An Interpretable XGBoost‑Based Mortality Prediction Model for ICU Heart Failure Patients: Leveraging the eICU Database and SHAP Analysis

Abstract

Artificial intelligence, particularly deep learning, is increasingly deployed across medical domains, driven by big data labeling, enhanced computing power, and cloud storage. While AI promises to address persistent healthcare challenges—including diagnostic errors, treatment inaccuracies, resource waste, workflow inefficiencies, inequities, and limited patient‑clinician interaction—its application in critical care remains under rigorous investigation. This study developed an interpretable mortality prediction model for intensive care unit (ICU) patients with heart failure, utilizing the freely accessible eICU Collaborative Research Database (eICU‑CRD). The extreme gradient boosting (XGBoost) algorithm was applied, with model interpretability enhanced via the SHapley Additive exPlanations (SHAP) method to identify and quantify key prognostic factors. The proposed framework offers a transparent, data‑driven tool for risk stratification and clinical decision support in heart failure management.