Predicting hospital mortality using APACHE II scores in neurocritically ill patients: a prospective study

Four versions of Acute Physiology and Chronic Health Evaluation are limited in predicting hospital mortality for neurocritically ill patients. This prospective study aimed to develop and assess the accuracy of a modified APACHE II model in predicting mortality in neurologic intensive care unit (N-ICU). A total of 653 patients entered the study. APACHE II scores on admission, and worst 24-, 48-, and 72-h scores were obtained. Neurologic diagnoses on admission were classified into five categories: cerebral infarction, intracranial hemorrhage, neurologic infection, neuromuscular disease, and other neurologic diseases. We developed a modified APACHE II model based on the variables of the 72-h APACHE II score and disease category using a multivariate logistic regression procedure to estimate probability of death. We assessed the calibration and discrimination of the modified APACHE II model using the Hosmer–Lemeshow goodness-of-fit chi-squared statistic and area under the receiver operating characteristic curve (AU-ROC). The modified APACHE II model had good discrimination (AU-ROC = 0.88) and calibration (Hosmer–Lemeshow statistic: χ 2 = 3.707, P = 0.834). The discrimination of the 72-h APACHE II score for cerebral infarction, intracerebral hemorrhage, and neurologic infection was satisfactory, with AU-ROC of 0.858, 0.863, and 1.000, respectively, but it was poor in discriminating for the categories of other neurologic diseases and neuromuscular disease. The results showed that our modified APACHE II model can accurately predict hospital mortality for patients in N-ICU. It is more applicable to clinical practice than the previous model because of its simplicity and ease of use.