Objective Interdisciplinary assessment, the gold standard for assessment of concussion and identification of rehabilitation targets, is a barrier for many due to geographical or financial limitations. This study aimed to explore the opportunity of leveraging AI beyond subtyping concussion to its application and accuracy in generating AI driven treatment plans.

Design A blinded chart review of 25 randomly selected de-identified subjects from the original algorithm sample based upon a machine learning approach that identified five distinct concussion subtypes along a complexity continuum.

Participants Subjects had attended concussion assessment and rehabilitation at Advance Concussion Clinic (ACC), an interdisciplinary clinic in Vancouver, Canada.

Main Results Assessment ‘prescriptions’ as generated by the AI model yielded system specific recommendations including physiotherapy queries of cervical, oculomotor, vestibular, and/or autonomic dysfunction, neuropsychology query of cognitive dysfunction, counselling query of emotional/mood dysfunction, and occupational therapy query of ADL dysfunction in school, work, or home. Comparison of clinical assessment recommendations to AI driven assessment queries matched at 99.2%, with the one discrepancy associated with human error not replicated by the AI model. A Permutation test to evaluate the accuracy of our AI model yielded a P_value≈0, demonstrating the efficacy of AI driven support in clinical decision making and prescription of targeted assessment and rehabilitation/treatment planning.

Conclusions AI driven treatment planning supports a comprehensive view of the complex, multi-system injury that is concussion, potentially offering recovery opportunities otherwise unavailable to those for whom best practice interdisciplinary assessment may not be accessible or available.