Determination of Global Registry of Acute Coronary Events score based on initial symptoms of coronary syndrome using multilayer perceptrón

Abstract

Introduction: the GRACE scale estimates the prognosis and mortality due to acute coronary syndrome.

Objective: predict the GRACE score according to admission characteristics for acute coronary syndrome in a Peruvian hospital using multilayer perceptron.

Materials and method: analytical and cross-sectional study of a secondary database of 106 patients from a Peruvian hospital admitted for acute coronary syndrome. The variables were GRACE score, heart rate, systolic blood pressure, left arm pain, neck pain, abdominal pain, age, and ST segment elevation. Multilayer perceptron-type neural networks were used.

Results: The neural network had a relative error of 0.168 and 0.139 in training and testing, respectively. The scatter plot had a homogeneous distribution, with an R2 coefficient of 0.847, indicating that 85% of the variation in the GRACE score can be explained by the GRACE score predicted by multilayer perceptron. The average GRACE score in patients without ST elevation was 104.74, while with ST elevation it was 142.14, similar to the GRACE score obtained using multilayer perceptron, where the average in the absence of ST elevation was 105.52, while with ST elevation was 140.36.

Conclusions: the use of neural networks is efficient for the prediction of GRACE score based on symptoms and signs of admission in patients with acute coronary syndrome.