Comparison of Concussion Risk Functions Generated from Two Different Datasets

To estimate risk of concussion, risk functions based on injuries occurring in sports are often used. A range of datasets have been used to develop injury risk functions for concussion based on either global kinematics or tissue-level predictors. Two such datasets are one from American football, and another one from Australian football and rugby. These two datasets constitute the largest published collections of video-verified concussive cases in sports with known kinematics suitable for constructing risk functions. The objective of this study was to analyze the differences between two datasets of concussion for injury predictions to better understand the influence on injury risk functions. The kinematics were applied to the KTH head model and risk functions for different kinematic- and tissue-based predictors were developed and compared. The accuracy, sensitivity, specificity, and AUC were also compared. The two datasets evaluated in this study generated different risk curves. The datasets had some similarities such as having no significant difference in resultant linear acceleration, but also some differences, for example having a significant difference in resultant angular velocity. The Australian cases had relatively equally distributed major x-, y-, and z-components for angular velocity while the majority (59%) of the NFL cases had a major x-component (coronal plane rotation) representing more than 50% of the resultant. The y-component of the linear acceleration (lateral direction) was the major component in 64% of the Australian cases and 72% of the NFL cases. The two datasets, from Australian football/rugby, and American football, generated different injury risk curves with a lower 50% risk of concussion for the Australian dataset. This indicates that the choice of data as input for the development of injury risk functions is important. Therefore, it is necessary to improve methodology with focus on sampling methods and reliable/valid data collection.