Skip to contents

Datopotamab deruxtecan ADC + DXd coupled population PK model (Hong 2025)

Source: vignettes/articles/Hong_2025_datopotamab.Rmd
Hong_2025_datopotamab.Rmd
library(nlmixr2lib)
library(PKNCA)
#> 
#> Attaching package: 'PKNCA'
#> The following object is masked from 'package:stats':
#> 
#>     filter
library(rxode2)
#> rxode2 5.0.2 using 2 threads (see ?getRxThreads)
#>   no cache: create with `rxCreateCache()`
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)
library(ggplot2)

Model and source

  • Citation: Hong Y, Peigne S, Pan Y, Friberg Hietala S, McLaughlin A, Tajima N, Uema D, Zebger-Gong H, Tang Z, Zhou D, Abutarif M, Garimella T. Population Pharmacokinetic Analysis of Datopotamab Deruxtecan (Dato-DXd), a TROP2-Directed Antibody-Drug Conjugate, in Patients With Advanced Solid Tumors. CPT Pharmacometrics Syst Pharmacol. 2025;14(12):2149-2160.
  • Article: https://doi.org/10.1002/psp4.70118 (PMID 41035281).

The model couples two submodels (paper Figure 1):

  1. Dato-DXd submodel – two-compartment with parallel linear and Michaelis-Menten elimination from the central compartment.
  2. DXd submodel – one-compartment with linear clearance. Its central compartment is fed at a rate equal to the total (linear + Michaelis-Menten) Dato-DXd elimination rate, scaled by the DXd-to-Dato-DXd molecular-weight ratio (493.5 / 150,000) and a time-and-cycle-dependent drug-to-antibody ratio DAR(tad, CYCLE).
DAR(tad, CYCLE) = 4 * (0.25 + 0.75 * exp(-beta * tad)) *
                  (1                if CYCLE = 1
                   Factor1 = 0.696  otherwise)            (Hong 2025 Eq. 13)

rate_DXd_in     = (CL_lin * C_DatoDXd + Vmax * C_DatoDXd / (Km + C_DatoDXd))
                  * DAR * (493.5 / 150000)               (Hong 2025 Methods)

tad is the time since the most recent dose; CYCLE is the integer treatment-cycle number, set to 1 over the first 21 days post-first-dose and incremented at each new Q3W dosing cycle.

Concentrations are in mg/L (= ug/mL) for both analytes inside the model; DXd is conventionally reported in ng/mL and is rescaled outside the model (Cdxd_ng_mL = Cdxd * 1000).

Population

Pooled analysis dataset (Hong 2025 Methods + Table S1):

  • Studies: Phase 1 TROPION-PanTumor01 (TP01), Phase 2 TL05, Phase 3 TL01.
  • Subjects: 729 patients with advanced or metastatic NSCLC and breast cancer.
  • Observations: 9036 Dato-DXd PK observations + 9012 DXd PK observations.
  • Body-weight range: 37.0 - 156 kg analysis range; reference 66 kg.
  • Reference subjects:
    • Dato-DXd: 66 kg male, age 62 yr, albumin 38 g/L, tumor size 66 mm, region != Japan (Hong 2025 Figure 4 caption).
    • DXd: 66 kg male, region US, albumin 38 g/L, AST 22 U/L, total bilirubin 0.4 mg/dL (Hong 2025 Figure 4 caption).
  • Dose regimens: Phase 1 included a 0.27 - 10 mg/kg dose-escalation; Phase 2 / 3 used the labeled 6 mg/kg Q3W. The approved label (Datroway) caps the flat dose at 540 mg for body weight >= 90 kg.
mod <- rxode2::rxode(readModelDb("Hong_2025_datopotamab"))
#> ℹ parameter labels from comments will be replaced by 'label()'
str(mod$meta$population, max.level = 1)
#> List of 14
#>  $ n_subjects                : int 729
#>  $ n_studies                 : int 3
#>  $ age_range                 : chr "Adults with advanced or metastatic solid tumors (NSCLC and breast cancer); reference subject 62 years per Hong "| __truncated__
#>  $ weight_range              : chr "37.0 - 156 kg analysis range (Hong 2025 Methods / Body Weight-Based Dosing); reference 66 kg."
#>  $ sex_female_pct            : num NA
#>  $ race_ethnicity            : chr "Multi-regional enrollment including Japan, US, Europe, and Rest of the World; race per se was not retained as a"| __truncated__
#>  $ disease_state             : chr "Advanced or metastatic non-small-cell lung cancer (NSCLC) and breast cancer (BC)."
#>  $ dose_range                : chr "Dato-DXd IV every 3 weeks. Phase 1 TROPION-PanTumor01 (TP01) dose range 0.27 - 10 mg/kg; Phase 2 TL05 and Phase"| __truncated__
#>  $ regions                   : chr "Multi-regional: Japan, US, Europe, and Rest of the World."
#>  $ n_observations            : chr "9036 Dato-DXd PK observations + 9012 DXd PK observations from 729 patients across three studies (Hong 2025 Methods)."
#>  $ studies                   : chr "Phase 1 TROPION-PanTumor01 (TP01), Phase 2 TL05, and Phase 3 TL01."
#>  $ reference_subject_dato_dxd: chr "66 kg male, age 62 years, albumin 38 g/L, tumor size 66 mm, region != Japan (Hong 2025 Figure 4 caption / Table 1 reference)."
#>  $ reference_subject_dxd     : chr "66 kg male, region US, albumin 38 g/L, AST 22 U/L, total bilirubin 0.4 mg/dL (Hong 2025 Figure 4 caption / Table 2 reference)."
#>  $ notes                     : chr "Pooled analysis dataset (Hong 2025 Methods). Final model fit jointly to all 729 subjects across the 3 studies, "| __truncated__

