion).
  • Remifentanil: the published Eleveld 2017 three-compartment model, held at its published values (only its residual error was re-estimated).
  • BIS: logistic Emax in which remimazolam and remifentanil add and CNS 7054 antagonises both competitively. The antagonism is the paper’s mechanism for acute tolerance.
  • MOAAS: proportional-odds model whose sigmoid remimazolam effect is antagonised by CNS 7054 and multiplied by 1 + Cr / 2.5.
  • TOL / TOTS: probability of no response to the stimulus, an Emax function of remimazolam whose C50 shrinks by 1 - Cr / (Cr + EC50R). No metabolite antagonism.

All drug effects are driven by arterial plasma concentrations; the models have no effect compartment.

Population

Twenty-four healthy volunteers (13 men, 11 women), ASA physical status I, aged 19 to 70 years (mean 43), weighing 51 to 106 kg (mean 74), height 159 to 192 cm (mean 175), BMI 20.2 to 29.4 kg/m^2 (Results). Enrolment was stratified by age (18-34, 35-49, 50-70 years) and sex at the University Medical Center Groningen, The Netherlands (NCT04670471). After an interim analysis the period-3 remifentanil targets were changed from 2.0 / 4.0 to 0.1 / 1.0 ng/mL; 8, 24, 10, 4 and 2 subjects received the 0.1, 0.5, 1, 2 and 4 ng/mL targets. The data comprise 1,652 remimazolam / CNS 7054 samples, 634 remifentanil samples, 891 BIS values, 1,025 MOAAS scores, 611 TOTS and 374 TOL observations.

The same information is available programmatically via readModelDb("Vellinga_2025_remimazolam")()$population.

Source trace

Every ini() value carries an in-file comment naming its source. The table collects them. “S5” is Supplemental Digital Content 5 (Table 1, Table 2 and the NONMEM control stream), “S6” is Supplemental Digital Content 6 (the four PD control streams), and “T1” is Table 1 of the main text.

Model Parameter Value Source
PK lvc, lvp, lvp2 log(5.31), log(12.81), log(40.04) L / 70 kg S5 Table 1 ‘All + interaction’ (Table 2: 5.3, 12.8, 40.0)
PK lcl, lq, lq2 log(1.15), log(0.91), log(0.29) L/min / 70 kg S5 Table 2
PK fm 0.8 (fixed) S5 Table 2 and footnote (Byfavo EPAR)
PK lktr, lcl_cns7054, lvc_cns7054 log(0.27) 1/min, log(0.05) L/min, log(7.77) L S5 Tables 1 and 2
PK lic50_remifentanil, lhill_remifentanil log(8.0) ng/mL, log(0.61) S5 Tables 1 and 2; main text equation 1
PK lvc_remifentanil … e_wt_vp2_remifentanil Eleveld 2017 values S5 control stream $THETA 1-12 (all FIX)
PK IIV etalcl, etalq2, etalcl_cns7054, etalktr 19.8%, 57.5%, 30.3%, 67.0% CV S5 Table 2; omega^2 = log(1 + CV^2) per its footnote
PK IIV eta*_remifentanil Eleveld 2017 variances S5 control stream $OMEGA 1-6 (all FIX)
PK propSd, propSd_cns7054, addSd_cns7054 0.242, 0.116, 25.28 ng/mL S5 Tables 1 and 2
PK propSd_remifentanil, addSd_remifentanil exp(-1.4506), exp(-3.6352) S5 control stream $THETA 13-14; S5 text 23.4% and 0.03 ng/mL
PK allometry, Al-Sallami FFM, ageing, sex terms equations S5 control stream $PK
PK ODEs, CNS 7054 inhibition, MW ratio 425.3/439.3 equations S5 control stream $DES, $ERROR; main text equation 1
BIS logitrbase logit(0.94) (BASEB = 94.0) T1; S6 LBASE = LOG(BASEB/(100-BASEB)) + ETA(1)
BIS lemax, lec50, lec50_cns7054, lec50_remifentanil log(3.5), log(202), log(9495), log(7.96) T1
BIS etalogitrbase, etalemax 0.07 SD (logit), 10.3% CV T1
BIS addSd_BIS, propSd_BIS 0.31, 0.187 (logit domain) T1; S6 SDB
BIS drug term, logit link equation main text equation 6; S6 BIS $ERROR
MOAAS logitbase0_moaas, lbase1_moaas … lbase4_moaas -9.7; log(2.0, 1.1, 0.8, 2.2) T1; S6 MOAAS logits
MOAAS lemax, lec50, lec50_cns7054, lec50_remifentanil, lhill log(11.7), log(211), log(3160), log(2.5), log(1.75) T1
MOAAS etalogitbase0_moaas 0.99 SD (logit) T1
MOAAS proportional odds, (1 + Cr/EC50R) equation main text equation 2; S6 MOAAS $ERROR
TOL lec50, lec50_remifentanil, etalec50 log(54128), log(0.02), 231.3% CV T1
TOTS lec50, lec50_remifentanil, etalec50 log(12133), log(0.04), 118.8% CV T1
TOL / TOTS zero baseline, C50 correction equation main text equations 3-5; S6 TOL / TOTS $ERROR

Rebuilding the paper’s TCI simulations

The paper’s simulations (Methods, Simulations) give remimazolam by effect-site TCI on this paper’s final PK model, with an effect-site equilibration rate constant of 0.135 1/min and a maximum infusion rate of 6 mg/min. Remifentanil is given by Eleveld-model plasma TCI started 15 minutes earlier. The subject is a typical 70 kg individual. The pump rates are not printed, so the maintainers rebuilt them with discrete TCI controllers. Every 10 seconds the controller picks a constant rate:

  • Plasma TCI (remifentanil): the rate that brings plasma to the target by the end of the 10-second step.
  • Effect-site TCI (remimazolam): the largest rate, capped at 6 mg/min, such that the effect-site concentration never overshoots the target after the step, whatever happens next. This is the Shafer-Gregg peak-effect criterion.

Both use the exact one-step propagation of the linear compartment model, by eigen-decomposition of its rate matrix. The remifentanil subject is the Eleveld reference individual (35 years, 70 kg, 170 cm, male), for whom every covariate scalar equals one; under plasma TCI the remifentanil concentration equals the target regardless.

