009 Big data in youth elite football: could machine learning help us to better understand injury risk?. (3rd March 2020)

• Record Type: Journal Article
• Authors: Rommers, Nikki; Rössler, Roland; Verhagen, Evert; Vandecasteele, Florian; Verstockt, Steven; Lenoir, Matthieu; D'Hondt, Eva; Witvrouw, Erik
• Is Part Of: British journal of sports medicine. Volume 54(2020)Supplement 1
• Page Start: A5
• Page End: A5
• Publication Date: 2020-03-03
• Subjects: Sports medicine -- Periodicals; 617.1027
• DOI: 10.1136/bjsports-2020-IOCAbstracts.9
• Languages: English
• ISSNs: 0306-3674

Abstract

Abstract : Background: Due to the large burden of injuries, injury prevention is utmost important in elite-level youth soccer. Targeting risk management strategies more efficiently, requires a better knowledge of injury risk. Objective: To identify the injury risk in elite-level youth soccer players based on a large number of anthropometric and motor performance measures using a data science approach. Design: Prospective cohort study. Setting: Youth academies of seven Belgian premier league soccer clubs. Participants: 734 male players of the U10 to U15 age categories (mean age: 11.7±1.7 years). Assessment of risk factors: Preseason anthropometric measurements (height, weight, and sitting height) and tests to assess physical fitness (strength, flexibility, speed, agility, and endurance) and motor coordination (Körperkoordinationstest für Kinder) were taken. Main outcome measurements: Injuries were monitored continuously during one competitive season by the academies' medical staff. Machine learning, using extreme gradient boosting algorithms (XGBoost), was used to differentiate injured from non-injured players based on their baseline test results. Subsequently, the same technique was used to classify injuries as either overuse or acute. Results: During the season, 609 injuries occurred in 368 players, sustaining 1 to 7 injuries each. Of these injuries, 290 were identified as overuse and 319 as acute injuries. The machine learning algorithm was able to classify the injured players with 92% precision, 92% recall and 92% accuracy. Injuries could be classified as overuse or acute with 95% precision, 94% recall and 94% accuracy. The features in the model were ranked according to their importance in the decision process. Conclusions: Machine learning is a promising approach to assess injury risk based on anthropometric and motor performance measures in this population. This knowledge could be applied in injury risk management strategies, aiming at youth players with the highest risk and targeting those features that have the largest influence on injury risk.