Skip to contents

Propofol (Chi 2018)

Source: vignettes/articles/Chi_2018_propofol.Rmd
Chi_2018_propofol.Rmd

Model and source

  • Citation: Chi X, Pan J, Cai J, Luo G, Li S, Yuan D, Rui J, Chen W, Hei Z. Pharmacokinetic Analysis of Propofol Target-Controlled Infusion Models in Chinese Patients with Hepatic Insufficiency. Med Sci Monit. 2018;24:6925-6933.
  • Article: https://doi.org/10.12659/MSM.910103

Chi et al. ran a single-centre prospective observational study at the Third Affiliated Hospital of Sun Yat-sen University (Guangzhou, China; ChiCTR-OCH-12002255) in 32 Chinese adults undergoing elective liver transplantation. Patients were stratified by their preoperative Child-Turcotte-Pugh (CTP) score into Class A (n = 11), B (n = 10), and C (n = 11) and induced with propofol via a Diprifusor target-controlled-infusion (TCI) pump (Marsh parameters) set to a 3 ug/mL plasma target for 30 minutes, followed by 30 minutes of washout sampling. The published TCI-performance analysis showed that the Marsh-parameter Diprifusor system was clinically acceptable in CTP A patients (MDPE 8.0%, MDAPE 23.3%) but accumulated substantial overshoot in CTP B and C (MDPE 21.8% and 45.5%; MDAPE 33.2% and 50.7%). Alongside this, the authors fitted a 2-compartment population PK model to the 32-subject cohort using NONMEM Version V with Wings for NONMEM bootstrap and reported the six final-model THETAs in the text (page 6930):

CL = theta_CL + BW * theta_BW V1 = theta_V1 Q = theta_Q V2 = theta_V2 * (CTP / 5)^theta_CTP

with theta_CL = 0.737 L/min, theta_V1 = 9.94 L, theta_Q = 1.2 L/min, theta_V2 = 52.9 L, theta_BW = 0.0163 L/min/kg, and theta_CTP = -0.848. These are the values packaged in modellib("Chi_2018_propofol").

The paper does NOT report any OMEGA (IIV) or SIGMA (residual error) values, no goodness-of-fit / VPC plots, and no individual parameter estimates. Per operator decision (extraction-task sidecar 320-chi_2018_medical_science_monitor request-002, q1 = C) the packaged model fixes every eta to zero and includes no residual-error term; it is a deterministic typical-value forward predictor. See the Assumptions and deviations section for the full provenance discussion and pointers to companion-paper extractions that DO report a full OMEGA / SIGMA structure for propofol in a Chinese cohort.

Population

The fit cohort is 32 Chinese adults (5 women, 27 men) with hepatic insufficiency scheduled for elective liver transplantation between May 2014 and March 2016. Age 18-65 years (group means 43.7 +/- 7.4, 48.3 +/- 8.5, 46.4 +/- 8.7 across CTP A / B / C). Body weights 58.6 +/- 11.2 kg (CTP A), 66.1 +/- 16.4 kg (CTP B), 63.1 +/- 8.4 kg (CTP C) per Table 1. Underlying liver disease: cirrhosis in 18 of 32 (56%); hepatic carcinoma in 14 of 32 (44%). ASA II-IV. Demographics were comparable across CTP groups (all P > 0.05). The packaged metadata (readModelDb("Chi_2018_propofol")$population) carries the per-group means and pooled cohort summary.

Source trace

The per-parameter origin is also recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/Chi_2018_propofol.R. The table below collects them in one place for review.

Equation / parameter Value Source location
Two-compartment IV linear model n/a Chi 2018 Methods, Pharmacokinetics (page 6929)
Additive WT effect on CL structural form Chi 2018 page 6930 final regression equation
Power CTP effect on V2 at CTP=5 structural form Chi 2018 page 6930 final regression equation
lcl = log(0.737) (theta_CL) 0.737 L/min Chi 2018 page 6930 prose listing
e_wt_cl (theta_BW) 0.0163 L/min/kg Chi 2018 page 6930 prose listing
lvc = log(9.94) (theta_V1) 9.94 L Chi 2018 page 6930 prose listing
lq = log(1.2) (theta_Q) 1.2 L/min Chi 2018 page 6930 prose listing
lvp = log(52.9) (theta_V2) 52.9 L Chi 2018 page 6930 prose listing
e_ctp_score_vp (theta_CTP) -0.848 Chi 2018 page 6930 prose listing
OMEGA (all etas) NOT REPORTED -
SIGMA (residual error) NOT REPORTED -
GOF / VPC / individual estimates NOT REPORTED -
Cohort demographics - Chi 2018 Table 1 (page 6928)
TCI MDPE / MDAPE / wobble / divergence - Chi 2018 Figure 4 (page 6930)
Linear regression Cm vs Cp by class per-class slopes Chi 2018 Results, “Propofol plasma concentration…” (page 6928)

