Precision medicine research has shown great promise to improve individual health and our understanding about how a person's genetics, environment, and lifestyle can help determine the best approach to prevent or treat disease. However, one serious challenge we are facing is the limited representation of minorities and disadvantaged populations in precision medicine research. Many studies have shown populations with low socioeconomic status as well as racial and ethnic minority groups tend to be underrepresented in biomedical research or biorepositories. To advance prediction medicine, it is crucial to improve the performance of statistical and machine learning models in underrepresented populations so as not to exacerbate health disparities.
In this paper, we address the lack of representation and disparities in model performance through two strategies: (1) leverage the shared knowledge from diverse populations, and (2) integrate larger bodies of data from multiple healthcare organizations. More specifically, we develop transfer learning strategies to transfer the shared knowledge learned from diverse populations to an underrepresented population, so that comparable model performance can be reached with much less data. On the other hand, we propose federated learning methods to increase the sample sizes of underrepresented populations and the diversity of the data through multi-center collaborative research via a safe and efficient way. Our methods have solid theoretical foundations, and we demonstrate the feasibility and validity of our methods through numerical experiments and a real application to a multi-center study for constructing polygenic risk prediction models for Type II Diabetes.