Skip to contents

Simvastatin and four metabolites in children and adolescents (Ogungbenro 2019)

Source: vignettes/articles/Ogungbenro_2019_simvastatin.Rmd
Ogungbenro_2019_simvastatin.Rmd

Model and source

  • Citation: Ogungbenro K, Wagner JB, Abdel-Rahman S, Leeder JS, Galetin A. A population pharmacokinetic model for simvastatin and its metabolites in children and adolescents. Eur J Clin Pharmacol. 2019;75(9):1227-1235. doi:10.1007/s00228-019-02697-y.
  • Description: Joint one-compartment population PK model for oral simvastatin (lactone parent, SV) and four plasma metabolites – simvastatin acid (SVA), 6-hydroxymethyl simvastatin (HMSV), 6-hydroxymethyl simvastatin acid (HMSVA) and 3,5-dihydrodiol simvastatin (DHSV) – in hyperlipidemic children and adolescents (Ogungbenro 2019). SV absorption is two parallel sequential zero- then first-order processes, the second lagged; each metabolite is a one-compartment species formed from its precursor’s concentration and described by apparent (CL/V) rate constants because metabolite volumes were not identifiable. HMSV elimination is a saturable sigmoid (Hill) process and HMSVA is formed from both SVA and HMSV. SLCO1B1 c.521T>C (rs4149056, additive allele count) raises SVA formation and lowers its apparent volume; HMSVA and DHSV elimination decline with chronological age through a fractional Imax function. Doses in nmol and concentrations in nmol/L, matching the source. The Table 1 estimates are encoded as printed; they reproduce the paper’s covariate relationships but give exposures 2- to 10-fold below its own VPC (Figure 3) – see the vignette.
  • Article (open access): https://doi.org/10.1007/s00228-019-02697-y

The model is one joint NONMEM fit of simvastatin (SV, the lactone prodrug) and four plasma metabolites: simvastatin acid (SVA), 6-hydroxymethyl simvastatin (HMSV), 6-hydroxymethyl simvastatin acid (HMSVA) and 3,5-dihydrodiol simvastatin (DHSV). Figure 1 of the paper gives the structure:

  • SV: one compartment, oral, with two parallel absorption processes, each a zero-order input followed by first-order absorption; the second process is lagged. The fractions are F1 = 1/(1 + BA) and F2 = BA/(1 + BA).
  • SVA, HMSV and DHSV are formed from SV; HMSVA is formed from both SVA and HMSV. Every metabolite has one compartment.
  • Metabolite volumes are not identifiable, so the paper estimated only the ratios CL/V (1/h). The metabolite states in the packaged model therefore hold concentrations (nmol/L) directly.
  • HMSV elimination is saturable (sigmoid, gamma = 0.86).

Population

Ogungbenro 2019 re-analysed the single-centre, open-label, genotype-stratified single-dose study of Wagner et al.: 32 hyperlipidemic children and adolescents aged 8.8-20.2 years (mean 14.1), 17 female and 15 male, mostly overweight for age (mean weight 79.4 kg, range 30.2-149.2 kg; mean BMI 30.2 kg/m^2). Subjects aged 8-17 years received 10 mg simvastatin (n = 28) and those aged 18 years or older 20 mg (n = 4). SLCO1B1 c.521T>C genotypes were TT/TC/CC = 15/15/2, and 30 of the 32 subjects were Caucasian. Plasma was sampled at 0, 0.5, 1, 1.5, 2, 3, 4, 6 and 8 h; the LLOQ was 0.5 nmol/L for every analyte and BLQ samples were handled with the M3 method (Supplemental Table S1 and Methods).

The same information is available programmatically via readModelDb("Ogungbenro_2019_simvastatin")()$population.