# exp(M h) and its integral over [0, h], by eigen-decomposition (the rate
# matrices here have distinct real eigenvalues).
lin_step <- function(M, h) {
  e <- eigen(M)
  vi <- solve(e$vectors)
  list(
    E = Re(e$vectors %*% diag(exp(e$values * h)) %*% vi),
    G = Re(e$vectors %*% diag((exp(e$values * h) - 1) / e$values) %*% vi)[, 1],
    vectors = e$vectors, vi = vi, values = e$values
  )
}

rate_matrix_3cmt <- function(k10, k12, k21, k13, k31) {
  rbind(
    c(-(k10 + k12 + k13), k21, k31),
    c(k12, -k21, 0),
    c(k13, 0, -k31)
  )
}

# Effect-site TCI. Concentrations in ng/mL, volumes in L, amounts in mg, so
# Cp = A1 / v1 * 1000. Returns one row per step with the constant rate (mg/min).
tci_effect <- function(pk, ke0, target, t_end, h = 1 / 6, umax = 6,
                       horizon = seq(0, 30, by = 0.1)) {
  M <- matrix(0, 4, 4)
  M[1:3, 1:3] <- rate_matrix_3cmt(pk$k10, pk$k12, pk$k21, pk$k13, pk$k31)
  M[4, ] <- c(ke0 * 1000 / pk$v1, 0, 0, -ke0)
  st <- lin_step(M, h)
  # Effect-site row of exp(M tau) over the look-ahead horizon.
  ce_row <- t(vapply(horizon, function(tau) {
    Re(st$vectors %*% diag(exp(st$values * tau)) %*% st$vi)[4, ]
  }, numeric(4)))
  n <- round(t_end / h)
  x <- numeric(4)
  rate <- numeric(n)
  for (i in seq_len(n)) {
    x_free <- st$E %*% x
    ce_free <- as.vector(ce_row %*% x_free)
    ce_unit <- as.vector(ce_row %*% st$G)
    pos <- ce_unit > 1e-12
    rate[i] <- min(max(min((target - ce_free[pos]) / ce_unit[pos]), 0), umax)
    x <- x_free + st$G * rate[i]
  }
  data.frame(time = (seq_len(n) - 1) * h, rate = rate)
}

# Plasma TCI. Amounts in ug and volumes in L, so Cp = A1 / v1 in ng/mL.
tci_plasma <- function(pk, target, t_end, h = 1 / 6) {
  st <- lin_step(rate_matrix_3cmt(pk$k10, pk$k12, pk$k21, pk$k13, pk$k31), h)
  n <- round(t_end / h)
  x <- numeric(3)
  rate <- numeric(n)
  for (i in seq_len(n)) {
    x_free <- st$E %*% x
    rate[i] <- max((target * pk$v1 - x_free[1]) / st$G[1], 0)
    x <- x_free + st$G * rate[i]
  }
  data.frame(time = (seq_len(n) - 1) * h, rate = rate)
}

The controllers use the typical parameters of the packaged PK model.

mod_pk <- rxode2::rxode2(readModelDb("Vellinga_2025_remimazolam"))
#> ℹ parameter labels from comments will be replaced by 'label()'
th <- mod_pk$theta
micro <- function(v1, cl, q2, v2, q3, v3) {
  list(v1 = v1, k10 = cl / v1, k12 = q2 / v1, k21 = q2 / v2, k13 = q3 / v1, k31 = q3 / v3)
}
pk_remimazolam <- micro(
  exp(th[["lvc"]]), exp(th[["lcl"]]), exp(th[["lq"]]), exp(th[["lvp"]]),
  exp(th[["lq2"]]), exp(th[["lvp2"]])
)
pk_remifentanil <- micro(
  exp(th[["lvc_remifentanil"]]), exp(th[["lcl_remifentanil"]]),
  exp(th[["lq_remifentanil"]]), exp(th[["lvp_remifentanil"]]),
  exp(th[["lq2_remifentanil"]]), exp(th[["lvp2_remifentanil"]])
)
ke0_tci <- 0.135 # 1/min, Methods (Drug Administration and Simulations)

t_pre <- 15 # remifentanil starts 15 min before remimazolam
step <- 1 / 6 # controller step, min

# One scenario: event rows for both infusions plus observation rows at the
# requested times after the start of remimazolam. The controllers run one
# minute past the last observation so that no infusion ends exactly on an
# observation record.
tci_events <- function(id, rmz_target, remi_target, obs_times) {
  t_last <- max(obs_times) + 1
  ev <- list()
  if (remi_target > 0) {
    r <- tci_plasma(pk_remifentanil, remi_target, t_pre + t_last, step)
    ev$remi <- data.frame(time = r$time, rate = r$rate, cmt = "central_remifentanil")
  }
  m <- tci_effect(pk_remimazolam, ke0_tci, rmz_target, t_last, step)
  ev$rmz <- data.frame(time = t_pre + m$time, rate = m$rate, cmt = "central")
  doses <- dplyr::bind_rows(ev) |>
    dplyr::filter(rate > 1e-8) |>
    dplyr::mutate(amt = rate * step, evid = 1L, dvid = NA_integer_)
  obs <- data.frame(
    time = t_pre + obs_times, rate = 0, cmt = NA_character_, amt = 0,
    evid = 0L, dvid = 1L
  )
  dplyr::bind_rows(doses, obs) |>
    dplyr::mutate(
      id = id, rmz_target = rmz_target, remi_target = remi_target,
      WT = 70, AGE = 35, HT = 170, SEXF = 0
    ) |>
    dplyr::arrange(time, dplyr::desc(evid))
}

# Typical-value solve of one PD model; every PD file carries the full PK model.
solve_typical <- function(name, events, cols) {
  mod <- rxode2::rxode2(readModelDb(name))
  sim <- rxode2::rxSolve(
    rxode2::zeroRe(mod), events,
    keep = c("rmz_target", "remi_target"), returnType = "data.frame"
  )
  sim$tmin <- sim$time - t_pre
  sim[, c("id", "tmin", "rmz_target", "remi_target", cols)]
}

solve_all_endpoints <- function(events) {
  solve_typical(
    "Vellinga_2025_remimazolam_bis", events,
    c("Cc", "Cc_cns7054", "Cc_remifentanil", "BIS")
  ) |>
    dplyr::left_join(
      solve_typical("Vellinga_2025_remimazolam_moaas", events, paste0("prob_moaas", 0:5)),
      by = c("id", "tmin", "rmz_target", "remi_target")
    ) |>
    dplyr::left_join(
      solve_typical("Vellinga_2025_remimazolam_tol", events, "prob_tol"),
      by = c("id", "tmin", "rmz_target", "remi_target")
    ) |>
    dplyr::left_join(
      solve_typical("Vellinga_2025_remimazolam_tots", events, "prob_tots"),
      by = c("id", "tmin", "rmz_target", "remi_target")
    )
}

