Secondary Logo

Institutional members access full text with Ovid®

Share this article on:

Employing Machine Learning to Predict Lower Extremity Injury in U.S. Special Forces

Connaboy, Chris; Eagle, Shawn R.; Johnson, Caleb; Flanagan, Shawn; Mi, Qi; Nindl, Bradley C.

doi: 10.1249/MSS.0000000000001881
Original Investigation: PDF Only

Introduction Musculoskeletal injury rates in military personnel remain unacceptably high. Application of machine learning algorithms could be useful in multivariate models to predict injury in this population. The purpose of this study was to investigate if interaction between individual predictors, using a decision tree model, could be used to develop a population-specific algorithm of lower extremity injury (LEI) risk.

Methods 140 Air Force Special Forces Operators (27.4±5.0 years, 177.6±5.8 cm, 83.8±8.4 kg) volunteered for this prospective cohort study. Baseline testing included body composition, isokinetic strength, flexibility, aerobic/anaerobic capacity, anaerobic power, and landing biomechanics. To evaluate unilateral landing patterns, subjects jumped off two-feet from a distance (40% of their height) over a hurdle and landing single-legged on a force plate. Medical chart reviews were conducted 365 days post-baseline. Chi-square automatic interaction detection (CHAID) was used, which compares predictor variables to LEI and assigns a population-specific “cut-point” for the most relevant predictors.

Results Twenty-seven percent of Operators (n=38) suffered a LEI. A maximum knee flexion angle difference of 25.1% had the highest association with injury in this population (p=0.006). Operators with >25.1% differences in max knee flexion angle (n=13) suffered LEI at a 69.2% rate. Seven of the 13 Operators with >25.1% difference in max knee flexion angle weighed >81.8 kg, and 100% of those operators suffered LEI (p=0.047; n=7). Only 33% of operators with >25.1% difference in max knee flexion angle that weighed <81.8 kg suffered LEI.

Conclusion This study demonstrated increased risk of LEI over a 365-day period in Operators with greater differences in single leg landing strategies and higher body mass. The CHAID approach can be a powerful tool to analyze population-specific risk factors for injury, along with how those factors may interact to enhance risk.

Neuromuscular Research Laboratory/Warrior Human Performance Research Center, Department of Sports Medicine and Nutrition, University of Pittsburgh, Pittsburgh, PA

Corresponding Author: Chris Connaboy, 3860 S. Water St, Pittsburgh, PA 15203. (p): 412-246-0460. (e):

These results are presented clearly, honestly, and without fabrication, falsification, or inappropriate data manipulation. Further, the results of the present study do not constitute endorsement by ACSM. CC and SE contributed to design of the work, analysis and interpretation of data, manuscript drafting, and final approval. CJ, SF, and BN contributed to interpretation of data, drafting and review of manuscript, and final approval. QM contributed to design of the work, analysis/interpretation of data, revising the work and final approval. This study was funded by Air Force Special Operations Command, #FA8650-12-2-6271. The authors have no competing interests to disclose.

Accepted for publication: 17 December 2018.

© 2019 American College of Sports Medicine