A data integration framework to improve the efficiency of survival analysis by leveraging multiple external studies with multifaceted heterogeneity.

In survival analysis, integrating multiple data sources holds significant potential to increase event counts, improve statistical efficiency, and enhance scientific discovery. However, challenges arise when external datasets differ in data forms, distributions, and covariate availability. We propose a novel learning framework that strengthens statistical inference of internal Cox regression coefficients through efficient data integration. Our framework is robust integration and accommodates multifaceted data heterogeneity through transformation, working models, penalized approach, and weighting that adaptively selects homogeneous components. The resulting estimators are consistent, asymptotically normal, more efficient than internal-only estimators, and robust to working model misspecification and data heterogeneity. Simulation studies confirm our framework's favorable properties. We applied our method to study dementia risk among black individuals in the Religious Orders Study and the Rush Memory and Aging Project, integrating information from external cohorts, including the Alzheimer's Disease Neuroimaging Initiative and the National Alzheimer's Coordinating Center. The analysis identified key risk factors with improved efficiency and demonstrated clinical relevance.