Toward Better Risk Stratification for Implantable Cardioverter-Defibrillator Recipients: Implications of Explainable Machine Learning Models

Background: Current guideline-based implantable cardioverter-defibrillator (ICD) implants fail to meet the demands for precision medicine. Machine learning (ML) designed for survival analysis might facilitate personalized risk stratification. We aimed to develop explainable ML models predicting mortality and the first appropriate shock and compare these to standard Cox proportional hazards (CPH) regression in ICD recipients. Methods and Results: Forty-five routine clinical variables were collected. Four fine-tuned ML approaches (elastic net Cox regression, random survival forests, survival support vector machine, and XGBoost) were applied and compared with the CPH model on the test set using Harrell’s C-index. Of 887 adult patients enrolled, 199 patients died (5.0 per 100 person-years) and 265 first appropriate shocks occurred (12.4 per 100 person-years) during the follow-up. Patients were randomly split into training (75%) and test (25%) sets. Among ML models predicting death, XGBoost achieved the highest accuracy and outperformed the CPH model (C-index: 0.794 vs. 0.760, p < 0.001). For appropriate shock, survival support vector machine showed the highest accuracy, although not statistically different from the CPH model (0.621 vs. 0.611, p = 0.243). The feature contribution of ML models assessed by SHAP values at individual and overall levels was in accordance with established knowledge. Accordingly, a bi-dimensional risk matrix integrating death and shock risk was built. This risk stratification framework further classified patients with different likelihoods of benefiting from ICD implant. Conclusions: Explainable ML models offer a promising tool to identify different risk scenarios in ICD-eligible patients and aid clinical decision making. Further evaluation is needed.