Traumatic Brain Injury Rehabilitation Outcome Prediction Using Machine Learning Methods.

Data were last collected in 2011, so the dataset is relatively aged. The current analysis used 1946 of 2130 originally enrolled participants, after excluding Canadian sites and people admitted and treated for a disorder of consciousness. The dataset may have missed important therapy features and confounders such as patient familiarity, patient preference, treatment target, bowel and bladder management, and social determinants of health. Therapy time was averaged weekly over the stay, so timing details were lost. FIM was used as a proxy for cognitive and motor function, and more specific motor or cognitive measures might change the findings. The study is a prediction analysis of usual-care records, so it cannot show that changing a therapy feature will change outcomes.

Do not read effort, age, days to rehabilitation or therapy activities as causes of better recovery. The study used observational usual-care records and prediction models, not a trial that assigned treatments.