Skip to contents

Tisotumab vedotin ADC and MMAE (Gibiansky 2022)

Source: vignettes/articles/Gibiansky_2022_tisotumab.Rmd
Gibiansky_2022_tisotumab.Rmd

Model and source

  • Citation: Gibiansky L, Passey C, Voellinger J, Gunawan R, Hanley WD, Gupta M, Winter H. Population pharmacokinetic analysis for tisotumab vedotin in patients with locally advanced and/or metastatic solid tumors. CPT Pharmacometrics Syst Pharmacol. 2022;11(10):1358-1370. doi:10.1002/psp4.12850. PMID 35932175; PMCID PMC9574719. Covariate coefficients and variance parameters, which the article does not print, are from the same final model as tabulated in the FDA BLA 761208 Multi-discipline Review (2021), Tables 49 and 50.
  • Description: Joint population PK model for the antibody-drug conjugate tisotumab vedotin (ADC) and its released payload monomethyl auristatin E (MMAE) in adults with locally advanced or metastatic solid tumors (Gibiansky 2022). The ADC is a two-compartment model with parallel linear and Michaelis-Menten elimination from the central compartment. Linear ADC elimination, multiplied by a drug-to-antibody ratio that decays mono-exponentially with time after dose from 4 to 1, feeds an amount-only MMAE delay compartment in full and the central MMAE compartment by an additional fraction FR1; a fraction FR2 of the Michaelis-Menten flux also feeds the delay compartment. The delay compartment drains first-order into a one-compartment MMAE model with apparent clearance and volume. Covariates: body weight, albumin and sex on ADC CL and Vc; weight on Q and Vp; weight, albumin, tumor type, eGFR, tumor size, ECOG performance status and hepatic impairment on MMAE clearance; weight, ECOG and albumin on MMAE volume; age and weight on the delay rate constant.
  • Article: CPT Pharmacometrics Syst Pharmacol 2022;11:1358-1370 (open access, PMC9574719)
  • Regulatory review: FDA BLA 761208 Multi-discipline Review, tisotumab vedotin (2021)

Tisotumab vedotin is an antibody-drug conjugate (ADC) of a tissue-factor-directed human IgG1 and the microtubule-disrupting payload monomethyl auristatin E (MMAE), attached through a protease-cleavable vc linker at an average of four MMAE per antibody. The model describes the conjugated antibody (ADC) and the unconjugated MMAE that is released from it.

The article prints the structural parameters (Table 2) and the model equations (Table S1), but neither the covariate coefficients theta17-theta36 nor the variance parameters. Those come from the sponsor’s population PK report as reproduced in the FDA BLA 761208 Multi-discipline Review (Table 49, covariate fixed effects; Table 50, variance parameters). The review’s Table 48 matches the article’s Table 2 to every printed digit, so both describe the same final model.

Population

Gibiansky 2022 pooled 399 adults with locally advanced or metastatic solid tumors known to express tissue factor from four phase I/II trials: innovaTV 201 (NCT02001623, n = 195), innovaTV 202 (NCT02552121, n = 33), innovaTV 204 (NCT03438396, n = 101, all cervical cancer) and innovaTV 207 (NCT03485209, n = 70). Mean age was 56.1 years, 74.2% were women and 92.2% were White; 70.7% were enrolled in Europe and 29.3% in the United States. Cervical cancer accounted for 43.1% of patients. Renal function was normal in 53.9%, mildly impaired in 35.6% and moderately impaired in 10.5%; 14.5% had mild hepatic impairment (NCI ODWG). ECOG performance status was 0 in 38.8% and 1 in 61.2% (article Table 1). Median (range) body weight was 70.1 (33-148) kg, albumin 40 (27-52) g/L, tumor size 57 (0-299) mm and CKD-EPI eGFR 91 (24.1-145) mL/min/1.73 m^2 (FDA review Table 45). Doses were 0.3-2.2 mg/kg every 3 weeks (Q3W) in innovaTV 201, 1.2 mg/kg on days 1, 8 and 15 of a 28-day cycle or 2.0 mg/kg Q3W in innovaTV 202, and 2.0 mg/kg Q3W (at most 200 mg) in innovaTV 204 and 207. The same information is stored in the model’s population metadata.

