Machine learning algorithms to predict heart failure with preserved ejection fraction among patients with premature myocardial infarction.

Background: Heart Failure with Preserved Ejection Fraction (HFpEF) in patients with Premature Myocardial Infarction (PMI) is a crucial factor affecting long-term prognosis. This study aims to develop a model based on a machine learning algorithm that can predict the risk of in-hospital HFpEF in patients with PMI early and quickly. Methods: This prospective study consecutively included PMI patients from January 2017 to December 2022. Lasso-Logistic, XGBoost, Random Forest, K-Nearest Neighbor, and Support Vector Machine models were constructed. The prediction performance of the models was compared through AUC, Accuracy, Precision, F1 score, and Brier score. Shapley Additive exPlanations is used to explain the model. A prediction system was developed to identify high-risk patients. Results: The study finally included 840 PMI patients. 268 (31.90%) developed in-hospital HFpEF. The XGBoost model has the best prediction performance (AUC 0.854; Accuracy 0.798; Precision 0.686; F1 score 0.586; Brier score 0.143). The final model included ten variables, which were Brain natriuretic peptide (BNP) > 100pg/ml, SYNTAX Score > 14.5, Age, Monocyte to Lymphocyte Ratio (MLR) > 0.3, Hematocrit (HCT) < 45%, Heart rate (HR) > 75 bpm, Body Mass Index (BMI) ≥ 24 kg/m Conclusions: The explainable prediction model established based on the XGBoost algorithm can accurately predict the risk of in-hospital HFpEF in PMI patients and is available at https://hfpefpmi.shinyapps.io/apppredict/. This system is expected to assist clinicians in decision-making by providing timely, prioritized, and precise interventions for PMI patients, ultimately reducing the incidence of HFpEF and improving long-term prognosis.