Source trace

Equation / parameter Value Source location
lcl (CLSLe/F) log(1300 L/h) Table 1
lvc (VSL/F) log(110 L) Table 1
ld1, ld2 (D1, D2) log(0.069 h), log(0.39 h) Table 1 (unit printed as h-1; see Assumptions)
lka_slow, lka_fast (ka1, ka2) log(0.030 1/h), log(0.41 1/h) Table 1
logitfrel (from BA) -log(0.78) Table 1 BA = 0.78 and footnote F1 = 1/(1 + BA)
ltlag (ALAG) log(0.18 h) Table 1
lkmet_acid, lkel_acid log(0.043), log(0.13) 1/h Table 1; Eqs 1-2
e_slco1b1_521_vc_acid -0.37 Table 1 theta561; Eqs 1-2
e_slco1b1_521_kmet_acid 0.93 Table 1 theta562; Eq 1
lkmet_hmsv log(16.1 1/h) Table 1
lvmax_hmsv, lkm_hmsv, hill_hmsv log(660 nmol/L/h), log(20 nmol/L), 0.86 Table 1
lkmet_hmsva, lkmet_hmsva_hmsv log(0.48), log(0.55) 1/h Table 1
lkel_hmsva, e_age_kel_hmsva, lage50_kel_hmsva log(4.91 1/h), 1 (fixed), log(5.34 y) Table 1; Eq 3
lkmet_dhsv log(4.11 1/h) Table 1
lkel_dhsv, e_age_kel_dhsv, lage50_kel_dhsv log(12.9 1/h), 0.90, log(6.3 y) Table 1; Eq 4
BSV (14 etas) log(1 + CV^2) Table 1, footnote b (CV)
expSd*, mRes* sqrt(variance), m Table 1 residual rows, footnote c
SV ODE and parallel absorption n/a Methods; Fig 1
Metabolite ODEs (concentration states) n/a Methods; Fig 1
Double-exponential residual error n/a Table 1 footnote c; Supplementary Method

Covariate equations against the paper’s own numbers

Eqs 1-4 are the whole covariate model. The paper and its supplement restate their consequences in prose, which gives exact checks of the encoding. The typical-value derived rate constants are read straight from a solve.

mod <- readModelDb("Ogungbenro_2019_simvastatin")
mw_sv <- 418.57 # simvastatin lactone, g/mol
dose_nmol <- function(mg) mg / mw_sv * 1e6

typical_rates <- function(age, het, hom) {
  ev <- data.frame(
    id = 1L, time = c(0, 0, 1), amt = c(dose_nmol(10), dose_nmol(10), 0),
    evid = c(1L, 1L, 0L), cmt = c("depot", "depot2", "central"),
    rate = c(-2, -2, 0), dvid = c(NA, NA, 1L),
    AGE = age, SNP_SLCO1B1_RS4149056_HET = het, SNP_SLCO1B1_RS4149056_HOM = hom
  )
  s <- rxode2::rxSolve(mod, ev, omega = NA, sigma = NA, returnType = "data.frame")
  s[1, c("kmet_acid", "kel_acid", "kel_hmsva", "kel_dhsv")]
}
tt <- typical_rates(14, 0, 0)
#> ℹ parameter labels from comments will be replaced by 'label()'
tc <- typical_rates(14, 1, 0)
cc <- typical_rates(14, 0, 1)
age4 <- typical_rates(4, 0, 0)
age18 <- typical_rates(18, 0, 0)

checks <- tibble::tribble(
  ~claim, ~published, ~model,
  "CLLA/VSVA, TC vs TT (% higher)", 206, 100 * (tc$kmet_acid / tt$kmet_acid - 1),
  "CLLA/VSVA, CC vs TT (% higher)", 1000, 100 * (cc$kmet_acid / tt$kmet_acid - 1),
  "CLHSVAe/VHSVA, age 4 to 18 y (% decrease)", 60, 100 * (1 - age18$kel_hmsva / age4$kel_hmsva),
  "CLDHSe/VDHSV, age 4 to 18 y (% decrease)", 49, 100 * (1 - age18$kel_dhsv / age4$kel_dhsv)
)
knitr::kable(checks, digits = 1, caption = "Supplementary Results and Results prose vs the packaged model.")
Supplementary Results and Results prose vs the packaged model.
claim published model
CLLA/VSVA, TC vs TT (% higher) 206 206.3
CLLA/VSVA, CC vs TT (% higher) 1000 1000.0
CLHSVAe/VHSVA, age 4 to 18 y (% decrease) 60 60.0
CLDHSe/VDHSV, age 4 to 18 y (% decrease) 49 48.8 

# The published figures are rounded to the nearest percent, and the model
# reproduces each to within that rounding.
stopifnot(all(abs(checks$model - checks$published) < 1))

The VSVA and CLLA components separately are 37%/74% lower and 93%/186% higher for TC/CC (Supplementary Results), which is -0.37 * rs56 and 0.93 * rs56 directly.

