Validation of an Artificial Intelligence Algorithm for Diagnostic Prediction of Coronary Disease: Comparison with a Traditional Statistical Model

Multivariate prognostic analysis has been traditionally performed by regression models. However, many algorithms capable of translating an infinity of patterns into probabilities have emerged. The comparative accuracy of artificial intelligence and traditional statistical models has not been established in the medical field. To test the artificial intelligence as an accurate algorithm for predicting coronary disease in the scenario of acute chest pain and evaluate whether its performance is superior to traditional statistical model. A consecutive sample of 962 patients admitted with chest pain was analyzed. Two probabilistic models of coronary disease were built using the first two-thirds of patients: a machine learning algorithm and a traditional logistic model. The performance of these two predictive strategies were evaluated in the remaining third of patients. The final logistic regression model had significant variables only, at the 5% significance level. The training sample had an average age of 59 ± 15 years, 58% males, and a 52% prevalence of coronary disease. The logistic model was composed of nine independent predictors. The machine learning algorithm was composed of all candidates for predictors. In the test sample, the area under the ROC curve for prediction of coronary disease was 0.81 (95% CI = 0.77 - 0.86) for the machine learning algorithm, similar to that obtained in logistic model (0.82; 95% CI = 0.77 - 0.87), p = 0.68. The present study suggests that an accurate machine learning prediction tool did not prove to be superior to the statistical model of logistic regression.