Interpreting the totality of clinical study data: is there a better way? Statistical methodology for quantitatively interpreting clinical study data across efficacy endpoints.

INTRODUCTION: Interpretation of the efficacy of an intervention in clinical trials often relies on whether the p-value for the primary endpoint is <0.05, with less emphasis on the pattern of results across efficacy endpoints. When results across endpoints are consistent and strongly supportive (or unsupportive) interpretation is straightforward. When results vary among endpoints, p-values are nominally statistically significant or slightly greater than 0.05, the standardized effect size is low, or there are divergent results across studies, interpretation is more complicated and relies on individual interpretation, which can introduce bias. Statistical approaches are needed to help interpret the strength of the totality of efficacy results (within a study and across studies). METHODS: We investigated the properties of a non-parametric approach to testing individual endpoints and a method of combining the individual endpoints into a single test of the overall shift that was then compared to parametric approaches using results from a recent Phase 3 analgesic clinical trial (CLEAR-1 Trial that assessed the efficacy of SP-102 for the treatment of lumbosacral radicular pain). RESULTS: Results from these applied methods allowed us to (1) confirm that the observed statistical significance of the primary endpoint in the CLEAR-1 Trial was not driven by pre-defined model assumptions, (2) that a strongly statistically significant result was observed across all endpoints, and (3) that the methods do not simply inflate alpha. CONCLUSION: Statistical approaches such as these are needed to help interpret the strength of the totality of efficacy results quantitatively to support more consistent interpretation of results.