Do models incorporating comorbidities outperform those incorporating vital signs and injury pattern for predicting mortality in geriatric trauma?

Abstract

Geriatric trauma is becoming a significant public health concern. The most commonly used prediction models for mortality benchmarking are based on vital signs and injury pattern, including the Trauma and Injury Severity Score (TRISS), which is less accurate in the elderly. The ICD-9-based prediction models incorporating injuries and comorbidities, such as the University Health System Consortium Expected Mortality (UHC-EM), may be more accurate for the elderly. We retrospectively studied all trauma admissions from January 2005 to June 2012 at an academic level I adult trauma center. This was an observational study comparing expected to actual in-hospital mortality for both geriatric (age ≥65 years) and nongeriatric populations. Predictive ability for TRISS and UHC-EM was determined by the area under the receiver operator characteristic curve (AUC). Geriatric patients had higher median TRISS predicted mortality (8.4% [interquartile range (IQR) 4.8%, 27.4%] vs 2.8% [IQR 1.1%, 30.2%], p < 0.001). Geriatric patients had a median UHC-EM 5 times higher than nongeriatric patients (5.0% [IQR 1.0%, 19.0%] vs 1.0% [IQR 0%, 7.0%], p < 0.001). In-hospital mortality was 3 times higher in geriatric patients (18.1% vs 6.0%, p < 0.001). The UHC-EM had superior AUC to TRISS in both geriatric (0.89 [95% CI 0.87, 0.91] vs 0.81 [95% CI 0.78, 0.84], p < 0.05) and nongeriatric (0.93 [95% CI 0.92, 0.94] vs 0.90 [95% CI 0.89, 0.91], p < 0.05) patients. An ICD-9-based algorithm, such as the UHC-EM, which incorporates injuries and comorbidities, may be superior to algorithms based on vital signs and injury patterns without comorbidities in predicting mortality after trauma in the geriatric population.