Artificial Neural Network to Predict ST-Segment Elevation Based on Symptoms and Signs of Acute Coronary Syndrome

Introduction: the presence or absence of ST elevation determines the prognostic and therapeutic approach to acute coronary syndrome. Objective: to predict, using neural networks, the presence or absence of ST-elevation according to symptoms and signs of acute coronary syndrome. Material and methods: analytical and cross-sectional study based on a database of 106 patients admitted for acute coronary syndrome in a Peruvian hospital. A series of symptoms and signs were analyzed prior to performing the electrocardiogram. Multilayer perceptron-type neural networks were used, from which the classification table of correct forecasts, the predictive capacity of the model, as well as the normalized importance of the predictor variables were evaluated. Results: The neural network had overall percentages of correct predictions in training and testing of 93.40% and 93.30%, respectively. The area under the curve was 0.982, indicating that the model has an outstanding predictive capacity. In the training stage, the percentages of correct predictions were 94% and 92.30% to rule out and detect ST elevation, respectively. In the test, the percentage of correct predictions to rule out and detect STelevation was 90% and 100%, respectively. The most influential predictor variables were heart rate, being a smoker, neck pain, nausea, and diaphoresis. Conclusions: the multilayer perceptron-type neural network was efficient for predicting ST-segment elevation according to symptoms and signs of patients admitted to the emergency room due to acute coronary syndrome.