Countermovement Jump Force‐Time Mechanics Differentiate ACL Injury Status in Elite Alpine Ski Racers

TL;DR

This study shows that machine learning can identify force-time patterns in jumps that distinguish skiers who have recovered from ACL injuries from healthy athletes.

Contribution

The novel use of machine learning to classify ACL injury status using CMJ force-time metrics in elite alpine ski racers.

Findings

Machine learning models achieved high accuracy in distinguishing ACLR and control athletes using CMJ metrics.

Propulsion phase features were most important for classification, indicating neuromuscular recovery differences.

Models provide a potential tool for tracking rehabilitation progress relative to healthy athletes.

Abstract

Biomechanical assessments of stretch‐shortening cycle (SSC) movements such as the countermovement jump (CMJ) are used to evaluate neuromuscular function in alpine ski racers after anterior cruciate ligament reconstruction (ACLR). However, this analysis yields multiple CMJ force‐time metrics that quantify SSC mechanics, creating challenges for data synthesis, interpretation, and return‐to‐sport decision making. Machine learning (ML) classification algorithms address this problem by determining patterns that distinguish healthy control athletes and athletes recovering from ACLR. ML classification algorithms were trained using CMJ force‐time metrics obtained from healthy control elite alpine ski racers (Control) and skiers tested after ACLR to identify features predictive of group membership. Participants (ACLR: n = 24, Control: n = 42) performed multiple CMJ testing sessions as part of a longitudinal athlete monitoring program (n = 836). ML algorithms (random forest, support vector machine, logistic regression, naïve Bayes, k‐nearest neighbors) were trained using 23 CMJ force‐time features with 5‐fold cross‐validation and evaluated using an independent test dataset. Classification performance was high with balanced accuracies ranging from 0.59 to 0.88 and areas under the receiver operating characteristic curve of 0.63–0.95. Features corresponding to the propulsion phase were most important for differentiating CMJ tests from ACLR and Control athletes. Recovery of neuromuscular function after ACLR may be inferred when the CMJ mechanics of athletes with ACLR become indistinguishable from those of healthy controls. In conclusion, ML classification models may assist interpretation of CMJ force‐time metrics after ACLR by identifying high‐information features related to injury status along with a potential indication of rehabilitation progression relative to healthy control athletes.

Figures