Source trace

Element Value Source
ADC two-compartment model with parallel linear and Michaelis-Menten elimination ODEs for A1, A2 Table S1; Figure 1
MMAE delay compartment and one-compartment MMAE model ODEs for A3, A4 Table S1; Figure 1
DAR = 1 + DAR0 exp(-beta tad), DAR0 = 3 fixed Table S1; Methods
MMAE output scale, MW_RATIO 4.8 (fixed) Table S1
lcl CL 1.42 L/day Table 2 theta1
lq Q 4.01 L/day Table 2 theta2
lvc Vc 3.10 L Table 2 theta3
lvp Vp 4.47 L Table 2 theta4
lvmax Vmax 3.35 ug/mL/day Table 2 theta5
lkm KM 3.44 ug/mL Table 2 theta6
propSd, addSd (ADC) 0.129, 0.0173 ug/mL Table 2 theta7, theta8
lktr ktr 0.271 1/day Table 2 theta9
lcl_mmae CL_MMAE 42.8 L/day Table 2 theta10
lvc_mmae V_MMAE 2.09 L Table 2 theta11
lbeta beta (DAR decay) 0.0189 1/day Table 2 theta12
propSd_mmae, addSd_mmae (MMAE) 0.282, 0.0113 ng/mL Table 2 theta13, theta14
fr1 FR1 0.0205 Table 2 theta15
fr2 FR2 0.0508 Table 2 theta16
Covariate functional forms and reference values (75 kg, 40 g/L, 90 mL/min/1.73 m^2, 60 mm, 60 years) – Table S1
e_wt_cl (CL and Q), e_wt_vc, e_wt_vp 0.495, 0.380, 0.622 FDA review Table 49 theta17-theta19
e_alb_cl, e_alb_vc -0.396, -0.197 FDA review Table 49 theta20-theta21
e_sexm_cl, e_sexm_vc 1.09, 1.13 FDA review Table 49 theta22-theta23
e_wt_cl_mmae, e_wt_vc_mmae 0.457, 0.895 FDA review Table 49 theta24-theta25
e_alb_cl_mmae 0.935 FDA review Table 49 theta26
e_tumtp_other_cl_mmae 1.22 FDA review Table 49 theta27
e_tumsz_cl_mmae -0.147 FDA review Table 49 theta28
e_crcl_cl_mmae 0.271 FDA review Table 49 theta29
e_ecog_cl_mmae, e_hepimp_cl_mmae 0.803, 0.853 FDA review Table 49 theta30-theta31
e_ecog_vc_mmae, e_alb_vc_mmae 0.827, 0.575 FDA review Table 49 theta32-theta33
e_age_ktr, e_wt_ktr -0.252, -0.175 FDA review Table 49 theta34-theta35
e_study_innovatv202_ruv 1.42 FDA review Table 49 theta36
IIV block CL, Vc, Vp 0.0538; 0.0166, 0.0296; 0, 0.0145, 0.0208 FDA review Table 50; Table S1 (CL-Vp covariance 0)
etalktr 0.0212 FDA review Table 50 Omega55
IIV block CL_MMAE, V_MMAE 0.299; 0.125, 0.215 FDA review Table 50
etaRUV, etaRUV_mmae (IIV on residual SD) 0.0561, 0.0712 FDA review Table 50 Omega44, Omega88; Table S1

The IIV coefficients of variation the article quotes in its Results (CL 23.2%, Vc 17.2%, Vp 14.4%, CL_MMAE 54.7%, V_MMAE 46.3%; correlations 0.415, 0.586 and 0.495) are the square roots of these variances and the implied correlations.

mod <- readModelDb("Gibiansky_2022_tisotumab")
# Only the between-subject variability is zeroed; the exposures below use the
# individual prediction columns, which residual error does not touch.
mod_typ <- rxode2::zeroRe(rxode2::rxode2(mod), which = "omega")
#> ℹ parameter labels from comments will be replaced by 'label()'

Derived quantities from the typical parameters

The article derives several secondary quantities from the Table 2 estimates. They are reproduced here in closed form for the 75-kg reference patient.

