Reconsidering some aspects of the two-trials paradigm

A common standard for the demonstration of efficacy in a clinical submission is a statistically significant outcome in at least two pivotal trials ("two-trials convention"). When the data structures in different trials are sufficiently similar to allow pooling of the data across trials for a combined analysis, we argue here that such an analysis is a more logical and efficient basis for a judgment regarding efficacy. Criteria for combined analyses may be established, which ensure the same false positive rate protection as the two-trials convention. A combined analysis will generally have much more power than the corresponding application of the two-trials approach that has the same false positive rate protection. In addition, we describe the behavior of modified versions of pure combined analysis, which incorporate a formal standard for reproducibility of trial results by limiting the larger of the individual trial p-values. These modifications are shown to maintain the desirable behavior of the pure combined analysis, namely, higher power compared to the two-trials convention.