The covariates retained in the final model were body weight (mechanistic, on all four Dato-DXd structural parameters and DXd CL/Vc), albumin and age (on Dato-DXd CL_lin), region Japan (on Dato-DXd CL_lin), female sex (on Dato-DXd CL_lin and Vc; DXd Vc), tumor size (on Dato-DXd Vmax), albumin / AST / total bilirubin (on DXd CL), and region Europe / Rest of World (on DXd CL). ADA, race, creatinine clearance, drug product, and broad hepatic-impairment classifications were tested and excluded.

Source trace

Every ini() entry in inst/modeldb/specificDrugs/Hong_2025_datopotamab.R carries an in-file comment pointing to Hong 2025 Table 1, Table 2, or Equations 8-15. The table below collates them.

Parameter (nlmixr2lib) Value Units Source
lcl_adc -> CL_lin 0.386 L/day Table 1, Eq. 8
lvc_adc -> Vc 3.06 L Table 1, Eq. 9
lq_adc -> Q 0.422 L/day Table 1
lvp_adc -> Vp 2.88 L Table 1, Eq. 10
lvmax -> Vmax 8.41 mg/day Table 1, Eq. 12 (8410 ug/day)
lkm -> Km 4.49 mg/L Table 1, Eq. 11 (4490 ng/mL)
e_wt_cl_adc 0.75 (fixed) - Table 1
e_wt_vc_adc 0.415 - Table 1
e_wt_vp_adc 0.311 - Table 1
e_age_cl_adc -0.306 - Table 1
e_alb_cl_adc -0.788 - Table 1
e_japan_cl_adc -0.219 - Table 1
e_sexf_cl_adc -0.263 - Table 1
e_sexf_vc_adc -0.160 - Table 1
e_tumsz_vmax 0.125 - Table 1, Eq. 12
lcl_dxd -> CL_DXd 63.84 L/day Table 2, Eq. 14 (2.66 L/h * 24)
lvc_dxd -> Vc_DXd 25.1 L Table 2, Eq. 15
lfactor1 -> Factor1 0.696 - Table 2, Eq. 13
lbeta -> beta 0.259 1/day Table 2, Eq. 13
e_wt_cl_dxd 0.298 (fixed) - Table 2
e_wt_vc_dxd 0.530 (fixed) - Table 2
e_alb_cl_dxd 0.343 - Table 2
e_ast_cl_dxd -0.154 - Table 2
e_tbili_cl_dxd -0.139 - Table 2
e_eu_cl_dxd 0.240 - Table 2
e_row_cl_dxd 0.196 - Table 2
e_sexf_vc_dxd -0.185 - Table 2
CcpropSd 0.121 fraction Table 1 (additive RUV log scale)
CdxdpropSd 0.283 fraction Table 2 (additive RUV log scale)

Inter-individual variability (log-normal; omega^2 = log(CV^2 + 1)):

Parameter %CV omega^2 Source
CL_lin 27.2 0.071375 Table 1
Vc 14.5 0.020807 Table 1
Q 31.1 0.092325 Table 1
Vp 31.2 0.092893 Table 1
Vmax 19.2 0.036201 Table 1
CL_DXd 31.4 0.094033 Table 2
Vc_DXd 36.3 0.123782 Table 2