Table 2 and Figure 5: 10 minutes of TCI

Table 2 of the paper tabulates the predicted endpoints after 10 minutes of remimazolam TCI, for six remimazolam and six remifentanil targets.

grid_t2 <- expand.grid(
  rmz_target = c(150, 300, 400, 800, 1300, 2000),
  remi_target = c(0, 0.1, 0.5, 1, 2, 4)
)
ev_t2 <- dplyr::bind_rows(lapply(seq_len(nrow(grid_t2)), function(i) {
  tci_events(i, grid_t2$rmz_target[i], grid_t2$remi_target[i], c(10, 60, 120))
}))
sim_t2 <- solve_all_endpoints(ev_t2)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalogitrbase', 'etalemax'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalogitbase0_moaas'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalec50'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalec50'
stopifnot(!anyNA(sim_t2$BIS), !anyNA(sim_t2$prob_moaas0))

# Vellinga 2025 Table 2 (10 min of TCI), transcribed. Percentages.
paper_t2 <- tibble::tribble(
  ~remi_target, ~rmz_target, ~BIS, ~TOL, ~TOTS, ~M5, ~M4, ~M3, ~M2, ~M1, ~M0,
  0, 150, 77, 0.3, 1.2, 38.4, 47.1, 7.7, 4.4, 2.1, 0.3,
  0, 300, 65, 0.6, 2.4, 2.3, 15.7, 15.7, 26.6, 31.5, 8.3,
  0, 400, 60, 0.7, 3.2, 0.7, 5.5, 7.1, 18.0, 45.7, 23.1,
  0, 800, 48, 1.5, 6.2, 0.1, 0.9, 1.2, 4.1, 26.6, 67.2,
  0, 1300, 43, 2.3, 9.7, 0.1, 0.4, 0.6, 2.2, 16.8, 79.9,
  0, 2000, 40, 3.1, 12.5, 0.0, 0.3, 0.5, 1.7, 13.9, 83.4,
  0.1, 150, 77, 1.5, 4.0, 34.8, 48.7, 8.7, 5.1, 2.4, 0.4,
  0.1, 300, 65, 3.0, 7.8, 1.7, 12.4, 13.5, 25.6, 36.1, 10.7,
  0.1, 400, 60, 4.0, 10.1, 0.5, 4.0, 5.3, 14.7, 45.9, 29.6,
  0.1, 800, 48, 7.6, 18.4, 0.1, 0.6, 0.8, 2.8, 20.2, 75.5,
  0.1, 1300, 43, 11.8, 26.8, 0.0, 0.3, 0.4, 1.4, 11.8, 86.0,
  0.1, 2000, 40, 15.2, 32.8, 0.0, 0.2, 0.3, 1.1, 9.6, 88.7,
  0.5, 150, 76, 6.2, 13.9, 22.1, 50.7, 13.3, 8.7, 4.4, 0.7,
  0.5, 300, 64, 11.7, 24.4, 0.5, 4.4, 5.8, 15.7, 46.0, 27.5,
  0.5, 400, 59, 15.0, 30.1, 0.1, 1.1, 1.6, 5.1, 30.5, 61.6,
  0.5, 800, 48, 26.1, 46.3, 0.0, 0.1, 0.2, 0.6, 5.1, 94.1,
  0.5, 1300, 42, 36.5, 58.3, 0.0, 0.1, 0.1, 0.3, 2.4, 97.2,
  0.5, 2000, 40, 43.4, 65.1, 0.0, 0.0, 0.1, 0.2, 1.8, 97.9,
  1, 150, 75, 11.5, 23.7, 11.4, 43.5, 19.0, 15.5, 9.0, 1.6,
  1, 300, 64, 20.6, 38.3, 0.1, 1.1, 1.6, 5.1, 30.7, 61.4,
  1, 400, 59, 25.7, 45.3, 0.0, 0.2, 0.3, 1.0, 8.9, 89.5,
  1, 800, 48, 40.9, 62.3, 0.0, 0.0, 0.0, 0.1, 0.7, 99.2,
  1, 1300, 42, 52.9, 72.9, 0.0, 0.0, 0.0, 0.0, 0.3, 99.7,
  1, 2000, 40, 60.0, 78.2, 0.0, 0.0, 0.0, 0.0, 0.2, 99.8,
  2, 150, 72, 20.4, 37.8, 2.6, 17.6, 16.8, 26.7, 29.1, 7.2,
  2, 300, 62, 33.9, 54.9, 0.0, 0.1, 0.1, 0.3, 2.9, 96.6,
  2, 400, 58, 40.6, 61.9, 0.0, 0.0, 0.0, 0.0, 0.4, 99.6,
  2, 800, 48, 57.8, 76.4, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  2, 1300, 42, 69.0, 84.1, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  2, 2000, 40, 74.8, 87.6, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  4, 150, 68, 33.8, 54.6, 0.1, 1.0, 1.4, 4.5, 28.5, 64.5,
  4, 300, 60, 50.5, 70.7, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  4, 400, 56, 57.7, 76.3, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  4, 800, 47, 73.1, 86.5, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  4, 1300, 42, 81.6, 91.3, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0,
  4, 2000, 39, 85.5, 93.3, 0.0, 0.0, 0.0, 0.0, 0.0, 100.0
)

sim10 <- sim_t2 |>
  dplyr::filter(tmin == 10) |>
  dplyr::transmute(
    remi_target, rmz_target, Cc, Cc_cns7054,
    BIS, TOL = 100 * prob_tol, TOTS = 100 * prob_tots,
    M5 = 100 * prob_moaas5, M4 = 100 * prob_moaas4, M3 = 100 * prob_moaas3,
    M2 = 100 * prob_moaas2, M1 = 100 * prob_moaas1, M0 = 100 * prob_moaas0
  )
endpoints <- c("BIS", "TOL", "TOTS", "M5", "M4", "M3", "M2", "M1", "M0")
cmp_t2 <- dplyr::inner_join(
  sim10, paper_t2,
  by = c("remi_target", "rmz_target"), suffix = c("_sim", "_paper")
)