cl <- 1.42; q <- 4.01; vc <- 3.10; vp <- 4.47
ktr <- 0.271; cl_mmae <- 42.8; vc_mmae <- 2.09
k10 <- cl / vc; k12 <- q / vc; k21 <- q / vp
a <- k10 + k12 + k21
lambda1 <- (a + sqrt(a^2 - 4 * k10 * k21)) / 2
lambda2 <- (a - sqrt(a^2 - 4 * k10 * k21)) / 2
derived <- tibble::tribble(
  ~quantity, ~published, ~computed,
  "ADC distribution half-life (day)", 0.28, log(2) / lambda1,
  "ADC terminal half-life, linear part (day)", 4.19, log(2) / lambda2,
  "MMAE delay half-life log(2)/ktr (day)", 2.56, log(2) / ktr,
  "MMAE half-life log(2)/K_MMAE (day)", 0.0339, log(2) / (cl_mmae / vc_mmae),
  "MMAE steady-state volume CL_MMAE/ktr (L)", 158, cl_mmae / ktr
) |>
  mutate(pct_diff = 100 * (computed - published) / published)
knitr::kable(derived, digits = 4,
             caption = "Derived quantities: Table 2 and Results of Gibiansky 2022; terminal half-life 4.19 days from FDA review Table 48.")
Derived quantities: Table 2 and Results of Gibiansky 2022; terminal half-life 4.19 days from FDA review Table 48.
quantity published computed pct_diff
ADC distribution half-life (day) 0.2800 0.2791 -0.3100
ADC terminal half-life, linear part (day) 4.1900 4.1887 -0.0314
MMAE delay half-life log(2)/ktr (day) 2.5600 2.5577 -0.0883
MMAE half-life log(2)/K_MMAE (day) 0.0339 0.0338 -0.1545
MMAE steady-state volume CL_MMAE/ktr (L) 158.0000 157.9336 -0.0420 
# Pure arithmetic on printed values, so only rounding separates the two columns.
stopifnot(all(abs(derived$pct_diff) < 1))

The article also states that after a 2-mg/kg dose about 40% of the ADC dose is eliminated by the Michaelis-Menten (target-mediated) route and 60% by linear clearance. The model’s own fluxes, integrated over a single dose for the reference patient, give:

ref_cov <- data.frame(
  WT = 75, SEXF = 1, ALB = 40, TUMTP_OTHER = 0, CRCL = 90, TUMSZ = 60,
  ECOG_GE1 = 0, HEPIMP = 0, AGE = 60, STUDY_INNOVATV202 = 0
)
inf_dur <- 30 / 60 / 24 # 30-minute infusion, in days

make_events <- function(dose, n_cycles, obs_times) {
  dosing <- data.frame(
    time = 21 * (seq_len(n_cycles) - 1), evid = 1L, amt = dose,
    rate = dose / inf_dur, cmt = "central", dvid = NA_integer_
  )
  obs <- data.frame(
    time = obs_times, evid = 0L, amt = 0, rate = 0, cmt = "central", dvid = 1L
  )
  bind_rows(dosing, obs) |> arrange(time, desc(evid))
}

sd_grid <- sort(unique(c(seq(0, 1, by = 0.005), seq(1, 150, by = 0.05))))
sd_events <- make_events(150, 1, sd_grid) |> mutate(id = 1L)
sd_sim <- rxode2::rxSolve(mod_typ, sd_events, params = ref_cov,
                          returnType = "data.frame", useLinCmt = FALSE)
#> ℹ omega/sigma items treated as zero: 'etalvc_mmae', 'etalcl_mmae', 'etalvp', 'etalvc', 'etalcl', 'etaRUV_mmae', 'etaRUV', 'etalktr'
# The drug-to-antibody ratio must follow 1 + 3 exp(-beta * tad); a time after
# dose that silently evaluated to 0 would hold it at 4.
day20 <- sd_sim[which.min(abs(sd_sim$time - 20)), ]
stopifnot(abs(day20$dar - (1 + 3 * exp(-0.0189 * day20$time))) < 1e-4)

trap <- function(t, y) sum(diff(t) * (head(y, -1) + tail(y, -1)) / 2)
mm_amount <- trap(sd_sim$time, sd_sim$mm_flux)
lin_amount <- trap(sd_sim$time, sd_sim$kel * sd_sim$central)
frac_mm <- mm_amount / (mm_amount + lin_amount)
cat(sprintf("Fraction of a 2-mg/kg dose eliminated by the Michaelis-Menten route: %.1f%%\n",
            100 * frac_mm))
#> Fraction of a 2-mg/kg dose eliminated by the Michaelis-Menten route: 38.5%
# The article reports "~40%"; the model gives about 38%.
stopifnot(abs(frac_mm - 0.40) < 0.05)

Typical-patient exposures (Figures 3 and 4)

