Does modification of the innsbruck and the Glasgow Coma Scales improve their ability to predict functional outcome?

Background: The accurate prediction of functional outcome requires the development of multivariate models. To enhance their contribution to such models, the predictive power of each component must be optimized. Objectives: To improve the predictive power of coma scales as the first step in building more sophisticated multivariate models to predict specific levels of functional outcome. Design: Prospective descriptive study. Setting: Neurology and neurosurgery intensive care unit (NNICU) in a tertiary care academic center. Patients: Eighty-four patients with acute traumatic brain injury, intracerebral hemorrhage, subarachnoid hemorrhage, or ischemic stroke. Interventions: None. Main Outcome Measures: The Glasgow Coma Scale (GCS) and Innsbruck Coma Scale (ICS) were administered within 24 hours of admission to the NNICU and then at 48-hour intervals until discharge of the patient from the NNICU. The assessments were performed by 3 occupational therapy graduate students working under the supervision of the medical director of the NNICU. The functional outcome at 3 months after discharge from the hospital was assessed by telephone by the same nurse using the following categories: (1) dead, (2) receiving nursing home or custodial care, (3) home with help, or (4) independent. Cronbach's alpha estimates of reliability for each scale were computed using all scores obtained during the study. The analyses indicated that the verbal response item of the GCS and the oral automatisms item of the ICS were less reliable in this patient population. The scales were modified by deleting those items, and predictive validity for the original and modified scales was computed using a discriminant function of the admission scores. Results: Before modification, both scales were best at predicting independence (GCS and ICS, 71% correct) and mortality (GCS, 60% correct, ICS, 56% correct). The modifications produced a modest improvement in the ability of both scales to better predict levels of outcome (modified GCS: home with help, 33% correct, independent, 71% correct; modified ICS: home with help, 0% correct, independent, 74% correct). Conclusions: By deleting items with low reliability from the ICS and the GCS we achieved improved reliability and predictive validity. The improvement in predictive power, however, was inadequate to accurately predict functional outcome. Combining clinical scales with other demographic, physiological, functional, and radiographic data will be needed to achieve useful predictions of functional outcome.