Bioequivalence and Bioavailability Forum

Dear Joshua!

❝ Here we intented for the same reference given to two groups and one test.

What do you mean by ‘two groups’ here? Do you mean that the final sample size of the ‘reference group’ is twice that of the ‘test group’ (N_R1=N_R2=N_T; N_R=2×N_T)? Your design will be extremely imbalanced.
In the calculation of the confidence interval different sample sizes will be accounted for (see how the pooled variance is calculated here). If you increase the sample size of the reference, its mean will be more precise – but why would you not want to estimate the mean of the test with the same precision?

If you want to be really creative you can discover the field of ‘matched pairs’. This concept is sometimes (!) used in clinical trials to adjust for confounding covariates. It’s a reasonable assumption that plasma concentration correlates with body weight (or surface or …). You can pair subjects according to this covariate and run a ‘matched pairs t-test’.^* I have never seen any BE study exploring this method; I would expect regulatory acceptance to be close to nil.

---

• David HA and JL Gunnink
  The Paired t Test Under Artificial Pairing
  The American Statistician 51/1, 9–12 (1997)
  Abstract