Developing A Fair Individualized Polysocial Risk Score (iPsRS) for Identifying Increased Social Risk of Hospitalizations in Patients with Type 2 Diabetes (T2D)

The tool successfully identified high-risk patients while maintaining fairness across racial and ethnic groups: The iPsRS achieved a C statistic of 0.72 in predicting 1-year hospitalizations after fairness optimization, demonstrating that it's possible to build predictive models that work equitably across different populations rather than perpetuating existing healthcare disparities.

Social determinants of health are powerful predictors of hospitalization risk: The model incorporated both neighborhood-level factors (like neighborhood deprivation) and individual-level social factors (like housing stability), suggesting that addressing social needs—not just medical symptoms—is crucial for preventing hospitalizations in diabetes patients.

The tool has dramatic real-world utility for identifying vulnerable patients: Patients in the top 5% of risk scores had approximately 13 times higher hospitalization rates than those in the bottom decile, indicating the iPsRS could help healthcare systems efficiently target interventions to those who need them most.

Explainability and fairness assessment were built into the development process, not added afterward: By incorporating explainable AI (XAI) techniques and fairness optimization from the start, the researchers demonstrated a responsible approach to AI development that prioritizes transparency and equity rather than treating them as afterthoughts.

EHR data can capture social risk if systems are designed to do so: The study shows that existing electronic health records contain valuable information about patients' social circumstances that can be leveraged to identify unmet social needs at the point of care.