A neural network improves the classification of high-risk intensive care patients

A two-layer feedforward neural network for classifying intensive care patients as normal or at risk for severe cardiorespiratory disorders was designed and compared with the best, previously investigated, Bayesian classifier of similar complexity. At three distinct observation times soon after heart surgery, the three variables most effective in separating the two classes were measured and used to test the performances of classifiers by the leave-one-out method. The results showed that the ability of the well-known neural network to describe input-output nonlinear behaviour made it possible to obtain a lower level of misclassification of new data without loss of generalization