The Injury Severity Score or the New Injury Severity Score for predicting intensive care unit admission and hospital length of stay?

OBJECTIVES: To compare the New Injury Severity Score (NISS) and the Injury Severity Score (ISS) as predictors of intensive care unit (ICU) admission and hospital length of stay (LOS) in an urban North American trauma population and in a subset of patients with head injuries.
METHODS: The study population consisted of 23,909 patients from three urban level I trauma centres in the province of Quebec, Canada. The predictive accuracies of the NISS and the ISS were compared using Receiver Operator Characteristic (ROC) curves and Hosmer-Lemeshow (H-L) statistics for the logistic regression model of ICU admission and using r2 for the linear regression model of LOS.
RESULTS: A total of 7660 (32%) patients were admitted to the ICU. Mean LOS was 8.2+/-2.5 days. In the whole sample, the NISS presented equivalent discrimination (area under ROC curve: NISS = 0.839 versus ISS = 0.843, p = 0.08) but better calibration (H-L statistic: 309 versus 611) for predicting ICU admission. In the subgroup patients with moderate to serious head injuries, the NISS was a better predictor of ICU admission in terms of both discrimination (area under ROC curve: NISS = 0.771 versus ISS = 0.747, p < 0.00001) and calibration (H-L statistic: 12 versus 21). The NISS explained more variation in LOS than the ISS for the whole sample (r2 = 0.254 versus 0.249, p = 0.0008) and in the sub-population with moderate to severe head injuries (r2 = 0.281 versus 0.263, p = 0.0002).
CONCLUSIONS: The NISS is a better choice for case mix control in trauma research than the ISS for predicting ICU admission and LOS, particularly among patients with moderate to severe head injuries.