Development of machine learning models for predicting 6-month readmission risk in patients with comorbid COPD and heart failure

Objectives This study aimed to develop and validate machine learning models for predicting 6-month readmission risk in hospitalized patients with comorbid COPD and heart failure. Methods This study utilized data from the MIMIC-IV database (version 3.1). Adult inpatients with concurrent diagnoses of COPD and HF were included. Seven ML algorithms were employed to develop predictive models for 6-month readmission risk. Model overall performance was evaluated using the area under the receiver operating characteristic curve (ROC), decision curve analysis (DCA), precision-recall curves (PRC), and calibration curves. SHAP (SHapley Additive exPlanations) analysis was applied to interpret the top-performing models. Results A total of 1,291 patients were included in the analysis. The Random Forest (RF) and Light Gradient Boosting Machine (LightGBM) models respectively demonstrated the best predictive performance, with AUCs of 0.756 and 0.733 on the test set. SHAP analysis revealed that NT-proBNP, pH, neutrophil count, Troponin-T, and comorbid diabetes were the top five features contributing most to the models’ predictions. Interpretation The RF and LightGBM models developed in this study can effectively identify patients with COPD and HF who are at high risk for readmission within six months. These models, coupled with the SHAP-based interpretability framework, could aid clinicians in risk stratification and guide targeted interventions for this high-risk population.