External validation and comparative performance of Marshall, Rotterdam, and Helsinki computed tomography scores in adult traumatic brain injury.

Abstract

Purpose: Computed tomography (CT)-based scoring systems enable standardized quantification of intracranial pathology following traumatic brain injury (TBI). Despite widespread clinical application, external validation studies comparing the Marshall, Rotterdam, and Helsinki CT scores remain limited, particularly in middle-income settings. This study aimed to externally validate these three CT scoring systems and evaluate their incremental prognostic value beyond clinical variables.

Methods: This retrospective cohort study included adult patients with traumatic brain injury requiring hospitalization beyond 24 h at a tertiary neurosurgical center between 2022 and 2024. Two neurosurgeons independently scored admission CT scans using the Marshall, Rotterdam, and Helsinki CT scoring systems. The primary outcome was in-hospital mortality, and the secondary outcome was poor functional outcome (Glasgow Outcome Scale 1-3) at hospital discharge. Discriminative performance was assessed using area under the receiver operating characteristic curve (AUC) with DeLong test comparisons. Multivariable logistic regression evaluated incremental prognostic value beyond age and admission Glasgow Coma Scale (GCS).

Results: Among 657 patients (mean age 45.9 years, 74.9% male; mild TBI 59.4%, moderate 25.1%, severe 15.5%), in-hospital mortality was 11.0% and poor functional outcome occurred in 23.9%. Among CT-based scores, the Helsinki CT score showed the highest standalone discriminative performance for mortality (AUC 0.793, 95% CI 0.731-0.855), followed by Rotterdam (AUC 0.766, 0.700-0.832) and Marshall (AUC 0.682, 0.625-0.739); Helsinki and Rotterdam both outperformed Marshall. However, when added to a baseline clinical model (age + GCS, AUC 0.876), all CT scores provided minimal to negligible incremental discrimination, with no clinically meaningful improvement in model performance (ΔAUC < 0.005 across all models, all p > 0.20). Similar patterns were observed for poor functional outcome.

Conclusion: The Helsinki CT score demonstrated the highest standalone discriminative performance; however, all three CT-based scoring systems provided limited incremental prognostic value beyond clinical assessment (age and GCS), supporting their role as complementary rather than independent tools in TBI evaluation. These findings support the use of CT-based scores primarily for early risk stratification and standardized communication in emergency settings, rather than as standalone prognostic tools.