Season-long head acceleration exposure, prior contact-sport history, and cognitive performance in British amateur American football players

Background: Head acceleration events (HAEs) are common in contact sports, but season-long exposure and its relationship with cognitive performance in amateur British American football remain poorly characterised. Methods: Thirty male players were recruited from three British American football teams. Cognitive performance was assessed pre-season, mid-season, and post-season using the Sport Concussion Assessment Tool 6 and computerised Stroop, 2-back, Go/No-go, and visual-search tasks. Seasonal cognitive analyses included 23 participants with available data. HAEs were quantified using helmet-mounted accelerometers; exposure-linked analyses were restricted to 19 participants who provided sufficiently complete sensor recordings across contact training sessions and competitive matches. Seasonal cognitive changes were assessed using mixed-effects models. Associations of in-season HAE metrics and previous contact-sport seasons with global cognitive change were examined using simple linear regression and elastic-net regression. Results: In the 19-participant exposure-linked cohort, prior contact-sport exposure showed the strongest nominal univariate association with worse global cognitive change (r squared = 0.337, p = 0.009), although this did not remain significant after correction for multiple comparisons (Benjamini-Hochberg-adjusted p = 0.055). Corrected nested cross-validation indicated little out-of-sample performance for the multivariable elastic-net model (r = 0.144, predictive R squared = -0.008). Although previous contact-sport seasons had the largest standardised coefficient in the final full-cohort fit, predictor retention varied substantially across resamples. Conclusion: Prior contact-sport exposure showed the strongest exploratory association with relatively worse global cognitive change, whereas in-season HAE metrics showed weaker associations. However, the multivariable model did not demonstrate useful out-of-sample predictive performance, and the small exposure-linked cohort precludes firm conclusions regarding the relative importance of individual predictors.