Virtual cohort

Match the reference-subject covariates and the labeled dosing regimen (6 mg/kg Q3W, 540 mg flat-dose cap for body weight >= 90 kg):

set.seed(20260427L)

n_subj   <- 200L
cycle_dy <- 21
n_cycles <- 5L

wt    <- exp(rnorm(n_subj, mean = log(66), sd = 0.20))
wt    <- pmin(pmax(wt, 37), 156)
sexf  <- rbinom(n_subj, 1, 0.50)
age   <- pmax(20, pmin(85, rnorm(n_subj, mean = 62, sd = 11)))
alb   <- pmax(20, pmin(50, rnorm(n_subj, mean = 38, sd = 4)))
ast   <- pmax(8,  pmin(80, rnorm(n_subj, mean = 22, sd = 6)))
tbili <- pmax(0.1, pmin(2, rlnorm(n_subj, log(0.4), 0.4)))
tumsz <- pmax(10, pmin(200, rlnorm(n_subj, log(66), 0.5)))
region <- sample(c("US", "JP", "EU", "RoW"), n_subj, replace = TRUE,
                 prob = c(0.45, 0.20, 0.20, 0.15))

make_subject <- function(id, wt_kg, sexf_ind, age_y, alb_g_L, ast_U_L,
                         tbili_mg_dL, tumsz_mm, reg) {
  dose_mg <- min(6 * wt_kg, 540)   # 6 mg/kg Q3W with 540 mg flat-dose cap

  doses <- rxode2::et() |>
    rxode2::et(amt = dose_mg, cmt = "adc_central", ii = cycle_dy,
               addl = n_cycles - 1L, time = 0)
  obs <- rxode2::et(seq(0.05, n_cycles * cycle_dy, length.out = 420),
                    cmt = "Cc")
  ev <- rbind(doses, obs)
  ev$id           <- id
  ev$WT           <- wt_kg
  ev$AGE          <- age_y
  ev$SEXF         <- sexf_ind
  ev$ALB          <- alb_g_L
  ev$AST          <- ast_U_L
  ev$TBILI        <- tbili_mg_dL
  ev$TUMSZ        <- tumsz_mm
  ev$REGION_JAPAN  <- as.integer(reg == "JP")
  ev$REGION_EUROPE <- as.integer(reg == "EU")
  ev$REGION_ROW    <- as.integer(reg == "RoW")
  ev$CYCLE        <- pmax(1L, as.integer(floor(ev$time / cycle_dy) + 1L))
  ev$dose_mg      <- dose_mg
  ev
}

events <- purrr::map_dfr(seq_len(n_subj),
  ~ make_subject(.x, wt[.x], sexf[.x], age[.x], alb[.x], ast[.x],
                 tbili[.x], tumsz[.x], region[.x])) |>
  as.data.frame()

stopifnot(length(unique(events$id)) == n_subj)

Typical-subject simulation

Single typical-subject curve (66 kg male, age 62, albumin 38 g/L, AST 22 U/L, total bilirubin 0.4 mg/dL, tumor 66 mm, region US, 6 mg/kg Q3W, 5 cycles) with all between-subject and residual variability turned off:

typ <- readModelDb("Hong_2025_datopotamab") |> rxode2::zeroRe()
#> ℹ parameter labels from comments will be replaced by 'label()'

typ_events <- rxode2::et() |>
  rxode2::et(amt = 6 * 66, cmt = "adc_central", ii = cycle_dy,
             addl = n_cycles - 1L, time = 0) |>
  rxode2::et(seq(0.05, n_cycles * cycle_dy, length.out = 1500), cmt = "Cc")

typ_events$WT            <- 66
typ_events$AGE           <- 62
typ_events$SEXF          <- 0
typ_events$ALB           <- 38
typ_events$AST           <- 22
typ_events$TBILI         <- 0.4
typ_events$TUMSZ         <- 66
typ_events$REGION_JAPAN  <- 0
typ_events$REGION_EUROPE <- 0
typ_events$REGION_ROW    <- 0
typ_events$CYCLE         <- pmax(1L, as.integer(floor(typ_events$time / cycle_dy) + 1L))

