Skip to contents

Melhem_2022_dostarlimab

Source: vignettes/articles/Melhem_2022_dostarlimab.Rmd
Melhem_2022_dostarlimab.Rmd

Model and source

  • Citation: Melhem M, Hanze E, Lu Z, Venkatakrishnan K, Gupta N, Vugmeyster Y. Population pharmacokinetics and exposure-response of anti-programmed cell death protein-1 monoclonal antibody dostarlimab in advanced solid tumours. Br J Clin Pharmacol. 2022;88(9):4142-4154. doi:[10.1111/bcp.15339](https://doi.org/10.1111/bcp.15339)
  • Description: Two-compartment population PK model for dostarlimab (anti-PD-1 IgG4) with time-dependent (sigmoid I_max) clearance in adults with advanced solid tumours.
  • Modality: Therapeutic monoclonal antibody (IgG4 hinge-stabilized), IV infusion.

Dostarlimab is an anti-PD-1 monoclonal antibody. The Melhem 2022 paper developed a population PK model from the Phase 1 GARNET trial (NCT02715284) supporting the recommended therapeutic dose regimen of 500 mg IV Q3W for four cycles followed by 1000 mg IV Q6W.

Structure: linear two-compartment IV model with time-dependent clearance via a sigmoid I_max function of time since first dose:

CL(t)=CLbase⋅exp(Imax⋅tHillT50Hill+tHill) \mathrm{CL}(t) = \mathrm{CL}_{\text{base}} \cdot \exp\!\left(I_{\max}\cdot \dfrac{t^{\mathrm{Hill}}} {T_{50}^{\mathrm{Hill}} + t^{\mathrm{Hill}}}\right)

with Imax=−0.161I_{\max} = -0.161 (a fractional decrease in log-CL at t≫T50t \gg T_{50}; 1−exp(Imax)=0.1491 - \exp(I_{\max}) = 0.149, matching the paper’s “maximum reduction of 14.9%”), T50=108T_{50} = 108 days, and Hill=5.29\mathrm{Hill} = 5.29. Allometric weight scaling uses reference weight 70 kg (exponent 0.470 on CL, 0.419 on Vc and Vp).

Population

The final-model population was 546 patients with 4783 PK observations (after exclusion of 21 records with absolute conditional weighted residual > 5) drawn from a single multinational Phase 1 trial across multiple expansion cohorts (Melhem 2022 Tables 1-2):

  • 23.4% mismatch repair-deficient (dMMR/MSI-H) endometrial cancer (EC),
  • 29.3% mismatch repair-proficient (MMRp/MSS) EC,
  • 12.3% non-small-cell lung cancer (NSCLC),
  • 28.8% non-EC dMMR/MSI-H or POLE-mutant tumours,
  • and an ovarian-cancer cohort.

Demographics: median age 64 years (range 24-86), 77.3% female, median weight 71.4 kg (range 34-182), median albumin 39 g/L, median ALT 17 U/L. Hepatic and renal impairment were present at baseline (mild-to-moderate in 11% and 61% of patients respectively) but did not affect dostarlimab PK. ADAs ever-positive in 18.5% of patients but had no impact on clearance.

The reference patient used in the published forest plot (Methods, PopPK model development, last paragraph) is female, 70 kg, 64 years old, with baseline albumin 39 g/L and ALT 18 U/L.

The same metadata is available programmatically via readModelDb("Melhem_2022_dostarlimab")$population.

Source trace

The per-parameter origin is recorded as an in-file comment next to each ini() entry in inst/modeldb/specificDrugs/Melhem_2022_dostarlimab.R. The table below collects them in one place for review.

