Skip to contents

Elotuzumab with pomalidomide/dexamethasone (Ide 2022)

Source: vignettes/articles/Ide_2022_elotuzumab.Rmd
Ide_2022_elotuzumab.Rmd

Model and source

  • Citation: Ide T, Osawa M, Sanghavi K, Vezina HE. Population pharmacokinetic and exposure-response analyses of elotuzumab plus pomalidomide and dexamethasone for relapsed and refractory multiple myeloma. Cancer Chemother Pharmacol. 2022;89(1):129-140.
  • Article: https://doi.org/10.1007/s00280-021-04365-4 (open access; the model equations and final estimates are in the Supplementary Material: the full NONMEM control stream in the Supplementary Methods and the parameter estimates in Supplementary Table S2).

Elotuzumab is a humanized IgG1 monoclonal antibody against SLAMF7, given with an immunomodulatory backbone in relapsed/refractory multiple myeloma. Ide 2022 updates the earlier elotuzumab population PK model (Gibiansky 2016; Ide 2020, available here as Ide_2020_elotuzumab) with data from ELOQUENT-3, in which elotuzumab was combined with pomalidomide/dexamethasone (Pd) and given as 10 mg/kg weekly for two 28-day cycles followed by 20 mg/kg every 4 weeks.

The structure is unchanged from the earlier model: two compartments, parallel linear and Michaelis-Menten elimination from the central compartment, and second-order target-mediated elimination from the peripheral compartment against a non-renewable target pool (initial concentration RMAX):

dcentraldt=−k12central+k21peripheral1−kelcentral−VmaxcentralCc+KM \frac{d\,\mathrm{central}}{dt} = -k_{12}\,\mathrm{central} + k_{21}\,\mathrm{peripheral1} - k_{el}\,\mathrm{central} - \frac{V_{max}\,\mathrm{central}}{C_c + K_M} dperipheral1dt=k12central−k21peripheral1−kintperipheral1⋅target \frac{d\,\mathrm{peripheral1}}{dt} = k_{12}\,\mathrm{central} - k_{21}\,\mathrm{peripheral1} - k_{int}\,\mathrm{peripheral1}\cdot\mathrm{target} dtargetdt=−kintperipheral1VPtarget,target(0)=Rmax \frac{d\,\mathrm{target}}{dt} = -k_{int}\,\frac{\mathrm{peripheral1}}{V_P}\,\mathrm{target}, \qquad \mathrm{target}(0) = R_{max}

Compared with Ide 2020, the update

  1. adds a pomalidomide/dexamethasone effect on linear CL (factor 0.811) and on KINT (factor 0.487), relative to the lenalidomide/dexamethasone (Ld) reference;
  2. keeps the monotherapy contrast (CL x 1/0.825, KINT x 1/9.78);
  3. switches off the CL covariates of the earlier model (age, eGFR, LDH, albumin, hepatic function, ECOG, sex, race and beta-2 microglobulin on CL are all 0 FIXED in the control stream), and keeps sex, Asian race and beta-2 microglobulin >= 3.5 mg/L on the central volume;
  4. uses time-varying serum M-protein on VMAX.
mod <- readModelDb("Ide_2022_elotuzumab")
mod_ui <- rxode2::rxode2(mod)
#> ℹ parameter labels from comments will be replaced by 'label()'
# zeroRe() warns that there are no sigma parameters: the residual SD is the
# model expression W, so there is nothing to zero on that side.
mod_typical <- suppressWarnings(rxode2::zeroRe(mod_ui))

Population

The model was fit to 8180 elotuzumab serum concentrations from 440 patients pooled from five trials (Ide 2022 Table 1): CA204-011 (elotuzumab monotherapy, high-risk smoldering myeloma), CA204-004 / ELOQUENT-2, CA204-005 and CA204-007 (with Ld), and CA204-125 / ELOQUENT-3 (with Pd). Median age was 66 years (37-88) and median body weight 75 kg (40-150); 41% were female and 13% Asian. Co-administration was monotherapy in 31 patients (7%), Ld in 349 (79%) and Pd in 60 (14%). Baseline serum M-protein had median 2.05 g/dL (0-7.7) and beta-2 microglobulin median 0.32 mg/dL (0.04-3.47).

