Association between the use of daily injury risk estimation feedback (I-REF) based on machine learning techniques and injuries in athletics (track and field): results of a prospective cohort study over an athletics season

Abstract

Objective: To analyse the association between the level of use of injury risk estimation feedback (I-REF) provided to athletes and the injury burden during an athletics season. Method: We conducted a prospective cohort study over a 38-week follow-up period on athletes competing at the French Federation of Athletics. Athletes completed daily questionnaires on their athletics activity, psychological state, sleep, self-reported level of I-REF use, and injuries. I-REF provided a daily estimation of the injury risk for the next day, ranging from 0% (no risk of injury) to 100% (maximum risk of injury). The primary outcome was the injury burden during the follow-up, defined as the number of days with injury per 1000 hours of athletics activity. A negative binomial regression model was used to analyse the association between self-reported I-REF use and the injury burden. Results: Of the 897 athletes who met the inclusion criteria, 112 (38% women) were included in the analysis. The mean daily response rate of the follow-up was 37%±30%. The primary analysis found no significant association between the self-reported I-REF use and the injury burden (n=112, e β: 0.992, 95% CI: 0.977 to 1.007; p=0.308). However, when considering athletes’ daily response rate in secondary analysis, for a response rate of at least 9%, we observed a significant association between the self-reported level of I-REF use and the injury burden (n=76, e β: 0.981, 95% CI: 0.965 to 0.998; p=0.027). Conclusions: Daily injury risk estimation feedback using machine learning was not associated with reducing injury burden.