Parameter (model name) Value Source
lcl (CL_base, L/day) log(0.179) Melhem 2022 Table 3, CL
lvc (Vc_base, L) log(2.98) Melhem 2022 Table 3, Vc
lq (Q, L/day) log(0.547) Melhem 2022 Table 3, Q
lvp (Vp, L) log(2.10) Melhem 2022 Table 3, Vp
lImax (log|I_max|) log(0.161) Melhem 2022 Table 3, I_max = -0.161
lt50 (T50, days) log(108) Melhem 2022 Table 3, T50
lhill (Hill) log(5.29) Melhem 2022 Table 3, Hill
e_wt_cl 0.470 Melhem 2022 Table 3, Effect of WT on CL
e_wt_v 0.419 Melhem 2022 Table 3, Effect of WT on Vc and Vp
e_age_cl -0.227 Melhem 2022 Table 3, Effect of age on CL
e_alb_cl -1.01 Melhem 2022 Table 3, Effect of ALB on CL
e_alt_cl -0.0585 Melhem 2022 Table 3, Effect of ALT on CL
e_alb_vc -0.153 Melhem 2022 Table 3, Effect of ALB on Vc
e_sex_cl 0.165 Melhem 2022 Table 3, Effect of male on CL
e_sex_vc 0.162 Melhem 2022 Table 3, Effect of male on Vc
IIV block etalcl + etalvc c(0.0551, 0.0210, 0.0258) Melhem 2022 Table 3 (omega^2 and covariance)
etalImax 0.537 Melhem 2022 Table 3, omega^2 I_max
propSd 0.133 Melhem 2022 Table 3, proportional residual
addSd (mg/L) 2.79 Melhem 2022 Table 3, additive residual

Equations:

  • d/dt(central) and d/dt(peripheral1): standard two-compartment IV micro-constant form (kel = cl/vc, k12 = q/vc, k21 = q/vp). Source: Melhem 2022 Results, equation block (“The final PopPK model is mathematically described by the following equations”).
  • Time-dependent CL via exp-Hill: same equation block.
  • Covariate effects on CL and Vc: power form (cov / ref)^theta for continuous covariates, multiplicative (1 + theta * SEXM) for sex.

Virtual cohort

Original observed data are not publicly available. The simulations below use a virtual cohort whose covariate distributions approximate the published trial demographics (Melhem 2022 Table 2).

set.seed(2022)
n_subj <- 200

cohort <- tibble(
  ID   = seq_len(n_subj),
  WT   = pmin(pmax(rlnorm(n_subj, log(71.4), 0.27), 34),  182),  # Table 2: median 71.4 kg, range 34-182
  AGE  = pmin(pmax(rnorm(n_subj,  62.5, 11.0),       24),  86),  # Table 2: mean 62.5, range 24-86
  ALB  = pmin(pmax(rnorm(n_subj,  38.2, 5.1),        19,        51)),  # Table 2: mean 38.2 g/L, range 19-51
  ALT  = pmin(pmax(rlnorm(n_subj, log(17), 0.5),     2.9),      120),  # Table 2: median 17, range 2.9-120
  SEXF = rbinom(n_subj, 1, 0.773)                                       # Table 2: 77.3% female
)

The recommended therapeutic dose (RTD) is 500 mg IV Q3W for four cycles followed by 1000 mg IV Q6W. The simulation runs for one year of dosing.

# RTD schedule (days from first dose).
loading_doses_d <- seq(0, by = 21, length.out = 4)              # 500 mg Q3W x 4
maint_start_d   <- max(loading_doses_d) + 21                    # first 1000 mg dose 21 d after the 4th 500 mg
maint_doses_d   <- seq(maint_start_d, by = 42, length.out = 12) # 1000 mg Q6W

obs_times_d <- sort(unique(c(seq(0, 365, by = 1), loading_doses_d, maint_doses_d)))

build_events <- function(pop) {
  d_load <- pop |>
    tidyr::crossing(TIME = loading_doses_d) |>
    mutate(AMT = 500, EVID = 1, CMT = "central", DUR = 0.5 / 24, DV = NA_real_,
           treatment = "RTD 500 mg Q3W x4 -> 1000 mg Q6W")
  d_maint <- pop |>
    tidyr::crossing(TIME = maint_doses_d) |>
    mutate(AMT = 1000, EVID = 1, CMT = "central", DUR = 0.5 / 24, DV = NA_real_,
           treatment = "RTD 500 mg Q3W x4 -> 1000 mg Q6W")
  d_obs <- pop |>
    tidyr::crossing(TIME = obs_times_d) |>
    mutate(AMT = NA_real_, EVID = 0, CMT = "central", DUR = NA_real_,
           DV = NA_real_, treatment = "RTD 500 mg Q3W x4 -> 1000 mg Q6W")
  dplyr::bind_rows(d_load, d_maint, d_obs) |>
    dplyr::arrange(ID, TIME, dplyr::desc(EVID)) |>
    as.data.frame()
}

events <- build_events(cohort)