The same information is available programmatically as readModelDb("Ide_2022_elotuzumab") metadata (population).

Source trace

Every value below is taken from Supplementary Table S2 (final estimates) of Ide 2022; the equations are from the Supplementary Methods NONMEM control stream.

Model element Value Source
lcl (CL_REF, L/day) log(0.0834) Table S2, CLREF = exp(theta1)
lvc (VC_REF, L) log(4.06) Table S2, VCREF = exp(theta2)
lq (Q_REF, L/day) log(0.512) Table S2, QREF = exp(theta3)
lvp (VP_REF, L) log(1.93) Table S2, VPREF = exp(theta4)
lrmax (ug/mL) log(849) Table S2, RMAX = exp(theta5)
lkint (mL/ug/day) log(0.216e-3) Table S2, KINTREF = 0.216 x 10^-3 /day/(ug/mL); control stream KINT = EXP(MU_6+ETA(6))/1000
lvmax (ug/mL/day) log(12.2) Table S2, VMAXREF at MCPROT = 0 g/dL
lkm (ug/mL) log(281) Table S2, KM
e_wt_cl, e_wt_vc, e_wt_q, e_wt_vp 1.32, 0.345, 0.75 (fixed), 0.696 Table S2, CLWT / VCWT / QWT / VPWT; reference 75 kg (VWT = WT/75)
e_mono_cl, e_mono_kint log(0.825), log(9.78) Table S2, CLMono / KINTMono, applied as (factor)^-Mono
e_combo_pom_dex_cl, e_combo_pom_dex_kint log(0.811), log(0.487) Table S2, CLPomDex / KINTPomDex
e_sexf_vc, e_race_asian_vc log(0.797), log(0.885) Table S2, VCSEX / VCRACE
e_b2m_ge35_vc log(1.11) Table S2, VCB2MICG>0.35 (theta36); control stream B2MICG.GE.0.35 mg/dL
e_mcprot_vmax 0.27 per g/dL Table S2, VMAXMCPROT (theta23)
IIV CL / VC / Q / VP / RMAX / KINT / KM 0.158 / 0.0361 / 0.454 / 0.133 / 0.189 / 1.69 / 0.385 Table S2, omega2
IIV VMAX 0.0001 (fixed) Table S2, footnote d
etaruv (IIV on residual magnitude) 0.183 Table S2, omega2 epsilon
sdL, sdH, sd50, sdPhase12 2.46, 0.0976, 6.17 ug/mL, 0.843 Table S2, SDL / SDH / SD50 / SDphase1,2
ODEs, target(0) = rmax see above control stream $DES, A_0(3)=RMAX
Residual error W = (sdL - (sdL - sdH) Cc/(sd50 + Cc)) sdPhase12^(1 - STUDY_PHASE3) exp(etaruv), Cc ~ lnorm(W) control stream $ERROR: W = (SDL-(SDL-SDH)*TY/(SD50+TY))*THETA(16)**STOTHER*EXP(ETA(9)), Y = LOG(TY) + W*EPS(1), $SIGMA 1 FIXED

Covariate effects: checks against the paper’s statements

The Results state that CL at 50.6 kg (5th percentile) and 105 kg (95th percentile) is about 41% lower and 55% higher than at the 75 kg reference, and that Pd lowers linear CL by 19% and KINT by 51% relative to Ld. These are deterministic functions of the typical-value parameters, so they are checked tightly.

