Bioequivalence and Bioavailability Forum

Dear kishpokuri,

❝ Can anyone suggest the appropriate sample size for a clinical endpoint bioequivalence study of Brinzolamide 1% ophthalmic suspension in chronic open angle glaucoma patients.

The FDA guidance on Brinzolamide gives you very specific details on how to establish BE with the clinical endpoint IOP (intra-ocular pressure.

Quote:
"To establish BE, the limits of each two-sided 95% confidence interval of the treatment difference (test – reference) for mean IOP of both eyes (continuous variable) at all four follow-up points (i.e., at approximately 8:00 a.m. (hour 0; before the morning drop) and 10:00 a.m. (hour 2) at the Day 14 (week 2) and Day 42 (week 6) visits must be within ± 1.5 mm Hg using the PP population for all time points measured and within ± 1.0 mm Hg using the PP population for the majority of time points measured."

Now you have:

• Statistical test = confidence interval inclusion rule, operationally identical with TOST (two one sided t-tests), but here for the difference on the original scale and with 95% CI (alpha=0.025, as usual with clinical endpoints)
  
• IUT (intersection-union) test of four time points, i.e. at all time points BE must be met. That allowes you to restrict your sample size estimation for one time point.
  
• Equivalence margin

What is missing for a successfull sample size estimation is the variability of the IOP. Try to find it in the literature.

Then look for a sample size estimation software which covers equivalence studies using the difference between means (f.i. PASS) with parallel group design.

You may also 'missuse' PowerTOST, which is intentionally programmed and described for ratios, but works also for differences using the argument logscale=FALSE. Set alpha=0.025, give theta0 as true difference (mm Hg) at which you will be able to accept BE, theta1 and theta2 to the stringenter equivalence margins as -1.0, 1.0 (mm Hg) and set CV to the standard deviation of the IOP (we don't know at moment).
Example use with a hypothetical IOP standard deviation and an assumed 'true' difference:
library(PowerTOST)
sampleN.TOST(alpha=0.025, logscale=FALSE, theta0=0.5, theta1=-1.0, theta2=1.0, CV=1.5, targetpower=0.8, design="parallel")

theta0 and standard deviation you must of course qualify from data in the literature.
The target power is ditto at your choice.

Hope this helps