Skip to contents

Model and source

  • Citation: Dari A, Boulton M, Neyens M, Le Gars M, Valenzuela B, Shukarev G, Cardenas V, Ruiz-Guinazu J, Sadoff J, Hoetelmans RMW, Perez Ruixo JJ. Quantifying Antibody Persistence After a Single Dose of COVID-19 Vaccine Ad26.COV2.S in Humans Using a Mechanistic Modeling and Simulation Approach. Clin Pharmacol Ther. 2023;113(2):380. doi:10.1002/cpt.2796.
  • Article: https://doi.org/10.1002/cpt.2796 (open access, PMC10107600)

Dari 2023 fit the same six-compartment mechanistic model of the humoral immune response to two antibody read-outs measured after a single 5 x 10^10 viral-particle (vp) dose of Ad26.COV2.S:

  • Dari_2023_ad26cov2s_selisa – spike-protein-binding antibody, S-ELISA (EU/mL), 220 participants. This model was fit first.
  • Dari_2023_ad26cov2s_wtvna – neutralizing antibody, wild-type virus neutralization assay (wtVNA, IC50 titer), 82 participants. The systemic parameters (antigen, memory-B-cell and plasma-cell kinetics, and the age effect) were held at the S-ELISA estimates and only the antibody-production parameters, the sex effect and the variability terms were re-estimated.

The two fits are sequential but separate (different datasets, different random effects), so they are packaged as two models. Both are the paper’s “final updated model”, which added the day-169 and day-239 visits and is the model behind every simulation in the paper.

The model (Dari 2023 Eq. 1-6, Figure 1): a K-PD virtual antigen compartment (antigen) decays at k_ag; antigen drives memory-B-cell (bcell_memory) production at kp_ag, which also self-replicates logistically (kp_m, ceiling n_mem) and differentiates, by a second-order antigen x memory-B-cell process, into short-lived (plasma_short) and long-lived (plasma_long) plasma cells. Plasma cells secrete antibody (kab_s, kab_l) into serum (antibody), which distributes to a peripheral compartment (antibody_periph) and is eliminated (ke_ab). The observation Cc is the serum antibody level.

mod_s <- readModelDb("Dari_2023_ad26cov2s_selisa")
mod_v <- readModelDb("Dari_2023_ad26cov2s_wtvna")
lloq <- c(selisa = 50.3, wtvna = 58) # Methods, 'Immunogenicity assays'

Population

Healthy SARS-CoV-2-seronegative adults (no prior infection or COVID-19 vaccination) aged 18-55 or >= 65 years from the phase I/IIa COV1001 study (cohorts 1a and 3) and the phase IIa COV2001 study (group 5), who received a single 5 x 10^10 vp dose of Ad26.COV2.S. Participants aged 56-64 years were not included. Samples were drawn on days 1, 15, 29, 57, (64,) 71, 85, 169 and 239.

From Table 1: the S-ELISA analysis set had 220 participants (1,185 titers), median age 52.5 years (range 18-82), 46.8% women, 47.3% aged >= 60 years, median weight 72.1 kg, 86.4% White non-Hispanic, enrolled in the United States, Belgium, the Netherlands, Spain and Germany. The wtVNA set had 82 participants (442 titers), median age 47.5 years (range 18-75), 42.7% women, 42.7% aged >= 60 years, median weight 72.7 kg, enrolled in Belgium, the Netherlands and Spain. The same information is in each model’s population metadata.

Source trace

Every value below is a Table 2 entry of Dari 2023 unless stated otherwise; the in-file comments in inst/modeldb/specificDrugs/Dari_2023_ad26cov2s_*.R carry the same trace.

