Developing a risk assessment system for heart failure during follow-up in acute myocardial infarction patients in combination with cardiopulmonary exercise testing

Background Heart failure (HF) significantly impacts the prognosis of acute myocardial infarction (AMI). However, current risk prediction tools remain limited. Purpose Developed and validated an HF risk prediction model for AMI patients using an ensemble learning and feature selection framework combined with cardiopulmonary exercise testing data and created a risk assessment system. Methods This study included 1738 AMI patients who underwent cardiopulmonary exercise testing between July 2018 and September 2021 (average follow-up=30.9 (22.8) months) from our medical university hospital. The primary outcomes were the occurrence of HFrEF and HFmrEF during follow-up. The performance of the constructed machine learning (ML) model was evaluated by calculating the area under the receiver operating characteristic curve (AUC). Results After applying inclusion and exclusion criteria, 760 patients were selected (median [IQR] age=53.0 [46.0, 60.0] years; 686 [90.3%] males), of whom 62 (8.2%) developed HFrEF or HFmrEF during follow-up. The optimal model constructed through the ensemble learning and feature selection framework was extreme gradient boosting, which achieved the following performance metrics: AUC=0.95 (95% CI=0.90–1.00), accuracy=0.89 (95% CI=0.85–0.93), sensitivity=0.91, and specificity=0.89. The risk assessment system built on this model can effectively differentiate patients at different risk levels, with significant HF risk differences between the low-, medium-, and high-risk groups (1.1% vs. 30.8% vs. 73.3%, P<0.001). Conclusions In this prognostic study, we developed an ML model that can accurately predict HF risk in patients with AMI and constructed a risk assessment system that may help guide early clinical decision-making to reduce the incidence of HF during follow-up.AUC of each model in the test set. User-friendly Web-Based Risk Calculator.