Integrating social determinants of health and genetic risk in disease risk models

Complex diseases are shaped by heritable factors and non-genetic environmental, behavioral, and social determinants of health, but these are rarely modeled together. The growing availability of large-scale, multimodal biobanks creates new opportunities to integrate diverse data types into more accurate disease risk models. Here, we apply multiple correspondence analysis (MCA) to over 100 environmental, behavioral, and social variables from the All of Us biobank (n = 413,457 individuals) to generate low-dimensional embeddings that quantify non-genetic risk for six common chronic conditions: asthma, chronic kidney disease, coronary heart disease, hypercholesterolemia, prostate cancer, and breast cancer. These embeddings recovered known risk factors such as economic status and smoking but also pointed to others such as loneliness and spirituality. Including MCA axes in addition to demographics and polygenic scores (PGSs) consistently improved disease-risk prediction, with improvement in area under the receiver operating characteristic curve (ΔROC-AUC) ranging from 0.007 to 0.027. For four of six diseases, the gains in model predictive power from MCA embeddings surpassed those attributable to PGS. Genetic and non-genetic risks combined additively, with little evidence of interaction effects (ΔROC-AUC ≤ 0.001) and highly stable variant effect sizes when embeddings were included in genetic association models (r > 0.98). In summary, we introduce a scalable, interpretable framework that summarizes survey-based environmental, behavioral, and social factors without prior assumptions about disease-specific variables. Our results demonstrate that these non-genetic contexts improve prediction but show limited evidence of interaction with genome-wide polygenic disease risk. Our results underscore the importance of incorporating social, behavioral, and environmental factors into clinical models of disease risk.