Parameter S-ELISA wtVNA Source
k_ag (1/day) 0.065 (fixed) 0.065 (fixed) Table 2; Balelli 2020
kp_ag (cells/antigen unit/day) 1 (fixed) 1 (fixed) Table 2; nonidentifiable, set to 1
kp_m, age < 60 y (1/day) 0.523 0.523 (from S-ELISA) Table 2
ke_m (1/day) 0 (fixed) 0 (fixed) Table 2 (0 /year)
n_mem (cells) 3.019 x 10^4 (fixed) 3.019 x 10^4 (fixed) Table 2; Goel 2021
kp_spl = kpS + kpL 0.691 x 10^-5 0.691 x 10^-5 (from S-ELISA) Table 2
r_kp_s_l = kpS/kpL 2.5/0.011 (fixed) 2.5/0.011 (fixed) Table 2; Balelli 2020
kp_s, kp_l derived derived Table 2 ‘Derived’ rows: kpS = kpSpL/(1 + 0.011/2.5), kpL = kpSpL - kpS
ke_s (1/day) 0.578 (fixed) 0.578 (fixed) Table 2; Pasin 2019
ke_l (1/day) 16.8 x 10^-4 0 (fixed) Table 2 (final updated model); Supplementary Materials 1.4
kab_s 0.530 0.603 Table 2
kab_l, males 4.43 1.86 Table 2
kab_p, kp_ab, ke_ab (1/day) 0.2075, 0.2716, 0.0556 (fixed) same Table 2; Davda 2014
e_age_ge60_kp_m -0.441 -0.441 (from S-ELISA) Table 2; form kpM * (1 + e * AGE_GE60), Supplementary Materials 1.7
e_sexf_kab_l 0.671 1.10 Table 2; form kAbL * (1 + e * SEXF)
IIV on kp_m (CV) 78.5% none Table 2 (final updated model); omega^2 = log(1 + CV^2)
IIV on kab_s (CV) 147% 102% Table 2 (final updated model)
IIV on kab_l (CV) 141% 104% Table 2 (final updated model)
Residual SD (log10 scale) 0.117 0.191 Table 2 ‘Proportional error’ (final updated model); see Assumptions
Antigen dose (amt) 500 (x 10^8 vp) 500 (x 10^8 vp) Not stated; see the next section
ODEs Eq. 1-6 Eq. 1-6 Methods, ‘Structural model’

Antigen dose scale

The paper gives kp_ag in cells/vp/day and kp_spl in 1/day/vp, but never states the number entered as the antigen dose. Entering the nominal 5 x 10^10 vp literally makes every titer about 10^11. The maintainers therefore solved the typical-value model (reference subject: man aged < 60 years) for each power-of-ten unit of the nominal dose and compared it with the observed median of Figure 2 (digitised by the maintainers to about +/- 0.1 log10).

obs_days <- c(14, 28, 56, 84, 168, 239)
# Figure 2 observed medians (red solid lines), log10 scale, digitised.
fig2_median <- tibble::tribble(
  ~assay,   ~time, ~fig2_log10,
  "selisa",    14, 2.30,
  "selisa",    28, 2.60,
  "selisa",    56, 2.70,
  "selisa",    84, 2.65,
  "selisa",   168, 2.55,
  "selisa",   239, 2.60,
  "wtvna",     14, 2.35,
  "wtvna",     28, 2.50,
  "wtvna",     56, 2.50,
  "wtvna",     84, 2.40,
  "wtvna",    168, 2.40,
  "wtvna",    239, 2.40
)

typical_profile <- function(mod, amt) {
  ev <- rxode2::et(amt = amt, cmt = "antigen") |>
    rxode2::et(obs_days, cmt = "antibody") |>
    as.data.frame() |>
    dplyr::mutate(SEXF = 0, AGE_GE60 = 0)
  sim <- rxode2::rxSolve(rxode2::zeroRe(mod), ev, returnType = "data.frame")
  sim[, c("time", "Cc")]
}

