Prognostic Value of MRI Anatomical Radiologic Scores for One-Year Neurological Outcome After Severe Traumatic Brain...

Severe traumatic brain injury (sTBI) is a major cause of mortality and long-term disability, and early prognostic evaluation is essential to support therapeutic decision-making in intensive care. MRI is the reference imaging modality for detecting diffuse axonal injury (DAI), yet the prognostic accuracy and reproducibility of MRI-based radiologic scoring systems remain uncertain. This study aimed to compare the prognostic performance of five published MRI-based DAI scores for predicting 1-year neurological outcome, in a cohort of sTBI patients admitted to intensive care.

We analyzed adult patients with sTBI included in the prospective MRI-COMA cohort (NCT00577954). Inclusion criteria were age ≥ 18 years, admission to ICU for sTBI, and absence of command following within 7 days after sedation withdrawal. Brain MRI was performed between day 7 and day 35 post-injury using standardized T1, FLAIR, T2*/SWI, and diffusion-weighted sequences. Three blinded evaluators (one neurointensivist, two neuroradiologists) independently applied five radiologic scores (Adams, Firsching, Hamdeh, Stockholm, Trondheim). The primary endpoint was the ability of each MRI-based radiologic score to predict 1-year neurological outcome, dichotomized as favorable (GOSE 5-8) or unfavorable (GOSE 1-4). Inter-rater reliability was quantified using Cohen's and Fleiss' kappa coefficients.

Between 2007 and 2023, 443 patients were screened and 185 met eligibility criteria. At one year, 111 patients (59%) had an unfavorable outcome. The prognostic performance of the five MRI scores was moderate, with AUCs ranging from 0.60 [CI95 0.52-0.68] to 0.70 [CI95 0.63-0.77], with no significant differences across scores or raters even between the intensivist and radiologists. Inter-rater reliability was fair to moderate (Fleiss' κ = 0.28 [CI95 0.22-0.34] to 0.38 [CI95 0.33-0.43]).

Conventional MRI-based scores have limited prognostic value and reproducibility in cases of sTBI and are therefore not suitable for accurate and objective neuroprognostication.