A model for predicting delayed intracranial hypertension following severe head injury

Abstract

✓ A graph is presented for predicting delayed intracranial hypertension (intracranial pressure (ICP) greater than 30 mm Hg) for severely head-injured patients, based on a logistic regression model. Data gathered during the first 24 hours of patient observation are used to predict patient status during the subsequent 48 hours. The best predictor out of 10 factors analyzed was the peak ICP level during the first 24 hours (p < 0.0001). Other predictors used in the final model were the presence of hypotension (p = 0.045) and abnormal ventricles — defined as ventricles which were either absent, small, or enlarged (p = 0.086). Error rates of 24% and 20% were obtained initially and by means of a separate cross-validation group, respectively. Use of a conservative cut point (25% estimated chance of developing excess ICP) for designating high-risk patients provided a procedure with sensitivity of 86% to 89% for the two groups.