dose_units <- c(
  "1e9 vp (amt = 50)" = 50,
  "1e8 vp (amt = 500)" = 500,
  "1e7 vp (amt = 5000)" = 5000,
  "vp (amt = 5e10)" = 5e10
)
scale_check <- dplyr::bind_rows(lapply(names(dose_units), function(u) {
  dplyr::bind_rows(
    typical_profile(mod_s, dose_units[[u]]) |> dplyr::mutate(assay = "selisa"),
    typical_profile(mod_v, dose_units[[u]]) |> dplyr::mutate(assay = "wtvna")
  ) |>
    dplyr::mutate(dose_unit = u)
})) |>
  dplyr::mutate(log10_Cc = log10(Cc)) |>
  dplyr::left_join(fig2_median, by = c("assay", "time"))
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkp_m', 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkp_m', 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkp_m', 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkp_m', 'etalkab_s', 'etalkab_l'
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ omega/sigma items treated as zero: 'etalkab_s', 'etalkab_l'

scale_check |>
  dplyr::select(dose_unit, assay, time, log10_Cc) |>
  tidyr::pivot_wider(names_from = time, values_from = log10_Cc) |>
  knitr::kable(digits = 2, caption = "Typical-value log10 titer by day after vaccination, for each candidate dose unit.")
Typical-value log10 titer by day after vaccination, for each candidate dose unit.
dose_unit assay 14 28 56 84 168 239
1e9 vp (amt = 50) selisa 0.87 1.40 1.48 1.45 1.37 1.31
1e9 vp (amt = 50) wtvna 0.90 1.39 1.37 1.25 1.09 1.07
1e8 vp (amt = 500) selisa 2.30 2.56 2.61 2.57 2.49 2.43
1e8 vp (amt = 500) wtvna 2.32 2.54 2.48 2.36 2.21 2.19
1e7 vp (amt = 5000) selisa 3.59 3.74 3.75 3.71 3.63 3.57
1e7 vp (amt = 5000) wtvna 3.60 3.70 3.61 3.49 3.35 3.34
vp (amt = 5e10) selisa 11.34 11.46 11.46 11.43 11.34 11.29
vp (amt = 5e10) wtvna 11.35 11.41 11.31 11.20 11.07 11.05

scale_fit <- scale_check |>
  dplyr::group_by(dose_unit, assay) |>
  dplyr::summarise(
    mean_abs_log10_diff = mean(abs(log10_Cc - fig2_log10)),
    .groups = "drop"
  )
knitr::kable(scale_fit, digits = 2, caption = "Mean absolute difference from the Figure 2 observed median (log10 units).")
Mean absolute difference from the Figure 2 observed median (log10 units).
dose_unit assay mean_abs_log10_diff
1e7 vp (amt = 5000) selisa 1.10
1e7 vp (amt = 5000) wtvna 1.09
1e8 vp (amt = 500) selisa 0.07
1e8 vp (amt = 500) wtvna 0.09
1e9 vp (amt = 50) selisa 1.25
1e9 vp (amt = 50) wtvna 1.25
vp (amt = 5e10) selisa 8.82
vp (amt = 5e10) wtvna 8.81

# Deterministic typical-value solves, so the bounds can be tight. Only the
# 10^8-vp unit lands within a few tenths of a log10 of the observed median;
# the neighbouring units miss by more than one log10 for both assays.
fit_of <- function(unit, a) {
  scale_fit$mean_abs_log10_diff[scale_fit$dose_unit == unit & scale_fit$assay == a]
}
stopifnot(
  fit_of("1e8 vp (amt = 500)", "selisa") < 0.2,
  fit_of("1e8 vp (amt = 500)", "wtvna") < 0.2,
  fit_of("1e9 vp (amt = 50)", "selisa") > 1,
  fit_of("1e7 vp (amt = 5000)", "selisa") > 0.8,
  fit_of("1e9 vp (amt = 50)", "wtvna") > 1,
  fit_of("1e7 vp (amt = 5000)", "wtvna") > 0.8
)

The vaccine dose is therefore entered as amt = 500 in units of 10^8 vp. The persistence projections further down (Table 3, S1 and S2) are an independent check of the same choice: they depend on the dose scale through the fraction of participants whose titers sit near the LLOQ.

