Purpose: To review the methods and guidelines on meta-analysis of rare events, and to propose practical guidance to select appropriate methods.
Meta-analysis involves the statistical synthesis of results from two or more studies, to arrive at a single estimate often using clinical data. By combining several (clinical) studies while maintaining randomization, meta-analyses increase the power of the estimation and improve precision. Many guidelines exist on how to conduct meta-analyses and correct for any bias. Most guidelines also address the issue of how to deal with sparse data, especially in the case of rare events. However, none of these guidelines provide practical guidance in terms of precisely what to do and when. At this workshop, we will present the different methods for statistical pooling, and the methods suggested in diverse guidelines on how to deal with rare events such as imputation, transformation, or the choice of specific models. We will also present evidence from the published literature that is helping to shape which method might lead to the least biased estimate and under which circumstances. We will also propose a practical guidance using examples from the literature. This guidance will offer straightforward advice and will propose a decision-aid, accounting for many issues, such as the reason for the rarity of data, the type of outcome considered the programming abilities, and the main objective of the meta-analysis. The audience will be encouraged to share their own experience with sparse data to improve on the suggested guidance.