Prediction of trauma mortality using a neural network.

A neural network is a computerized construct consisting of input neurons (which process input data) connected to hidden neurons (to mathematically manipulate values they receive from all the input neurons) connected to output neurons (to output a prediction). Neural networks are created and trained via multiple iterations over data with known results. In 1993, 897 trauma patients were either declared dead in the emergency room (ER; 76 cases), admitted to the intensive care unit (427 cases, 36 deaths), or taken directly to the operating room (394 cases, 29 deaths). Using only data available from the ER, a neural network was created, and 628 cases were randomly selected for training. After 268 iterations, the network was trained to correctly predict death or survival in all 628 cases. This trained network was then tested on the other 269 cases without our providing the death or survival result. Its overall accuracy was 91 per cent (244 of 269 cases). It was able to predict correctly 60 per cent (12 of 20 cases) of the postoperative or post-intensive care unit admission deaths and 90 per cent (26 of 29 cases) of the deaths in the ER. Computerized neural networks can accurately predict a trauma patient's fate based on inital ER presentation. The theory and use of neural networks in predicting clinical outcome will be presented.