Figures 3 and 4 of the article give the cycle-6 exposures of the reference patient after 2 mg/kg Q3W: ADC Cmax 41.8 ug/mL and Cavg 2.8 ug/mL, MMAE Cmax 4.4 ng/mL and Cavg 1.8 ng/mL. The article does not list the reference patient’s covariates. Two profiles are shown below: the normalisation values of Table S1 (75 kg, female, cervical cancer, ECOG 0), and a median patient (70 kg, the median weight; female, the majority sex; non-cervical tumor and ECOG 1, the majority categories; albumin 40 g/L, eGFR 91, tumor size 57 mm, age 57 years).

cyc_grid <- c(0, inf_dur, 0.05, 0.1, 0.25, 0.5, 0.75, seq(1, 4, by = 0.25),
              seq(4.5, 21, by = 0.5))
grid6 <- sort(unique(as.vector(outer(cyc_grid, 21 * (0:5), "+"))))
profiles <- bind_rows(
  ref_cov |> mutate(profile = "Table S1 reference (75 kg)"),
  data.frame(
    WT = 70, SEXF = 1, ALB = 40, TUMTP_OTHER = 1, CRCL = 91, TUMSZ = 57,
    ECOG_GE1 = 1, HEPIMP = 0, AGE = 57, STUDY_INNOVATV202 = 0,
    profile = "Median patient (70 kg)"
  )
)
typ6 <- lapply(seq_len(nrow(profiles)), function(i) {
  p <- profiles[i, ]
  ev <- make_events(2 * p$WT, 6, grid6) |> mutate(id = i)
  s <- rxode2::rxSolve(mod_typ, ev, params = p[, names(ref_cov)],
                       returnType = "data.frame", useLinCmt = FALSE)
  c6 <- s[s$time >= 105 & s$time <= 126, ]
  data.frame(
    profile = p$profile,
    adc_cmax = max(c6$Cc), adc_cavg = trap(c6$time, c6$Cc) / 21,
    mmae_cmax = max(c6$Cc_mmae), mmae_cavg = trap(c6$time, c6$Cc_mmae) / 21,
    mmae_cmax_ctrough = max(c6$Cc_mmae) / tail(c6$Cc_mmae, 1)
  )
}) |> bind_rows()
#> ℹ omega/sigma items treated as zero: 'etalvc_mmae', 'etalcl_mmae', 'etalvp', 'etalvc', 'etalcl', 'etaRUV_mmae', 'etaRUV', 'etalktr'
#> ℹ omega/sigma items treated as zero: 'etalvc_mmae', 'etalcl_mmae', 'etalvp', 'etalvc', 'etalcl', 'etaRUV_mmae', 'etaRUV', 'etalktr'
typ6 |>
  dplyr::rename(
    "Profile" = profile,
    "ADC Cmax (ug/mL)" = adc_cmax, "ADC Cavg (ug/mL)" = adc_cavg,
    "MMAE Cmax (ng/mL)" = mmae_cmax, "MMAE Cavg (ng/mL)" = mmae_cavg,
    "MMAE Cmax/Ctrough" = mmae_cmax_ctrough
  ) |>
  knitr::kable(digits = 2,
               caption = "Typical cycle-6 exposures after 2 mg/kg Q3W. Published reference-patient values: ADC Cmax 41.8, Cavg 2.8; MMAE Cmax 4.4, Cavg 1.8 (Figures 3 and 4). MMAE trough about 40 times below Cmax (Results).")
Typical cycle-6 exposures after 2 mg/kg Q3W. Published reference-patient values: ADC Cmax 41.8, Cavg 2.8; MMAE Cmax 4.4, Cavg 1.8 (Figures 3 and 4). MMAE trough about 40 times below Cmax (Results).
Profile ADC Cmax (ug/mL) ADC Cavg (ug/mL) MMAE Cmax (ng/mL) MMAE Cavg (ng/mL) MMAE Cmax/Ctrough
Table S1 reference (75 kg) 47.47 3.11 4.79 2.00 40.39
Median patient (70 kg) 45.49 2.97 4.70 1.93 45.21 

med_pt <- typ6[typ6$profile == "Median patient (70 kg)", ]
typ_pct <- 100 * (c(med_pt$adc_cmax, med_pt$adc_cavg, med_pt$mmae_cmax, med_pt$mmae_cavg) /
  c(41.8, 2.8, 4.4, 1.8) - 1)
round(typ_pct, 1)
#> [1] 8.8 6.2 6.9 7.2
# Deterministic solves, so the bound is exact; a mis-transcribed volume,
# clearance or unit moves these by tens of percent.
stopifnot(all(abs(typ_pct) < 15))
stopifnot(all(typ6$mmae_cmax_ctrough > 30 & typ6$mmae_cmax_ctrough < 55))

The median-patient profile lands 6-9% above the four published baselines, and the 75-kg profile 9-14% above. The article does not list the reference patient’s covariates, so part of this gap may be the choice of profile; all four exposures move together, as expected from a shared covariate offset rather than a mis-specified parameter.

