Prediction and Prognosis of Acute Myocardial Infarction in Patients with Previous Coronary Artery Bypass Grafting Using Neural Network Model (Preprint)

BACKGROUND Patients with previous coronary artery bypass grafting represent a substantial percentage of the total population of patients with acute myocardial infarction. Prognosis of the future disease expression is an important part in the follow-up of patients with previous CABG. It is well known that outcome of patients with previous CABG influenced with a lot of abnormalities. Neural networks are a form of artificial intelligence and they may obviate some of the problems associated with traditional statistical techniques, and they are representing a major advance in predictive modeling. OBJECTIVE The purpose of this study was to assess the usefulness and accuracy of artificial neural network in the prediction and prognosis of acute myocardial infarction in patients with previous coronary artery bypass surgery. METHODS The baseline characteristics and clinical data were recorded in 2180 consecutive patients. The data set contains 13 predictor variables per patient. It was first randomly split into training (1090 cases) and test sets (1090 cases). Artificial neural network performance was evaluated using the original data set for each network, as well as its complementary test data set, containing patient data not used for training the network. The program compared actual with predict outcome for each patient, generating a file of comparative results. At the end, results from this file were analyzed and compared, on the basis of a 2x2 contingency table constructed from expected or obtained statistics (accuracy, sensitivity, specificity and positive/negative predictivity). RESULTS Linear discriminant analysis was not efficient for prediction and prognosis of acute myocardial infarction in patients with prior CABG. The results show that a statistical linear model is not able to perform class separation in multidimensional space and that a nonlinear approach is justified. In analyzing the performance of neural network in outcome prognosis of AMI in patients with previous CABG it is clear that neural network method was better for almost all statistic parameters for all analyzed prediction variables. CONCLUSIONS In this clinical situation, artificial intelligence appears to be superior to linear methods for prediction and prognosis of AMI in patients with previous CABG.