Which brain injury metrics are suitable for supporting sports head injury assessment? A multi-sport brain strain evaluation

Objective: To identify brain injury metrics suitable for supporting sports head injury assessment by evaluating their association with brain tissue strain and consistency across sports. Methods: Head kinematics from 3,139 impacts in boxing, mixed martial arts, and rugby matches were recorded using instrumented mouthguards and used to calculate nine brain injury metrics. Impacts were simulated using an anatomically detailed finite element brain model to estimate peak 95th-percentile maximum principal strain (MPS) in the brain and brainstem, a measure of tissue deformation associated with long-term pathology. Sport-specific ordinary least squares models estimated xE, the metric value equivalent to a reference MPS of 0.21. Metric-MPS correlations and xE uncertainties were quantified using 5000 bootstrap resamples. Cross-sport consistency was assessed using the coefficient of variation (CV) of sport-specific median xE values, and uncertainty using the normalised confidence interval size (NCIS). Results: XGB, an extreme gradient boosting strain-prediction model, showed the strongest and most consistent correlations with whole-brain (r=0.924-0.974) and brainstem MPS (r=0.887-0.954) across all sports. PRV, BrIC and UBrIC also correlated strongly with whole-brain (r=0.724-0.930) and brainstem MPS (r=0.739-0.900), whereas HIC15 and HARM showed weaker correlation with MPS, particularly in rugby. XGB showed the lowest cross-sport variability (CV=0.034) and uncertainty (median NCIS=0.056). HARM, DAMAGE and HIC15 showed the greatest sport dependence (CV=0.575-0.588) and uncertainty (median NCIS=0.331-0.791). Conclusions: XGB, BrIC, and UBrIC demonstrated the strongest associations with brain tissue strain and the greatest consistency across sports. This study provides a biomechanically informed framework for selecting suitable metrics for sports HIA protocols.