ref_cov <- data.frame(
  WT = 75, SEXF = 0, RACE_ASIAN = 0, B2M = 3, MCPROT = 0,
  COMBO_LEN_DEX = 1, COMBO_POM_DEX = 0, STUDY_PHASE3 = 1
)
scenario <- function(label, ...) {
  out <- ref_cov
  changes <- list(...)
  for (nm in names(changes)) out[[nm]] <- changes[[nm]]
  out$scenario <- label
  out
}
scen <- dplyr::bind_rows(
  scenario("Reference (75 kg, Ld)"),
  scenario("WT 50.6 kg", WT = 50.6),
  scenario("WT 105 kg", WT = 105),
  scenario("Pd", COMBO_LEN_DEX = 0, COMBO_POM_DEX = 1),
  scenario("Monotherapy", COMBO_LEN_DEX = 0),
  scenario("Female", SEXF = 1),
  scenario("Asian", RACE_ASIAN = 1),
  scenario("B2M 3.5 mg/L", B2M = 3.5)
)
scen$id <- seq_len(nrow(scen))
ev_cov <- dplyr::bind_rows(
  dplyr::mutate(scen, time = 0, amt = 750, evid = 1L, cmt = "central"),
  dplyr::mutate(scen, time = 1, amt = 0, evid = 0L, cmt = "central")
) |>
  dplyr::arrange(id, time)

par_cov <- rxode2::rxSolve(mod_typical, events = ev_cov, keep = "scenario") |>
  as.data.frame() |>
  dplyr::distinct(scenario, cl, vc, kint)
#> ℹ omega/sigma items treated as zero: 'etalcl', 'etalvc', 'etalq', 'etalvp', 'etalrmax', 'etalkint', 'etalvmax', 'etalkm', 'etaruv'
#> Warning: multi-subject simulation without without 'omega'
ref_row <- par_cov[par_cov$scenario == "Reference (75 kg, Ld)", ]
par_cov <- par_cov |>
  dplyr::mutate(
    cl_ratio = cl / ref_row$cl,
    vc_ratio = vc / ref_row$vc,
    kint_ratio = kint / ref_row$kint
  )
par_cov |>
  dplyr::mutate(kint = kint * 1000) |>
  dplyr::rename(
    Scenario = scenario, `CL (L/day)` = cl, `VC (L)` = vc,
    `KINT (1e-3 mL/ug/day)` = kint, `CL ratio` = cl_ratio,
    `VC ratio` = vc_ratio, `KINT ratio` = kint_ratio
  ) |>
  knitr::kable(digits = 4, caption = "Typical-value parameters relative to the reference patient.")
Typical-value parameters relative to the reference patient.
Scenario CL (L/day) VC (L) KINT (1e-3 mL/ug/day) CL ratio VC ratio KINT ratio
Reference (75 kg, Ld) 0.0834 4.0600 0.2160 1.0000 1.0000 1.0000
WT 50.6 kg 0.0496 3.5446 0.2160 0.5948 0.8730 1.0000
WT 105 kg 0.1300 4.5597 0.2160 1.5592 1.1231 1.0000
Pd 0.0676 4.0600 0.1052 0.8110 1.0000 0.4870
Monotherapy 0.1011 4.0600 0.0221 1.2121 1.0000 0.1022
Female 0.0834 3.2358 0.2160 1.0000 0.7970 1.0000
Asian 0.0834 3.5931 0.2160 1.0000 0.8850 1.0000
B2M 3.5 mg/L 0.0834 4.5066 0.2160 1.0000 1.1100 1.0000 

ratio <- function(s, col) par_cov[[col]][par_cov$scenario == s]
stopifnot(
  nrow(par_cov) == nrow(scen),
  abs(ratio("WT 50.6 kg", "cl_ratio") - (1 - 0.41)) < 0.01,
  abs(ratio("WT 105 kg", "cl_ratio") - (1 + 0.55)) < 0.01,
  abs(ratio("Pd", "cl_ratio") - (1 - 0.19)) < 0.005,
  abs(ratio("Pd", "kint_ratio") - (1 - 0.51)) < 0.005,
  abs(ratio("Monotherapy", "cl_ratio") - 1 / 0.825) < 1e-6,
  abs(ratio("Monotherapy", "kint_ratio") - 1 / 9.78) < 1e-6,
  abs(ratio("Female", "vc_ratio") - 0.797) < 1e-6,
  abs(ratio("Asian", "vc_ratio") - 0.885) < 1e-6,
  abs(ratio("B2M 3.5 mg/L", "vc_ratio") - 1.11) < 1e-6
)