Virtual cohorts

Two cohorts of 200 virtual patients each are simulated at 2 mg/kg Q3W with a 200-mg dose cap for 6 cycles:

  • Pooled: covariates drawn to match the pooled analysis population (article Table 1; FDA review Table 45).
  • Cervical: covariates drawn to match innovaTV 204 (all women with cervical cancer; FDA review Tables 45 and 46).

Continuous covariates are drawn independently (normal or log-normal, truncated to the observed range), and binary covariates independently by their observed frequencies; the joint distribution of the source data is not available.

set.seed(20221010)
rxode2::rxSetSeed(20221010)

rtnorm <- function(n, mean, sd, lo, hi) {
  pmin(pmax(rnorm(n, mean, sd), lo), hi)
}

make_cohort <- function(n, cohort, id_offset, cervical) {
  if (cervical) {
    covs <- data.frame(
      WT = pmin(pmax(rlnorm(n, log(65), 0.22), 33), 110),
      SEXF = 1L,
      ALB = rtnorm(n, 41.8, 4.18, 31, 52),
      TUMTP_OTHER = 0L,
      CRCL = rtnorm(n, 84.2, 21.4, 24.1, 129),
      TUMSZ = pmin(pmax(rlnorm(n, log(64), 0.6), 10), 221),
      ECOG_GE1 = rbinom(n, 1, 0.416),
      HEPIMP = rbinom(n, 1, 0.079),
      AGE = rtnorm(n, 50.7, 10.7, 31, 78)
    )
  } else {
    covs <- data.frame(
      WT = pmin(pmax(rlnorm(n, log(70.1), 0.24), 33), 148),
      SEXF = rbinom(n, 1, 0.742),
      ALB = rtnorm(n, 39.4, 4.66, 27, 52),
      TUMTP_OTHER = rbinom(n, 1, 0.569),
      CRCL = rtnorm(n, 87.5, 19.7, 24.1, 145),
      TUMSZ = pmin(pmax(rlnorm(n, log(57), 0.7), 5), 299),
      ECOG_GE1 = rbinom(n, 1, 0.612),
      HEPIMP = rbinom(n, 1, 0.145),
      AGE = rtnorm(n, 56.1, 11.6, 21, 81)
    )
  }
  covs$id <- id_offset + seq_len(n)
  covs$STUDY_INNOVATV202 <- 0L
  covs$cohort <- cohort
  covs$dose <- pmin(2 * covs$WT, 200)

  ev <- lapply(seq_len(n), function(i) {
    make_events(covs$dose[i], 6, grid6) |> mutate(id = covs$id[i])
  }) |> bind_rows()
  left_join(ev, covs, by = "id")
}

events <- bind_rows(
  make_cohort(200, "Pooled", 0L, cervical = FALSE),
  make_cohort(200, "Cervical", 200L, cervical = TRUE)
)
stopifnot(!anyDuplicated(unique(events[, c("id", "time", "evid")])))

Simulation 

sim <- rxode2::rxSolve(
  mod, events = events,
  keep = c("cohort", "WT", "dose"),
  returnType = "data.frame",
  useLinCmt = FALSE
)
#> ℹ parameter labels from comments will be replaced by 'label()'
stopifnot(!anyNA(sim$Cc), !anyNA(sim$Cc_mmae))

Replicate Figure 2

Figure 2 of the article shows the median and the 5th and 95th percentiles of predicted ADC and MMAE concentrations over time for 2 mg/kg Q3W.

pct <- sim |>
  filter(cohort == "Pooled") |>
  select(time, ADC = Cc, MMAE = Cc_mmae) |>
  pivot_longer(c(ADC, MMAE), names_to = "analyte", values_to = "conc") |>
  group_by(analyte, time) |>
  summarise(
    p05 = quantile(conc, 0.05), p50 = median(conc), p95 = quantile(conc, 0.95),
    .groups = "drop"
  ) |>
  mutate(analyte = ifelse(analyte == "ADC", "ADC (ug/mL)", "MMAE (ng/mL)"))
# Pre-dose zeros cannot be drawn on a log axis
pct <- pct[pct$p05 > 0, ]

ggplot(pct, aes(time)) +
  geom_line(aes(y = p50), colour = "darkblue", linewidth = 0.8) +
  geom_line(aes(y = p05), colour = "lightblue3") +
  geom_line(aes(y = p95), colour = "lightblue3") +
  facet_wrap(~analyte, scales = "free_y") +
  scale_y_log10() +
  labs(x = "Time (day)", y = "Concentration",
       title = "Replicates Figure 2 of Gibiansky 2022",
       caption = "2 mg/kg Q3W (at most 200 mg), pooled virtual cohort, n = 200: median and 5th/95th percentiles.") +
  theme_bw()

As in the article, neither ADC nor MMAE accumulates across cycles.

