E. Niclas Jonsson1,2 and Lewis B. Sheiner2
1Department of Pharmacy, Uppsala University, Uppsala, Sweden; 2Department of Laboratory Medicine, School of Medicine, UCSF, San Francisco,CA
To approve a new drug for marketing, regulatory authorities generally require “substantial evidence” that it is effective relative to some reference treatment, preferably placebo. This requirement has traditionally been met by finding that the null hypothesis of no treatment difference; i.e., no drug effect (H0), can be rejected in two separate placebo-controlled clinical trials, each at p<0.05.
The sponsors of the drug in question, had indeed performed two double-blind, placebo controlled Phase 3 clinical trials including approximately 500 patients. The problem was that the first study (Study A) yielded a p-value of 0.031, but the second (Study B) yielded only p= 0.13.
Recently, regulatory authorities have been formally encouraged to consider “data from one adequate and well-controlled clinical investigation and confirmatory evidence” as constituting “substantial evidence.” While Study A might have provided sufficient evidence for approval if Study B had not been conducted, the fact is that Study B failed to reach statistical significance, and, in the present case, the sponsors deemed it infeasible to perform a third tie-breaking study. Instead it was agreed, with the regulatory authorities, that it would be possible to regard a significant concentration-response (CR) relationship in the active treatment arms of (one or both of) the two studies as corroborative evidence.
This situation poses two important questions: (1) Which standard and/or concentration response analyses should be considered as evidence of efficacy? (2) How can the total evidence provided by the various analyses be measured, and how does it contribute to an assessment of efficacy?
To address the above questions we suggest a strategy that uses two new (but not original with us) methodological tools. The first is statistical simulations, which is necessary to, e.g., obtain the reference distributions to which the statistical significance criteria in the CR analysis (the likelihood ratio) is to be compared. The second is the Bayes factor. The Bayes factor is an alternative to the p-value as a measure of statistical evidence. It has the advantage that it considers the evidence both for and against H0, in contrast to the p-value, which only considers evidence against H0.
Reference: PAGE 8 (1999) Abstr 133 [www.page-meeting.org/?abstract=133]
Poster: oral presentation