cmp_t2 |>
  dplyr::transmute(
    "Remifentanil (ng/mL)" = remi_target,
    "Remimazolam (ng/mL)" = rmz_target,
    "Cp at 10 min (ng/mL)" = round(Cc),
    "BIS sim / paper" = sprintf("%.0f / %.0f", BIS_sim, BIS_paper),
    "TOL % sim / paper" = sprintf("%.1f / %.1f", TOL_sim, TOL_paper),
    "TOTS % sim / paper" = sprintf("%.1f / %.1f", TOTS_sim, TOTS_paper),
    "MOAAS 0 % sim / paper" = sprintf("%.1f / %.1f", M0_sim, M0_paper),
    "MOAAS 2 % sim / paper" = sprintf("%.1f / %.1f", M2_sim, M2_paper)
  ) |>
  knitr::kable(caption = "Replicates Table 2 of Vellinga 2025 (selected columns): typical 70 kg individual after 10 min of TCI.")
Replicates Table 2 of Vellinga 2025 (selected columns): typical 70 kg individual after 10 min of TCI.
Remifentanil (ng/mL) Remimazolam (ng/mL) Cp at 10 min (ng/mL) BIS sim / paper TOL % sim / paper TOTS % sim / paper MOAAS 0 % sim / paper MOAAS 2 % sim / paper
0.0 150 149 78 / 77 0.3 / 0.3 1.2 / 1.2 0.3 / 0.3 4.4 / 4.4
0.0 300 299 67 / 65 0.5 / 0.6 2.4 / 2.4 8.1 / 8.3 26.7 / 26.6
0.0 400 397 62 / 60 0.7 / 0.7 3.2 / 3.2 22.5 / 23.1 18.4 / 18.0
0.0 800 799 50 / 48 1.5 / 1.5 6.2 / 6.2 66.7 / 67.2 4.1 / 4.1
0.0 1300 1299 44 / 43 2.3 / 2.3 9.7 / 9.7 79.5 / 79.9 2.2 / 2.2
0.0 2000 2867 38 / 40 5.0 / 3.1 19.1 / 12.5 86.1 / 83.4 1.4 / 1.7
0.1 150 149 78 / 77 1.6 / 1.5 4.1 / 4.0 0.4 / 0.4 5.0 / 5.1
0.1 300 299 67 / 65 3.2 / 3.0 7.9 / 7.8 10.5 / 10.7 25.8 / 25.6
0.1 400 397 61 / 60 4.2 / 4.0 10.3 / 10.1 28.9 / 29.6 15.0 / 14.7
0.1 800 799 50 / 48 8.1 / 7.6 18.7 / 18.4 75.3 / 75.5 2.8 / 2.8
0.1 1300 1299 44 / 43 12.6 / 11.8 27.3 / 26.8 85.8 / 86.0 1.5 / 1.4
0.1 2000 2867 38 / 40 24.1 / 15.2 45.3 / 32.8 90.8 / 88.7 0.9 / 1.1
0.5 150 149 77 / 76 6.7 / 6.2 14.2 / 13.9 0.7 / 0.7 8.7 / 8.7
0.5 300 299 66 / 64 12.5 / 11.7 24.9 / 24.4 27.2 / 27.5 15.8 / 15.7
0.5 400 397 61 / 59 16.0 / 15.0 30.6 / 30.1 61.1 / 61.6 5.1 / 5.1
0.5 800 799 50 / 48 27.7 / 26.1 47.0 / 46.3 94.1 / 94.1 0.6 / 0.6
0.5 1300 1299 44 / 42 38.4 / 36.5 59.1 / 58.3 97.3 / 97.2 0.3 / 0.3
0.5 2000 2867 38 / 40 57.9 / 43.4 76.1 / 65.1 98.4 / 97.9 0.1 / 0.2
1.0 150 149 76 / 75 12.3 / 11.5 24.2 / 23.7 1.6 / 1.6 15.5 / 15.5
1.0 300 299 65 / 64 22.0 / 20.6 39.0 / 38.3 61.5 / 61.4 5.1 / 5.1
1.0 400 397 60 / 59 27.2 / 25.7 46.0 / 45.3 89.5 / 89.5 1.0 / 1.0
1.0 800 799 50 / 48 42.9 / 40.9 63.1 / 62.3 99.2 / 99.2 0.1 / 0.1
1.0 1300 1299 44 / 42 55.0 / 52.9 73.6 / 72.9 99.7 / 99.7 0.0 / 0.0
1.0 2000 2867 38 / 40 73.0 / 60.0 86.0 / 78.2 99.8 / 99.8 0.0 / 0.0
2.0 150 149 74 / 72 21.8 / 20.4 38.5 / 37.8 7.1 / 7.2 26.8 / 26.7
2.0 300 299 64 / 62 35.8 / 33.9 55.7 / 54.9 96.7 / 96.6 0.3 / 0.3
2.0 400 397 59 / 58 42.6 / 40.6 62.5 / 61.9 99.6 / 99.6 0.0 / 0.0
2.0 800 799 49 / 48 59.8 / 57.8 77.0 / 76.4 100.0 / 100.0 0.0 / 0.0
2.0 1300 1299 44 / 42 70.8 / 69.0 84.5 / 84.1 100.0 / 100.0 0.0 / 0.0
2.0 2000 2867 38 / 40 84.3 / 74.8 92.3 / 87.6 100.0 / 100.0 0.0 / 0.0
4.0 150 149 70 / 68 35.6 / 33.8 55.4 / 54.6 64.7 / 64.5 4.5 / 4.5
4.0 300 299 61 / 60 52.6 / 50.5 71.3 / 70.7 100.0 / 100.0 0.0 / 0.0
4.0 400 397 57 / 56 59.6 / 57.7 76.8 / 76.3 100.0 / 100.0 0.0 / 0.0
4.0 800 799 49 / 47 74.8 / 73.1 86.9 / 86.5 100.0 / 100.0 0.0 / 0.0
4.0 1300 1299 44 / 42 82.8 / 81.6 91.5 / 91.3 100.0 / 100.0 0.0 / 0.0
4.0 2000 2867 38 / 39 91.4 / 85.5 96.0 / 93.3 100.0 / 100.0 0.0 / 0.0