Simulation 

mod <- readModelDb("Melhem_2022_dostarlimab")
sim <- rxSolve(mod, events = events, returnType = "data.frame")

Concentration-time profiles

Median and 5-95th percentile envelope across the virtual cohort under the recommended therapeutic dose:

sim_summary <- sim |>
  dplyr::filter(time > 0) |>
  dplyr::group_by(time) |>
  dplyr::summarise(
    median = stats::median(Cc, na.rm = TRUE),
    lo     = stats::quantile(Cc, 0.05, na.rm = TRUE),
    hi     = stats::quantile(Cc, 0.95, na.rm = TRUE),
    .groups = "drop"
  )

ggplot(sim_summary, aes(time, median)) +
  geom_ribbon(aes(ymin = lo, ymax = hi), alpha = 0.2, fill = "steelblue") +
  geom_line(linewidth = 1, colour = "steelblue") +
  labs(
    x = "Time (days)",
    y = "Dostarlimab serum concentration (mg/L)",
    title = "Simulated dostarlimab PK at the recommended therapeutic dose",
    subtitle = paste0("Median and 90% prediction interval (N = ",
                      n_subj, " virtual patients)"),
    caption = "Model: Melhem 2022 Br J Clin Pharmacol 88(9):4142-4154"
  ) +
  theme_bw()

Time-dependent clearance

Melhem 2022 reports time-dependent clearance with a sigmoid I_max function of time since first dose. The typical-value CL profile below reproduces the time course at the reference patient (deterministic, etas = 0):

t_grid <- seq(0, 365, by = 5)
events_cl <- data.frame(
  ID   = 1, WT = 70, AGE = 64, ALB = 39, ALT = 18, SEXF = 1,
  TIME = c(0, t_grid),
  AMT  = c(500, rep(NA_real_, length(t_grid))),
  EVID = c(1,  rep(0, length(t_grid))),
  CMT  = "central",
  DUR  = c(0.5 / 24, rep(NA_real_, length(t_grid))),
  DV   = NA_real_
)
sim_cl <- rxSolve(
  mod, events = events_cl,
  params = c(WT = 70, AGE = 64, ALB = 39, ALT = 18, SEXF = 1,
             etalcl = 0, etalvc = 0, etalImax = 0),
  omega  = NA,
  returnType = "data.frame"
)
sim_cl <- sim_cl[sim_cl$time > 0, ]

ggplot(sim_cl, aes(time, cl / cl_base)) +
  geom_line(linewidth = 1, colour = "steelblue") +
  geom_hline(yintercept = exp(-0.161), linetype = "dashed", colour = "grey40") +
  labs(
    x = "Time since first dose (days)",
    y = "CL(t) / CL(0)",
    title = "Time-dependent clearance in Melhem 2022",
    subtitle = "Reference patient; asymptotes at exp(I_max) = 0.851 (14.9% reduction)"
  ) +
  theme_bw()

PKNCA validation

NCA computed over the third 21-day dosing interval (2nd-3rd 500 mg dose window, days 21-42 after first dose) for the simulated reference cohort:

interval_start <- 21
interval_end   <- 42

sim_nca <- sim |>
  dplyr::filter(!is.na(Cc),
                time >= interval_start,
                time <= interval_end) |>
  dplyr::mutate(time_rel  = time - interval_start,
                treatment = "RTD 500 mg Q3W x4 -> 1000 mg Q6W") |>
  dplyr::select(id, treatment, time_rel, Cc)

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

dose_df <- data.frame(
  id        = cohort$ID,
  treatment = "RTD 500 mg Q3W x4 -> 1000 mg Q6W",
  time_rel  = 0,
  amt       = 500
)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time_rel | treatment + id)

intervals <- data.frame(
  start     = 0,
  end       = 21,
  cmax      = TRUE,
  tmax      = TRUE,
  auclast   = TRUE,
  half.life = TRUE
)

nca_data <- PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals)
nca_res  <- PKNCA::pk.nca(nca_data)
#>  ■■■■■■■■■■■                       35% |  ETA:  3s
knitr::kable(summary(nca_res),
             caption = "Simulated NCA parameters (3rd dosing interval, days 21-42 after first dose)")
Simulated NCA parameters (3rd dosing interval, days 21-42 after first dose)
start end treatment N auclast cmax tmax half.life
0 21 RTD 500 mg Q3W x4 -> 1000 mg Q6W 200 2340 [20.7] 188 [18.9] 1.00 [1.00, 1.00] 25.9 [4.86]

