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.

Upper quadrant musculoskeletal injuries (UQI) are common in civilians worldwide and in the Canadian Armed Forces (CAF) as well. The point prevalence of upper extremity and neck disorders in civilians ranges from 1.6% to 53%, and the 12-month prevalence ranges from 2.3% to 41% (¹). This wide range of prevalence is due to variability in the definition of upper extremity disorders with respect to the types and duration of symptoms, physical examination findings, and relationship to overuse, repetitive strain, or work. Upper quadrant musculoskeletal (MSK) conditions have been defined as pain symptoms in the arm, shoulder, or neck that are not due to trauma or underlying systemic disease (²). In 2013–2014, upper limb muscle or joint problems were the third most common chronic condition among CAF Regular Forces personnel, accounting for 11.4% of chronic conditions (³). Furthermore, repetitive strain injuries of the upper quadrant affected 23.6% of CAF regular forces personnel (³), and the incidence of neck pain in CAF aircrew ranged from 53% to 85% (⁴).

MSK injuries result in economic costs in both civilian and military populations. In 2010, MSK injury resulted in a total cost of $8.7 billion in Canada (⁵). In the 2014 CAF Surgeon General’s Reports, MSK injuries were the most common cause of medical releases (⁶). In the Canadian military, MSK injury is the main cause of days off and restricted duties, as well as the most common reason for being unable to deploy (⁷). In the US military, MSK injury is the leading cause of disability and has been reported to result in 2.4 million annual health care visits and US$548 million in direct patient care costs (⁸). The indirect cost of MSK injuries also include loss of training time resulting in impaired combat readiness and defense capabilities (⁹). The psychosocial stress created by disability and potentially losing an occupational role also negatively affects military personnel (⁹).

Many military occupations involve physically demanding tasks for the upper extremity that are associated with an increased risk of UQI (⁸). Prevention strategies for UQI could be critical to reduce health care expenses and morbidity, enhance physical performance, and improve quality of life. However, the multifactorial nature of MSK injury makes prediction challenging, and recent attention has stressed modifiable risk factors such as core stability, balance, and strength, as potentially effective injury prevention initiatives (¹⁰). In particular, athletes and military personnel with suboptimal movement patterns have an increased risk of MSK injury (^11,12), as these aberrant movements expose them to high joint forces and dangerous loading patterns (¹³). Addressing identified movement dysfunction and other modifiable risk factors could potentially reduce the risk of MSK injuries through a prevention program (¹⁴).

To ensure preventive strategies are cost-effectively offered to those at risk, an effective field-accessible screening tool that identifies modifiable risk factors must be established. MSK injury prediction models that combine demographic and movement-based risk factors have been developed (^15–17) but are limited to prediction of general or lower extremity MSK injuries. Few, if any, studies have investigated whether a movement-based screening tool is predictive of UQI specifically.

Our research question was to determine whether self-reported characteristics and clinical movement-based tests would predict the 12-month occurrence of first episode or recurrent noncontact MSK UQI in CAF personnel with no UQI at baseline. Operationally, an upper quadrant MSK injury was defined as the onset of pain in the head, upper back, neck, shoulder, elbow, wrist, or hand unrelated to a traumatic contact event, which caused ≥2/10 pain for at least 3 d, limited ability to work or exercise for 24 h, a self-reported function ≤90%, and resulted in seeking medical care or being placed on medical employment limitations (MEL). Contact trauma (i.e., motor vehicle accident) was specifically excluded from the definition as these events cannot be predicted based on movement-based tests. The rationale for not including contact trauma was that our goal was to identify predictors of MSK injury based on movement features. Traumatic contact injuries, such as being in a car accident, would not provide insight into MSK predictors because of the unpredictable nature of an event that would reasonably injure any person regardless of movement symmetry or fitness. However, we included sudden onset pain such as lifting a heavy object and getting sudden shoulder pain (“strain”), which would be encapsulated in our definition of a “noncontact” injury. We hypothesized that a combination of movement-based clinical tests would predict future noncontact UQI.

METHODS

Study design

This was a prospective observational cohort study of 494 healthy CAF personnel. To determine a sufficient sample size, we referred to Monte Carlo simulations, which showed that based upon considering up to 45 candidate variables in the logistic regression model, we would require a sample size of 450 participants. These simulations suggested that 10 participants were required for each candidate variable when developing the logistic prediction model. Such a sample size would be sufficient when using an α of 0.05 to obtain a power of 0.80 when planning to detect a medium effect size (odd ratio, 2.0) using a one-tailed test.