Figure 4: elimination parameters against age 

ages <- seq(8, 21, by = 0.5)
fig4 <- bind_rows(lapply(ages, function(a) cbind(AGE = a, typical_rates(a, 0, 0))))

# Digitised from the LOWESS lines of Figure 4 (middle and right panels).
fig4_pub <- tibble::tribble(
  ~AGE, ~param, ~value,
  10, "CLHSVAe/VHSVA", 1.7,
  18, "CLHSVAe/VHSVA", 1.15,
  10, "CLDHSe/VDHSV", 5.8,
  18, "CLDHSe/VDHSV", 4.3
)

fig4 |>
  select(AGE, `CLHSVAe/VHSVA` = kel_hmsva, `CLDHSe/VDHSV` = kel_dhsv) |>
  pivot_longer(-AGE, names_to = "param") |>
  ggplot(aes(AGE, value)) +
  geom_line() +
  geom_point(data = fig4_pub, shape = 21, size = 2.5, fill = "grey70") +
  facet_wrap(~param, scales = "free_y") +
  labs(
    x = "Age (years)", y = "Typical value (1/h)",
    caption = "Replicates the middle and right panels of Figure 4 of Ogungbenro 2019 (points: digitised LOWESS line)."
  )


fig4_chk <- fig4_pub |>
  mutate(model = ifelse(
    param == "CLHSVAe/VHSVA",
    fig4$kel_hmsva[match(AGE, fig4$AGE)],
    fig4$kel_dhsv[match(AGE, fig4$AGE)]
  ))
stopifnot(all(abs(fig4_chk$model / fig4_chk$value - 1) < 0.1))

The model curves are the typical values; the paper’s points are individual empirical Bayes estimates, which scatter by the BSV around them.

Virtual cohort

The cohort mirrors Supplemental Table S1: a 10 mg arm aged 8.8-17.9 years with the 10 mg genotype mix (TT/TC/CC = 13/13/2), and a 20 mg arm aged 17.8-20.2 years with the 20 mg mix (2/2/0). Age is drawn uniformly over each arm’s range. Each subject has one observation row per analyte per scheduled time (dvid 1-5 in the order Cc, Cc_acid, Cc_hmsv, Cc_hmsva, Cc_dhsv), so the sim column carries the residual-error draw for that analyte.

set.seed(2019)
rxode2::rxSetSeed(2019)

sample_times <- c(0.5, 1, 1.5, 2, 3, 4, 6, 8)
analytes <- c("Cc", "Cc_acid", "Cc_hmsv", "Cc_hmsva", "Cc_dhsv")

make_cohort <- function(n, dose_mg, age_range, geno_prob, id_offset = 0L) {
  geno <- sample(c("TT", "TC", "CC"), n, replace = TRUE, prob = geno_prob)
  subj <- tibble(
    id = id_offset + seq_len(n),
    AGE = runif(n, age_range[1], age_range[2]),
    SNP_SLCO1B1_RS4149056_HET = as.integer(geno == "TC"),
    SNP_SLCO1B1_RS4149056_HOM = as.integer(geno == "CC"),
    genotype = geno,
    treatment = paste(dose_mg, "mg")
  )
  doses <- tidyr::crossing(subj, cmt = c("depot", "depot2")) |>
    mutate(time = 0, amt = dose_nmol(dose_mg), evid = 1L, rate = -2, dvid = NA_integer_)
  obs <- tidyr::crossing(subj, time = c(0, sample_times), dvid = seq_along(analytes)) |>
    mutate(cmt = "central", amt = 0, evid = 0L, rate = 0, analyte_id = dvid)
  bind_rows(doses, obs) |>
    arrange(id, time, desc(evid), dvid)
}

events <- bind_rows(
  make_cohort(200, 10, c(8.8, 17.9), c(13, 13, 2), id_offset = 0L),
  make_cohort(100, 20, c(17.8, 20.2), c(2, 2, 0), id_offset = 200L)
)
stopifnot(!anyDuplicated(unique(events[, c("id", "time", "evid", "dvid", "cmt")])))

