Repeat CT-derived time–volume hematoma dynamics improve 6-month functional outcome prediction in traumatic intracranial hematoma

Prognostic assessment after traumatic brain injury with intracranial hematoma relies mainly on admission variables. We evaluated whether repeat CT-derived hematoma dynamic variables add value for 6-month functional outcome prediction beyond a strong baseline model. This retrospective study included 639 adults with traumatic intracranial hematoma admitted between March 2021 and March 2023. Unfavorable outcome was defined as a Glasgow Outcome Scale score < 4 at 6 months. Sequential logistic models were developed according to information availability. The admission model included age, sex, Glasgow Coma Scale score, baseline hematoma volume, and hematoma type. Laboratory variables were then added, followed by repeat CT-derived hematoma dynamics, including volume change, CT interval, average hourly hematoma volume change, and a spline interaction between volume change and CT interval. Performance was assessed using AUC, Brier score, calibration plots, bootstrap validation, and decision curve analysis. Among 639 patients, 144 had unfavorable outcomes. The admission model showed high discrimination, with an AUC of 0.902 and a Brier score of 0.098. Adding laboratory variables improved the AUC to 0.918 and reduced the Brier score to 0.092. Incorporating Repeat CT-derived hematoma dynamics further improved performance. The spline interaction model performed best, with an AUC of 0.937, an optimism-corrected AUC of 0.911, and a Brier score of 0.080. Average hourly hematoma volume change showed a nonlinear association with unfavorable outcome, and decision curve analysis indicated greater net benefit after adding repeat CT dynamics. Repeat CT-derived hematoma dynamics improved prediction of 6-month functional outcome beyond admission variables and laboratory results. Time–volume representations of hematoma change may provide interpretable imaging markers for early risk stratification.