Neural networks for mortality prognosis in cardiac patients:
This article was originally published in Clinica
Executive Summary
Neural networks can reliably predict mortality in patients with heart failure, says Dr Juarez Oriz of the Centro de Cardiologia N*o Invasiva, S*o Paulo, Brazil (Journal of the American College of Cardiology, December 1995). Dr Ortiz' team gathered clinical and Doppler-derived ECG data from 95 cardiac patients and, after one year, collected data on survival or death, to derive a prognostic variable. In separating survivors from non-survivors, linear discriminant analysis gave accuracy at the ideal cutoff value of only 67.4%, sensitivity of 67.5%, positive predictive value of 27.8%, and a 91.5% negative predictive value. Conversely, all "artificial neural networks were able to predict outcome with an accuracy of 90%", with 83% specificity and 71.4% sensitivity for the best artificial neural network.