Predictive model for survival among neurosurgical intensive care patients.

Background: Models for prediction of outcome of intensive care patients greatly help the physician to make decisions and are also important for risk stratification in clinical research and quality improvement. At present, there are no major predictive models for neurosurgical intensive care unit (NSICU) patients. This study aimed to develop a predictive model for survival in NSICU patients.

Methods: This is a prospective observational study in the NSICU at a tertiary-care university hospital. The data were collected within 24 hours of admission in all patients admitted to the NSICU. The parameters collected were demographic variables, systolic blood pressure, arterial oxygen tension after resuscitation (PaO2), Glasgow coma score (GCS) and pupillary signs, blood urea, creatinine, albumin, glucose, sodium, potassium, serum glutamic oxaloacetic transaminase, serum glutamic pyruvic transaminase, alkaline phosphatase, bilirubin, hemoglobin concentration, leukocyte count, platelet count, temperature, and evidence of infection. Mortality or discharge from NSICU was the primary outcome variable. All patients were provided full care until death or discharge from the ICU. Life support was not withdrawn in any of the patient based on the perception of outcome by the treating physician. All variables were compared between survivors and nonsurvivors. Significant variables were analyzed by multivariate logistic regression and a prediction model was developed.

Results: Four hundred six patients were included in the study. Three hundred two patients survived and 104 died (mortality of 25.6%). Significant variables on univariate analysis include primary reason for admission, GCS, pupillary reaction, systolic blood pressure, serum albumin, glucose, serum sodium concentration, hypothermia, and infection at the time of admission. Multivariate analysis showed that the significant independent factors for predicting outcome in NSICU patients are age, diagnosis, GCS, pupillary status, albumin, and serum sodium concentration. The predictive model has good discrimination (receiver operating characteristic curve=0.796) and good calibration (P=0.937). The overall accuracy of the model was 81%.

Conclusions: In the current model of prediction of survival in a neurosurgical ICU, age, diagnosis, GCS, pupillary status, serum albumin, and serum sodium are independent predictors of survival in NSICU patients.