Linear versus target-mediated elimination (Figure 1)

Figure 1 of Ide 2022 contrasts the constant linear clearance with the Michaelis-Menten clearance from the central compartment, which falls as the concentration rises. The Michaelis-Menten term VMAX * central / (Cc + KM) is a clearance of VMAX * VC / (Cc + KM).

ref_par <- ref_row
conc_grid <- 10^seq(0, 3, length.out = 200)
theta <- mod_ui$theta
clear <- dplyr::bind_rows(lapply(c(0, 2.05), function(mp) {
  vmax_i <- exp(theta[["lvmax"]]) * exp(theta[["e_mcprot_vmax"]] * mp)
  data.frame(
    conc = conc_grid,
    `Linear CL` = ref_par$cl,
    `Michaelis-Menten` = vmax_i * ref_par$vc / (conc_grid + exp(theta[["lkm"]])),
    MCPROT = paste(mp, "g/dL"),
    check.names = FALSE
  )
})) |>
  tidyr::pivot_longer(c(`Linear CL`, `Michaelis-Menten`), names_to = "pathway", values_to = "clearance")

ggplot(clear, aes(conc, clearance, colour = pathway, linetype = MCPROT)) +
  geom_line() +
  scale_x_log10("Elotuzumab concentration (ug/mL)") +
  scale_y_continuous("Clearance (L/day)") +
  theme_minimal()
Replicates the idea of Figure 1 of Ide 2022: typical linear (nonspecific) clearance and Michaelis-Menten clearance from the central compartment versus elotuzumab concentration, at M-protein 0 and 2.05 g/dL (the dataset median).

Replicates the idea of Figure 1 of Ide 2022: typical linear (nonspecific) clearance and Michaelis-Menten clearance from the central compartment versus elotuzumab concentration, at M-protein 0 and 2.05 g/dL (the dataset median).

Virtual cohort

The observed data are not public. The cohort below approximates Table 1 of Ide 2022: body weight normal (mean 75.6, SD 16.7 kg) redrawn until inside the observed 40-150 kg range; 41% female; 13% Asian; beta-2 microglobulin log-normal with median 3.2 mg/L (0.32 mg/dL); baseline M-protein gamma with the Table 1 mean 2.25 and SD 1.58 g/dL. M-protein is held at each subject’s baseline value for the whole simulation; see “Assumptions and deviations”. Two arms of 200 subjects share the same covariates:

  • E-Ld: 10 mg/kg IV weekly for two 28-day cycles, then 10 mg/kg every 2 weeks (ELOQUENT-2 regimen);
  • E-Pd: 10 mg/kg IV weekly for two 28-day cycles, then 20 mg/kg every 4 weeks (ELOQUENT-3 regimen).
rxode2::rxSetSeed(20220101)
set.seed(20220101)
n_per_arm <- 200L

draw_wt <- function(n) {
  out <- rnorm(n, 75.6, 16.7)
  bad <- out < 40 | out > 150
  while (any(bad)) {
    out[bad] <- rnorm(sum(bad), 75.6, 16.7)
    bad <- out < 40 | out > 150
  }
  out
}
mp_shape <- (2.25 / 1.58)^2
cohort <- data.frame(
  subj = seq_len(n_per_arm),
  WT = draw_wt(n_per_arm),
  SEXF = rbinom(n_per_arm, 1, 0.41),
  RACE_ASIAN = rbinom(n_per_arm, 1, 0.13),
  B2M = rlnorm(n_per_arm, log(3.2), 0.77),
  MCPROT = rgamma(n_per_arm, shape = mp_shape, scale = 2.25 / mp_shape)
)

