Effect of an unsupervised multidomain intervention integrating education, exercises, psychological techniques and machine learning feedback, on injury risk reduction in athletics (track and field): protocol of a randomised controlled trial (I-ReductAI).

Abstract

The primary aim is to assess the impact of a multidomain intervention that integrates education, exercise, psychological techniques and machine learning feedback on the duration athletes remain free from injury complaints leading to participation restriction (ICPR) during a 20-week summer competitive athletics season. The secondary aims are to assess the intervention's effect on reducing (i) the incidence, (ii) the burden, (iii) the period prevalence and (iv) the weekly prevalence of ICPR during the same timeframe. We will perform a two-arm randomised controlled trial. This study will involve an intervention group and a control group of competitive athletes licensed with the French Federation of Athletics, aged between 18 and 45, over an outdoor athletics competitive season lasting 20 weeks (March to July 2025). Data will be collected before the start (demographic, training and injury history) and one time per day (training and competition volume/intensity, perceived physical and psychological state, and illness and injury incidents) for both groups. The intervention group will be required to (i) view a series of 12 educational videos on injury prevention, (ii) engage in discipline-specific exercise programmes, (iii) implement stress and anxiety management techniques and (iv) view daily the injury prognostic feedback generated by the athlete's collected data based on machine learning. Outcomes will be analysed over the final 14 weeks of follow-up to allow time for the intervention to establish any potential efficacy. The primary outcome will be the time-to-event for each ICPR. Secondary outcomes will include (i) incidence, (ii) burden, (iii) period prevalence and (iv) weekly prevalence of ICPR. The primary outcome will be analysed using a Prentice-Williams-Peterson gap-time model. In contrast, the secondary outcomes will employ Poisson (i, ii), logistic (iii) and generalised estimating equations (iv) regression models, respectively.