Outcome after severe head injury: An analysis of prediction based upon comparison of neural network versus logistic regression analysis

More reliable prediction of outcome would be helpful for clinicians who treat severely head-injured patients. To determine if neural network modeling would improve outcome prediction compared with standard logistic regression analysis and to determine if data available 24 h after severe head injury allows better prediction than data obtained within 6 h, we tested the ability of both techniques at these two times to predict outcome (dead versus alive) at 6 months. One thousand sixty-six consecutive patients with Glasgow Coma Scale scores of 8 or less during the first 24 h after injury were randomly divided into two groups. Data from the first group (n = 799) were used to develop the models; data from the second group (n = 267) were used to test the accuracy, sensitivity, and specificity of the models by comparing predicted and actual outcomes. The 6-month mortality race was 63.5%. Our findings confirm the importance of age, Glasgow Coma Scale scores, and hypotension in predicting outcome. Using data available al 24 h improved the predictive power of both models compared with admission data; at both time points, however, the differences in the results obtained with the two models were negligible. We conclude that outcome (dead versus alive) at 6 months after severe head injury can be predicted with logistic regression or neural network models based on data available at 24 h. Critical therapeutic decisions, such as cessation of therapy should be based on the patient's status 1 day after injury and only rarely on admission status alone.