typ_sim <- rxode2::rxSolve(typ, typ_events, returnType = "data.frame") |>
  mutate(
    Cc_ug_mL    = Cc,                # mg/L = ug/mL already
    Cdxd_ng_mL  = Cdxd * 1000        # mg/L -> ng/mL
  )
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'
typ_sim |>
  tidyr::pivot_longer(c(Cc_ug_mL, Cdxd_ng_mL),
                      names_to = "analyte", values_to = "conc") |>
  mutate(analyte = recode(analyte,
                          Cc_ug_mL   = "Dato-DXd (ug/mL)",
                          Cdxd_ng_mL = "DXd (ng/mL)")) |>
  ggplot(aes(time, conc)) +
  geom_line(colour = "steelblue", linewidth = 0.8) +
  facet_wrap(~ analyte, ncol = 1, scales = "free_y") +
  scale_y_log10() +
  labs(x = "Time (day)", y = "Concentration",
       caption = "Replicates the median trend of Hong 2025 Figure 2.")
Typical-subject Dato-DXd (top, ug/mL) and DXd (bottom, ng/mL) concentration-time profiles for a 66 kg male reference patient receiving 6 mg/kg Q3W for 5 cycles. Replicates the median trend of Hong 2025 Figure 2 (upper / lower panels).

Typical-subject Dato-DXd (top, ug/mL) and DXd (bottom, ng/mL) concentration-time profiles for a 66 kg male reference patient receiving 6 mg/kg Q3W for 5 cycles. Replicates the median trend of Hong 2025 Figure 2 (upper / lower panels). 

typ_sim |>
  ggplot(aes(time, dar)) +
  geom_line(colour = "darkred", linewidth = 0.7) +
  labs(x = "Time (day)", y = "DAR(t)",
       caption = "Hong 2025 Eq. 13.")
Internal DAR(t) trajectory across five cycles for the typical subject. DAR resets at each dose (4 in cycle 1; 4 * 0.696 = 2.78 in cycles 2 onwards) and decays toward DAR0 * 0.25 in cycle 1 (= 1) and DAR0 * 0.25 * 0.696 in later cycles (= 0.696).

Internal DAR(t) trajectory across five cycles for the typical subject. DAR resets at each dose (4 in cycle 1; 4 * 0.696 = 2.78 in cycles 2 onwards) and decays toward DAR0 * 0.25 in cycle 1 (= 1) and DAR0 * 0.25 * 0.696 in later cycles (= 0.696).

Linear vs nonlinear elimination across the dose range

The paper’s Figure 3 visualizes the contributions of CL_lin and CL_nonlin to total Dato-DXd clearance versus the typical-subject Dato-DXd plasma concentration. We replicate that plot from the typical-value parameters:

cl_grid <- tibble(
  C_ug_mL    = exp(seq(log(0.1), log(500), length.out = 200)),
  CL_lin     = 0.386,
  rate_lin   = CL_lin * C_ug_mL,
  rate_mm    = 8.41 * C_ug_mL / (4.49 + C_ug_mL),
  CL_nonlin  = rate_mm / C_ug_mL,
  CL_total   = CL_lin + CL_nonlin,
  pct_nl     = 100 * CL_nonlin / CL_total
)

ggplot(cl_grid, aes(C_ug_mL, CL_total)) +
  geom_line(aes(colour = "Total"), linewidth = 0.8) +
  geom_line(aes(C_ug_mL, CL_lin,    colour = "Linear"), linewidth = 0.8) +
  geom_line(aes(C_ug_mL, CL_nonlin, colour = "Nonlinear (MM)"), linewidth = 0.8) +
  scale_x_log10() +
  labs(x = "Dato-DXd plasma concentration (ug/mL)",
       y  = "Apparent clearance (L/day)",
       colour = "Clearance",
       caption = "Replicates Hong 2025 Figure 3 (typical-subject parameters).")
Typical-subject Dato-DXd linear, nonlinear, and total clearance (left axis) and percentage nonlinear (right axis) versus Dato-DXd plasma concentration. Replicates the qualitative message of Hong 2025 Figure 3: CL_lin dominates above ~50 ug/mL while nonlinear MM clearance dominates at sub-therapeutic concentrations.

Typical-subject Dato-DXd linear, nonlinear, and total clearance (left axis) and percentage nonlinear (right axis) versus Dato-DXd plasma concentration. Replicates the qualitative message of Hong 2025 Figure 3: CL_lin dominates above ~50 ug/mL while nonlinear MM clearance dominates at sub-therapeutic concentrations.

