Min Jung Kim
UConn Health
Title: From Prediction to Explanation: Explainable Machine Learning for 30-Day Heart Failure Readmission
Abstract: Thirty-day readmission after heart failure hospitalization remains a major clinical and public health challenge. Traditional risk scores rely on a limited number of predictors and linear assumptions, while many machine learning approaches improve predictive performance but lack interpretability. In addition, most existing models focus only on patients with heart failure as the primary diagnosis, excluding a large portion of real-world patients in whom heart failure appears as a secondary diagnosis.
We conducted a retrospective cohort study using statewide health information exchange data from the Kansas Health Information Network. After resolving EHR record fragmentation and applying a one index admission per patient rule, the final analytic cohort included 2,734 patients with heart failure as either a primary or secondary diagnosis. A Random Forest model was trained using stratified sampling and cross validation. Model performance was evaluated using AUROC, AUPRC, Brier score, calibration, and decision curve analysis. Model interpretability was assessed using SHAP, with clustering used to identify risk phenotypes and counterfactual simulations used to evaluate modifiable risk factors.
The model achieved an AUROC of 0.75 with good calibration (0.16). Prior healthcare utilization, chronic kidney disease, and anemia were key drivers of risk, while good kidney disease management was protective. SHAP based clustering identified distinct patient phenotypes, and simulations suggested that improving behavioral and kidney related factors could reduce predicted readmission risk.
This work integrates prediction and interpretability to identify heterogeneous risk pathways and potential intervention targets, supporting more personalized and actionable strategies for reducing heart failure readmissions.
Bio: Minjung Kim, PhD, is a faculty member at the University of Connecticut School of Medicine, with appointments in the Family Medicine Department, and Calhoun Cardiology Center. Her research focuses on the application of statistical and machine learning methods to clinical and healthcare data, with an emphasis on predictive modeling, explainable artificial intelligence, and real-world data analysis. Her work aims to develop interpretable and actionable models to improve clinical decision making and patient outcomes, particularly in cardiovascular disease and medical education.
Date: Wednesday, April 29, 2026, 3:30 PM, AUST 434
Webex link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m5e85dd0874f87c7e92842a501383d535
Coffee will be available at 3:00 PM in the Noether Lounge (AUST 326)
For more information, contact: Yuwen Gu at yuwen.gu@uconn.edu