La prédiction des blessures en sport : fiction ou réalité ?

Today, injury in sport represents a major problem for athletes and their entourage. Prevention measures are developed and are available to the sports community. Among them, the emergence of new technologies and data analysis approaches offer new opportunities. Given the fact that these prediction methods tend to develop, it seemed important to us that the actors around the athlete, and in particular health professionals, have notions to better understand these approaches and to be able to interpret the work presenting injury prediction (risk estimation) analyses. Through this article, based on a narrative review of the literature, we have presented Machine Learning (ML), as well as its applications and limitations. ML, or “machine learning”, is a tool derived from statistics, related to artificial intelligence, which makes it possible to build, from input data (predictive variables) and output data (variables to be predicted), models capable of predicting an event. Thus, like any analysis, ML can present certain limitations and risks that should be avoided, but also to know and detect when reading articles/works using ML, or when you want to use it. In conclusion, in sports traumatology, Machine Learning models offer the opportunity: 1) to help diagnose injuries or; 2) to optimize athletes’ training by estimating their risk of injury, both in a screening and monitoring context. However, this prediction tool cannot adapt to all situations without risk and can sometimes lead to false predictions. Thus, Machine Learning offers interesting perspectives with the possibility of having a decision support tool for field actors, but it is necessary to take into account the limits and risks of this approach in order to use them best and get the best benefits. Machine Learning is not a crystal ball that allows us to see the future, but a method of data analysis that relies on measured data and therefore depends on the quality of the latter.