Cluster analysis of cutting technique-a valuable approach for assessing anterior cruciate ligament injury risk?

Abstract

Background: Despite extensive efforts to pinpoint singular biomechanical risk factors for anterior cruciate ligament (ACL) injuries, research findings are still inconclusive. By combining multiple biomechanical variables, cluster analyses could help us identify safe and risky cutting technique strategies.

Purpose: To identify common movement strategies during cutting maneuvers and to assess their association with ACL injury risk.

Methods: A total of 754 female elite handball and football players, including 59 with a history of ACL injury, performed a sport-specific cutting task while 3D biomechanics were recorded. Over an 8-year follow-up period, 43 of these players sustained a primary ACL injury and 13 players a secondary ACL injury. Cutting technique was described using 36 discrete kinematic variables. To identify different cutting techniques, we employed a K-means clustering algorithm on data subsets involving different numbers of kinematic variables (36, 13 and 5 variables) and different sports (handball, football, and both combined). To assess the impact of the identified cutting technique clusters on ACL injury risk, we compared the proportion of injured players between these clusters using the Fisher-Freeman-Halton Exact test and adjusted rand indices (ARI).

Results: We identified two distinguishable cutting technique clusters in the subset involving both sports and five kinematics variables (average silhouette score, ASS = 0.35). However, these clusters were formed based on sport- or task-related differences (Fisher's p < 0.001, ARI = 0.83) rather than injury-related differences (Fisher's p = 0.417, ARI = 0.00). We also found two cutting technique clusters in the handball (ASS = 0.23) and football (ASS = 0.30) subsets with five kinematic variables. However, none of these clusters appeared to be associated with ACL injury risk (Fisher's p > 0.05, ARI = 0.00).

Conclusion: No safe or risky cutting technique strategies could be discerned among female elite handball and football players. Cluster analysis of cutting technique, using a K-means algorithm, did not prove to be a valuable approach for assessing ACL injury risk in this dataset.

Keywords: ACL; K-means; biomechanics; football; handball; kinematics; kinetics; return to sport.

Figure 1

Flow diagram of the tested players, including their injury status at baseline and follow-up, as well as the number of analyzed players in each group. ACL, anterior cruciate ligament.

Figure 2

Visualization of the two clusters (cluster 0 and cluster 1) as well as the true labels of (a) the injury groups and (b) the sport groups for the core 5 subset. The first two principal components of the principal component analysis, which explain the highest amount of variance in the original data, are plotted against each other. Prev/New ACL group, players with a previous ACL injury who went on to sustain a new secondary ACL injury; Prev ACL group, players with a previous ACL injury only; new ACL group, players without a previous ACL injury who went on to sustain a new primary ACL injury; no ACL group, injury free players.

Figure 3

Animated figures for the handball core 5 subset, illustrating the mean differences in cutting technique between cluster 0 (grey) and cluster 1 (color) in (a) the frontal plane, (b) the sagittal plane, and (c) the horizontal plane. The players nearest to the respective cluster centers, established by the Euclidean norm, were chosen as the basis. Created by Muscle Animations.

Figure 4

Animated figures for the Football Core 5 subset, illustrating the mean differences in cutting technique between Cluster 0 (grey) and Cluster 1 (color) in (a) the frontal plane, (b) the sagittal plane, and (c) the horizontal plane. The players nearest to the respective cluster centers, established by the Euclidean norm, were chosen as the basis. Created by Muscle Animations.