For every remimazolam target up to 1,300 ng/mL, the controller has brought both plasma and effect site to the target by 10 minutes. The probabilities (TOL, TOTS and all six MOAAS categories) then agree with Table 2 to within 2.6 percentage points, most of them within one. The largest gaps are at the steepest part of the curves (MOAAS 5 versus 4 at 150 ng/mL). There the simulation is slightly less sedated than the table, in the same direction as BIS, which runs one to two units above the table throughout. The Assumptions section discusses this.

low <- cmp_t2 |> dplyr::filter(rmz_target <= 1300)
prob_cols <- setdiff(endpoints, "BIS")
max_abs_diff <- vapply(prob_cols, function(e) {
  max(abs(low[[paste0(e, "_sim")]] - low[[paste0(e, "_paper")]]))
}, numeric(1))
round(max_abs_diff, 2)
#>  TOL TOTS   M5   M4   M3   M2   M1   M0 
#> 2.13 0.82 2.58 1.84 1.14 0.38 0.47 0.72
bis_diff <- low$BIS_sim - low$BIS_paper
summary(bis_diff)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#>  0.9051  1.3889  1.5878  1.6258  1.9323  2.3211

stopifnot(
  # Typical-value solve with no random draws, so the differences are
  # deterministic; the measured maximum is 2.6 percentage points.
  all(max_abs_diff < 3.5),
  # BIS: a constant 1-2 unit offset (see Assumptions), never more than 3.
  all(bis_diff >= -1 & bis_diff <= 3)
)

At 2,000 ng/mL the 6 mg/min rate cap is still binding at 10 minutes: plasma has overshot to about 2,870 ng/mL while the effect site has reached only about 1,740 ng/mL. Inverting the paper’s own TOTS and TOL probabilities for that row (no remifentanil) shows which concentration drove its simulation. For TOTS, P = C / (C + EC50) gives C = EC50 * P / (1 - P).

pk_mat <- matrix(0, 4, 4)
pk_mat[1:3, 1:3] <- with(pk_remimazolam, rate_matrix_3cmt(k10, k12, k21, k13, k31))
pk_mat[4, ] <- c(ke0_tci * 1000 / pk_remimazolam$v1, 0, 0, -ke0_tci)
st10 <- lin_step(pk_mat, 10)
x10 <- st10$G * 6 # 10 min at the 6 mg/min cap
c_plasma_10 <- x10[1] / pk_remimazolam$v1 * 1000
c_effect_10 <- x10[4]
c_from_tots <- 12133 * 0.125 / (1 - 0.125)
c_from_tol <- 54128 * 0.031 / (1 - 0.031)
data.frame(
  quantity = c(
    "Plasma after 10 min at 6 mg/min", "Effect site after 10 min at 6 mg/min",
    "Driver implied by Table 2 TOTS 12.5%", "Driver implied by Table 2 TOL 3.1%"
  ),
  ng_per_mL = round(c(c_plasma_10, c_effect_10, c_from_tots, c_from_tol))
) |> knitr::kable()
quantity ng_per_mL
Plasma after 10 min at 6 mg/min 2867
Effect site after 10 min at 6 mg/min 1738
Driver implied by Table 2 TOTS 12.5% 1733
Driver implied by Table 2 TOL 3.1% 1732
stopifnot(
  abs(c_from_tots / c_effect_10 - 1) < 0.02,
  abs(c_from_tol / c_effect_10 - 1) < 0.02,
  c_plasma_10 > 1.5 * c_from_tots
)

Both implied drivers agree with the TCI effect-site concentration to within 1%, and neither is anywhere near the plasma concentration. The paper’s simulation therefore evaluated the remimazolam effect at the TCI effect-site concentration (ke0 = 0.135 1/min). The packaged models follow the fitted model code, which uses arterial plasma (CP = A(1)/V1*1000). Since the fitting data were collected at steady state, where the two coincide, the difference shows only during the loading phase of the 2,000 ng/mL row. That row is excluded from the check above.

sim10 |>
  dplyr::select(remi_target, rmz_target, BIS, TOL, TOTS, M0) |>
  tidyr::pivot_longer(c(BIS, TOL, TOTS, M0), names_to = "endpoint") |>
  dplyr::mutate(endpoint = factor(
    endpoint,
    levels = c("BIS", "M0", "TOTS", "TOL"),
    labels = c("BIS", "P(MOAAS 0) (%)", "P(TOTS) (%)", "P(TOL) (%)")
  )) |>
  ggplot(aes(rmz_target, value, colour = factor(remi_target))) +
  geom_line() +
  geom_point(size = 1) +
  facet_wrap(~endpoint, scales = "free_y") +
  labs(
    x = "Remimazolam effect-site target (ng/mL)", y = NULL,
    colour = "Remifentanil\ntarget (ng/mL)",
    caption = "Replicates Figure 5 of Vellinga 2025 (typical individual, 10 min of TCI)."
  ) +
  theme_bw()

Acute tolerance at 60 and 120 minutes

Results: at a 2,000 ng/mL remimazolam target without remifentanil, the probability of MOAAS 0 falls from 83.4% to 74.4% and 64.3% after 10, 60 and 120 minutes, and BIS rises from 40 to 45 and 50. With 4 ng/mL remifentanil, BIS rises from 39 to 45 and 51 while MOAAS 0 stays at 100%. The rise comes from CNS 7054 accumulating and antagonising the parent. At 60 and 120 minutes the TCI is at steady state, so plasma equals the target.

tol_cmp <- sim_t2 |>
  dplyr::filter(rmz_target == 2000, remi_target %in% c(0, 4), tmin %in% c(60, 120)) |>
  dplyr::arrange(remi_target, tmin) |>
  dplyr::transmute(
    remi_target, tmin,
    Cc_cns7054 = round(Cc_cns7054),
    BIS_sim = round(BIS, 1),
    BIS_paper = c(45, 50, 45, 51),
    M0_sim = round(100 * prob_moaas0, 1),
    M0_paper = c(74.4, 64.3, 100, 100)
  )
tol_cmp |>
  dplyr::rename(
    "Remifentanil (ng/mL)" = remi_target, "Time (min)" = tmin,
    "CNS 7054 (ng/mL)" = Cc_cns7054, "BIS sim" = BIS_sim, "BIS paper" = BIS_paper,
    "MOAAS 0 % sim" = M0_sim, "MOAAS 0 % paper" = M0_paper
  ) |>
  knitr::kable(caption = "Results text and Supplemental Digital Content 7: 2,000 ng/mL remimazolam target.")
Results text and Supplemental Digital Content 7: 2,000 ng/mL remimazolam target.
Remifentanil (ng/mL) Time (min) CNS 7054 (ng/mL) BIS sim BIS paper MOAAS 0 % sim MOAAS 0 % paper
0 60 11065 47.0 45 74.2 74.4
0 120 18931 51.6 50 64.4 64.3
4 60 11859 46.8 45 100.0 100.0
4 120 21660 52.3 51 100.0 100.0
stopifnot(
  all(abs(tol_cmp$M0_sim - tol_cmp$M0_paper) < 1.5),
  all(abs(tol_cmp$BIS_sim - tol_cmp$BIS_paper) < 3),
  # Tolerance develops: BIS rises and P(MOAAS 0) falls between 60 and 120 min.
  diff(tol_cmp$BIS_sim[tol_cmp$remi_target == 0]) > 0,
  diff(tol_cmp$M0_sim[tol_cmp$remi_target == 0]) < 0
)