infusion_days <- 2 / 24
end_day <- 392
regimen <- function(arm) {
  weekly <- seq(0, 49, by = 7)
  if (arm == "E-Ld") {
    maint <- seq(56, end_day - 14, by = 14)
    data.frame(time = c(weekly, maint), mg_per_kg = 10)
  } else {
    maint <- seq(56, end_day - 28, by = 28)
    data.frame(time = c(weekly, maint), mg_per_kg = c(rep(10, 8), rep(20, length(maint))))
  }
}

build_arm <- function(arm, id_offset) {
  doses <- regimen(arm)
  obs_times <- sort(unique(c(seq(0, end_day, by = 0.5), doses$time + infusion_days)))
  dplyr::bind_rows(lapply(seq_len(nrow(cohort)), function(i) {
    cov_i <- cohort[i, ]
    dose_rows <- data.frame(
      time = doses$time, amt = doses$mg_per_kg * cov_i$WT,
      dur = infusion_days, evid = 1L
    )
    obs_rows <- data.frame(time = obs_times, amt = 0, dur = 0, evid = 0L)
    dplyr::bind_rows(dose_rows, obs_rows) |>
      dplyr::mutate(
        id = id_offset + i, cmt = "central", treatment = arm,
        WT = cov_i$WT, SEXF = cov_i$SEXF, RACE_ASIAN = cov_i$RACE_ASIAN,
        B2M = cov_i$B2M, MCPROT = cov_i$MCPROT,
        COMBO_LEN_DEX = as.integer(arm == "E-Ld"),
        COMBO_POM_DEX = as.integer(arm == "E-Pd"),
        STUDY_PHASE3 = as.integer(arm == "E-Ld")
      )
  })) |>
    dplyr::arrange(id, time, dplyr::desc(evid))
}
events <- dplyr::bind_rows(build_arm("E-Ld", 0L), build_arm("E-Pd", 1000L))
stopifnot(all(table(unique(events[, c("id", "treatment")])$treatment) == n_per_arm))

Simulation 

sim <- rxode2::rxSolve(mod_ui, events = events, keep = "treatment") |>
  as.data.frame()
stopifnot(all(is.finite(sim$Cc)), all(sim$Cc >= 0))

Concentration-time profiles (Figure 2) 

prof <- sim |>
  dplyr::group_by(treatment, time) |>
  dplyr::summarise(
    median = median(Cc), p05 = quantile(Cc, 0.05), p95 = quantile(Cc, 0.95),
    .groups = "drop"
  )
ggplot(prof, aes(time / 7, median, colour = treatment, fill = treatment)) +
  geom_ribbon(aes(ymin = p05, ymax = p95), alpha = 0.2, colour = NA) +
  geom_line() +
  scale_x_continuous("Time since first dose (weeks)", breaks = seq(0, 56, by = 8)) +
  scale_y_continuous("Elotuzumab concentration (ug/mL)") +
  theme_minimal()
Replicates Figure 2 of Ide 2022: median (line) and 5th-95th percentile band of simulated elotuzumab concentrations for 10 mg/kg weekly (cycles 1-2) followed by 10 mg/kg Q2W with Ld, or 20 mg/kg Q4W with Pd.

Replicates Figure 2 of Ide 2022: median (line) and 5th-95th percentile band of simulated elotuzumab concentrations for 10 mg/kg weekly (cycles 1-2) followed by 10 mg/kg Q2W with Ld, or 20 mg/kg Q4W with Pd.

