Return to Play Prediction Accuracy of the MLG-R Classification System for Hamstring Injuries in Football Players: A Machine Learning Approach
Xavier Valle,Sandra Mechó,Eduard Alentorn-Geli,Tero A H Järvinen,Lasse Lempainen,Ricard Pruna,Joan C Monllau,Gil Rodas,Jaime Isern-Kebschull,Mourad Ghrairi,Xavier Yanguas,Ramon Balius,Adrian Martinez-De la Torre
DOI: https://doi.org/10.1007/s40279-022-01672-5
Abstract:Background and objective: Muscle injuries are one of the main daily problems in sports medicine, football in particular. However, we do not have a reliable means to predict the outcome, i.e. return to play from severe injury. The aim of the present study was to evaluate the capability of the MLG-R classification system to grade hamstring muscle injuries by severity, offer a prognosis for the return to play, and identify injuries with a higher risk of re-injury. Furthermore, we aimed to assess the consistency of our proposed system by investigating its intra-observer and inter-observer reliability. Methods: All male professional football players from FC Barcelona, senior A and B and the two U-19 teams, with injuries that occurred between February 2010 and February 2020 were reviewed. Only players with a clinical presentation of a hamstring muscle injury, with complete clinic information and magnetic resonance images, were included. Three different statistical and machine learning approaches (linear regression, random forest, and eXtreme Gradient Boosting) were used to assess the importance of each factor of the MLG-R classification system in determining the return to play, as well as to offer a prediction of the expected return to play. We used the Cohen's kappa and the intra-class correlation coefficient to assess the intra-observer and inter-observer reliability. Results: Between 2010 and 2020, 76 hamstring injuries corresponding to 42 different players were identified, of which 50 (65.8%) were grade 3r, 54 (71.1%) affected the biceps femoris long head, and 33 of the 76 (43.4%) were located at the proximal myotendinous junction. The mean return to play for grades 2, 3, and 3r injuries were 14.3, 12.4, and 37 days, respectively. Injuries affecting the proximal myotendinous junction had a mean return to play of 31.7 days while those affecting the distal part of the myotendinous junction had a mean return to play of 23.9 days. The analysis of the grade 3r biceps femoris long head injuries located at the free tendon showed a median return to play time of 56 days while the injuries located at the central tendon had a shorter return to play of 24 days (p = 0.038). The statistical analysis showed an excellent predictive power of the MLG-R classification system with a mean absolute error of 9.8 days and an R-squared of 0.48. The most important factors to determine the return to play were if the injury was at the free tendon of the biceps femoris long head or if it was a grade 3r injury. For all the items of the MLG-R classification, the intra-observer and inter-observer reliability was excellent (k > 0.93) except for fibres blurring (κ = 0.68). Conclusions: The main determinant for a long return to play after a hamstring injury is the injury affecting the connective tissue structures of the hamstring. We developed a reliable hamstring muscle injury classification system based on magnetic resonance imaging that showed excellent results in terms of reliability, prognosis capability and objectivity. It is easy to use in clinical daily practice, and can be further adapted to future knowledge. The adoption of this system by the medical community would allow a uniform diagnosis leading to better injury management.