Primary Submission Category: Generalizability/Transportability
Toward Personalized and Sample-bounded Meta-analyses
Authors: Wenqi Shi, José Zubizarreta,
Presenting Author: Wenqi Shi*
Meta-analysis is a powerful tool for synthesizing evidence across multiple studies, often yielding more precise estimates than those of individual studies. However, traditional methodologies often lack the flexibility to tailor inferences to a well-defined target population and do not ensure that the final estimates remain sample-bounded. This paper presents a novel framework for personalized and sample-bounded meta-analyses, which extends traditional approaches by explicitly incorporating target covariate profiles and enforcing sample-boundedness, thereby enhancing both personalization and reliability. This framework includes variance estimation, allows for partial overlap between study and target populations, and supports both individual patient data (IPD) and aggregate-level data. Through personalization, this method mitigates selection bias and ensures that effect estimates apply to the intended population. By explicitly identifying study contributions and enforcing sample-boundedness, this method selects reliable study donors under appropriate assumptions, improves robustness to overlap violations, enhances transparency, and prevents extrapolation bias. We establish theoretical properties of this approach, including multiple consistency conditions and asymptotic normality, and provide diagnostics for covariate balance and study selection. Simulations demonstrate the robustness of this method to overlap violations and a case study highlights its practical application.