Prediction of Lower Extremity Musculoskeletal Injuries for Naval Special Warfare Operators

Abstract

Lower extremity (LE) musculoskeletal injuries (MSI) are a common and costly occurrence in US NAVY Sea, Air, and Land (SEALs) Operators. Understanding the risk factors associated with LE MSI is an important step in designing injury prevention programs. PURPOSE: To develop a robust mathematical model to predict LE MSI in SEAL Operators. METHODS: 285 subjects (age: 26 ± 5 yrs, height: 179 ± 7 cm, weight: 85 ± 9 kg) participated in testing, including: LE muscular strength and flexibility; balance; body composition; anaerobic power/capacity; and aerobic capacity (VO2max). Medical charts were reviewed for LE MSI 365 days following laboratory testing. The correlated variable sets were identified using Hierarchical Clustering Analysis (HCA). Important features then were selected from the clusters and modeled with regression trees wherein output (predictions) were interpreted as the probability of injury for each individual. To classify observations, a decision threshold was defined that minimized the false positive rate (FPR) conditional on a true positive rate (TPR) of approximately 90% whenever all available variables were utilized. Individuals with predicted probabilities above this threshold were classified as injured. Variables selected in the final models were chosen in a forward fashion, with individual predictors that reduced the FPR without significantly lowering the TPR added to the model. The procedure stopped when no remaining predictor variables were able to produce a model that outperformed the current iteration. RESULTS: LE MSI rate was 13/285 or 4.5%. Each cluster of feature sets from HCA consisted of variables mostly from the same laboratory test category. The final regression tree model contained knee flexion and left knee extension strength (normalized to body weight), fat-free mass (kg), and hamstring flexibility, as the best predictors (TPR of 92.3% and FPR of 2.9%). CONCLUSION: Knee strength, fat-free mass, and hamstring flexibility were important risk factors identified in the machine learning algorithm that accurately classified SEAL Operators with LE MSKI. Alternative high prediction models also can be created using this modeling framework on different variable sets. Supported by ONR N00014-11-1-0929