Population (VPC-style) simulation

Full population simulation with all between-subject variability and residual error preserved.

mod <- readModelDb("Hong_2025_datopotamab")
pop_sim <- rxode2::rxSolve(mod, events,
                           keep = c("WT", "SEXF", "CYCLE", "dose_mg"),
                           returnType = "data.frame") |>
  mutate(
    Cc_ug_mL   = Cc,
    Cdxd_ng_mL = Cdxd * 1000
  )
#> ℹ parameter labels from comments will be replaced by 'label()'
pop_sim |>
  group_by(time) |>
  summarise(
    p05 = quantile(Cc_ug_mL, 0.05, na.rm = TRUE),
    p50 = quantile(Cc_ug_mL, 0.50, na.rm = TRUE),
    p95 = quantile(Cc_ug_mL, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  ggplot(aes(time, p50)) +
  geom_ribbon(aes(ymin = p05, ymax = p95), fill = "steelblue", alpha = 0.3) +
  geom_line(colour = "steelblue4", linewidth = 0.8) +
  scale_y_log10() +
  labs(x = "Time (day)", y = "Dato-DXd concentration (ug/mL)",
       title = "Dato-DXd VPC -- 6 mg/kg Q3W",
       caption = "Replicates the upper panel of Hong 2025 Figure 2.")
VPC-style 5/50/95th-percentile ribbon of simulated Dato-DXd concentrations vs time across 200 virtual subjects on 6 mg/kg Q3W (540 mg flat-dose cap >= 90 kg) for 5 cycles. Reproduces the Dato-DXd panel of Hong 2025 Figure 2.

VPC-style 5/50/95th-percentile ribbon of simulated Dato-DXd concentrations vs time across 200 virtual subjects on 6 mg/kg Q3W (540 mg flat-dose cap >= 90 kg) for 5 cycles. Reproduces the Dato-DXd panel of Hong 2025 Figure 2. 

pop_sim |>
  group_by(time) |>
  summarise(
    p05 = quantile(Cdxd_ng_mL, 0.05, na.rm = TRUE),
    p50 = quantile(Cdxd_ng_mL, 0.50, na.rm = TRUE),
    p95 = quantile(Cdxd_ng_mL, 0.95, na.rm = TRUE),
    .groups = "drop"
  ) |>
  ggplot(aes(time, p50)) +
  geom_ribbon(aes(ymin = p05, ymax = p95), fill = "tomato", alpha = 0.3) +
  geom_line(colour = "tomato4", linewidth = 0.8) +
  scale_y_log10() +
  labs(x = "Time (day)", y = "DXd concentration (ng/mL)",
       title = "DXd VPC -- 6 mg/kg Q3W",
       caption = "Replicates the lower panel of Hong 2025 Figure 2.")
VPC-style 5/50/95th-percentile ribbon of simulated DXd concentrations vs time across the same 200 virtual subjects. Reproduces the DXd panel of Hong 2025 Figure 2; cycle-1 to later-cycle scaling reflects the DAR Factor1 = 0.696 step.

VPC-style 5/50/95th-percentile ribbon of simulated DXd concentrations vs time across the same 200 virtual subjects. Reproduces the DXd panel of Hong 2025 Figure 2; cycle-1 to later-cycle scaling reflects the DAR Factor1 = 0.696 step.

PKNCA validation

Per-cycle NCA on the simulated typical-subject (no IIV) curves – the paper reports relative metrics (R_ac for AUC, Cmax, Ctrough) rather than absolute single-cycle Cmax / AUC values, so the validation target is the cycle-1 to cycle-3 ratio and the qualitative ordering between cycles.

typ_nca_input <- typ_sim |>
  filter(time > 0) |>
  mutate(
    cycle         = pmax(1L, as.integer(floor((time - 1e-9) / cycle_dy) + 1L)),
    time_in_cycle = time - (cycle - 1) * cycle_dy
  ) |>
  filter(cycle <= n_cycles) |>
  mutate(treatment = "Dato-DXd 6 mg/kg Q3W") |>
  transmute(id = 1L, treatment, cycle = as.factor(cycle),
            time = time_in_cycle, Cc = Cc_ug_mL, Cdxd = Cdxd_ng_mL)

dose_typ <- tibble(
  id        = 1L,
  treatment = "Dato-DXd 6 mg/kg Q3W",
  cycle     = factor(seq_len(n_cycles)),
  time      = 0,
  amt       = 6 * 66
)

intervals_q3w <- data.frame(
  start   = 0,
  end     = cycle_dy,
  cmax    = TRUE,
  tmax    = TRUE,
  auclast = TRUE
)