Targets maximising MOAAS 2 or 3

Results: after 10 minutes of TCI the probability of MOAAS 2 or 3 is highest at remimazolam targets of 275, 250 and 200 ng/mL combined with 0, 0.1 and 0.5 ng/mL remifentanil (45%, 45% and 44%), where BIS is 67, 68 and 71.

grid_m23 <- expand.grid(rmz_target = seq(150, 400, by = 25), remi_target = c(0, 0.1, 0.5))
ev_m23 <- dplyr::bind_rows(lapply(seq_len(nrow(grid_m23)), function(i) {
  tci_events(i, grid_m23$rmz_target[i], grid_m23$remi_target[i], 10)
}))
sim_m23 <- solve_all_endpoints(ev_m23) |>
  dplyr::mutate(p23 = 100 * (prob_moaas2 + prob_moaas3))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalogitrbase', 'etalemax'
#> Warning: multi-subject simulation without without 'omega'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalogitbase0_moaas'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalec50'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> Warning: No sigma parameters in the model
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil', 'etalec50'
best_m23 <- sim_m23 |>
  dplyr::group_by(remi_target) |>
  dplyr::slice_max(p23, n = 1) |>
  dplyr::ungroup() |>
  dplyr::transmute(
    remi_target, rmz_sim = rmz_target, rmz_paper = c(275, 250, 200),
    p23_sim = round(p23, 1), p23_paper = c(45, 45, 44),
    BIS_sim = round(BIS), BIS_paper = c(67, 68, 71)
  )
best_m23 |>
  dplyr::rename(
    "Remifentanil (ng/mL)" = remi_target, "Best remimazolam target, sim" = rmz_sim,
    "Best remimazolam target, paper" = rmz_paper, "P(MOAAS 2-3) % sim" = p23_sim,
    "P(MOAAS 2-3) % paper" = p23_paper, "BIS sim" = BIS_sim, "BIS paper" = BIS_paper
  ) |>
  knitr::kable()
Remifentanil (ng/mL) Best remimazolam target, sim Best remimazolam target, paper P(MOAAS 2-3) % sim P(MOAAS 2-3) % paper BIS sim BIS paper
0.0 275 275 44.1 45 68 67
0.1 250 250 44.2 45 70 68
0.5 200 200 43.4 44 73 71
stopifnot(
  all(abs(best_m23$rmz_sim - best_m23$rmz_paper) <= 25),
  all(abs(best_m23$p23_sim - best_m23$p23_paper) < 2)
)

Stochastic simulation of a trial period

The figure below simulates the period-2 protocol in a virtual cohort with between-subject variability: remifentanil plasma TCI at 0.5 ng/mL, then remimazolam effect-site targets stepped through 150, 300, 400, 800, 1,300 and 2,000 ng/mL and back down, 25 minutes per step. Demographics approximate the trial’s: ages 19-70 years, weight around 74 kg, height around 175 cm, 11 of 24 female. The TCI rates are those for a typical 70 kg individual (one-minute controller steps). In the trial each pump was driven by the Zhou 2018 model and the subject’s own covariates, so this is an illustration of variability, not a visual predictive check against the trial data.

set.seed(20250110)
rxode2::rxSetSeed(20250110)
n_sub <- 100
targets_p2 <- c(150, 300, 400, 800, 1300, 2000, 1300, 800, 400, 300, 150)
step_len <- 25
step_p2 <- 1

# Effect-site TCI through a sequence of targets: run the controller one step
# at a time, carrying the state across target changes.
tci_effect_sequence <- function(pk, ke0, targets, step_len, h, umax = 6) {
  M <- matrix(0, 4, 4)
  M[1:3, 1:3] <- rate_matrix_3cmt(pk$k10, pk$k12, pk$k21, pk$k13, pk$k31)
  M[4, ] <- c(ke0 * 1000 / pk$v1, 0, 0, -ke0)
  st <- lin_step(M, h)
  horizon <- seq(0, 30, by = 0.1)
  ce_row <- t(vapply(horizon, function(tau) {
    Re(st$vectors %*% diag(exp(st$values * tau)) %*% st$vi)[4, ]
  }, numeric(4)))
  n_per <- round(step_len / h)
  tg <- rep(targets, each = n_per)
  x <- numeric(4)
  rate <- numeric(length(tg))
  for (i in seq_along(tg)) {
    x_free <- st$E %*% x
    ce_free <- as.vector(ce_row %*% x_free)
    ce_unit <- as.vector(ce_row %*% st$G)
    pos <- ce_unit > 1e-12
    rate[i] <- min(max(min((tg[i] - ce_free[pos]) / ce_unit[pos]), 0), umax)
    x <- x_free + st$G * rate[i]
  }
  data.frame(time = (seq_along(tg) - 1) * h, rate = rate, target = tg)
}

rmz_p2 <- tci_effect_sequence(pk_remimazolam, ke0_tci, targets_p2, step_len, step_p2)
t_total <- max(rmz_p2$time) + step_p2
remi_p2 <- tci_plasma(pk_remifentanil, 0.5, t_pre + t_total + 1, step_p2)