PKNCA validation

Cycle 1 (days 0-21) and cycle 6 (days 105-126) are analysed separately for each analyte. Cc is the ADC concentration (ug/mL) and Cc_mmae the MMAE concentration (ng/mL).

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

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

intervals <- data.frame(
  start = c(0, 105), end = c(21, 126),
  cmax = TRUE, tmax = TRUE, cav = TRUE, auclast = TRUE
)

run_nca <- function(conc_col) {
  conc_df <- sim_nca |> rename(conc = all_of(conc_col))
  conc_obj <- PKNCA::PKNCAconc(conc_df, conc ~ time | cohort + id)
  dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | cohort + id)
  res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))
  as.data.frame(res) |>
    mutate(cycle = ifelse(start == 0, "Cycle 1", "Cycle 6"))
}
nca_adc <- run_nca("Cc")
nca_mmae <- run_nca("Cc_mmae")

nca_adc |>
  bind_rows(nca_mmae, .id = "analyte") |>
  mutate(analyte = ifelse(analyte == "1", "ADC", "MMAE")) |>
  filter(PPTESTCD %in% c("cmax", "cav", "auclast")) |>
  group_by(analyte, cohort, cycle, PPTESTCD) |>
  summarise(median = median(PPORRES), .groups = "drop") |>
  pivot_wider(names_from = PPTESTCD, values_from = median) |>
  dplyr::rename(
    "Analyte" = analyte, "Cohort" = cohort, "Cycle" = cycle,
    "Cmax" = cmax, "Cavg" = cav, "AUC (21 days)" = auclast
  ) |>
  knitr::kable(digits = 2,
               caption = "Median simulated NCA by cohort and cycle (ADC ug/mL, ug*day/mL; MMAE ng/mL, ng*day/mL).")
Median simulated NCA by cohort and cycle (ADC ug/mL, ugday/mL; MMAE ng/mL, ngday/mL).
Analyte Cohort Cycle AUC (21 days) Cavg Cmax
ADC Cervical Cycle 1 59.05 2.81 42.38
ADC Cervical Cycle 6 59.07 2.81 42.39
ADC Pooled Cycle 1 61.32 2.92 45.13
ADC Pooled Cycle 6 61.34 2.92 45.14
MMAE Cervical Cycle 1 43.98 2.09 5.21
MMAE Cervical Cycle 6 44.40 2.11 5.25
MMAE Pooled Cycle 1 42.27 2.01 4.64
MMAE Pooled Cycle 6 42.81 2.04 4.71

Comparison against published exposures

The article reports median individual exposures (empirical Bayes estimates) for the pooled population by weight group in cycle 1 (Results and Figure S7) and the median steady-state ADC Cmax in innovaTV 204 (Results, Japanese sub-analysis). The remaining innovaTV 204 medians were read by the maintainers from the box plots of Figure S12 (to about 0.1 units) and are marked “(Fig S12)”.

wt_cuts <- quantile(sim_nca$WT[sim_nca$cohort == "Pooled"], c(1, 2) / 3)
wt_group <- sim_nca |>
  distinct(id, cohort, WT) |>
  mutate(group = case_when(
    cohort == "Cervical" ~ "innovaTV 204-like, cycle 6",
    WT < wt_cuts[1] ~ "Weight tertile 1, cycle 1",
    WT < wt_cuts[2] ~ "Weight tertile 2, cycle 1",
    TRUE ~ "Weight tertile 3, cycle 1"
  ))

pick <- function(nca) {
  nca |>
    left_join(wt_group, by = c("id", "cohort")) |>
    filter((cohort == "Pooled" & cycle == "Cycle 1") |
             (cohort == "Cervical" & cycle == "Cycle 6"))
}
pooled_lt100 <- function(nca) {
  nca |>
    left_join(wt_group |> select(id, WT), by = "id") |>
    filter(cohort == "Pooled", cycle == "Cycle 1", WT < 100) |>
    mutate(group = "Pooled, weight < 100 kg, cycle 1")
}