Simulation 

sim <- rxode2::rxSolve(
  mod, events = events,
  keep = c("treatment", "genotype", "analyte_id"),
  returnType = "data.frame"
)
sim <- sim |> mutate(analyte = analytes[analyte_id])

Figure 3: VPC after a 10 mg dose 

lloq <- 0.5
vpc <- sim |>
  filter(treatment == "10 mg", time %in% sample_times) |>
  group_by(analyte, time) |>
  summarise(
    Q05 = quantile(sim, 0.05), Q50 = quantile(sim, 0.50), Q95 = quantile(sim, 0.95),
    .groups = "drop"
  ) |>
  mutate(analyte = factor(analyte, analytes, c("SV", "SVA", "HMSV", "HMSVA", "DHSV")))

ggplot(vpc, aes(time, Q50)) +
  geom_ribbon(aes(ymin = Q05, ymax = Q95), alpha = 0.25) +
  geom_line() +
  geom_hline(yintercept = lloq, linetype = "dashed") +
  facet_wrap(~analyte, nrow = 1) +
  scale_y_log10() +
  labs(
    x = "Time (h)", y = "Concentration (nmol/L)",
    caption = "Replicates the upper panels of Figure 3 of Ogungbenro 2019 (10 mg; median and 5th-95th percentiles with residual error)."
  )

Known discrepancy: the printed estimates do not reproduce Figure 3

The model reproduces every covariate relationship the paper states, but its absolute exposures fall well below the paper’s own VPC. The table compares the simulated medians with the medians digitised from Figure 3 (10 mg).

# Digitised from the median (50th percentile) lines of Figure 3, 10 mg.
fig3_pub <- tibble::tribble(
  ~analyte, ~time, ~published,
  "Cc", 1, 7.5, "Cc", 2, 7.0, "Cc", 4, 4.0, "Cc", 8, 1.2,
  "Cc_acid", 1, 0.8, "Cc_acid", 2, 1.2, "Cc_acid", 4, 1.45, "Cc_acid", 8, 1.25,
  "Cc_hmsv", 1, 10, "Cc_hmsv", 2, 10, "Cc_hmsv", 4, 4.0, "Cc_hmsv", 8, 0.9,
  "Cc_hmsva", 1, 2.5, "Cc_hmsva", 2, 3.5, "Cc_hmsva", 4, 2.8, "Cc_hmsva", 8, 0.95,
  "Cc_dhsv", 1, 4.5, "Cc_dhsv", 2, 5.3, "Cc_dhsv", 4, 3.0, "Cc_dhsv", 8, 0.85
)
fig3_cmp <- vpc |>
  mutate(analyte = analytes[as.integer(analyte)]) |>
  select(analyte, time, simulated = Q50) |>
  inner_join(fig3_pub, by = c("analyte", "time")) |>
  mutate(published_over_simulated = published / simulated)
knitr::kable(fig3_cmp, digits = 2, caption = "Median concentration (nmol/L): simulated vs Figure 3.")
Median concentration (nmol/L): simulated vs Figure 3.
analyte time simulated published published_over_simulated
Cc 1 1.93 7.50 3.89
Cc 2 2.01 7.00 3.47
Cc 4 1.16 4.00 3.44
Cc 8 0.47 1.20 2.57
Cc_acid 1 0.11 0.80 7.52
Cc_acid 2 0.24 1.20 4.95
Cc_acid 4 0.36 1.45 4.02
Cc_acid 8 0.33 1.25 3.81
Cc_dhsv 1 1.76 4.50 2.56
Cc_dhsv 2 1.65 5.30 3.21
Cc_dhsv 4 1.00 3.00 2.99
Cc_dhsv 8 0.41 0.85 2.05
Cc_hmsv 1 0.59 10.00 16.95
Cc_hmsv 2 0.62 10.00 16.22
Cc_hmsv 4 0.33 4.00 12.30
Cc_hmsv 8 0.11 0.90 8.26
Cc_hmsva 1 0.16 2.50 15.87
Cc_hmsva 2 0.34 3.50 10.38
Cc_hmsva 4 0.32 2.80 8.81
Cc_hmsva 8 0.19 0.95 5.11