dose_p2 <- dplyr::bind_rows(
  data.frame(time = t_pre + rmz_p2$time, rate = rmz_p2$rate, cmt = "central"),
  data.frame(time = remi_p2$time, rate = remi_p2$rate, cmt = "central_remifentanil")
) |>
  dplyr::filter(rate > 1e-8) |>
  dplyr::mutate(amt = rate * step_p2, evid = 1L, dvid = NA_integer_)
obs_p2 <- data.frame(
  time = t_pre + seq(2.5, t_total - 2.5, by = 2.5), rate = 0, amt = 0,
  cmt = NA_character_, evid = 0L, dvid = 1L
)
cohort <- data.frame(
  id = seq_len(n_sub),
  AGE = round(runif(n_sub, 19, 70)),
  SEXF = rbinom(n_sub, 1, 11 / 24)
) |>
  dplyr::mutate(
    WT = pmin(pmax(rnorm(n_sub, ifelse(SEXF == 1, 67, 80), 10), 51), 106),
    HT = pmin(pmax(rnorm(n_sub, ifelse(SEXF == 1, 168, 182), 6), 159), 192)
  )
events_p2 <- dplyr::bind_rows(lapply(seq_len(n_sub), function(i) {
  dplyr::bind_rows(dose_p2, obs_p2) |>
    dplyr::mutate(id = i) |>
    dplyr::arrange(time, dplyr::desc(evid))
})) |>
  dplyr::left_join(cohort, by = "id")
mod_bis <- rxode2::rxode2(readModelDb("Vellinga_2025_remimazolam_bis"))
#> ℹ parameter labels from comments will be replaced by 'label()'
sim_p2 <- rxode2::rxSolve(mod_bis, events_p2, returnType = "data.frame") |>
  dplyr::mutate(tmin = time - t_pre)
stopifnot(!anyNA(sim_p2$BIS))

sim_p2 |>
  dplyr::select(id, tmin, Cc, Cc_cns7054, BIS) |>
  tidyr::pivot_longer(c(Cc, Cc_cns7054, BIS), names_to = "output") |>
  dplyr::group_by(output, tmin) |>
  dplyr::summarise(
    p05 = quantile(value, 0.05), p50 = median(value), p95 = quantile(value, 0.95),
    .groups = "drop"
  ) |>
  dplyr::mutate(output = factor(
    output,
    levels = c("Cc", "Cc_cns7054", "BIS"),
    labels = c("Remimazolam (ng/mL)", "CNS 7054 (ng/mL)", "BIS (individual prediction)")
  )) |>
  ggplot(aes(tmin, p50)) +
  geom_ribbon(aes(ymin = p05, ymax = p95), alpha = 0.25) +
  geom_line() +
  facet_wrap(~output, scales = "free_y", ncol = 1) +
  labs(
    x = "Time since start of remimazolam (min)", y = NULL,
    caption = "Median and 90% interval, 100 virtual subjects, period-2 protocol (remifentanil 0.5 ng/mL)."
  ) +
  theme_bw()

Remimazolam tracks its targets closely, since only clearance and Q3 carry between-subject variability. CNS 7054 accumulates throughout the step-down phase (its half-life is about log(2) * 7.77 / 0.05, roughly 108 minutes, for a 70 kg subject without remifentanil). BIS on the way down is therefore higher than at the same target on the way up. That hysteresis is the acute tolerance the paper attributes to the metabolite.

# Same target (400 ng/mL) on the way up (step 3) and down (step 9): the
# remimazolam concentration is matched, the metabolite is higher, BIS is higher.
mid_step <- function(k) (k - 0.5) * step_len
hyst <- sim_p2 |>
  dplyr::filter(tmin %in% c(mid_step(3), mid_step(9))) |>
  dplyr::mutate(phase = ifelse(tmin < 150, "up", "down")) |>
  dplyr::group_by(phase) |>
  dplyr::summarise(
    Cc = median(Cc), Cc_cns7054 = median(Cc_cns7054), BIS = median(BIS),
    .groups = "drop"
  )
knitr::kable(hyst, digits = 1)
phase Cc Cc_cns7054 BIS
down 367.1 13428.0 76.3
up 384.1 1561.7 62.9
up <- hyst[hyst$phase == "up", ]
down <- hyst[hyst$phase == "down", ]
stopifnot(
  abs(down$Cc / up$Cc - 1) < 0.15,
  down$Cc_cns7054 > 2 * up$Cc_cns7054,
  down$BIS > up$BIS + 3
)

PKNCA: remifentanil and the CNS 7054 exposure

The paper reports no non-compartmental results. The check here is analytic instead. A remimazolam infusion of 10 mg over 1 minute is given to a typical 70 kg subject, alone or under a constant remifentanil plasma concentration of 4 ng/mL (plasma TCI started 15 minutes earlier and run throughout). Remimazolam AUC should equal dose / CL in both arms. The CNS 7054 AUC should equal fm * dose * (425.3 / 439.3) / CLM, inflated by 1 / (1 - INH) with INH = 4^0.61 / (4^0.61 + 8^0.61), about 0.40.

t_obs <- c(0, 0.5, seq(1, 20, by = 1), seq(25, 600, by = 5), seq(620, 2880, by = 20))
nca_events <- function(id, remi_target) {
  ev <- data.frame(time = t_pre, amt = 10, rate = 10, cmt = "central", evid = 1L, dvid = NA_integer_)
  if (remi_target > 0) {
    r <- tci_plasma(pk_remifentanil, remi_target, t_pre + max(t_obs) + 1, 1)
    ev <- dplyr::bind_rows(ev, data.frame(
      time = r$time, amt = r$rate, rate = r$rate, cmt = "central_remifentanil",
      evid = 1L, dvid = NA_integer_
    ) |> dplyr::filter(rate > 1e-8))
  }
  obs <- data.frame(time = t_pre + t_obs, amt = 0, rate = 0, cmt = NA_character_, evid = 0L, dvid = 1L)
  dplyr::bind_rows(ev, obs) |>
    dplyr::mutate(
      id = id, treatment = ifelse(remi_target > 0, "remifentanil 4 ng/mL", "remimazolam alone"),
      WT = 70, AGE = 35, HT = 170, SEXF = 0
    ) |>
    dplyr::arrange(time, dplyr::desc(evid))
}
ev_nca <- dplyr::bind_rows(nca_events(1, 0), nca_events(2, 4))
sim_nca <- rxode2::rxSolve(
  rxode2::zeroRe(mod_pk), ev_nca,
  keep = "treatment", returnType = "data.frame", rtol = 1e-10, atol = 1e-12
) |>
  dplyr::mutate(time = time - t_pre)
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalq2', 'etalcl_cns7054', 'etalktr', 'etalvc_remifentanil', 'etalvp_remifentanil', 'etalvp2_remifentanil', 'etalcl_remifentanil', 'etalq_remifentanil', 'etalq2_remifentanil'
#> Warning: multi-subject simulation without without 'omega'
nca_long <- dplyr::bind_rows(
  sim_nca |> dplyr::transmute(id, time, treatment, analyte = "remimazolam", Cc),
  sim_nca |> dplyr::transmute(id, time, treatment, analyte = "CNS 7054", Cc = Cc_cns7054)
) |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::mutate(Cc = pmax(Cc, 0))
nca_long <- dplyr::bind_rows(
  nca_long,
  nca_long |> dplyr::distinct(id, treatment, analyte) |> dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::distinct(id, treatment, analyte, time, .keep_all = TRUE) |>
  dplyr::arrange(id, treatment, analyte, time)

