Predicting Return Home After Moderate-to-Severe Traumatic Brain Injury.

Abstract

Days alive and at home (DAH) is a validated outcome measure that captures health care transitions between time spent at home vs various nonhome care settings, offering a more nuanced patient-centered understanding of recovery. We aimed to (1) characterize long-term recovery trajectories for adults with moderate-to-severe traumatic brain injury (msTBI) using DAH and (2) develop and internally validate a clinical prediction model for favorable vs unfavorable recovery. This multicenter retrospective population-based cohort study was performed in Ontario, Canada; we identified adults hospitalized with isolated msTBI between 2009 and 2021 that survived beyond 72 hours postadmission. DAH was calculated in 30-day intervals from index admission through 3 years postinjury. Latent class mixed modeling identified unique recovery trajectories. Sociodemographic, clinical, and injury variables were then evaluated as candidate predictors. Logistic regression, penalized regression, and machine learning approaches were compared for discrimination of favorable (early, intermediate, or late recovery) vs unfavorable (delayed deterioration or poor recovery) trajectories. Model performance was evaluated in a held-out test set using area under the receiver operating characteristic curve (AUC) with 95% CIs. There were 3,004 adults included in the cohort with an overall male predominance (n = 2,284, 76.1%), and the average age was 49.6 years (SD 21.4). Latent class mixed modeling revealed 5 distinct DAH trajectories: early recovery (38.6%), intermediate recovery (27.5%), and late recovery (3.2%) groups were characterized by return to the community at varying time intervals; by contrast, delayed deterioration (3.4%) and poor recovery (27.3%) groups experienced limited time at home at final follow-up. Patients in the unfavorable recovery trajectory groups were older, frailer, and more likely to sustain fall-related injuries. Radiographic predictors of poor outcome included subdural hematoma, large intraparenchymal hematoma, diffuse axonal injury, and cerebral edema. Using various demographic, injury and radiographic characteristics, a prediction model (DAH-TBI calculator) could discriminate trajectories with an AUC of 0.812 (95% CI 782-0.842) on a held-out internal validation data set. Our findings emphasize a longer period of postinjury outcome observation may be necessary to capture societal reintegration after msTBI. We developed and internally validated the DAH-TBI calculator to predict long-term postinjury outcomes.