Validation and comparison of models predicting survival following intracerebral hemorrhage

Abstract

To compare the performance of two previously reported logistic regression models using data independent from those data used to derive the models. Prospective. Acute stroke unit of a tertiary care hospital. One hundred twenty-nine patients with supratentorial intracerebral hemorrhage. Model 1 contains the initial Glasgow Coma Scale score, hemorrhage size, and pulse pressure. The more complex model 2 includes, in addition to those three variables, the presence or absence of intraventricular hemorrhage and a term representing the interaction of intraventricular hemorrhage and Glasgow Coma Scale score. The areas under the receiver operating characteristic curves generated for each model were statistically indistinguishable. Model 1 predicts 30-day patient status as well as the more complex model 2. Model 1 provides a valid, easy-to-use means of categorizing supratentorial intracerebral hemorrhage patients in terms of their probability of survival.