Re-using Mini-Sentinel data following rapid assessments of potential safety signals via modular analytic programs

The U.S. Food and Drug Administration (FDA)'s Mini-Sentinel pilot has created a distributed data system with over 125 million lives and nearly 350 million person-years of observation time. The pilot allows the FDA to use modular analytic programs to assess suspected safety signals quickly. The FDA convened a committee to assess the implications of such rapid assessments on subsequent analyses of the same product-outcome pair using the same data. The committee offers several non-binding recommendations based on the strength of the knowledge of the suspected association before running the analysis: signal generation (an analysis with no prior), signal refinement (an analysis with a weak or moderate prior), and signal evaluation (an analysis with a strong prior). The committee believes that modular programs (MPs) are most useful for signal refinement. If MPs are used for analyses with no or weak/moderate priors, the committee members generally agree that the data may be re-used if certain conditions are met. When there is a strong prior, the committee recommends that a protocol-based assessment be used; Mini-Sentinel data may be analyzed by MPs and re-used only under very uncommon circumstances. The committee agrees that any subsequent assessment of the same product-outcome pair that follows an MP analysis should not be interpreted as independent confirmation of the association, such as would be established via replication of the same product-outcome association in two different populations. Instead, the follow-up assessment should be interpreted as an analysis that has reduced insofar as possible systematic errors that may have been present or residual in the original MP analysis. The committee also discussed how this general framework may apply to two completed rapid assessments of dabigatran and bleeding risk and of olmesartan and celiac disease risk.