Machine learning approaches for improved understanding of factors associated with history of sport-related concussion

Sport-related concussion (SRC), which accounts for a significant portion of all mild traumatic brain injuries in the United States, can adversely affect quality of life and long-term cognitive function. Identifying the persisting effects of concussion is vital for developing interventions that may reduce the risk of concussion recurrence and progressive neurodegeneration. Development of improved prognostic and therapeutic procedures might be achieved through an increased understanding of interrelationships among self-reported health and wellness status indicators, demographic and anthropometric data, and perceptual-motor performance metrics. This study aims to identify key factors that are associated with (a) a lifetime history of at least one concussion, (b) a lifetime history of more than one concussion, and (c) the number of years since the most recent concussion occurrence. We developed numerous analytical models from the set of disparate data. We addressed the class imbalance problem in objectives one and two of the study using the synthetic minority oversampling technique method and extracted the most important features relating to our three objectives using the random forest (RF) method. The results demonstrated that perceptual-motor performance capabilities play an important role in confirming that a concussion was previously sustained. RF, artificial neural networks, and decision trees demonstrated the best performance in this regard, whereas having a history of more than one previous concussion was best identified by K-nearest neighbors (KNNs). Multivariate adaptive regression splines and general linear model provided the best retrospective association with the number of years since the most recent occurrence of concussion. This study demonstrates that computational models have the potential to inform the development of individualized interventions for optimal health and wellness outcomes.