Virtual cohorts

# One dosing record per subject plus observation records on the serum
# antibody state; each subject carries its own sex and age indicators.
make_events <- function(SEXF, AGE_GE60, obs_times, id_offset = 0L) {
  n <- length(SEXF)
  ids <- id_offset + seq_len(n)
  covs <- data.frame(id = ids, SEXF = SEXF, AGE_GE60 = AGE_GE60)
  dose <- data.frame(id = ids, time = 0, evid = 1L, amt = 500, cmt = "antigen")
  obs <- expand.grid(time = obs_times, id = ids)
  obs$evid <- 0L
  obs$amt <- 0
  obs$cmt <- "antibody"
  dplyr::bind_rows(dose, obs[, names(dose)]) |>
    dplyr::left_join(covs, by = "id") |>
    dplyr::arrange(id, time, dplyr::desc(evid))
}

Figure 2: visual predictive check

Each assay’s cohort has 200 participants with the Table 1 proportions of women and of participants aged >= 60 years. sim includes the residual error, as the paper’s VPC does.

rxode2::rxSetSeed(2023)
set.seed(2023)
vpc_days <- c(14, 21, 28, 42, 56, 70, 84, 112, 140, 168, 196, 224, 239, 252, 280)
n_vpc <- 200
ev_s <- make_events(
  SEXF = sample(rep(0:1, c(n_vpc - round(0.468 * n_vpc), round(0.468 * n_vpc)))),
  AGE_GE60 = sample(rep(0:1, c(n_vpc - round(0.473 * n_vpc), round(0.473 * n_vpc)))),
  obs_times = vpc_days
)
ev_v <- make_events(
  SEXF = sample(rep(0:1, c(n_vpc - round(0.427 * n_vpc), round(0.427 * n_vpc)))),
  AGE_GE60 = sample(rep(0:1, c(n_vpc - round(0.427 * n_vpc), round(0.427 * n_vpc)))),
  obs_times = vpc_days
)
sim_vpc <- dplyr::bind_rows(
  rxode2::rxSolve(mod_s, ev_s, returnType = "data.frame") |> dplyr::mutate(assay = "selisa"),
  rxode2::rxSolve(mod_v, ev_v, returnType = "data.frame") |> dplyr::mutate(assay = "wtvna")
)
#> ℹ parameter labels from comments will be replaced by 'label()'
#> ℹ parameter labels from comments will be replaced by 'label()'

vpc <- sim_vpc |>
  dplyr::group_by(assay, time) |>
  dplyr::summarise(
    q05 = quantile(log10(sim), 0.05),
    q50 = quantile(log10(sim), 0.50),
    q95 = quantile(log10(sim), 0.95),
    .groups = "drop"
  )

ggplot(vpc, aes(time, q50)) +
  geom_ribbon(aes(ymin = q05, ymax = q95), fill = "steelblue", alpha = 0.25) +
  geom_line() +
  geom_point(data = fig2_median, aes(time, fig2_log10), colour = "red") +
  geom_hline(
    data = data.frame(assay = names(lloq), y = log10(lloq)),
    aes(yintercept = y), linetype = "dashed", colour = "navy"
  ) +
  facet_wrap(~assay, ncol = 1, labeller = as_labeller(c(selisa = "S-ELISA (EU/mL)", wtvna = "wtVNA (IC50)"))) +
  labs(
    x = "Time since vaccination (days)", y = "log10 titer",
    title = "Figure 2 - simulated median and 5th-95th percentiles",
    caption = "Replicates Figure 2 of Dari 2023. Red points: digitised observed medians; dashed line: LLOQ."
  )


