Bioequivalence and Bioavailability Forum

Hi Yung-jin,

❝ ❝ Sample size: 18

❝ ❝ Runtime (seconds):

❝ ❝ user system elapsed

❝ ❝ 0.14 0.01 0.15

❝

❝ I run the codes three times under R v4.0.3 and got:

❝

❝ Runtime (seconds):

❝    user  system elapsed

❝    0.11    0.00    0.13

❝    0.09    0.01    0.10

❝    0.11    0.00    0.11

❝

❝ a little bit faster. I guess that you have not upgraded to R 4.0.3 yet.

Nope, was on 4.0.3 as well.

❝ R v4.0.3 is simply faster (still my gut feeling ).

Cannot check. Deleted 4.0.2 already and not in the mood to reinstall it.

❝ I remember that your Xeon computer runs faster than my Windows, even than my Linux notebook last time. Is it the same computer?

It is. However, median of 15 runs 0.16 seconds. No idea.

❝ ❝ BTW, in v2.5 of last Tuesday Dave removed all reference-scaled methods and gives the code of PowerTOST in a box.

❝

❝ Thanks for the information.

Welcome.

Not for an initiate like you but for FARTSSIE-adepts…
With the script below you can reproduce the results of FARTSSIE ≤2.4; for the EMA’s ABEL you have to change cell U16 from 5 (which is for Health Canada’s TPD) to 3.

library(PowerTOST)
CV     <- seq(0.25, 0.65, 0.05)
theta0 <- 0.90
power  <- 0.80
design <- "2x2x4"
pers   <- as.numeric(substr(design, 5, 5))
n      <- data.frame(CV = CV)
theta1 <- numeric()
for (j in seq_along(CV)) {
  theta1[j] <- exp(-0.76*CV2se(CV[j]))
  if (CV[j] <= 0.3) theta1[j] <- 0.8
  if (CV[j] >= 0.5) theta1[j] <- exp(-0.76*CV2se(0.5))
  n.2x2x2       <- sampleN.TOST(CV = CV[j], design = "2x2x2",
                                targetpower = power, theta0 = theta0,
                                theta1 = theta1[j], theta2 = 1/theta1[j],
                                method = "nct", details = FALSE,
                                print = FALSE)[["Sample size"]]
  n$FARTSSIE[j] <- ceiling(n.2x2x2*(-0.25*pers+1.5))
  n$ABEL[j]     <- sampleN.scABEL(CV = CV[j], design = design,
                                  targetpower = power, theta0 = theta0,
                                  details = FALSE,
                                  print = FALSE)[["Sample size"]]
}
n$CV        <- n$CV*100
names(n)[1] <- "CV (%)"
cat("design:", design, "\n"); print(n, row.names = FALSE)

design: 2x2x4
 CV (%) FARTSSIE ABEL
     25       28   28
     30       40   34
     35       32   34
     40       27   30
     45       25   28
     50       23   28
     55       27   30
     60       31   32
     65       35   36

design: 2x2x3
 CV (%) FARTSSIE ABEL
     25       42   42
     30       60   50
     35       48   50
     40       41   46
     45       38   42
     50       35   42
     55       41   44
     60       47   48
     65       53   54

Flaws:

• Direct expansion of the limits instead of simulations, restriction of the point estimate (within 80.00–125.00%) not taken into account.
  
• Sample size estimated for 2×2×2 design and adapted (i.e., 50% for 4-period full replicates, 75% for three period replicates).

PS: Do you plan to implement my -code for ABE of last May in a future release of bear?