Abstract
Return to work is a critical indicator of recovery after acute myocardial infarction (AMI), and accurate identification of patients with low return-to-work rates is critical for timely intervention. The aim of this study was to develop a machine learning (ML) model for predicting the return to work in AMI patients. A retrospective study of data from 1473 patients was conducted using the Incidence Rate of Heart Failure After Acute Myocardial Infarction With Optimal Treatment database. Patients were randomly divided into a training cohort and a validation cohort (7:3). A total of five ML models were developed based on the training cohort to predict return to work. Model performance was assessed according to numerous learning metrics, including the area under the receiver operating characteristic curve (AUC), accuracy, F1-score, and Brier score. The ML models were constructed using 12 features, including age, occupation, income, anterior wall AMI, hypertension, fasting plasma glucose (FPG), beta-blockers, marriage, aspartate transaminase (AST), body mass index (BMI), TG (triglyceride) and phase II cardiac rehabilitation (CR). Among the five ML models, the LR model achieved the best performance, with an AUC of 0.793 (95% CI, 0.712-0.874), an accuracy of 0.719 (95% CI, 0.642-0.787), an F1 score of 0.800, and a Brier score of 0.135, and was subsequently transformed into a nomogram. A new return-to-work prediction model was developed based on a machine learning algorithm, which may help identify patients with low return-to-work rates and may become an effective management tool for AMI patients.
Clinical trial registration: Clinical Trials.gov ID: NCT03297164.