Novel Screening Methods for Binary Outcomes with Rare Events

Background: Epigenome-wide association studies often involve high-dimensional DNA methylation data. Efficient methods are needed to screen CpG sites associated with binary outcomes, particularly when the event rate is below 5%. Existing screening methods, such as ttScreening, are effective for filtering out unimportant variables, such as CpG sites, in epigenome-wide studies when event rates are moderate. However, ttScreening and many other screening approaches face challenges in handling rare events. Therefore, we proposed two novel methods, Rare-Screening and Firth-ttScreening , designed specifically for binary outcomes with rare events. Results: We evaluated the performance of the proposed approaches and benchmark methods, including FDR-based and Bonferroni approaches for multiple testing correction, through extensive simulations across varying sample sizes and numbers of independent variables. Conclusions: The Rare-Screening and Firth-ttScreening methods outperformed FDR and Bonferroni methods with higher sensitivities. The proposed functions have been incorporated into the ttScreening R package, which is available through the Comprehensive R Archive Network (CRAN).