adc_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(typ_nca_input, Cc ~ time | treatment + id/cycle),
  PKNCA::PKNCAdose(dose_typ,      amt ~ time | treatment + id + cycle),
  intervals = intervals_q3w
))
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.05) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0540027) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0580053) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.062008) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0660107) is not allowed
adc_nca_summary <- summary(adc_nca)

dxd_nca <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(typ_nca_input, Cdxd ~ time | treatment + id/cycle),
  PKNCA::PKNCAdose(dose_typ,       amt ~ time | treatment + id + cycle),
  intervals = intervals_q3w
))
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.05) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0540027) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0580053) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.062008) is not allowed
#> Warning: Requesting an AUC range starting (0) before the first measurement
#> (0.0660107) is not allowed
dxd_nca_summary <- summary(dxd_nca)

adc_nca_summary
#>  start end            treatment cycle N auclast cmax   tmax
#>      0  21 Dato-DXd 6 mg/kg Q3W     1 1      NC  128 0.0500
#>      0  21 Dato-DXd 6 mg/kg Q3W     2 1      NC  131 0.0540
#>      0  21 Dato-DXd 6 mg/kg Q3W     3 1      NC  132 0.0580
#>      0  21 Dato-DXd 6 mg/kg Q3W     4 1      NC  132 0.0620
#>      0  21 Dato-DXd 6 mg/kg Q3W     5 1      NC  132 0.0660
#> 
#> Caption: auclast, cmax: geometric mean and geometric coefficient of variation; tmax: median and range; N: number of subjects
dxd_nca_summary
#>  start end            treatment cycle N auclast cmax  tmax
#>      0  21 Dato-DXd 6 mg/kg Q3W     1 1      NC 8.34 0.820
#>      0  21 Dato-DXd 6 mg/kg Q3W     2 1      NC 6.06 0.824
#>      0  21 Dato-DXd 6 mg/kg Q3W     3 1      NC 6.04 0.828
#>      0  21 Dato-DXd 6 mg/kg Q3W     4 1      NC 6.06 0.832
#>      0  21 Dato-DXd 6 mg/kg Q3W     5 1      NC 6.07 0.836
#> 
#> Caption: auclast, cmax: geometric mean and geometric coefficient of variation; tmax: median and range; N: number of subjects

Cycle 1 to cycle 3 accumulation comparison 

adc_pp <- as.data.frame(adc_nca$result) |>
  filter(PPTESTCD %in% c("auclast", "cmax")) |>
  mutate(analyte = "Dato-DXd")
dxd_pp <- as.data.frame(dxd_nca$result) |>
  filter(PPTESTCD %in% c("auclast", "cmax")) |>
  mutate(analyte = "DXd")

rac_tbl <- bind_rows(adc_pp, dxd_pp) |>
  group_by(analyte, PPTESTCD, cycle) |>
  summarise(value = median(PPORRES, na.rm = TRUE), .groups = "drop") |>
  pivot_wider(names_from = cycle, values_from = value,
              names_prefix = "cycle_") |>
  mutate(R_ac_3 = cycle_3 / cycle_1)

knitr::kable(
  rac_tbl,
  digits  = c(NA, NA, 3, 3, 3, 3, 3, 2),
  caption = "Cycle-1 vs cycle-3 NCA metrics for the typical subject. Hong 2025 Results report R_ac (AUC) < 1.2 and R_ac (Cmax) ~ 1 at 6 mg/kg Q3W. Differences > 20% would indicate a structural problem with the model and would be investigated rather than tuned away."
)
Cycle-1 vs cycle-3 NCA metrics for the typical subject. Hong 2025 Results report R_ac (AUC) < 1.2 and R_ac (Cmax) ~ 1 at 6 mg/kg Q3W. Differences > 20% would indicate a structural problem with the model and would be investigated rather than tuned away.
analyte PPTESTCD cycle_1 cycle_2 cycle_3 cycle_4 cycle_5 R_ac_3
DXd auclast NA NA NA NA NA NA
DXd cmax 8.339 6.061 6.044 6.061 6.066 0.72
Dato-DXd auclast NA NA NA NA NA NA
Dato-DXd cmax 127.586 130.754 131.806 132.003 131.955 1.03