Virtual cohort

The packaged model has no IIV and no residual error – the typical-value forward prediction is the deliverable. The simulation below dosed one representative typical patient per CTP class (Class A, B, C) at the cohort-mean body weight 62 kg (the weighted mean of the three group means in Table 1) with a constant-rate IV infusion at the model-derived rate that targets a 3 ug/mL steady-state plasma propofol concentration. The 30-minute-infusion + 30-minute-washout protocol matches Chi 2018 Methods (Anesthesia and Blood sampling).

mean_wt <- 62               # kg; weighted cohort mean across the three CTP groups

# Representative CTP scores for each class: Class A = 5 (lower bound),
# Class B = 8 (midpoint of 7-9), Class C = 12 (midpoint of 10-15 in the
# cohort range observed up to 14).
class_table <- tibble::tribble(
  ~cohort,  ~ctp_score,
  "CTP A",  5,
  "CTP B",  8,
  "CTP C", 12
)

# Model-derived typical CL at this WT (additive linear in BW, CTP-independent
# per Chi 2018 final equation): CL = 0.737 + 0.0163 * 62 = 1.7476 L/min.
cl_typ_lmin <- 0.737 + 0.0163 * mean_wt

# Maintenance infusion rate to achieve steady-state Cp ~ 3 ug/mL:
# rate (mg/min) = CL (L/min) * Cp_target (mg/L) = 1.7476 * 3 = 5.243 mg/min.
# Convert to mg/h for the rxode2 rate column (rate in dose-units per
# model-time-unit; here dose mg and time minute, so rate is mg/min).
maint_rate_mg_min <- cl_typ_lmin * 3   # mg/min

# 30-minute constant infusion, followed by 30-minute washout. The Chi 2018
# sampling grid (1, 2, 5, 10, 20, 30 min during, then 1, 2, 5, 10, 20, 30
# min after stop) is included in the observation grid plus dense
# intervening times for smooth plotting.
obs_times <- sort(unique(c(
  seq(0, 60, by = 1),
  c(1, 2, 5, 10, 20, 30),           # paper's during-infusion grid
  30 + c(1, 2, 5, 10, 20, 30)       # paper's post-discontinuation grid
)))

make_cohort <- function(i, ctp_score, cohort_label) {
  # Total dose delivered = rate * 30 min; rate is in mg/min so amt is the
  # cumulative bolus equivalent; we use the rate column instead so rxode2
  # treats it as a zero-order infusion. Stop after 30 minutes.
  amt_total <- maint_rate_mg_min * 30   # mg over the 30-min infusion
  dose_row <- tibble::tibble(
    id   = i,
    time = 0,
    cmt  = "central",
    evid = 1L,
    amt  = amt_total,
    rate = maint_rate_mg_min,         # mg/min
    WT   = mean_wt,
    CTP_SCORE = ctp_score,
    cohort = cohort_label
  )
  obs_rows <- tibble::tibble(
    id   = i,
    time = obs_times,
    cmt  = NA_character_,
    evid = 0L,
    amt  = 0,
    rate = 0,
    WT   = mean_wt,
    CTP_SCORE = ctp_score,
    cohort = cohort_label
  )
  dplyr::bind_rows(dose_row, obs_rows)
}

events <- dplyr::bind_rows(
  make_cohort(1L, class_table$ctp_score[1], class_table$cohort[1]),
  make_cohort(2L, class_table$ctp_score[2], class_table$cohort[2]),
  make_cohort(3L, class_table$ctp_score[3], class_table$cohort[3])
)
stopifnot(!anyDuplicated(unique(events[, c("id", "time", "evid")])))