vpc_check <- vpc |>
  dplyr::inner_join(fig2_median, by = c("assay", "time")) |>
  dplyr::mutate(diff = q50 - fig2_log10)
knitr::kable(vpc_check, digits = 2, caption = "Simulated vs digitised observed median (log10).")
Simulated vs digitised observed median (log10).
assay time q05 q50 q95 fig2_log10 diff
selisa 14 1.60 2.23 3.00 2.30 -0.07
selisa 28 2.02 2.57 3.27 2.60 -0.03
selisa 56 2.17 2.68 3.30 2.70 -0.02
selisa 84 2.05 2.65 3.31 2.65 0.00
selisa 168 1.80 2.54 3.19 2.55 -0.01
selisa 239 1.67 2.45 3.20 2.60 -0.15
wtvna 14 1.62 2.24 2.87 2.35 -0.11
wtvna 28 2.00 2.56 3.13 2.50 0.06
wtvna 56 2.09 2.56 3.02 2.50 0.06
wtvna 84 1.96 2.48 3.00 2.40 0.08
wtvna 168 1.70 2.32 2.90 2.40 -0.08
wtvna 239 1.55 2.28 3.07 2.40 -0.12

# The simulated median has a sampling SD of about 0.05 log10 at n = 200. A
# mis-transcribed production rate, dose or covariate shifts it by > 0.3.
stopifnot(
  abs(median(vpc_check$diff[vpc_check$assay == "selisa"])) < 0.2,
  abs(median(vpc_check$diff[vpc_check$assay == "wtvna"])) < 0.2,
  quantile(abs(vpc_check$diff), 0.9) < 0.3
)

Tables 3, S1 and S2: persistence above the LLOQ

The paper defines individual persistence as the time from vaccination until the last titer above the LLOQ before two consecutive monthly titers below it, and reports the share of participants still persisting at 8, 12, 18 and 24 months. The simulation below follows that rule on monthly simulated observations (including residual error), with 200 participants in each sex x age stratum.

rxode2::rxSetSeed(3)
months <- 1:24
strata <- expand.grid(SEXF = 0:1, AGE_GE60 = 0:1)
n_stratum <- 200
ev_p <- make_events(
  SEXF = rep(strata$SEXF, each = n_stratum),
  AGE_GE60 = rep(strata$AGE_GE60, each = n_stratum),
  obs_times = months * 30.4375
)
stopifnot(!anyDuplicated(ev_p[, c("id", "time", "evid")]))

persistence_months <- function(below) {
  # First month m with titers below the LLOQ at both m and m + 1; persistence
  # ends at the month before it. Never two in a row: persists beyond 24.
  k <- which(below[-length(below)] & below[-1])
  if (length(k) == 0) Inf else k[1] - 1
}

pers <- dplyr::bind_rows(
  rxode2::rxSolve(mod_s, ev_p, returnType = "data.frame", keep = c("SEXF", "AGE_GE60")) |>
    dplyr::mutate(assay = "selisa"),
  rxode2::rxSolve(mod_v, ev_p, returnType = "data.frame", keep = c("SEXF", "AGE_GE60")) |>
    dplyr::mutate(assay = "wtvna")
) |>
  dplyr::arrange(assay, id, time) |>
  dplyr::group_by(assay, id, SEXF, AGE_GE60) |>
  dplyr::summarise(
    persist = persistence_months(sim < lloq[[dplyr::first(assay)]]),
    .groups = "drop"
  )

by_stratum <- tidyr::expand_grid(pers, month = c(8, 12, 18, 24)) |>
  dplyr::group_by(assay, SEXF, AGE_GE60, month) |>
  dplyr::summarise(pct = 100 * mean(persist >= month), .groups = "drop")

The age strata of Table S1 have a 1:1 sex ratio, as in the paper’s simulation. The paper sampled age from the phase III COV3001 age distribution for Table 3 and Table S2; the share of participants >= 60 years implied by Table 3 and Table S1 is about 0.35 (for example S-ELISA at 8 months: (95.9 - 95.2) / (95.9 - 93.9) = 0.35), which is used to weight the age strata.