Body-weight-based dose-cap simulation

Hong 2025 Figure 5 evaluates the 540 mg flat-dose cap above 90 kg by simulating WT subgroups [37,46), [46,60), [60,80), [80,90), [90,156] kg. We replicate the Cmax3 and AUC3 boxplots for cycle 3 (using the typical covariate vector, no IIV):

wt_groups <- tibble(
  group    = factor(c("[37,46)", "[46,60)", "[60,80)", "[80,90)", "[90,156]"),
                    levels = c("[37,46)", "[46,60)", "[60,80)", "[80,90)",
                               "[90,156]")),
  wt_med   = c(42, 53, 70, 85, 110)
)

simulate_typ_wt <- function(wt_kg) {
  dose_mg <- if (wt_kg >= 90) 540 else 6 * wt_kg
  ev <- rxode2::et() |>
    rxode2::et(amt = dose_mg, cmt = "adc_central", ii = cycle_dy,
               addl = 2L, time = 0) |>            # 3 cycles total
    rxode2::et(seq(0.05, 3 * cycle_dy, length.out = 1000), cmt = "Cc")
  ev$WT            <- wt_kg
  ev$AGE           <- 62
  ev$SEXF          <- 0
  ev$ALB           <- 38
  ev$AST           <- 22
  ev$TBILI         <- 0.4
  ev$TUMSZ         <- 66
  ev$REGION_JAPAN  <- 0
  ev$REGION_EUROPE <- 0
  ev$REGION_ROW    <- 0
  ev$CYCLE         <- pmax(1L, as.integer(floor(ev$time / cycle_dy) + 1L))
  out <- rxode2::rxSolve(typ, ev, returnType = "data.frame") |>
    filter(time > 2 * cycle_dy & time <= 3 * cycle_dy) |>
    summarise(
      auc3  = sum(diff(time) * (head(Cc, -1) + tail(Cc, -1)) / 2),
      cmax3 = max(Cc, na.rm = TRUE)
    )
  out$wt_kg <- wt_kg
  out$dose  <- dose_mg
  out
}

wt_sim <- purrr::map_dfr(wt_groups$wt_med, simulate_typ_wt) |>
  left_join(wt_groups, by = c("wt_kg" = "wt_med"))
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'
#> ℹ omega/sigma items treated as zero: 'etalcl_adc', 'etalvc_adc', 'etalq_adc', 'etalvp_adc', 'etalvmax', 'etalcl_dxd', 'etalvc_dxd'

wt_sim |>
  pivot_longer(c(auc3, cmax3), names_to = "metric", values_to = "value") |>
  mutate(metric = recode(metric,
                         auc3  = "Cycle-3 AUC (ug*day/mL)",
                         cmax3 = "Cycle-3 Cmax (ug/mL)")) |>
  ggplot(aes(group, value)) +
  geom_col(fill = "steelblue", alpha = 0.7) +
  facet_wrap(~ metric, scales = "free_y") +
  labs(x = "Body-weight group (kg)", y = NULL,
       caption = "Replicates Hong 2025 Figure 5 (typical-value approximation; full IIV simulation produces the boxplots in the paper).")
Cycle-3 Dato-DXd AUC (left) and Cmax (right) by weight subgroup using typical-value parameters. Subjects in the >= 90 kg group receive the 540 mg cap. Replicates the qualitative message of Hong 2025 Figure 5: cap-vs-mg/kg dosing yields broadly similar exposures across subgroups.

Cycle-3 Dato-DXd AUC (left) and Cmax (right) by weight subgroup using typical-value parameters. Subjects in the >= 90 kg group receive the 540 mg cap. Replicates the qualitative message of Hong 2025 Figure 5: cap-vs-mg/kg dosing yields broadly similar exposures across subgroups.

Assumptions and deviations

  • Coupled parent + payload in one model file. The paper presents a single joint analysis with two interconnected sub-models (Dato-DXd two-compartment ADC, DXd one-compartment payload). We extract them together in Hong_2025_datopotamab.R, mirroring the Li 2017 brentuximab precedent.
  • Concentration units mg/L (= ug/mL) for both analytes. Dato-DXd is conventionally reported in ug/mL and DXd in ng/mL. Inside the ODE, both are integrated in mg/L. The vignette converts DXd to ng/mL after solve.
  • CL_DXd unit conversion. Hong 2025 Table 2 reports CL_DXd as 2.66 L/h; the model file stores it as 2.66 * 24 = 63.84 L/day so that every clearance / rate constant in the ODE is on the day timescale (matching beta = 0.259 /day, the cycle-dosing interval, and Dato-DXd CL_lin = 0.386 L/day). The abstract reports CL_DXd as 2.66 L/day; this is a unit typo in the abstract – a half-life of 6.5 days is implausible for the small-molecule payload DXd, whereas 2.66 L/h gives a half-life of ~6.5 h consistent with the published exatecan congener. The Table 2 / Eq. 14 unit (L/h) is treated as authoritative.
  • AST unit clarification. The Methods text reports the AST reference value as 22 g/L; this is implausible for a transaminase activity (normal range ~10-40 U/L; 22 g/L would be a higher concentration than total serum albumin). The reference numeric value 22 is consistent with the canonical clinical-PK unit U/L (= IU/L), and that is what we use. The parameter point estimate e_ast_cl_dxd = -0.154 is unaffected by the unit label.
  • No IIV on Dato-DXd Q (CL-Vc covariance). Hong 2025 Tables 1 / 2 do not report a correlation block for any pair of structural parameters; every IIV term is implemented as a diagonal omega.
  • No eta on the residual error (RUV). Paper Eq. 3 places an exponential IIV on the RUV magnitude with %CV reported as IIV RUV in Tables 1 and 2. nlmixr2’s residual-error model does not accept a subject-level random effect on the SD, so this term is omitted from the implementation; the reported additive-on-log-scale SD (which equals proportional-in-linear- space SD) is used as the typical-subject residual SD. This will under- reproduce the tails of the residual distribution but match the median.
  • Region-grouping discrepancy between Dato-DXd and DXd CL covariates. Hong 2025 evaluates Dato-DXd CL with reference = “any region != Japan” (so REGION_JAPAN = 1 implies the -0.219 fractional shift) but DXd CL with reference = “US or Japan” (so REGION_JAPAN does not enter the DXd CL expression at all – Japan is grouped with US for the DXd reference). The model file uses the three indicators REGION_JAPAN, REGION_EUROPE, REGION_ROW with US as the implicit reference (all three = 0); REGION_JAPAN enters the Dato-DXd CL term only and not the DXd CL term, faithfully reproducing the paper’s parameterization.
  • Cycle covariate. CYCLE is a 1-based integer treatment cycle counter (1 in days [0, 21), 2 in [21, 42), and so on for Q3W dosing). The DAR equation reads factor_dar = 1 in cycle 1 and factor_dar = 0.696 thereafter, implemented with an if test so the simulation user must populate CYCLE >= 1 for every observation row.
  • tad() tracks time since the most recent dose. Within the first 21 days post-first-dose, tad() increases monotonically from 0; at the next dose it resets to 0 again. The DAR equation therefore decays during each cycle and reverts to 4 (cycle 1) or 4 * 0.696 = 2.78 (cycles 2+) at each new dose.
  • No bioavailability term. Dato-DXd is administered IV; doses are written directly into adc_central and there is no depot compartment.
  • Population metadata. Median age, median weight, sex distribution, and per-region patient counts are not enumerated in the publicly-available trimmed text used during extraction; the population block reports the pooled-cohort summary directly cited (n = 729, 3 studies, 9036 + 9012 observations) and notes the omissions in population$notes. The virtual-cohort weight distribution is anchored to the reference 66 kg with log-normal sigma 0.20 to span ~37 - 156 kg.

Errata

No published erratum or correction was located for PMID 41035281 as of 2026-04-27 (PubMed metadata and the journal landing page carry no correction-notice link). The items below are implementation-level notational inconsistencies in the source paper that the extraction had to resolve; they do not affect the final-model parameter point estimates.

  • Abstract vs Table 2 / Eq. 14: CL_DXd units. The abstract reports CL_DXd as 2.66 L/day; Table 2 and Eq. 14 report the same numerical value as 2.66 L/h. The Table 2 / Eq. 14 unit (L/h) is consistent with the small-molecule disposition of DXd (half-life ~5-7 h) and with the exatecan congener cited in the Discussion; the abstract value is interpreted as a typographical error.
  • Methods text: AST unit label g/L. The Methods reference value for AST is given as 22 g/L. AST is an enzyme activity reported in U/L (= IU/L); the canonical AST unit per inst/references/covariate-columns.md is U/L. The numerical value 22 is consistent with U/L only.
  • DAR equation typesetting. Equation 1 / Equation 13 uses a piecewise inline notation for the Factor1-vs-1 cycle scaling that flattens ambiguously in plain-text extracts. The MathML in the source XML resolves it as 4 * (0.25 + 0.75 * exp(-beta * tad)) * (1 if cycle 1 else Factor1) – the cycle scaling multiplies the entire DAR(t) curve.