sim_adc <- bind_rows(pick(nca_adc), pooled_lt100(nca_adc)) |>
  select(group, PPTESTCD, PPORRES)
sim_mmae <- bind_rows(pick(nca_mmae), pooled_lt100(nca_mmae)) |>
  select(group, PPTESTCD, PPORRES)

ref_adc <- tibble::tribble(
  ~group, ~cmax, ~cav,
  "Weight tertile 1, cycle 1", NA, 2.34,
  "Weight tertile 2, cycle 1", NA, 2.84,
  "Weight tertile 3, cycle 1", NA, 2.99,
  "Pooled, weight < 100 kg, cycle 1", NA, 2.70,
  "innovaTV 204-like, cycle 6", 40.3, 2.8
)
ref_mmae <- tibble::tribble(
  ~group, ~cmax, ~cav,
  "Pooled, weight < 100 kg, cycle 1", NA, 1.92,
  "innovaTV 204-like, cycle 6", 4.8, 1.9
)

cmp_adc <- nlmixr2lib::ncaComparisonTable(
  simulated = sim_adc, reference = ref_adc, by = "group",
  params = c("cmax", "cav"),
  units = c(cmax = "ug/mL", cav = "ug/mL"), tolerance_pct = 20
)
cmp_mmae <- nlmixr2lib::ncaComparisonTable(
  simulated = sim_mmae, reference = ref_mmae, by = "group",
  params = c("cmax", "cav"),
  units = c(cmax = "ng/mL", cav = "ng/mL"), tolerance_pct = 20
)
cmp <- bind_rows(
  cmp_adc |> mutate(Analyte = "ADC"),
  cmp_mmae |> mutate(Analyte = "MMAE")
) |>
  relocate(Analyte) |>
  filter(!is.na(Reference), Reference != "\u2014")
knitr::kable(cmp, caption = "Simulated (median of virtual cohort) vs. published median exposures. For the innovaTV 204-like row, the ADC Cmax reference is printed in the article; the ADC Cavg and both MMAE references were read from Figure S12. * differs from the published value by more than 20%.")
Simulated (median of virtual cohort) vs. published median exposures. For the innovaTV 204-like row, the ADC Cmax reference is printed in the article; the ADC Cavg and both MMAE references were read from Figure S12. * differs from the published value by more than 20%.
Analyte NCA parameter group Reference Simulated % diff
ADC Cmax (ug/mL) innovaTV 204-like, cycle 6 40.3 42.4 +5.2%
ADC Cavg (ug/mL) Weight tertile 1, cycle 1 2.34 2.42 +3.4%
ADC Cavg (ug/mL) Weight tertile 2, cycle 1 2.84 2.99 +5.2%
ADC Cavg (ug/mL) Weight tertile 3, cycle 1 2.99 3.31 +10.8%
ADC Cavg (ug/mL) Pooled, weight < 100 kg, cycle 1 2.7 2.83 +4.8%
ADC Cavg (ug/mL) innovaTV 204-like, cycle 6 2.8 2.81 +0.5%
MMAE Cmax (ng/mL) innovaTV 204-like, cycle 6 4.8 5.25 +9.3%
MMAE Cavg (ng/mL) Pooled, weight < 100 kg, cycle 1 1.92 2.01 +4.8%
MMAE Cavg (ng/mL) innovaTV 204-like, cycle 6 1.9 2.11 +11.3% 
# Medians of 200-subject cohorts: robust to which subjects land in the tails.
# A mis-transcribed clearance, volume or unit moves these by tens of percent.
gate <- bind_rows(
  sim_adc |> mutate(analyte = "ADC") |> inner_join(
    ref_adc |> pivot_longer(c(cmax, cav), names_to = "PPTESTCD", values_to = "ref") |>
      filter(!is.na(ref)) |> mutate(analyte = "ADC"),
    by = c("analyte", "group", "PPTESTCD")
  ),
  sim_mmae |> mutate(analyte = "MMAE") |> inner_join(
    ref_mmae |> pivot_longer(c(cmax, cav), names_to = "PPTESTCD", values_to = "ref") |>
      filter(!is.na(ref)) |> mutate(analyte = "MMAE"),
    by = c("analyte", "group", "PPTESTCD")
  )
) |>
  group_by(analyte, group, PPTESTCD, ref) |>
  summarise(sim = median(PPORRES), .groups = "drop") |>
  mutate(pct_diff = 100 * (sim / ref - 1))
gate$pct_diff
#> [1]  4.8214310  3.4100994  5.1702785 10.7570574  0.4552878  5.1788844  4.8120524
#> [8] 11.2720847  9.2827994
stopifnot(nrow(gate) == 9, all(abs(gate$pct_diff) < 20))