The SV shortfall cannot be a matter of reading the figure. With CL/F = 1300 L/h and F1 + F2 = 1, the AUC from zero to infinity of a 10 mg dose is at most dose / (CL/F); the area under the Figure 3 median line over 0-8 h alone exceeds that bound.

auc_bound <- dose_nmol(10) / 1300
fig3_sv <- fig3_pub |>
  filter(analyte == "Cc") |>
  bind_rows(tibble::tibble(analyte = "Cc", time = 0, published = 0)) |>
  arrange(time)
auc_fig3_0_8 <- sum(diff(fig3_sv$time) * (head(fig3_sv$published, -1) + tail(fig3_sv$published, -1)) / 2)
c(auc_inf_bound_nM_h = auc_bound, fig3_median_auc_0_8_nM_h = auc_fig3_0_8)
#>       auc_inf_bound_nM_h fig3_median_auc_0_8_nM_h 
#>                 18.37759                 32.40000

# Typical-value check that the packaged model obeys the bound, so the
# shortfall against Figure 3 comes from the printed estimates, not the encoding.
ev_typ <- data.frame(
  id = 1L, time = c(0, 0, seq(0, 200, by = 0.25)),
  amt = c(dose_nmol(10), dose_nmol(10), rep(0, 801)),
  evid = c(1L, 1L, rep(0L, 801)), cmt = c("depot", "depot2", rep("central", 801)),
  rate = c(-2, -2, rep(0, 801)), dvid = c(NA, NA, rep(1L, 801)),
  AGE = 14, SNP_SLCO1B1_RS4149056_HET = 0, SNP_SLCO1B1_RS4149056_HOM = 0
)
typ <- rxode2::rxSolve(mod, ev_typ, omega = NA, sigma = NA, returnType = "data.frame")
auc_typ <- sum(diff(typ$time) * (head(typ$Cc, -1) + tail(typ$Cc, -1)) / 2)
stopifnot(abs(auc_typ / auc_bound - 1) < 0.02)

The covariate relationships therefore reproduce and the absolute levels do not: SV and DHSV run about 2- to 4-fold below Figure 3, SVA about 4- to 7-fold, and HMSV and HMSVA, downstream of the saturable HMSV elimination, about 5- to 17-fold. No single misprinted value explains both: lowering CL/F alone raises SV, SVA and DHSV toward the figure but leaves HMSV and HMSVA several-fold low. The packaged model keeps the estimates exactly as printed in Table 1; it is not tuned to the figure.

Fraction below the LLOQ

Supplemental Table S1 reports, per analyte, the percentage of all samples (both dose groups) below the LLOQ.

blq_pub <- tibble::tribble(
  ~analyte, ~published_pct,
  "Cc", 7, "Cc_acid", 38, "Cc_hmsv", 10, "Cc_hmsva", 19, "Cc_dhsv", 14
)
blq <- sim |>
  filter(time %in% sample_times) |>
  group_by(analyte) |>
  summarise(simulated_pct = 100 * mean(sim < lloq), .groups = "drop") |>
  left_join(blq_pub, by = "analyte")
knitr::kable(blq, digits = 1, caption = "Percentage of samples below the LLOQ (0.5 nmol/L).")
Percentage of samples below the LLOQ (0.5 nmol/L).
analyte simulated_pct published_pct
Cc 17.1 7
Cc_acid 69.5 38
Cc_dhsv 23.0 14
Cc_hmsv 51.6 10
Cc_hmsva 64.5 19

The simulated BLQ fractions are two to five times the observed ones, for the same reason as the Figure 3 shortfall above; the paper’s own VPC (lower panels of Figure 3) tracks the observed fractions closely.

SLCO1B1 c.521T>C and SVA exposure

The Results state that the observed and simulated median SVA profiles were up to 6.3-fold higher in c.521CC than in c.521TT subjects (VPC stratified by genotype, Online Resource).

sva_geno <- sim |>
  filter(treatment == "10 mg", analyte == "Cc_acid", time %in% sample_times) |>
  group_by(genotype, time) |>
  summarise(median_ipred = median(Cc_acid), .groups = "drop") |>
  pivot_wider(names_from = genotype, values_from = median_ipred) |>
  mutate(ratio_CC_TT = CC / TT)
knitr::kable(sva_geno, digits = 3, caption = "Median SVA IPRED (nmol/L) by genotype, 10 mg arm.")
Median SVA IPRED (nmol/L) by genotype, 10 mg arm.
time CC TC TT ratio_CC_TT
0.5 0.112 0.036 0.008 14.851
1.0 0.443 0.172 0.040 10.989
1.5 0.694 0.301 0.076 9.104
2.0 0.950 0.412 0.101 9.425
3.0 1.053 0.518 0.143 7.377
4.0 0.946 0.609 0.165 5.727
6.0 0.737 0.644 0.181 4.079
8.0 0.442 0.572 0.169 2.615 

ggplot(
  sim |> filter(treatment == "10 mg", analyte == "Cc_acid", time %in% sample_times) |>
    group_by(genotype, time) |> summarise(Q50 = median(Cc_acid), .groups = "drop"),
  aes(time, Q50, linetype = genotype)
) +
  geom_line() +
  scale_y_log10() +
  labs(x = "Time (h)", y = "Median SVA (nmol/L)",
       caption = "Genotype-stratified SVA median profile (Online Resource, Supplemental Figure S3).")

PKNCA validation

The paper reports no NCA of its own; the noncompartmental analysis of the same data (Wagner et al.) found each copy of the c.521C allele raised SVA exposure about 2.5-fold (Introduction). Cmax, Tmax and AUC0-8 are computed per analyte and dose group from the individual predictions.

sim_nca <- sim |>
  filter(!is.na(Cc)) |>
  mutate(conc = case_when(
    analyte == "Cc" ~ Cc, analyte == "Cc_acid" ~ Cc_acid,
    analyte == "Cc_hmsv" ~ Cc_hmsv, analyte == "Cc_hmsva" ~ Cc_hmsva,
    analyte == "Cc_dhsv" ~ Cc_dhsv
  )) |>
  select(id, time, conc, treatment, analyte, genotype)

conc_obj <- PKNCA::PKNCAconc(sim_nca, conc ~ time | treatment + analyte + id)
dose_df <- events |>
  filter(evid == 1, cmt == "depot") |>
  select(id, time, amt, treatment)
dose_obj <- PKNCA::PKNCAdose(dose_df, amt ~ time | treatment + id)
intervals <- data.frame(start = 0, end = 8, cmax = TRUE, tmax = TRUE, auclast = TRUE)
nca_res <- PKNCA::pk.nca(PKNCA::PKNCAdata(conc_obj, dose_obj, intervals = intervals))

nca_tbl <- as.data.frame(nca_res$result) |>
  group_by(treatment, analyte, PPTESTCD) |>
  summarise(median = median(PPORRES), .groups = "drop") |>
  pivot_wider(names_from = PPTESTCD, values_from = median) |>
  mutate(analyte = factor(analyte, analytes, c("SV", "SVA", "HMSV", "HMSVA", "DHSV"))) |>
  arrange(treatment, analyte)
nca_tbl |>
  rename(
    "Dose" = treatment, "Analyte" = analyte, "Cmax (nmol/L)" = cmax,
    "Tmax (h)" = tmax, "AUC0-8 (nmol*h/L)" = auclast
  ) |>
  knitr::kable(digits = 2, caption = "Median simulated NCA parameters (individual predictions).")
Median simulated NCA parameters (individual predictions).
Dose Analyte AUC0-8 (nmol*h/L) Cmax (nmol/L) Tmax (h)
10 mg SV 9.52 2.76 1.0
10 mg SVA 2.15 0.40 4.0
10 mg HMSV 2.99 0.94 1.0
10 mg HMSVA 2.18 0.42 3.0
10 mg DHSV 8.35 2.10 1.5
20 mg SV 18.16 5.25 1.5
20 mg SVA 4.55 0.83 6.0
20 mg HMSV 6.73 2.30 1.5
20 mg HMSVA 5.77 1.18 3.0
20 mg DHSV 17.81 4.63 1.5 

