Predicting Who Gets Readmitted Before It Happens

Social Determinants of Health · 30-Day Readmission Risk · Sai Manasa Adduru, PharmD, MPH

SDOH Analysis Risk Stratification Health Equity Patient Navigation AUC 0.642
~75%
Low Risk · 8.2% readmission
~20%
Medium Risk · 18.5% readmission
~5%
High Risk · 42.3% readmission
88.7%
Critical Tier · Top 5%
0.642
AUC · model discriminates high-risk patients significantly above chance
88.7%
Of critical-tier patients (top 5%) were actually readmitted within 30 days
10+
SDOH features engineered · housing, insurance, transport, food, social support
5%
Critical tier · the model's actionable target for care intervention

Readmission Rate by Risk Tier

The model creates a clear, actionable separation · the critical tier (top 5%) shows an 88.7% readmission rate, making targeted intervention both feasible and high-impact.

Top SDOH Features Driving Readmission Risk

Prior admissions and ED visits are the strongest predictors · but social factors (housing, transport, food) collectively account for over 40% of model signal.

Risk Score Distribution · All Patients

Most patients cluster at low risk · the distribution's long right tail represents the critical minority the model is designed to surface for care teams.

ROC Curve · Model Discrimination (AUC 0.642)

AUC 0.642 reflects a meaningful signal from SDOH data alone · without clinical vitals or lab values · demonstrating that social context predicts hospital return.