Objective: This study was to develop a machine learning model which can predict and diagnose CAD in patients complaining of chest pain based on a large real-world prospective registry database and computing power.
Method: A total of 10,177 subjects with typical or atypical chest pain who underwent a coronary angiography at the cardiovascular center of our University Hospital, South Korea between November 2004 and May 2014 were evaluated in this study. . The generation of the diagnostic prediction model for CAD used the classification application by technical support of MATLAB R2017a . The performance evaluation of the learning model generated by machine learning was evaluated by the area under the curve (AUC) of the receiver-operating characteristic (ROC) analysis.
Results: The diagnostic prediction model of CAD had been generated according to the user’s accessibility such as the general public or clinician (Model 1 - 4). The performance of the models has ranged from 0.78 to 0.96 by the AUC of ROC analysis. The prediction accuracy of the models ranged from 70.4% to 88.9%. The performance of the diagnostic prediction model of CAD by machine learning improved as the input information increased.
Conclusion: A diagnostic prediction model of CAD using the machine learning method and the registry database was developed. Further studies are needed to verify our results.