Associations and Predictive Value of Neuroimaging Characteristics for Extubation Failure in Moderate to Severe Traumatic Brain Injury: An Exploratory Analysis.

BACKGROUND: Predicting successful extubation in traumatic brain injury (TBI) remains challenging. We examined associations between structural neuroimaging characteristics and extubation failure in TBI, and the ability to enhance prognostic models based on clinical characteristics alone. METHODS: Retrospective cohort study of patients with moderate-severe TBI [Glasgow Coma Scale (GCS) ≤ 12] who were mechanically ventilated ≥ 24 h and underwent ≥ 1 planned extubation attempt. We fit multivariable logistic regression models (adjusted for age, GCS, and cardiopulmonary disease) to examine associations between validated head computed tomography (CT)-based scores (on both the admission CT and pre-extubation CT) and extubation failure (reintubation ≤ 5 days from extubation). We used DeLong's test to compare the area under the curve (AUC) for two predictive clinical models (model 1: cough, gag, age; model 2: temperature, hypertension, respiratory rate) with and without neuroimaging characteristics. RESULTS: Of 2044 patients screened, 181 (74% male, 79% White) were included, and 32 (17.7%) experienced extubation failure. No neuroimaging characteristics on admission CT were associated with extubation failure; however, the following pre-extubation CT characteristics were statistically significantly associated with extubation failure: cisternal compression [adjusted odds ratio (aOR) 3.40, 95% confidence interval (CI) 1.46-8.10], cerebral edema score > 2 (aOR 4.02, 95% CI 1.74-9.90), Marshall CT score III-IV-VI (aOR 2.91, 95% CI 1.15-7.21), Rotterdam score > 2 (aOR 2.74, 95% CI 1.16-6.43), and Helsinki score > 5 (aOR 3.43, 95% CI 1.37-8.50). Model prediction improved modestly after adding either cisternal compression or cerebral edema score, although it was not statistically significant (AUCs for models with vs. without cisternal compression: model 1: 0.558 vs. 0.669, p = 0.07 and model 2: 0.727 vs. 0.764, p = 0.20; AUCs for models with vs. without cerebral edema score: model 1: 0.558 vs. 0.664, p = 0.05 and model 2: 0.727 vs. 0.767, p = 0.28). CONCLUSIONS: Several neuroimaging characteristics on pre-extubation CT, but not admission CT, were associated with extubation failure. The predictive ability of neuroimaging characteristics warrants further exploration in larger cohorts.