w60 <- 0.35
sim_tab <- dplyr::bind_rows(
  by_stratum |>
    dplyr::group_by(assay, month, AGE_GE60) |>
    dplyr::summarise(pct = mean(pct), .groups = "drop") |>
    dplyr::mutate(group = ifelse(AGE_GE60 == 1, "age >= 60 (Table S1)", "age < 60 (Table S1)")),
  by_stratum |>
    dplyr::group_by(assay, month, SEXF) |>
    dplyr::summarise(pct = sum(pct * ifelse(AGE_GE60 == 1, w60, 1 - w60)), .groups = "drop") |>
    dplyr::mutate(group = ifelse(SEXF == 1, "female (Table S2)", "male (Table S2)")),
  by_stratum |>
    dplyr::group_by(assay, month) |>
    dplyr::summarise(pct = sum(pct * 0.5 * ifelse(AGE_GE60 == 1, w60, 1 - w60)), .groups = "drop") |>
    dplyr::mutate(group = "overall (Table 3)")
) |>
  dplyr::select(assay, group, month, pct)

# Predicted % above the LLOQ, Dari 2023 Table 3, Table S1 and Table S2.
paper_tab <- tibble::tribble(
  ~group, ~month, ~selisa, ~wtvna,
  "overall (Table 3)", 8, 95.2, 88.4,
  "overall (Table 3)", 12, 92.4, 84.9,
  "overall (Table 3)", 18, 87.6, 82.1,
  "overall (Table 3)", 24, 81.1, 80.4,
  "age < 60 (Table S1)", 8, 95.9, 89.9,
  "age < 60 (Table S1)", 12, 93.4, 86.7,
  "age < 60 (Table S1)", 18, 88.9, 84.1,
  "age < 60 (Table S1)", 24, 82.8, 82.5,
  "age >= 60 (Table S1)", 8, 93.9, 85.3,
  "age >= 60 (Table S1)", 12, 90.5, 81.3,
  "age >= 60 (Table S1)", 18, 84.8, 78.2,
  "age >= 60 (Table S1)", 24, 77.7, 76.3,
  "male (Table S2)", 8, 93.0, 81.1,
  "male (Table S2)", 12, 89.2, 75.8,
  "male (Table S2)", 18, 82.9, 71.8,
  "male (Table S2)", 24, 74.9, 69.4,
  "female (Table S2)", 8, 97.4, 95.6,
  "female (Table S2)", 12, 95.6, 93.9,
  "female (Table S2)", 18, 92.2, 92.4,
  "female (Table S2)", 24, 87.2, 91.4
) |>
  tidyr::pivot_longer(c(selisa, wtvna), names_to = "assay", values_to = "paper_pct")

persist_cmp <- sim_tab |>
  dplyr::inner_join(paper_tab, by = c("assay", "group", "month")) |>
  dplyr::mutate(diff = pct - paper_pct)
stopifnot(nrow(persist_cmp) == nrow(paper_tab))

