Marrying Generative Model of Healthcare Events with Digital Twin of Social Determinants of Health for Disease Reasoning

Researchers develop a generative AI model that integrates social determinants of health (SDoH) with multi-organ sensor data and medical events to improve disease prediction and personalized clinical decision support. Tested on UK Biobank data spanning nearly 500,000 medical histories, the model outperforms existing autoregressive disease prediction systems by explicitly modeling socioeconomic factors alongside imaging and biomarker data.

This research addresses a critical gap in computational healthcare by incorporating social determinants of health into disease prediction models. Traditional generative models for disease rely heavily on clinical events and biomarkers while ignoring the socioeconomic and environmental factors that substantially influence health outcomes. The authors propose a novel framework combining geometric diffusion models for complex data types—such as brain network connectivity represented as graphs—with standard diffusion models for tabular organ-system data.

The significance lies in the model's architecture, which treats SDoH not as auxiliary information but as a core component of disease reasoning. By using ICD-coded proxies for socioeconomic factors and conditioning a latent diffusion framework on multi-organ imaging data, the model enables more granular personalized medicine. The validation on UK Biobank's extensive dataset—encompassing brain, heart, liver, and kidney imaging from tens of thousands of participants paired with 500,000 medical histories—demonstrates clinical-scale viability.

For healthcare AI development, this work establishes a template for integrating traditionally siloed data sources into unified predictive models. The ability to simulate interventions on future disease trajectories creates opportunities for preventive medicine and healthcare resource allocation. For digital health platforms and clinical decision support systems, such models could enable more equitable predictions by explicitly accounting for socioeconomic disparities in health outcomes.

Future research should focus on validating these predictions in prospective clinical settings and ensuring model interpretability for clinician adoption. The framework's scalability to additional health determinants and validation across diverse populations remains an open challenge.

• →Novel generative model integrates social determinants of health with multi-organ imaging data for improved disease prediction
• →Geometric diffusion models enable explicit representation of complex graph-based data like brain connectivity networks
• →Validation on UK Biobank demonstrates improvements over existing autoregressive disease models at scale
• →Framework enables simulated interventions to reason about future disease trajectories and personalized treatment pathways
• →Addresses healthcare AI's historical blindspot by treating socioeconomic factors as core to disease modeling rather than auxiliary variables