Cross-Study Learning for Generalist and Specialist Predictions

• Activity Number: 268 - Replicability and the Narrative of Scientific Research
• Type: Invited
• Date/Time: Wednesday, August 11, 2021 : 1:30 PM to 3:20 PM
• Sponsor: American Association for the Advancement of Science
• Author(s): Boyu Ren* and Prasad Patil and Francesca Dominici and Giovanni Parmigiani and Lorenzo Trippa
• Companies: McLean Hospital and Boston University School of Public Health and Harvard University and Harvard University and Harvard University
• Keywords: statistical replicability and reproducibility; hierarchical model; environmental health

Abstract

The integration and use of data from multiple studies, for the development of prediction models is an important task in several scientific fields. We propose a framework for generalist and specialist predictions that leverages multiple datasets, with potential differences in the relationships between predictors and outcomes. Our framework uses stacking, and it includes three major components: 1) training of the ensemble members using one or more datasets, 2) task-specific utility functions and 3) a no-data-reuse technique for estimating stacking weights. We illustrate that under mild regularity conditions the framework produces stacked prediction function with oracle properties. In particular we show that the stacking weights are nearly optimal. We also provide sufficient conditions under which the proposed no-data-reuse technique increases prediction accuracy compared to stacking with data reuse. We perform a simulation study to illustrate these results. We apply our framework to predict mortality using a collection of datasets on long-term exposure to air pollutants.