Simulation 

mod <- readModelDb("Chi_2018_propofol")
sim <- rxode2::rxSolve(
  rxode2::rxode2(mod),
  events = events,
  keep   = c("cohort", "WT", "CTP_SCORE")
) |>
  as.data.frame()
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalq', 'etalvp'

Replicate Figure 2 – predicted Cc by CTP class

Chi 2018 Figure 2 plots the measured propofol concentration Cm for the three CTP groups across the 60-minute observation period (the first 30 minutes are during TCI to a 3 ug/mL plasma target; the next 30 minutes are after discontinuation). The packaged model is a population-PK fit to those same observations under the additive-CL / power-V2 covariate structure described above; the panel below shows the deterministic typical-value predictions at the cohort-mean body weight 62 kg, one representative CTP score per class.

The packaged structural model captures the qualitative pattern shown in the published figure: Class A converges to the 3 ug/mL target during infusion and decays after stop, while Class C retains higher concentrations during the post-infusion phase because its peripheral volume V2 is smaller (V2 = 52.9 * (12/5)^(-0.848) = 24.4 L vs Class A’s 52.9 L), leading to slower redistribution. The packaged model cannot reproduce the measured Cm overshoot pattern in Figure 2 because that pattern reflects the mismatch between the Marsh-parameter Diprifusor TCI algorithm (which sets dose rates from a population-typical Caucasian-cohort PK model) and the Chinese hepatic-impairment cohort’s actual PK – whereas the packaged model is the post-hoc PK fit to the latter cohort itself.

sim |>
  dplyr::filter(time > 0) |>
  ggplot2::ggplot(ggplot2::aes(time, Cc, colour = cohort)) +
  ggplot2::geom_line(linewidth = 1) +
  ggplot2::geom_vline(xintercept = 30, linetype = "dashed", alpha = 0.5) +
  ggplot2::geom_hline(yintercept = 3, linetype = "dotted", alpha = 0.5) +
  ggplot2::labs(
    x = "Time (min)",
    y = "Predicted plasma propofol concentration (ug/mL)",
    colour = "CTP class",
    title = "Chi 2018 Figure 2 -- typical-value prediction by CTP class",
    caption = "Constant-rate IV infusion at the model-derived maintenance rate (5.24 mg/min) for 30 min then off, at WT = 62 kg. Dashed line: infusion stop. Dotted line: 3 ug/mL TCI target."
  ) +
  ggplot2::theme_minimal()

Replicate Figure 4 – typical CL by CTP class

Chi 2018 Figure 4 reports the TCI performance summary across CTP A / B / C (MDPE, MDAPE, wobble, divergence). The packaged model does not include the Marsh-vs-Chi TCI controller and so cannot reproduce these system-performance metrics directly. What it CAN show is the underlying typical-value disposition parameters across the three classes, which is what drives the published TCI-performance differences.

cmp_table <- class_table |>
  dplyr::mutate(
    `Typical CL (L/min)` = round(0.737 + 0.0163 * mean_wt, 3),
    `Typical V1 (L)`     = round(9.94, 3),
    `Typical Q  (L/min)` = round(1.2, 3),
    `Typical V2 (L)`     = round(52.9 * (ctp_score / 5)^(-0.848), 3)
  )
knitr::kable(
  cmp_table,
  caption = "Chi 2018 typical-value disposition parameters by CTP class at WT = 62 kg.",
  align   = c("l", "r", "r", "r", "r", "r")
)
Chi 2018 typical-value disposition parameters by CTP class at WT = 62 kg.
cohort ctp_score Typical CL (L/min) Typical V1 (L) Typical Q (L/min) Typical V2 (L)
CTP A 5 1.748 9.94 1.2 52.900
CTP B 8 1.748 9.94 1.2 35.511
CTP C 12 1.748 9.94 1.2 25.179

The CL column is identical across classes because Chi 2018’s final model does not include CTP on CL; only V2 carries a CTP effect. The diminishing V2 with worse hepatic function is what drives the higher persistent concentrations in CTP C after infusion stop.

PKNCA validation

Run NCA on the simulated typical-value profiles to summarise Cmax (at the end of the 30-min infusion), AUC over the 60-minute observation window, and the apparent terminal half-life of the post-infusion decay. The paper does not publish per-class Cmax / AUC / half-life values (its published validation is the TCI-performance MDPE / MDAPE / wobble / divergence table in Figure 4), so this section is a self-consistency check on the packaged model rather than a comparison against Chi 2018 numbers.

