EMA: α-inflation – the suspicion begins to mount [RSABE / ABEL]
maybe some of you remember this
Surprisingly, when ‘scaled’ bioequivalence limits were set as bioequivalence limits, the highest rejection rate observed was at the lowest variability investigated. With CVANOVA of 30%, it was 7.05%, and it was still 5.39% with a variability of 40%. Therefore, taking into account a simulation error of roughly 0.5%, and the fact that our simulations are based on uncorrelated data and do not consider interindividual variabilities, one may doubt that an α-error of 5% is controlled even with the pre-set ‘scaled’ limits, at least for variabilities close to the cut-off point of CVANOVA of 30%.
- Table II Rejection Rate After 10000 Simulations […] and Empirical α-Error Rate at the ‘Scaled’ BE Limits […] According to EMA with Increasing Intraindividual Variability
CVANOVA [%] N GMR Empirical α-error rate [%]
────────────────────────────────────────────────
30 22 1.250 7.05 *
35 25 1.295 5.58 *
40 27 1.340 5.39 *
45 27 1.386 4.25
50 28 1.432 3.51
────────────────────────────────────────────────
*
significantly >0.05; my addition)Try this code:
library(PowerTOST)
CV <- seq(30, 50, 5)
res <- data.frame(CV = CV, N = c(22, 25, 27, 27, 28),
GMR = scABEL(CV/100)[, "upper"], pBE = NA, sig = "",
stringsAsFactors = FALSE)
for (j in seq_along(CV)) {
res$pBE[j] <- round(100*power.scABEL(CV = CV[j]/100, theta0 = res$GMR[j],
n = res$N[j], design = "2x3x3",
nsims = 1e6), 2)
}
sig <- binom.test(0.05*1e6, 1e6, alternative = "less")$conf.int[[2]]
res$sig[res$pBE/100 > sig] <- "*"
names(res)[5] <- ""
print(res, row.names = FALSE)
Do these results look familiar?
CV N GMR pBE
30 22 1.250000 6.88 *
35 25 1.294796 5.42 *
40 27 1.340165 5.04 *
45 27 1.385915 4.34
50 28 1.431910 3.32
If you are planning a study for evaluation with EMA’s ABEL-method think about it.
- Wonnemann M, Frömke C, Koch A. Inflation of the Type I Error: Investigations on Regulatory Recommendations for Bioequivalence of Highly Variable Drugs. Pharm Res. 31 (preprint published 18 July 2014) doi:10.1007/s11095-014-1450-z.
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
Complete thread:
- EMA: α-inflation – the suspicion begins to mountHelmut 2014-07-21 02:04 [RSABE / ABEL]
- EMA: α-inflation – the suspicion begins to mount ElMaestro 2014-07-21 09:24
- Why simulate a simple AB|BA at all? Helmut 2014-07-21 16:14
- Why simulate a simple AB|BA at all? ElMaestro 2014-07-21 20:50
- Fancy smoothing? Helmut 2014-07-21 23:13
- Arbitrary smoothing? d_labes 2014-07-22 08:43
- Noise… Helmut 2014-07-22 13:31
- TSD Japonica… d_labes 2014-07-22 16:06
- TSD Japonica… Helmut 2014-07-22 17:50
- Noise debugged d_labes 2014-07-24 13:47
- Noise debugged Helmut 2014-07-24 15:07
- Shit happens d_labes 2014-07-24 16:09
- Noise debugged Helmut 2014-07-24 15:07
- TSD Japonica… d_labes 2014-07-22 16:06
- Noise… Helmut 2014-07-22 13:31
- Arbitrary smoothing? d_labes 2014-07-22 08:43
- Commentary d_labes 2014-07-22 08:18
- Commentary nobody 2014-09-15 08:24
- Commentary Helmut 2014-09-15 10:12
- Commentary nobody 2014-09-15 08:24
- IBE/PBE = Two-Stage Helmut 2014-09-29 03:33
- IBE/PBE = Two-Stage ElMaestro 2014-09-29 15:16
- Sequential Design = Inflation likely… Helmut 2014-09-30 15:03
- IBE/PBE = Two-Stage ElMaestro 2014-09-29 15:16
- Fancy smoothing? Helmut 2014-07-21 23:13
- Why simulate a simple AB|BA at all? ElMaestro 2014-07-21 20:50
- Why simulate a simple AB|BA at all? Helmut 2014-07-21 16:14
- EMA: α-inflation – the suspicion begins to mount ElMaestro 2014-07-21 09:24