sva_auc <- as.data.frame(nca_res$result) |>
  filter(analyte == "Cc_acid", treatment == "10 mg", PPTESTCD == "auclast") |>
  left_join(distinct(sim_nca, id, genotype), by = "id") |>
  group_by(genotype) |>
  summarise(median_auc = median(PPORRES), .groups = "drop")
knitr::kable(sva_auc, digits = 2, caption = "Median SVA AUC0-8 (nmol*h/L) by c.521T>C genotype, 10 mg arm.")
Median SVA AUC0-8 (nmol*h/L) by c.521T>C genotype, 10 mg arm.
genotype median_auc
CC 5.68
TC 4.13
TT 1.08

The genotype effect on SVA exposure has the published direction and size: TC and CC raise the median SVA AUC0-8 several-fold over TT, in line with the roughly 2.5-fold increase per c.521C allele of the noncompartmental analysis. The ratio of genotype medians over time (table above) peaks higher than the paper’s “up to 6.3-fold” at the earliest times, where TT concentrations are far below the LLOQ and the ratio is of two very small numbers. The absolute values in the NCA table inherit the shortfall described under “Known discrepancy”, so no published-value comparison table is given.

Assumptions and deviations

  • Absolute exposure does not reproduce Figure 3 (see “Known discrepancy” above). The estimates are kept exactly as printed in Table 1. The maintainers have not adopted any back-solved value; a correction needs the authors’ control stream or an erratum.
  • Typical-value reference (Table 1 footnote a). The footnote calls the tabulated values those of a typical 14-year-old, 80 kg, c.521CC subject. Eqs 1-4 multiply the tabulated values by the genotype and age factors, which makes them the TT (rs56 = 0) and AGE = 0 values. The maintainers followed the equations: they reproduce every percentage the paper and supplement state (206% and 1000%; 60% and 49%) and the Figure 4 age trends, whereas the footnote’s reading reproduces none of them.
  • D1 and D2 units. Table 1 prints (h-1); the Methods define D1 and D2 as durations of zero-order input, so they are encoded in hours.
  • Bioavailability split. BA is encoded as logitfrel = -log(BA), which makes expit(logitfrel) = 1/(1 + BA) = F1 exactly. BA had no BSV. The split is applied to each depot’s first-order outflow rather than through f(), and each depot receives the full dose: in rxode2 5.1.8, f() on two depots together with a lagged modelled duration is mis-solved in any model that has more than one residual-error endpoint. The central and metabolite profiles are identical between the two forms.
  • Residual error. Table 1 footnote c gives the double-exponential model ln(y) = ln(f + m) + f/(f + m) * eps1 + m/(f + m) * eps2, with the variances of eps1 and eps2 tabulated; for SV, HMSV and DHSV, eps1 has separate variances for 0-2 h and 2-8 h after the dose. The packaged model uses that log-scale SD exactly (lnorm() with sqrt((f/(f+m))^2 var(eps1) + (m/(f+m))^2 var(eps2)), switching at 2 h after the dose), but centres the log-normal on ln(f) rather than ln(f + m): nlmixr2’s lnorm() takes the prediction itself. The offset m is 0.0001 nmol/L (fixed) for HMSV, HMSVA and DHSV and 0.012 nmol/L for SV, so only SVA (m = 0.22 nmol/L, below the 0.5 nmol/L LLOQ) is materially affected, and only at concentrations near or below the LLOQ.
  • Molecular weight. Doses are converted to nmol with the simvastatin lactone molecular weight 418.57 g/mol, which the paper does not state.
  • Virtual cohort. Age is drawn uniformly over each dose group’s range and genotype from each group’s observed frequencies (Supplemental Table S1); the observed ages and genotypes were not published per subject. Body weight was not retained in the model and is not simulated.
  • Errata. No correction notice for this article was found on Europe PMC as of 2026-10-16.