READER REACTION Bioequivalence TestiIlg-A Need to Rethink

The symmetric confidence interval method for bioequivalence testing, proposed by Westlake (1976, 1979), is founded on a premise which is highly questionable, and in his description of the method, Westlake confuses two distinct statistical issues. The purpose of this note is to clarify the apparent flaws in Westlake's method and to point out a formal similarity between testing drugs for bioequivalence and checking that their potencies conform to specified levels. It is suggested that the adoption of a common statistical approach to the two problems may be advantageous. A pair of drugs or, more commonly, two alternative formulations of the same drug are said to be 'bioequivalent' when equal amounts of them produce equal therapeutic effects. In place of the extensive clinical trials that would be needed to investigate equality of therapeutic effect directly, decisions on bioequivalence are usually made by comparing univariate biological responses (e.g. area under drug blood-level curve) after administration of supposedly equivalent single doses of the drugs (for a fuller account see Westlake, 1979). Such a test is known as a 'comparative bioavailability trial', and discussion here is confined to this simple case. Because of experimental error and intrinsic biological variability, true bioequivalence can never be demonstrated exactly. Nor is it meaningful simply to conduct a conventional test of the null hypothesis that the drugs are bioequivalent. As pointed out in this context by Westlake (1972, 1979), a difference which is statistically significant may, nevertheless, be trivially small, while lack of significance may merely be the result of poor reproducibility. A more useful approach is to require that the confidence interval for the mean difference 6 between the responses to the drugs is completely contained within some defined range of tolerance about zero (see Westlake, 1972; Metzler, 1974). In the usual statistical approach a confidence interval for 6 would be centred dn the sample mean difference, b. However, Westlake (1976) proposed a modification to this method, which involves calculating a 'confidence interval' that is constrained to be symmetrical about zero. He claimed that the modification would have the dual advantages of (i) decreasing the 'effective' length of the confidence interval, and (ii) increasing the confidence coefficient. However, Westlake's main argument for the adoption of this approach seems to be based on a misconception. Furthermore, in a later part of his paper, Westlake switched his symmetric-interval method from the context in which it was first developed, namely