Comparison of APACHE III, II and the Glasgow Coma Scale for prediction of mortality in a neurosurgical intensive care unit.

OBJECTIVES: This study examined the efficacy of predicting power for hospital mortality of three different scoring systems in a neurosurgical intensive care unit (NICU).

SETTING: An eight-bed NICU in a 1,270-bed medical centre (Taichung Veterans General Hospital).

SUBJECTS: Two hundred patients with head injury, brain tumour, hypertensive intracerebral haemorrhage, rupture of aneurysm or arteriovenous malformation, or other categories were included in our study in a consecutive period of 14 months. Patients less than 14 years old were not included.

DESIGN: On the first day of admission, data were collected from each patient to compute the Acute Physiology and Chronic Health (APACHE) II and III, and Glasgow Coma Scale (GCS) scores. Hospital mortality was defined as when death occurred before discharge from hospital.

INTERVENTIONS: none.

MEASUREMENTS: Sensitivity, specificity and correct prediction outcomes were measured by logistic regression in three scoring systems. The Youden index was also obtained. The best cutoff point in each scoring system was determined by logistic regression or by the Youden index. Data obtained by logistic regression were compared by McNemar's test. The differences in Youden index were calculated by the Student's t-test. The area under the Receiver Operating Characteristic (ROC) curve was computed and the area of each scoring system was then compared by the Wilcoxon Mann-Whitney test.

MAIN RESULTS: The correct prediction of outcome was 85.5% in APACHE III, 77.5% in APACHE II and 75.0% in GCS. The area under the Receiver Operating Characteristic (ROC) curve was 0.892 in APACHE III, 0.826 in APACHE II and 0.868 in GCS. For the prediction of dead patients at the best cutoff point, APACHE III and GCS were better than APACHE II, (both p < 0.01 respectively). For the prediction of alive patients at the best cutoff point, APACHE III was better than GCS and APACHE II (p < 0.01 respectively).

CONCLUSION: The APACHE III system seems to be the most reliable. The results reveal that the APACHE III system is better in predicting power for hospital mortality than either the GCS or APACHE II systems in our NICU patients.