Bioequivalence and Bioavailability Forum

Dear Durva!

❝ Can you help me how to design sample size in simple language?

I’ll try. The sample size of a study depends on:

1. α; the probability of error type I (patient’s risk to be a treated with a bioinequivalent formulation). This value is commonly fixed to 0.05 (or 5%).^*
   
2. β; the probability of error type II (producer’s risk of not being able to demonstrate BE for a bioequivalent formulation). Power π (the chance to show BE) is 1–β. This value is can be freely chosen though most regulations require power of at least 80%. A study with lower power is not ethical since the risk of failure will be too high.
   
3. The expected deviation of test from reference.
   
4. The expected variability (CV).

Note that the last two points are only – more or less accurate – educated guesses. Therefore one should avoid the term “sample size calculation” and use “sample size estimation” instead. It is not possible to directly obtain the sample size – only power. By an iterative process the sample size is increased until at least the target power is reached. Example (cross-over, T/R 95%, α 0.05, CV 20%, target power 80%):
  n  % power
 18  79.12  ⇒ we start at n=18; power too low and increase the sample size
 19  81.43  ⇒ target power reached but we want a balanced design (even number)
 20  83.47  ⇒ here we are; with one drop-out in the study power will still be >80%
Hope that’s an appetizer. For more details see my presentations on the topic.

---

• Lower values have to be used if multiplicity is an issue (simultaneous testing), in certain two-stage designs, or for NTIDs (Brazil’s ANVISA only).