Evaluation of the Efficacy of Head and Brain Injury Risk Functions

Head and brain injury risk functions have been proposed over the years in order to estimate the probability of brain injury from head impact using kinematic and injury data from various sources, such as PMHS and animal testing. Yet, researchers have reiterated the need for human head kinematic data for creating and evaluating injury risk functions and risk assessment values given that other types of data may lack the physiologic and anthropometric response required.
To this end, this thesis aimed to collect human head kinematics to evaluate existing injury risk functions. A literature review revealed what measures of human head motion have been previously published, including direct measures of head kinematics (e.g., wearable sensors used on-field with contact athletes) and indirect measures of head motion (e.g., computational and experimental reconstructions of real-life impacts). Three data sources (n = 443) were selected from the literature based on inclusion criteria. Data were evaluated for consistency and used to calculate kinematic and strain-based injury risks.
The injury risk values were used to evaluate the efficacy of each of 16 injury risk functions using four separate analytical tools. Correlations between the injury risk functions and strain metrics showed that strain- and rotationally-based injury risk functions had the strongest correlations. Area under ROC curves assessed each function’s ability to separate injurious and non-injurious impacts; all risk functions were better than random guessing. Likelihood estimates ranked the injury risk functions on their ability to correctly predict the dataset’s injury outcomes, with GAMBIT showing the best predictive capability according to this measure. The number of expected injuries was calculated for each risk function; however, most did not correctly estimate the number of observed injuries (n = 31). Results from these four assessment tools, however, were mixed, with no single risk function performing best.
The lack of consensus in the assessment tools may be a result of the data used to develop the risk functions. Studies have noted unbiased exposure data is needed to estimate the absolute injury risk from impact; however, most injury risk functions (and the data in this thesis) were based on case-control data. Statistical simulations were conducted to create injury risk functions based on several sampling scenarios. These results from these simulations demonstrated that researchers should be wary of risk curves derived solely from case-controlled data given that these may be over-predictive of injury probabilities compared to the absolute risk of a population. Despite the issues with the risk functions, this thesis provides a verified data set that is a necessary step for injury risk function evaluation.