Predicting Upper Quadrant Musculoskeletal Injuries in the Military: A Cohort Study

Abstract

Purpose: This study aimed to identify characteristics and movement-based tests that predict upper quadrant musculoskeletal injury (UQI) in military personnel over a 12-month follow-up. Methods: A prospective observational cohort study of military members (n = 494; 91.9% male) was conducted. Baseline predictors associated with UQI were gathered through surveys and movement-based tests. Survey data included demographic information, injury history, and biosocial factors. Movement-based tests include the following: Y Balance Tests (YBT), Functional Movement Screen, Selective Functional Movement Assessment lumbar multisegmental mobility, modified-modified Schober, side bridge, ankle mobility, modified Sorensen, and passive lumbar extension. Self-reported UQI was collected through monthly online surveys, and 87% completed the follow-up. Univariate associations were determined between potential predictors and UQI. A forward, stepwise logistic regression model was used to identify the best combination of predictors for UQI. Results: Twenty-seven had UQI. Univariate associations existed with three demographic (smoking, >1 previous UQI, baseline upper quadrant function ≤90%), three pain-related (Selective Functional Movement Assessment rotation, side bridge, hurdle step), and six movement-based variables (YBT upper quarter (UQ) superolateral worst score ≤57.75 cm, YBT-UQ composite worst score ≤81.1%, failed shoulder clearance, Sorenson <72.14 s, in-line lunge total score <15, and in-line lunge asymmetry >1). Smoking, baseline upper quadrant function ≤90%, and YBT-UQ composite score ≤81.1% predicted UQI in the logistic regression while controlling for age and sex. Presenting two or more predictors resulted in good specificity (85.6%; odds ratio, 4.8; 95% confidence interval, 2.2-10.8), and at least one predictor resulted in 81.5% sensitivity (odds ratio, 3.2; 95% confidence interval, 1.2-8.7). Conclusions: A modifiable movement-based test (YBT-UQ), perceived upper limb function, and smoking predicted UQI. A specific (two or more) and sensitive (at least one predictor) model could identify persons at higher risk.

Associations between upper quarter Y-balance test performance and sport-related injuries in adolescent handball players

“…A maximum superolateral (SL) reach distance of ≤80.1% arm length (AL) as well as an inferolateral (IL) reach asymmetry of ≥7.75% AL was found to increase the likelihood of a future time-loss injury. Further, specific cut-off values for military personnel ( N = 494, m = 454, f = 40; age: 28.6 ± 6.8 years) were reported by Campbell et al ( 22 ). In this study, the authors identified an SL reach distance of ≤57.75 cm and a composite score (CS) of less than ≤81.1% AL in the YBT-UQ as risk factors for suffering an upper quadrant musculoskeletal injury.…”

Improving Clinical Prediction Model Methods

“…Sample size is a key design feature in prediction model studies; obtaining a suitable sample size is important to mitigate against overfitting. The authors opaquely performed a sample size calculation using Monte Carlo Simulation with 450 participants and ultimately assessed 45 variables (1). Sample size calculations for developing a prediction model should be based on outcome prevalence, the number of parameters, level of shrinkage (measure of overfitting, higher values implying less overfitting), and anticipated overall model fit ( R 2 ) (8).…”

“…Our motivation is not to critique the time and effort undertaken by the authors (1), but rather help improve the model, as prediction models can have direct effect on patient health. We would be interested in collaborating to improve these models and ultimately patient care.…”

Response

“…The backward approach selected the same variables and detected the same number of upper quadrant injuries (12/27, or 44.4%) as our forward selection for a prediction accuracy of 83%. Bootstrap coefficient estimates for selected variables remained as published (2). Finally, we acknowledge that dichotomizing continuous predictors for clinical simplicity can lead to information loss and may limit prediction performance (6).…”

“…Also, a univariate filtering strategy presents limitations (6). Nevertheless, after applying this strategy and eliminating variables presenting collinearity or limited theoretical relevance, we only entered five history/personal characteristics, two pain provocation, and two movement-based variables in the multivariable analysis (2). Overfitting is still possible given the number of predictors considered, albeit much lower than suggested in the letter (1).…”