PKNCA validation

Table 2 of Ide 2022 reports geometric means of the average, minimum and maximum concentration over the first dosing interval (days 0-7) and at steady state. For the first interval the minimum is the trough just before the second dose on day 7 (PKNCA ctrough); the interval minimum itself would be the pre-dose zero at time 0. The paper states steady state is reached by about 16 weeks; the steady-state interval used here is the maintenance interval starting on day 336 (week 48), which is 14 days for E-Ld and 28 days for E-Pd.

conc_df <- sim |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::select(id, time, Cc, treatment)
dose_df <- events |>
  dplyr::filter(evid == 1L) |>
  dplyr::select(id, time, amt, treatment)

conc_obj <- PKNCA::PKNCAconc(conc_df, Cc ~ time | treatment + id, concu = "ug/mL", timeu = "day")
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | treatment + id, doseu = "mg")

intervals <- data.frame(
  treatment = c("E-Ld", "E-Pd", "E-Ld", "E-Pd"),
  start = c(0, 0, 336, 336),
  end = c(7, 7, 350, 364),
  cmax = TRUE, cav = TRUE,
  ctrough = c(TRUE, TRUE, FALSE, FALSE),
  cmin = c(FALSE, FALSE, TRUE, TRUE)
)
nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))

nca_geo <- as.data.frame(nca_res$result) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "cmin", "ctrough", "cav")) |>
  dplyr::mutate(
    interval = ifelse(start == 0, "First dose", "Steady state"),
    # Table 2 calls the first-interval trough Cmin1.
    PPTESTCD = ifelse(PPTESTCD == "ctrough", "cmin", PPTESTCD)
  ) |>
  dplyr::group_by(treatment, interval, PPTESTCD) |>
  dplyr::summarise(
    n = dplyr::n(),
    PPORRES = exp(mean(log(PPORRES))),
    .groups = "drop"
  )
stopifnot(all(nca_geo$n == n_per_arm), nrow(nca_geo) == 12L)

Comparison against published exposures 

published <- data.frame(
  treatment = c("E-Ld", "E-Pd", "E-Ld", "E-Pd"),
  interval = c("First dose", "First dose", "Steady state", "Steady state"),
  cav = c(114, 113, 260, 266),
  cmin = c(63.4, 69.7, 179, 124),
  cmax = c(195, 185, 394, 543)
)
cmp <- nlmixr2lib::ncaComparisonTable(
  simulated = dplyr::select(nca_geo, -n),
  reference = published,
  by = c("treatment", "interval"),
  units = c(cav = "ug/mL", cmin = "ug/mL", cmax = "ug/mL")
)
knitr::kable(cmp, caption = "Geometric means: Ide 2022 Table 2 versus the simulated cohort.")
Geometric means: Ide 2022 Table 2 versus the simulated cohort.
NCA parameter treatment interval Reference Simulated % diff
Cmax (ug/mL) E-Ld First dose 195 194 -0.4%
Cmax (ug/mL) E-Ld Steady state 394 349 -11.4%
Cmax (ug/mL) E-Pd First dose 185 195 +5.6%
Cmax (ug/mL) E-Pd Steady state 543 527 -3.0%
Cmin (ug/mL) E-Ld First dose 63.4 64.9 +2.4%
Cmin (ug/mL) E-Ld Steady state 179 126 -29.8%*
Cmin (ug/mL) E-Pd First dose 69.7 72.6 +4.2%
Cmin (ug/mL) E-Pd Steady state 124 92.5 -25.4%*
Cavg (ug/mL) E-Ld First dose 114 114 -0.2%
Cavg (ug/mL) E-Ld Steady state 260 209 -19.4%
Cavg (ug/mL) E-Pd First dose 113 118 +4.4%
Cavg (ug/mL) E-Pd Steady state 266 236 -11.2% 
if (!is.null(attr(cmp, "footnote"))) cat(attr(cmp, "footnote"))
#> * differs from reference by more than ±20%.