Comparison against published NCA

Melhem 2022 Table 4 reports per-patient cycle-1 (Day 21) exposure summary statistics for the efficacy ER subset (N = 362, all from the RTD Part 2B arm):

Metric Published (Table 4) Notes
Cmin (mg/L) mean 41.3 (SD 10.5), range 13.7-70.2 Cycle 1 trough, RTD
Cmax (mg/L) mean 177 (SD 32.8), range 103-284 Cycle 1 peak, RTD
AUC (mg*h/L) mean 34 400 (SD 6090), range 19 100-51 000 Cycle 1 over 21 d, RTD

Note the published AUC is reported in mgh/L but is computed over the 21-day cycle 1 interval; the simulated auclast from PKNCA in the table above is in mgday/L because the simulation time variable is in days. Multiplying simulated auclast by 24 converts to mg*h/L for comparison. Simulated cycle-1 Cmax in the 150-200 mg/L range (single 500 mg dose into ~3 L Vc, with a small fraction redistributed peripherally by day 1) is consistent with the published Table 4 mean of 177 mg/L.

Assumptions and deviations

  • Residual error convention. Melhem 2022 Table 3 reports the proportional residual variability as 0.133 (unitless) and the additive residual variability as 2.79 mg/L. The additive entry carries the units of the observation (mg/L, not mg2/L2), indicating that the table column is the standard deviation, not the variance. Both entries are therefore used directly as propSd and addSd in nlmixr2’s combined add(addSd) + prop(propSd) form. A 13.3% CV is consistent with mAb assay residuals reported in comparable anti-PD-1 papers.
  • Imax parameterization. I_max is always negative in the source (-0.161). To keep every individual I_max strictly negative under log-normal IIV, the model stores log|I_max| and applies the negative sign in the model() block: Imax_i <- -exp(lImax + etalImax). The table’s omega^2_Imax = 0.537 maps to omega = sqrt(0.537) = 0.733, which the paper labels “73.3% CV” — i.e., omega is reported as a CV without the log-normal CV = sqrt(exp(omega^2) - 1) correction. The packaged model uses 0.537 directly as the variance of etalImax, preserving the paper’s table value at the cost of a ~10 percentage- point divergence between “labelled CV” and “true log-normal CV”; the same parameterization is used by Masters_2022_avelumab.R.
  • Albumin units typo in Methods. The Methods text describes the reference patient with “ALB 39 g dL-1”, but Table 2 lists albumin in g/L (median 39, range 19-51). 39 g/dL is physiologically impossible (normal serum albumin is 35-50 g/L = 3.5-5 g/dL). The model uses g/L as the canonical unit and 39 g/L as the reference, consistent with Table 2 and with the same range reported elsewhere in the paper.
  • Time variable. The packaged model uses rxode2’s t (simulation time in days) for the time-dependent CL term. Simulations must therefore start at t = 0 with the first dose so that the time-on-CL profile aligns with the source’s “time since first dose” convention.
  • Time-varying covariates. ALB and ALT enter the published model as time-varying covariates. The simulations in this vignette hold each subject’s ALB and ALT fixed at their baseline value over the dosing window. Modelling on-treatment ALB/ALT trajectories is out of scope for the packaged structural model; users with longitudinal lab data can pass time-varying ALB / ALT columns to rxSolve directly via the event table.
  • Race / ethnicity in the cohort. The simulated cohort does not include explicit race / ethnicity columns because race did not enter the final population PK model in Melhem 2022. The race distribution in the published analysis set (Table 2: 75.1% White, 3.5% Black, 2.4% Asian, 16.3% Not reported) is recorded in the model’s population metadata for users who need to stratify simulations.
  • Race covariates not declared. No RACE_* covariate is registered in covariateData because the structural model has no race effect.
  • Erratum. A 15 June 2022 publisher correction was issued for the copyright line only; no parameter, equation, or unit was changed.