Reappraisal of Computerized Tomography Scoring Systems for Outcome Prediction in Traumatic Brain Injury: A...

Several computerized tomography (CT)-based scoring systems have been developed to grade traumatic brain injury (TBI) and predict outcome. However, most were derived from younger populations, and their performance in older adults-who represent an increasing proportion of TBI patients-remains unclear. This study evaluated the prognostic performance of different CT-based models for short- and long-term outcomes in younger versus older adults (≥65 years).

We retrospectively analyzed 1935 consecutive TBI patients admitted between 2013 and 2024. Individual components of each CT scoring system were recorded from the first CT scan obtained within 24 h after injury. The Marshall CT classification, Neuroimaging Radiological Interpretation System (NIRIS), Rotterdam, Helsinki, and Stockholm CT scores were evaluated. Outcomes included short-term outcomes (TBI-related in-hospital mortality, surgical evacuation of intracranial lesions, intracranial pressure monitoring, and intracranial lesion progression) and long-term outcomes (mortality and unfavorable outcome [Glasgow Outcome Scale {GOS} 1-3] at 6 months). Model performance was assessed using discrimination, calibration, and overall fit. Explanatory contribution of CT findings was evaluated through proportional explained variance. Published equations were applied when available to validate long-term outcome predictions. Internal validation was performed using optimism-corrected bootstrap resampling (1000 iterations).

Long-term outcome data were available for 1798 patients (1243 younger and 555 older adults). Overall, prognostic performance was significantly better in younger than in older adults across all outcome measures. The Helsinki CT score demonstrated the best overall performance, with area under the receiver operating characteristic curve values ranging from 0.805 to 0.877. While the Marshall and Rotterdam scores performed comparably to the Helsinki score in predicting in-hospital and 6-month mortality in younger adults, they were significantly less accurate for predicting unfavorable outcomes. The Rotterdam score showed particularly poor performance in older adults. For identifying patients at risk of Marshall CT class deterioration or requiring intracranial pressure monitoring, the NIRIS and Stockholm CT scores performed better, respectively, and maintained noninferior performance in older adults.

CT-based prognostic models show reduced accuracy in older adults compared with younger patients. Among the evaluated tools, the Helsinki CT score provided the most reliable prediction of mortality and unfavorable outcome across age groups, including the older adults.