persist_cmp |>
  dplyr::arrange(assay, group, month) |>
  dplyr::rename(
    "Assay" = assay, "Group" = group, "Month" = month,
    "Simulated %" = pct, "Paper %" = paper_pct, "Difference" = diff
  ) |>
  knitr::kable(digits = 1, caption = "Predicted percentage of participants with persisting titers above the LLOQ.")
Predicted percentage of participants with persisting titers above the LLOQ.
Assay Group Month Simulated % Paper % Difference
selisa age < 60 (Table S1) 8 96.8 95.9 0.8
selisa age < 60 (Table S1) 12 94.0 93.4 0.6
selisa age < 60 (Table S1) 18 89.5 88.9 0.6
selisa age < 60 (Table S1) 24 84.5 82.8 1.7
selisa age >= 60 (Table S1) 8 95.2 93.9 1.3
selisa age >= 60 (Table S1) 12 93.0 90.5 2.5
selisa age >= 60 (Table S1) 18 85.0 84.8 0.2
selisa age >= 60 (Table S1) 24 79.5 77.7 1.8
selisa female (Table S2) 8 98.3 97.4 0.9
selisa female (Table S2) 12 97.3 95.6 1.7
selisa female (Table S2) 18 93.4 92.2 1.2
selisa female (Table S2) 24 88.6 87.2 1.4
selisa male (Table S2) 8 94.1 93.0 1.1
selisa male (Table S2) 12 90.0 89.2 0.8
selisa male (Table S2) 18 82.4 82.9 -0.5
selisa male (Table S2) 24 76.9 74.9 2.0
selisa overall (Table 3) 8 96.2 95.2 1.0
selisa overall (Table 3) 12 93.6 92.4 1.2
selisa overall (Table 3) 18 87.9 87.6 0.3
selisa overall (Table 3) 24 82.8 81.1 1.7
wtvna age < 60 (Table S1) 8 93.2 89.9 3.3
wtvna age < 60 (Table S1) 12 91.8 86.7 5.0
wtvna age < 60 (Table S1) 18 89.5 84.1 5.4
wtvna age < 60 (Table S1) 24 87.8 82.5 5.2
wtvna age >= 60 (Table S1) 8 89.2 85.3 4.0
wtvna age >= 60 (Table S1) 12 85.0 81.3 3.7
wtvna age >= 60 (Table S1) 18 82.0 78.2 3.8
wtvna age >= 60 (Table S1) 24 79.5 76.3 3.2
wtvna female (Table S2) 8 95.8 95.6 0.2
wtvna female (Table S2) 12 95.1 93.9 1.2
wtvna female (Table S2) 18 93.9 92.4 1.5
wtvna female (Table S2) 24 91.8 91.4 0.4
wtvna male (Table S2) 8 87.9 81.1 6.8
wtvna male (Table S2) 12 83.7 75.8 7.9
wtvna male (Table S2) 18 79.8 71.8 8.0
wtvna male (Table S2) 24 77.9 69.4 8.5
wtvna overall (Table 3) 8 91.8 88.4 3.4
wtvna overall (Table 3) 12 89.4 84.9 4.5
wtvna overall (Table 3) 18 86.9 82.1 4.8
wtvna overall (Table 3) 24 84.9 80.4 4.5

ggplot(persist_cmp, aes(month, pct, colour = group)) +
  geom_line() +
  geom_point(aes(y = paper_pct), shape = 4, size = 2) +
  facet_wrap(~assay) +
  labs(
    x = "Months after vaccination", y = "% of participants above LLOQ",
    caption = "Lines: simulation; crosses: Dari 2023 Table 3, S1 and S2 (Figures 3 and 4 plot the same strata)."
  )


# With 200 participants per stratum a single cell has a sampling SD of 2-3
# percentage points, and the cells of one assay move together because they
# share subjects. The wtVNA cells also sit about 3 points above the paper
# even in a 4,000-subject run (see the text below). One draw gave a median
# difference of 1.7 and a 90th percentile of about 6. A dose unit one power
# of ten off moves every cell by tens of points.
stopifnot(
  abs(median(persist_cmp$diff)) < 5,
  quantile(abs(persist_cmp$diff), 0.9) < 10
)
# The sex contrast for wtVNA at 24 months is the paper's headline (69.4% men
# vs 91.4% women); its size, not just its sign, is checked.
wt24 <- sim_tab |> dplyr::filter(assay == "wtvna", month == 24)
stopifnot(
  wt24$pct[wt24$group == "female (Table S2)"] - wt24$pct[wt24$group == "male (Table S2)"] > 10
)

The S-ELISA projections agree with Tables 3, S1 and S2 to within a few percentage points. The wtVNA projections run high, by about 3 points in a 4,000-subject run and most in men, where more titers sit close to the LLOQ. The likely cause is that the paper also sampled parameter uncertainty, which widens the titer distribution and moves more participants below the LLOQ; this vignette simulates the point estimates only.