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

# Guarantee a time-zero row per (id, cohort); for IV infusion pre-dose Cc = 0
# is the correct value.
sim_nca <- dplyr::bind_rows(
  sim_nca,
  sim_nca |> dplyr::distinct(id, cohort) |>
    dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::distinct(id, cohort, time, .keep_all = TRUE) |>
  dplyr::arrange(id, cohort, time)

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

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

dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | cohort + id)

intervals <- data.frame(
  start       = 0,
  end         = 60,
  cmax        = TRUE,
  tmax        = TRUE,
  auclast     = TRUE,
  half.life   = TRUE
)

nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))
nca_res <- as.data.frame(nca$result)
knitr::kable(
  nca_res,
  caption = "PKNCA NCA over the 60-minute window for the typical-value profiles, by CTP class."
)
PKNCA NCA over the 60-minute window for the typical-value profiles, by CTP class.
cohort id start end PPTESTCD PPORRES exclude
CTP A 1 0 60 auclast 67.5566670 NA
CTP A 1 0 60 cmax 2.0971803 NA
CTP A 1 0 60 tmax 30.0000000 NA
CTP A 1 0 60 tlast 60.0000000 NA
CTP A 1 0 60 lambda.z 0.0134800 NA
CTP A 1 0 60 r.squared 0.9999295 NA
CTP A 1 0 60 adj.r.squared 0.9999154 NA
CTP A 1 0 60 lambda.z.time.first 54.0000000 NA
CTP A 1 0 60 lambda.z.time.last 60.0000000 NA
CTP A 1 0 60 lambda.z.n.points 7.0000000 NA
CTP A 1 0 60 clast.pred 0.2920492 NA
CTP A 1 0 60 half.life 51.4202562 NA
CTP A 1 0 60 span.ratio 0.1166855 NA
CTP B 2 0 60 auclast 72.0600592 NA
CTP B 2 0 60 cmax 2.2154995 NA
CTP B 2 0 60 tmax 30.0000000 NA
CTP B 2 0 60 tlast 60.0000000 NA
CTP B 2 0 60 lambda.z 0.0195467 NA
CTP B 2 0 60 r.squared 0.9999433 NA
CTP B 2 0 60 adj.r.squared 0.9999352 NA
CTP B 2 0 60 lambda.z.time.first 52.0000000 NA
CTP B 2 0 60 lambda.z.time.last 60.0000000 NA
CTP B 2 0 60 lambda.z.n.points 9.0000000 NA
CTP B 2 0 60 clast.pred 0.3419320 NA
CTP B 2 0 60 half.life 35.4610919 NA
CTP B 2 0 60 span.ratio 0.2255994 NA
CTP C 3 0 60 auclast 76.2343889 NA
CTP C 3 0 60 cmax 2.3370738 NA
CTP C 3 0 60 tmax 30.0000000 NA
CTP C 3 0 60 tlast 60.0000000 NA
CTP C 3 0 60 lambda.z 0.0269544 NA
CTP C 3 0 60 r.squared 0.9999132 NA
CTP C 3 0 60 adj.r.squared 0.9999045 NA
CTP C 3 0 60 lambda.z.time.first 49.0000000 NA
CTP C 3 0 60 lambda.z.time.last 60.0000000 NA
CTP C 3 0 60 lambda.z.n.points 12.0000000 NA
CTP C 3 0 60 clast.pred 0.3619669 NA
CTP C 3 0 60 half.life 25.7155046 NA
CTP C 3 0 60 span.ratio 0.4277575 NA

