Bioequivalence and Bioavailability Forum

Hi Imph

❝ can you please enlighten me on the most accepted way to handle missing points around tmax?

Sorry, cards are stacked against you. I tried a lot with data imputation by PK modeling in the past, but nothing was accepted by agencies. However, if that doesn’t happen in many subjects, the loss in power when you exclude the subject(s) is generally not large (unless the number of eligible subjects becomes really small).

Examples:

library(PowerTOST)
# 2x2x2 design defaults: T/R 0.9, target power 0.8
excl <- 0.05 # fraction of excluded subjects
CV   <- seq(0.15, 0.35 0.05)
x    <- data.frame(CV = CV, n = NA_integer_, power = NA_real_,
                   elig = NA_real_, pow.elig = NA_real_, loss = NA_real_)
for (j in 1:nrow(x)) {
  tmp           <- sampleN.TOST(CV = CV[j], print = FALSE)
  x$n[j]        <- tmp[["Sample size"]]
  x$power[j]    <- tmp[["Achieved power"]]
  x$elig[j]     <- floor(x$n[j] * (1- excl))
  x$pow.elig[j] <- suppressMessages(power.TOST(CV = CV[j], n = x$elig[j]))
  x$loss[j]     <- 100 * (1 - x$pow.elig[j] / x$power[j])
}
names(x)[4:6] <- c("eligibe", "power", "loss (%)")
print(signif(x, 4), row.names = FALSE)

   CV  n  power eligibe  power loss (%)
 0.15 12 0.8305      11 0.7877    5.150
 0.20 20 0.8347      19 0.8132    2.569
 0.25 28 0.8074      26 0.7761    3.887
 0.30 40 0.8158      38 0.7953    2.515
 0.35 52 0.8075      49 0.7831    3.017