A Scalable Framework to Integrate Social Determinants of Health into Disease Risk Models using Biobank Survey Data

Area of Science:

• Genomics and Epidemiology
• Computational Biology
• Public Health

Background:

• Complex diseases pose a significant global health challenge, with limited ability to predict individual risk.
• Disease risk is influenced by a combination of genetic and non-genetic factors (environmental, behavioral, social), which are seldom analyzed together.
• Large-scale biobanks offer opportunities to integrate diverse data for improved disease risk prediction.

Purpose of the Study:

• To develop and validate a scalable framework using Multiple Correspondence Analysis (MCA) to model non-genetic determinants of complex diseases.
• To assess the contribution of non-genetic factors, summarized by MCA embeddings, to disease risk prediction.
• To investigate the interplay between genetic and non-genetic risk factors in complex diseases.

Main Methods:

• Applied Multiple Correspondence Analysis (MCA) to over 100 environmental, behavioral, and social variables from the All of Us biobank (N=171,614).
• Generated low-dimensional embeddings to quantify non-genetic risk for six common chronic conditions.
• Integrated MCA embeddings with demographic data and polygenic scores (PGS) to evaluate prediction performance (ROC-AUC).

Main Results:

• MCA embeddings identified known and novel risk factors for chronic diseases.
• Non-genetic risk embeddings consistently improved prediction accuracy beyond demographics and PGS (ROC-AUC increase: 0.03-0.05).
• For five of six diseases, MCA embeddings outperformed PGS in prediction improvement; genetic and non-genetic risks largely acted additively.

Conclusions:

• Introduced a scalable, interpretable MCA framework for summarizing complex survey-based non-genetic factors.
• Demonstrated that non-genetic factors substantially enhance complex disease risk prediction, acting largely additively with genetic risk.
• Clarified gene-environment interplay, supporting more equitable and robust risk modeling for diverse populations.