Bioequivalence and Bioavailability Forum

Sample size? [RSABE / ABEL]

posted by Helmut  – Vienna, Austria, 2013-02-19 18:25 (4451 d 15:33 ago) – Posting: # 10062
Views: 12,572

Hi John,

❝ ❝ Use the unscaled average bioequivalence procedure to determine BE for individual PK parameter(s). Every study should pass the scaled average bioequivalence limits and also regular unscaled bioequivalence limits of 80.00-125.00%

❝

❝ If a parameter passes ABE, it should pass SABE correct?

Not necessarily for NTIDs according to FDA’s requirement. For any CV_WR <21.42% the acceptance range will be narrower than 80–125%. In other words, the test might pass ABE but fail rSABE.

❝ Okay it doesn't necessary work the other way around. But in this case, since the intrasubject variability is less than 30% (as per FDA's info on the guidance), a properly powered study with an acceptable test formulation should have no issue passing both correct?

Define “properly powered study”.
Right now we don’t have an algo estimating the sample size taking all three conditions (rSABE, ABE, equal variances) into account. For CV_WR ≥21.42% the decision is triggered by conventional ABE, for CV_WR <21.42% one might use (as an approximation!) the downscaled acceptance ranges.*

require(PowerTOST)
sigma0 <- 0.1
for(CVwr in seq(0.05, 0.25, by=0.01)){
  sigmawr <- CV2se(CVwr)
  L       <- exp(-log(1.11111)/sigma0*sigmawr)
  U       <- exp(+log(1.11111)/sigma0*sigmawr)
  if(sigmawr > 0.211792){L <- 0.8; U <-1.25} # no scaling
  n       <- as.numeric(sampleN.TOST(
             CV=CVwr, theta0=0.95, theta1=L, theta2=U,
             design="2x2x4", print=F, details=F)[7])
  cat(round(100*CVwr,2), round(100*L,2), round(100*U,2), n, "\n")
}

%CVwr    L      U      n
 5     94.87 105.41 8426
 6     93.88 106.52  160
 7     92.90 107.64   62
 8     91.93 108.78   38
 9     90.97 109.93   28
10     90.02 111.08   22
11     89.09 112.25   20
12     88.16 113.43   18
13     87.25 114.61   16
14     86.35 115.81   14
15     85.46 117.02   14
16     84.58 118.24   14
17     83.71 119.46   12
18     82.85 120.70   12
19     82.00 121.95   12
20     81.17 123.20   12
21     80.34 124.47   12
22     80.00 125.00   12
23     80.00 125.00   12
24     80.00 125.00   14
25     80.00 125.00   14

Realistically demonstrating BE for CV <7% with T/R 0.95 is impossible. For NTIDs FDA suggested tighter limits on content which would make a smaller difference more likely. For T/R 0.975 I got:

%CVwr    L      U     n
 5     94.87 105.41  22
 6     93.88 106.52  18
 7     92.90 107.64  14
 8     91.93 108.78  12
 9     90.97 109.93  12
10     90.02 111.08  12
11     89.09 112.25  10
…
23     80.00 125.00  10
24     80.00 125.00  12
25     80.00 125.00  12

OK, that works. For the equality of variances cross fingers.

---

• \(\sigma_0=0.1 \;(CV \sim 10.03\%)\)
  \(\theta_s=(\log{(1.11111)})^2/{\sigma_{0}}^{2}\sim 1.11006\)
  \(\sigma_{wR}=\sqrt{\log{(CV{_{wR}}^{2}+1)}}\)
  \([L,U]=e^{\mp \log{(1.11111)}\cdot \sigma_{wR}/\sigma_0}\)

—
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:

• SABE or Normal BE? luvblooms 2013-02-19 05:07 [RSABE / ABEL] 
  • FDA: rSABE for NTIDs Helmut 2013-02-19 11:05
  • SABE or Normal BE? jag009 2013-02-19 15:23
    • Sample size?Helmut 2013-02-19 17:25
    • SABE or Normal BE? luvblooms 2013-02-20 05:26
      • SABE or Normal BE? jag009 2013-02-20 15:20
      • SABE or Normal BE? Helmut 2013-02-20 15:45
        • SABE or Normal BE? ElMaestro 2013-02-20 16:08
          • SABE or Normal BE? jag009 2013-02-21 17:10
            • FDA’s example data sets Helmut 2013-02-21 17:18
              • FDA’s example data sets jag009 2013-02-21 20:14
                • FDA’s example data sets Helmut 2013-02-21 22:42
                  • FDA’s example data sets jag009 2013-02-22 15:46
                    • FDA’s example data sets Helmut 2013-02-22 16:33