PKNCA: peak antibody response

The paper reports no non-compartmental summaries, so this block only records the simulated individual peak titer and its timing (no residual error) for the Figure 2 cohorts, by assay and sex.

nca_in <- sim_vpc |>
  dplyr::filter(!is.na(Cc)) |>
  dplyr::mutate(group = paste(assay, ifelse(SEXF == 1, "female", "male"))) |>
  dplyr::select(id, time, Cc, group)
nca_in <- dplyr::bind_rows(
  nca_in,
  nca_in |> dplyr::distinct(id, group) |> dplyr::mutate(time = 0, Cc = 0)
) |>
  dplyr::arrange(group, id, time)
stopifnot(all(nca_in$time[!duplicated(nca_in[, c("group", "id")])] == 0))

dose_in <- nca_in |>
  dplyr::distinct(id, group) |>
  dplyr::mutate(time = 0, amt = 500)

nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(
  PKNCA::PKNCAconc(nca_in, Cc ~ time | group + id),
  PKNCA::PKNCAdose(dose_in, amt ~ time | group + id),
  intervals = data.frame(start = 0, end = 280, cmax = TRUE, tmax = TRUE)
))
nca_sum <- as.data.frame(nca_res) |>
  dplyr::filter(PPTESTCD %in% c("cmax", "tmax")) |>
  dplyr::group_by(group, PPTESTCD) |>
  dplyr::summarise(median = median(PPORRES), .groups = "drop") |>
  tidyr::pivot_wider(names_from = PPTESTCD, values_from = median)
nca_sum |>
  dplyr::rename("Assay and sex" = group, "Median Cmax (titer)" = cmax, "Median Tmax (day)" = tmax) |>
  knitr::kable(digits = 0, caption = "Simulated peak antibody response.")
Simulated peak antibody response.
Assay and sex Median Cmax (titer) Median Tmax (day)
selisa female 718 84
selisa male 479 56
wtvna female 503 42
wtvna male 371 42

The median peak falls on the plateau that the Figure 2 medians show between days 28 and 84, and women peak higher than men because of the sex effect on long-lived-plasma-cell antibody production.

Assumptions and deviations

  • Antigen dose scale (not stated in the paper). The vaccine dose enters the K-PD antigen compartment as amt = 500, i.e. the 5 x 10^10 vp dose in units of 10^8 vp. The paper gives kp_ag and kp_spl per vp but never the amount entered; the literal 5 x 10^10 gives titers near 10^11. The maintainers compared each power-of-ten unit of the nominal dose and kept the only one that reproduces both the Figure 2 medians and the Table 3 / S1 / S2 persistence (see above). No continuous fitting was done.
  • Residual error scale. Supplementary Materials 1.6 states an additive error on log10 titers; Table 2 reports it as “Proportional error” 11.7 (S-ELISA) and 19.1 (wtVNA) in a column headed CV%. The maintainers read these as log10-scale SDs x 100 (0.117 and 0.191) and encode them as lnorm() SDs on the natural-log scale (0.2694 and 0.4398). Reading them as CVs on the original scale (natural-log SD 0.117 and 0.189) leaves the simulated wtVNA persistence almost flat after month 8 (about 89% at 24 months), against 80.4% in Table 3; the log10 reading gives 83-85%.
  • Final model vs final updated model. Only the final updated model (day-239 data, ke_l estimated for S-ELISA, IIV and residual error re-estimated) is packaged. The day-85 final model shares the same structural estimates with ke_l = 0; its IIV and residual-error values are recorded in the model-file comments.
  • Persistence simulation. The paper simulated 250,000 participants with parameter uncertainty and drew age from the COV3001 distribution; this vignette uses 200 participants per sex x age stratum, no parameter uncertainty, and weights the age strata 0.65 / 0.35, the split implied by Table 3 and Table S1.
  • Observation name. The serum antibody output is named Cc, as for the anti-RBD IgG output of Dasti_2025_mrna1273_qsp, although it is an antibody response and not a concentration of the administered vaccine.
  • Screened covariates. Body weight, race and country were screened and not retained; they appear in covariatesDataExcluded (body weight) and in the model notes.
  • No erratum or correction notice for Dari 2023 was found in Europe PMC (search and citation list) as of 2026-10-26.