National validation of an artificial intelligence algorithm to predict risk of acute coronary syndrome - ANGINA II

Abstract

Introduction: Previous work by the team demonstrated an excellent ability of an artificial intelligence tool to predict the risk of suffering an acute coronary syndrome (ACS) in patients who consulted the emergency department (ED) for thoracic pain, with an area under the ROC curve of 0.8991. On this occasion, the group validated the Cardio TrIAge machine learning algorithm for use in ED. Methods: The team designed a prospective, observational, multicenter study that analyzed data corresponding to 165 patients from 3 centers in different regions of Argentina who were admitted to the Cardio TrIAge platform by physicians from the emergency department. The Random Forest algorithm predicted the risk of suffering an acute coronary syndrome (ACS) 30 days after medical evaluation. Follow-up was carried out by telephone and through clinical history data to evaluate the presence of events. Results: The Random Forest Classifier presented an area under the ROC curve (AUC) of 0.961 (CI: 0.919 – 0.985, P < 0.001) considering the probability of certainty in the prediction made by the algorithm, the sensitivity of 87.01% and specificity of 89.77%, which generates a negative predictive value of 91.56% and a positive predictive value of 82.92%. Conclusion: Machine learning classifiers are a useful tool for predicting the risk of acute coronary syndrome during a 30-day follow-up period in different populations that consult ED in Argentina, with a high predictive capacity.