A machine learning model to predict the in-hospital mortality risk of elderly patients with acute exacerbation of chronic heart failure

Objective This study aimed to construct a machine learning model to predict the in-hospital mortality risk of elderly patients with chronic heart failure who experienced acute exacerbation. Materials and methods A primary complete-case cohort of 650 patients was retrospectively included for model development and evaluation. The primary outcome was all-cause index-hospitalization mortality within 28 days after admission. All patients were randomly divided into a training set and a test set in a ratio of 7:3. Lasso regression was used to select variables in the training set. Building multiple machine learning models and using AUC, calibration curves, and decision curve analysis in both the training set and the test set to complete the performance evaluation of the models. SHAP analysis and LIME explanations were used to analyze the contributions of the main features. Results The Lasso regression identified 15 important variables. Nine machine learning models were constructed and compared using the selected variables. Among them, LightGBM achieved the highest AUC in both the training set and the test set: the AUC in the training set was 0.95 (95% CI: 0.92–0.97), and the AUC in the test set was 0.93 (95% CI: 0.88–0.98). The calibration curve of LightGBM indicated a good consistency between the predicted probabilities and the actual outcomes. The decision curve analysis of LightGBM demonstrated its potential value in clinical decision support. SHAP and LIME analyses identified LVEF, CRP, SBP, NT-proBNP, and creatinine as influential contributors to model predictions. Conclusions This study developed an interpretable machine-learning model for predicting 28-day index-hospitalization mortality in older patients with acute exacerbation of chronic heart failure. The model showed promising performance in internal validation; however, independent multicenter external validation is required before clinical implementation.