The replication of non-inferiority and equivalence studies
Replication studies are increasingly conducted to assess credibility of scientific findings. Most of these replication attempts target studies with a superiority design, but there is a lack of methodology regarding the analysis of replication studies with alternative types of designs. In order to fill this gap, we adapt three approaches used for superiority settings to non-inferiority and equivalence designs: the two-trials rule, the sceptical p-value approach and the meta-analysis criterion. While the adaptation to the non-inferiority design is relatively straightforward, the equivalence design requires more effort. We propose to use the 'two one-sided test' (TOST) procedure and tailor it to the replication setting. In addition, we formulate the different criteria in terms of a 'success interval' for the relative effect size (replication to original). The properties of the different approaches are studied in detail. In particular, the two-trials rule does not penalize replication effect estimates very close to the margin for large sample sizes. The meta-analysis criterion is convenient as it combines the two estimates into one, but can lead to replication success in odd situations. Finally, the sceptical p-value approach seems well-suited as it penalizes a too large increase of the replication effect estimate as compared to the original one, while taking into account the direction of the latter in the calculation of the success interval.
READ FULL TEXT