geo <- function(trt, int, p) {
  nca_geo$PPORRES[nca_geo$treatment == trt & nca_geo$interval == int & nca_geo$PPTESTCD == p]
}
stopifnot(
  # First-dose exposure depends on the dose, VC and linear/MM clearance but
  # hardly on the M-protein trajectory, so it pins the structural model.
  abs(geo("E-Ld", "First dose", "cav") / 114 - 1) < 0.15,
  abs(geo("E-Pd", "First dose", "cav") / 113 - 1) < 0.15,
  abs(geo("E-Ld", "First dose", "cmin") / 63.4 - 1) < 0.15,
  abs(geo("E-Pd", "First dose", "cmin") / 69.7 - 1) < 0.15,
  # The Q4W regimen gives a higher peak and a lower trough than Q2W at steady
  # state (Table 2: +38% and -31%).
  geo("E-Pd", "Steady state", "cmax") / geo("E-Ld", "Steady state", "cmax") > 1.2,
  geo("E-Pd", "Steady state", "cmin") < geo("E-Ld", "Steady state", "cmin")
)

All three first-dose geometric means (Cavg1, Cmin1, Cmax1) are within about 6% of Table 2 for both regimens, and the steady-state regimen contrast (higher peak and lower trough with 20 mg/kg Q4W) is reproduced. The steady-state values fall below the published geometric means, by about 3-11% for Cmax, 11-19% for Cavg and 25-30% for Cmin. The trough is the most sensitive to the Michaelis-Menten pathway, which is largest at low concentrations. The shortfall is expected because the cohort holds M-protein at its baseline value. In the trials M-protein fell during treatment in most patients (Discussion of Ide 2022), which lowers VMAX and raises steady-state exposure, and the paper derived its exposures from each patient’s measured, linearly interpolated M-protein.

Assumptions and deviations

  • M-protein held constant. MCPROT is time-varying in the model and the paper interpolated each patient’s measurements. No M-protein trajectory is published, so the virtual cohort keeps each subject at a baseline value drawn to match Table 1. This is the main reason the simulated steady-state exposures are lower than Table 2. To simulate a responding patient, supply MCPROT as a declining time series on the event rows.
  • Infusion duration. The paper models a zero-order infusion but does not state the infusion duration; 2 hours is assumed for all doses.
  • Backbone columns. The source control stream’s LENDEX flag is 1 for every non-monotherapy study, including the Pd study, so it means “any immunomodulatory backbone”. The model uses two mutually exclusive indicators instead, COMBO_LEN_DEX (Ld) and COMBO_POM_DEX (Pd), with monotherapy when both are 0; the monotherapy factors (CLMono, KINTMono) are applied as (factor)^-mono exactly as in Table S2.
  • Residual error. The saturable log-scale residual error is encoded as in the control stream, including the study multiplier (sdPhase12 = 0.843 for the phase 1/2 studies, selected by STUDY_PHASE3 = 0) and the inter-individual variability on the residual magnitude (etaruv). The control stream’s floor LTY = -2.5 for non-positive predictions is not needed in simulation.
  • Fixed-to-zero covariates. The control stream carries twelve CL and VC covariate terms of the predecessor model as 0 FIXED (theta24-theta35). They contribute nothing and are listed in the model’s covariatesDataExcluded metadata rather than in the model.
  • Beta-2 microglobulin units. The dataset column is in mg/dL with the threshold at 0.35 mg/dL; the model takes the canonical B2M in mg/L with the equivalent 3.5 mg/L threshold. Table S2 labels the effect “VCB2MICG>0.35” while the control stream uses .GE.; the model follows the control stream (>=).
  • Exposure-response models not included. The paper’s exposure-response analyses are semi-parametric Cox proportional-hazards models (unspecified baseline hazard) of progression-free survival and grade 3+ adverse events on daily time-varying average concentration. Without a parametric baseline hazard they cannot be simulated, so only the population PK model is provided.