Assumptions and deviations

  • No OMEGA / SIGMA / GOF in the source. Chi 2018 reports the six final-model THETAs in the text on page 6930 but provides no inter-individual variability values, no residual-error variances, no goodness-of-fit plots, and no individual parameter estimates. The Methods describe NONMEM final-model estimation with Wings for NONMEM bootstrap validation in the 32-patient cohort, so the OMEGA and SIGMA values DO exist in the original NONMEM output – they were simply not transcribed into the paper. The packaged model is therefore a deterministic typical-value predictor; VPC-style validation is not possible from the packaged model alone. Operator decision (extraction-task sidecar 320-chi_2018_medical_science_monitor request-002, q1 = C) is to fix every eta to zero rather than borrow from a companion paper or insert placeholders.
  • No propSd placeholder. Operator-specific direction in the sidecar response: “do NOT add a propSd placeholder. If the residual error is not reported in Chi 2018, leave it out – do not invent a value.” The packaged model therefore declares the observation as Cc <- central / vc with no Cc ~ prop(...) line; nlmixr2 / rxode2 simulation treats Cc as a deterministic algebraic output. The Oniki 2018 NAFLD-risk and Zou 2012 MI-219 entries follow the same no-residual-error pattern for typical-value-only forward predictors.
  • No GOF or VPC available. Without an OMEGA / SIGMA structure, neither the packaged model nor the source paper supports a VPC. The PKNCA section above is a self-consistency check on the packaged structural model rather than a comparison against published quantitative summary statistics.
  • CL covariate form: additive linear, not multiplicative. Chi 2018 page 6930 reports CL = theta_CL + BW * theta_BW (an additive equation on the linear scale). This deviates from the nlmixr2lib convention of cl <- exp(lcl + etalcl) * <multiplicative_factor>. The packaged model encodes the additive form as cl <- (exp(lcl) + e_wt_cl * WT) * exp(etalcl), faithful to the paper but unusual for nlmixr2lib. Precedent for the (exp(lcl) + e_cov_cl * (cov - ref)) * exp(etalcl) additive-on-linear pattern: Royer 2010 HuHMFG1 (AST on CL) and Weatherley 2018 fosdagrocorat (AGE on CL).
  • CTP score encoded as the continuous canonical CTP_SCORE. This is a new canonical covariate registered alongside this extraction in inst/references/covariate-columns.md. The existing HEPIMP_SEV (binary Class C indicator) and HEPIMP_MODSEV (binary Class B+C indicator) entries cannot represent the continuous power-form encoding used by Chi 2018 on V2. Scope: general; founding example: Chi_2018_propofol.R. See inst/references/covariate-columns.md for the full entry.
  • Companion 3-compartment popPK paper (Ye 2012 / Rui 2012) is on disk but not extracted into this entry. The same first author (Xinjin Chi) co-authored a multicenter Chinese-cohort propofol popPK paper that DOES report a full OMEGA / SIGMA structure: Ye HB, Li JH, Rui JZ et al, “Propofol pharmacokinetics in China: A multicentric study”, Indian J Pharmacol 2012; 44(3):393-7 (PMID 22701253). Ye 2012 uses a 3-compartment model with age and sex on V1 and body weight on Q3 (n = 220 across four hospitals) and is therefore structurally distinct from Chi 2018’s 2-compartment model with body weight on CL and CTP on V2 (n = 32, hepatic-insufficiency subgroup of the same Sun Yat-sen hospital). Because the two papers describe structurally distinct final models, OMEGA / SIGMA values are not directly transferable between them; the operator (sidecar request-002, q2 = A) queued Ye 2012 for its own standalone extraction rather than treating it as a depends_on of this task. Users wanting between-subject variability for propofol in a Chinese cohort should consult modellib("Ye_2012_propofol") when it becomes available.
  • TCI controller not encoded. Chi 2018’s TCI-performance analysis (Figures 2 / 3 / 4 and the published MDPE / MDAPE / wobble / divergence summary) compares the Marsh-parameter Diprifusor TCI algorithm against the cohort’s measured propofol concentrations. The packaged model is the post-hoc PK fit to the same measured concentrations – it is not a TCI controller. Reproducing Chi 2018’s Figure 2 measured-Cm overshoot pattern would require simulating the Marsh-TCI algorithm as a forcing function on top of the packaged Chi 2018 PK model; that is out of scope for this extraction.
  • Representative CTP scores in the simulation. The packaged covariate column CTP_SCORE is a continuous integer 5-15; the published cohort spanned 5-14. The simulation panel above used representative values 5 / 8 / 12 for Classes A / B / C. The model is monotonic in CTP_SCORE so intermediate values interpolate predictably.
  • Cohort-mean body weight 62 kg. Chi 2018 Table 1 reports group means 58.6 / 66.1 / 63.1 kg across CTP A / B / C; their weighted-by-group-size mean is 62.5 kg. The simulation panel rounds this to 62 kg for the representative patient in each class.