conc_obj <- PKNCA::PKNCAconc(nca_long, Cc ~ time | treatment + analyte + id)
dose_obj <- PKNCA::PKNCAdose(
  data.frame(id = 1:2, time = 0, amt = 10, treatment = c("remimazolam alone", "remifentanil 4 ng/mL")),
  amt ~ time | treatment + id
)
intervals <- data.frame(start = 0, end = Inf, cmax = TRUE, tmax = TRUE, aucinf.obs = TRUE, half.life = TRUE)
nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))

auc <- as.data.frame(nca_res) |>
  dplyr::filter(PPTESTCD == "aucinf.obs") |>
  dplyr::select(treatment, analyte, aucinf = PPORRES)

inh4 <- 4^0.61 / (4^0.61 + 8^0.61)
auc_theory <- tibble::tibble(
  treatment = rep(c("remimazolam alone", "remifentanil 4 ng/mL"), each = 2),
  analyte = rep(c("remimazolam", "CNS 7054"), 2),
  theory = c(
    10 * 1000 / 1.15, 0.8 * 10 * 1000 * 425.3 / 439.3 / 0.05,
    10 * 1000 / 1.15, 0.8 * 10 * 1000 * 425.3 / 439.3 / (0.05 * (1 - inh4))
  )
)
auc_cmp <- dplyr::left_join(auc, auc_theory, by = c("treatment", "analyte")) |>
  dplyr::mutate(pct_diff = 100 * (aucinf / theory - 1))
auc_cmp |>
  dplyr::rename(
    Treatment = treatment, Analyte = analyte,
    "AUC0-inf, PKNCA (ng*min/mL)" = aucinf, "AUC0-inf, closed form (ng*min/mL)" = theory,
    "Difference (%)" = pct_diff
  ) |>
  knitr::kable(digits = 1)
Treatment Analyte AUC0-inf, PKNCA (ng*min/mL) AUC0-inf, closed form (ng*min/mL) Difference (%)
remifentanil 4 ng/mL CNS 7054 256370.0 256391.6 0
remifentanil 4 ng/mL remimazolam 8693.2 8695.7 0
remimazolam alone CNS 7054 154880.8 154901.0 0
remimazolam alone remimazolam 8693.2 8695.7 0
# Deterministic solve at tight tolerances; the residual is trapezoid and
# tail-extrapolation error on the observation grid.
stopifnot(all(abs(auc_cmp$pct_diff) < 2))

Remifentanil leaves remimazolam exposure unchanged. At 4 ng/mL it raises the CNS 7054 AUC by 1 / (1 - INH), which is 1.66-fold. That is the roughly 40% clearance reduction quoted in the Discussion.

Assumptions and deviations

  • Simulation driver at 2,000 ng/mL. The packaged PD models are driven by arterial plasma, exactly as in the fitted control streams of Supplemental Digital Content 6. Table 2’s 10-minute row at 2,000 ng/mL is instead reproduced only with the TCI effect-site concentration as the driver (shown above). Every other row, and the 60- and 120-minute values, is at steady state, where plasma and effect site coincide.
  • BIS offset. The typical BIS is 1 to 2 units above Table 2 at every target, with or without remifentanil. The difference cannot be closed with the printed values: at 800 ng/mL and no remifentanil, even with no CNS 7054 the printed EMAXB = 3.5 and EC50B = 202 ng/mL give a BIS of about 49, against Table 2’s 48. An EMAXB about 3% higher (printed to two significant figures, 95% CI 3.3 to 3.8) would remove the offset. The printed values are kept.
  • TCI schedules. The pump rates of the paper’s simulations are not printed. They were rebuilt with the controllers above, using the paper’s ke0 (0.135 1/min) and rate cap (6 mg/min). The remifentanil covariates (35 years, 170 cm, male) are the Eleveld reference individual’s. They are unstated in the paper and irrelevant under plasma TCI.
  • Precision of printed estimates. CNS 7054 clearance is printed as 0.05 L/min (95% CI 0.04-0.05) in both supplementary tables. The TOL and TOTS remifentanil C50s are printed as 0.02 and 0.04 ng/mL. These are used as printed. Where Supplemental Digital Content 5 Table 1 (‘Model (All + interaction)’, the same final model) prints an extra digit (V1, V2, V3, V5, KTR, the Hill factor, the CNS 7054 additive error), that value is used.
  • Residual-error scale. The Supplement 5 footnote says the proportional CVs were computed as the square root of the residual error. The control stream, however, fixes $SIGMA to 1 and uses EXP(THETA) directly as the standard deviation. Its initial estimates (EXP(-1.40) = 0.25, EXP(-2.1) = 0.12, EXP(3.2) = 24.5) and the remifentanil EXP(-1.4506) = 0.234 reported as “23.4%” show that the tabulated values are SDs. They are encoded as such.
  • Between-subject variability on the PD models. Table 1 gives the additive logit-domain etas (BIS baseline, MOAAS baselines) as SDs, which are squared. The exponential etas (BIS EMAX, TOL and TOTS C50) are given as CV%, converted with the rule Supplemental Digital Content 5 states for the PK model, omega^2 = log(1 + CV^2).
  • Below-quantification data. The fitted model handled censored samples with the M3 method during development and excluded them in the final run. This affects only estimation, not simulation.
  • Sequential fit. The four PD models were estimated on individual post hoc PK predictions. Each PD file reproduces the full PK model, including its between-subject variability, so it can be simulated on its own. Simulating two endpoints for the same virtual subject therefore needs the same PK etas supplied to both models.
  • Errata. No correction notice for the article was found on PubMed, EuropePMC or the journal site as of 2026-11-03.