Machine learning integration of echocardiographic and clinical data to improve prediction of survival following myocardial infarction.

AIMS

Machine learning (ML) could improve risk stratification following myocardial infarction (MI). However, previous ML studies for risk prediction following MI did not incorporate comprehensive echocardiographic data. This study sought to use machine learning (ML) to integrate comprehensive echocardiographic and clinical data for the predicting all-cause mortality following MI.

METHODS AND RESULTS

Retrospective study of consecutive patients admitted with MI to a tertiary referral hospital, with echocardiography performed within 24 h of admission. The cohort was randomly split into training (70%) and test (30%) sets. Two ML models (XGBoost and a neural network) were developed using echocardiographic and clinical data, and then compared with conventional logistic regression. The Shapley Additive exPlanations method was used for ML model interpretation. In the final study population of 3202 patients (mean age, 63.2 ± 12.5 years; 29.2% females), ST-elevation MI was present in 28.8%, and the mean cohort LVEF was 52.5 ± 11.2%. At a median follow-up of 4.5 years, there were 465 deaths. In the test set, XGBoost achieved the highest performance (AUC, 0.854), compared with logistic regression (AUC, 0.824; = 0.006) and the neural network (AUC, 0.808; = <0.001) for the prediction of death. In the XGBoost model, the highest-ranked predictors included age, renal function, echocardiographic left ventricular outflow velocity time integral, and diastolic parameters. Further, in nested ML models, the addition of echocardiographic parameters provided incremental value beyond clinical variables alone (AUC, 0.854 vs. 0.820; = 0.002).

CONCLUSION

ML integration of comprehensive echocardiographic data with clinical data could lead to improved prediction of survival following MI. Clinical implementation should be considered.