## Now I got it! [Power / Sample Size]

Hi Sereng,

❝ Many thanks for your response.

Welcome – though I missed the target.

❝ I am not even sure if I need to reply before or after your text.

See there.

❝ Is it possible you misunderstood my question?

Given what you posted in the following, yes, indeed.

❝ What I meant to ask is if I do replicate crossover, i.e., 50 subjects on Test x 2 periods and 50 subjects on Reference x 2 periods (replicate crossover, 2X2X4) as opposed to 100 subjects on test x 1 period and 100 subjects on Reference x 1 period (2X2X2), do I gain any sample size (or power) efficiency using FDA 3-tests (per PSG) for Levothyroxine?

That’s hypothetical cause the FDA will not accept a 2×2×2 crossover. Study cost hinges mainly on the number of treatments (which drives the number of samples and hence, costs of bioanalytics). Peanuts: In a replicate (less subjects) you safe some costs of pre-/post study lab exams which might be outweighed by a higher chance of dropouts.
See Fig.1.
Anyway: Let’s compare the FDA’s RSABE and the EMA’s fixed limits of 90.00–111.11% (in 2×2×2 and 2×2×4 crossovers) to conventional ABE with fixed limits of 80.00–125.00% (2×2×2 crossover)* based on data assessed by the FDA in 2011.

library(PowerTOST) # Yu (2011) https://www.fda.gov/media/82940/Download # 9 ANDAs of Levothyroxine: Cmax CV     <- c(0.052, 0.096, 0.186) # min, mean, max) theta0 <- 0.975                  # assumed T/R-ratio target <- 0.80                   # target power ≥80% x      <- data.frame(CV = CV, n.FDA = NA_integer_, cost.FDA = NA_real_,                      n.EMA2 = NA_integer_, cost.EMA2 = NA_real_,                      n.EMA4 = NA_integer_, cost.EMA4 = NA_real_,                      n.ABE = NA_integer_, cost.ABE = 1) for (j in seq_along(CV)) {   # RSABE for NTIDs, 2x2x4 design mandatory acc. to the guidance   x$n.FDA[j] <- sampleN.NTIDFDA(CV = CV[j], theta0 = theta0, targetpower = target, details = FALSE, print = FALSE)[["Sample size"]] # EMA for NTIDs, fixed limits 90.00-111.11% # 2x2x2 design (in product-specific guidance for NTIDs) x$n.EMA2[j] <- sampleN.TOST(CV = CV[j], theta0 = theta0, theta1 = 0.90,                               targetpower = target, design = "2x2x2",                               print = FALSE)[["Sample size"]]   # 2x2x4 design (optional)   x$n.EMA4[j] <- sampleN.TOST(CV = CV[j], theta0 = theta0, theta1 = 0.90, targetpower = target, design = "2x2x4", print = FALSE)[["Sample size"]] # conventional ABE, 2x2x2 design, fixed limits 80.00-125.00% x$n.ABE[j] <- sampleN.TOST(CV = CV[j], theta0 = theta0, theta1 = 0.80,                           targetpower = target, design = "2x2x2",                           print = FALSE)[["Sample size"]]   # minimum sample size acc. to the guideline   if (x$n.EMA2[j] < 12) x$n.EMA2[j] <- 12   if (x$n.EMA4[j] < 12) x$n.EMA4[j] <- 12   if (x$n.ABE[j] < 12) x$n.ABE[j] <- 12 } # cost relative to ABE 2×2×2 design with fixed limits 80.00-125.00% x$cost.FDA <- x$n.FDA * 2 / x$n.ABE x$cost.EMA2 <- x$n.EMA2 / x$n.ABE x$cost.EMA4 <- x$n.EMA4 * 2 / x\$n.ABE names(x)[c(3, 5, 7, 9)] <- rep("cost", 4) print(signif(x, 4), row.names = FALSE)     CV n.FDA  cost n.EMA2  cost n.EMA4  cost n.ABE cost  0.052    30 5.000     12 1.000     12 2.000    12    1  0.096    18 3.000     20 1.667     12 2.000    12    1  0.186    16 2.286     70 5.000     34 4.857    14    1

In short: For low variability RSABE is more costly than the EMA’s fixed limits. If the CV is larger than ~12% it is the other way ’round.
Don’t forget the comparisons of variabilities. Whereas for the EMA’s approaches we assume homoscedasticity $$\small{(s_\textrm{wT}^2\equiv s_\textrm{wR}^2),}$$ in RSABE a test for unequal variances is part of the procedure (see Fig.3). Hence, I recommend a pilot study to avoid surprises.

• The EMA recommends 90.00–111.11% for AUC0–48 and 80.00–125.00% for Cmax. A replicate design is not required (though always acceptable).

Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes