|
Abstract:
|
In evidence synthesis from different electronic health records (EHRs) datasets, the feature of privacy-preserving is important as patient-level data are often protected against sharing across clinical sites. Conventional meta-analysis might suffer from substantial bias when studying rare conditions due to the limited number of events in single clinical sites. In this talk, I will introduce distributed algorithms for commonly used regression models to study binary and time-to-event outcomes. With the help of patient-level data from a single site, and aggregated information from other sites, the proposed distributed algorithms avoid sharing patient-level information across sites, and provide estimates on par with the estimates obtained by analyzing, in a single dataset, the combined patient-level data from all sites. The algorithms are also communication-efficient in the sense that they require no iterative communication across clinical sites. Through real data applications to large research networks such as the Observational Health Data Sciences and Informatics (OHDSI) consortium, our algorithms were shown to outperform the meta-analysis estimator for studying rare conditions.
|