# Weight trend of cycle-1 ADC Cavg: the article reports +5% (tertile 3) and
# -18% (tertile 1) against the middle tertile. Assert the magnitude of the
# low-weight deficit, which is the effect the article calls out.
tert <- sim_adc |>
  filter(PPTESTCD == "cav", grepl("tertile", group)) |>
  group_by(group) |>
  summarise(med = median(PPORRES), .groups = "drop")
low_vs_mid <- tert$med[1] / tert$med[2] - 1
low_vs_mid
#> [1] -0.1898462
stopifnot(low_vs_mid < -0.08, low_vs_mid > -0.30)

The pooled cohort’s cycle-1 Cavg rises with weight (lowest tertile about 18% below the middle tertile in the article), the expected consequence of a clearance that scales with weight to the power 0.495 under weight-proportional dosing. The innovaTV 204-like cohort reproduces the published median steady-state ADC Cmax of 40.3 ug/mL and the Figure S12 MMAE medians.

Assumptions and deviations

  • Covariate coefficients and variances are from the FDA review, not the article. The article’s Table S1 gives the covariate model as equations in theta17-theta36, and its Results give the IIV coefficients of variation and correlations, but no table in the article or its supplements prints the coefficients or the variances. The FDA BLA 761208 Multi-discipline Review (Tables 49 and 50, sourced from the sponsor’s population PK report) prints all of them for the same final ADC-MMAE model, and its Table 48 matches the article’s Table 2 exactly. The IIV CVs and correlations quoted in the article’s Results agree with the review’s variances.
  • Weight exponent on ADC clearance. The article’s Results mention a “power of 0.487” for the dependence of clearance on weight; the final-model estimate in the review is 0.495 (and 0.45 in the ADC-only model). The model uses 0.495, the final-model value.
  • MMAE output scale factor. Table S1 gives MW_RATIO = 4.8, while the Methods quote the molecular-weight ratio as 718/152 = 4.72. The equation table is used. Because CL_MMAE and V_MMAE are apparent parameters estimated with this constant, the choice does not change predicted MMAE concentrations relative to the fitted data.
  • Peripheral ODE sign. Table S1 prints dA2/dt = -K12 A1 - K21 A2. Figure 1 and mass balance require the inflow +K12 A1, which the model uses.
  • Covariate centring. Table S1 normalises weight to 75 kg and eGFR to 90; the FDA review’s covariate-effect table (Table 51) displays effects against 70 kg. The model follows Table S1. The typical values in Table 2 therefore describe a 75-kg patient.
  • MMAE additive residual SD units. Table 2 (and the FDA review) print the MMAE additive residual SD 0.0113 with ug/mL units; MMAE is measured and modelled in ng/mL (LLOQ 0.025 ng/mL), so the value is taken as ng/mL.
  • Residual error with IIV. The source model is Y = IPRED + SD * eps with sigma^2 = 1 fixed and SD = sqrt(IPRED^2 theta_prop^2 + theta_add^2) * theta36^STUDY202 * exp(eta). This is encoded as a combined2() additive-plus-proportional error whose two SDs are both multiplied by the study factor and exp(etaRUV), which is algebraically identical. STUDY_INNOVATV202 affects only this residual SD; set it to 0 when simulating a new population.
  • Time after dose before the first dose. The drug-to-antibody ratio uses rxode2’s tad(central), which is undefined before the first dose; it is set to 0 there, where no ADC is present.
  • Infusion duration. The article does not state it; 30 minutes is used, as in the US prescribing information and the review’s description of the pivotal study.
  • Virtual cohorts. Covariates are drawn independently from marginal summaries (article Table 1; FDA review Tables 45 and 46). Correlations between covariates (for example weight and sex) are not reproduced.
  • Below-quantification data. The source fit handled BLQ records with the M3 method; this does not affect simulation.
  • Japanese sub-analysis. The article re-estimated individual parameters for 18 Japanese patients with the final model fixed; no separate model was produced, so nothing further is extracted.
  • No erratum or correction to Gibiansky 2022 was found as of 2026-10-09.