A Machine Learning Algorithm for Risk Prediction of Acute Coronary Syndrome

pp. 9-13

Abstract

Background: Chest pain represents one of the most common reasons for consultation in emergency medical services (EMS). A diagnostic strategy using objective and subjective information about the characteristics of chest pain has not been identified yet.

Objective: The aim of this study was to evaluate the performance of a machine learning classifier to predict the risk for non-STsegment elevation acute coronary syndrome (ACS) in patients consulting an EMS due to chest pain.

Methods: A total of 161 patients consulting the EMS due to chest pain were analyzed. Both objective and subjective variables about the characteristics of chest pain were recorded using a machine learning classifier.

Results: Mean age was 57.43±12 years, 75% were men and 16% had history of cardiovascular disease. Acute coronary syndrome was present in 57.8% of cases with an incidence of acute myocardial infarction of 29.8%. Among the latter 35% required percutaneous coronary intervention and 9.9% myocardial revascularization surgery during the 30-day follow-up. A Random Forest Classifier was used as model of classification, with an area under the ROC curve of 0.8991, sensitivity of 0.8552, specificity of 0.8588 and accuracy of 0.8441. The most significant predictors in the model were weight (p = 0.002), age (p = 5.011e-07), pain intensity (p = 3.0679e-05), systolic blood pressure (p = 0.6068) and the subjective characteristics of pain (p = 1.590 e-04).

Conclusions: Machine learning classifiers are a useful and effective tool to predict an acute coronary syndrome at 30-day follow-up.