Prediction of outcome in the critically ill using an artificial neural network synthesised by a genetic algorithm

Summary The uncertain fate of individual patients in intensive care results from the heterogeneity of case-mix, life threatening insults, and the host response to such insults. These variables preclude an accurate prediction of likely outcome. Logistic regression models incorporating large numbers of those measurements which vary from patient to patient, and relate to prognosis, have been designed in an effort to predict accurately clinical outcome. A neural net, whose structure was selected by means of a genetic algorithm, predicted hospital mortality more accurately than a logistic regression model.