Volunteer participants were recruited from combat and support roles at a single Canadian Forces Base in 2016. An initial briefing was provided to a large group in a classroom setting, after which volunteers self-identified to approach the researchers to proceed with screening and informed consent (Fig. 1). The study was approved by the Health Research Ethics Board at the University of Alberta (Pro00065519), and participants provided written informed consent.

FIGURE 1: Flow diagram showing recruitment and retention of enrolled participants.

Inclusion criteria were as follows: age 18–50 yr, English fluency, active member of CAF, and free of MSK pain. Exclusion criteria were as follows: an episode of MSK pain ≥2/10 or that limited work or activity and caused the participant to seek health care within the last 3 months, MSK or abdominal surgery within 12 months, MSK injections within 3 months, current prescription pain medications, MEL or a temporary medical category that restricted the ability to participate in exercise as a result of MSK injury within the last 3 months, medical or mental health conditions, a history of fractures (stress or traumatic), pregnancy, a planned deployment, retirement, or travel outside the local area during the follow-up period. An MEL is defined as an administrative constraint on CAF members’ work schedule, tasks, roles, environment(s), or geographical location imposed as a result of a medical condition determined through a formal medical assessment. Temporary medical category is defined as a temporary medical change for a CAF member providing information about a member’s employability and deployability.

Baseline data collection

Potential predictors of UQI were identified based on previous research or theoretical rationale. Self-reported surveys were used to collect participant age, gender, smoking history, height, weight, trade, years in CAF, physical activity level, baseline upper quadrant function on a scale from 0 (unable to function) to 100 (fully functional) (¹⁸), history of MEL, and history of prior MSK injury, using the same operational definition of the main outcome (MSK UQI).

Participants completed 17 tests during a single baseline testing session lasting approximately 60 min. All participants were tested over a 6-wk period by protocol-trained staff including one Selective Functional Movement Assessment (SFMA)–certified physiotherapist (PT), six PTs, two PT assistants, eight Functional Movement Screen (FMS)– and Y Balance Test (YBT)–certified fitness and sport instructors, and a physical exercise specialist.

The clinical test battery included the upper and lower quarter YBT (YBT-UQ and YBT-LQ) (¹⁹), the 100-point FMS (²⁰), SFMA lumbar multisegmental mobility (²¹), modified–modified Schober (²²), double-leg side bridge (²³), ankle mobility (²⁴), modified Sorensen (²⁵), and passive lumbar extension (²⁶). The tests and published reliability are presented in Supplemental Digital Content (see Tables S1 and S2, Supplemental Digital Content, supplemental tables, https://links.lww.com/MSS/C432) (^36–40).

MSK injury surveillance

Over the 12-month follow-up, participants completed monthly online questionnaires about MSK injuries sustained over the previous month and their impact on work and function. Participants received up to two email reminders and one phone call for each monthly questionnaire. The questionnaire was adapted for the CAF from the survey developed by Teyhen et al. (¹⁵) for US rangers.

Data analysis

Univariate associations between potential predictors and UQI were examined to narrow the number of predictors of interest. For continuous variables, independent t-tests were used to detect differences in means between those with and without UQI. Continuous variables with P < 0.20 were retained for further analysis. This liberal cutoff was used to protect against type I errors in this early stage of analysis as per Teyhen et al. (¹⁵). Receiver operator characteristic (ROC) curves were used to identify meaningful cut points to dichotomize continuous predictors. Cut points were determined based on the largest area under the ROC curve, indicating that they had the best sensitivity and specificity (¹⁵). Categorical predictors with an odds ratio (OR) ≥2.0 and a Pearson’s χ²P < 0.20 or Fisher’s exact P < 0.20 were retained for further analysis. Fisher’s exact P value was used if fewer than five observations were observed in a cell from the 2 × 2 table between predictor and UQI status.

A forward, stepwise logistic regression model was used to identify the best combination of predictors for UQI while controlling for age and sex. Variables were entered into the equation if P < 0.05 and removed if P > 0.10. Weaker predictors from variables with high multicollinearity were excluded. Final model estimates were obtained by entering the selected variables into the logistic regression analysis and using bootstrapping generating 1000 sample from the existing data using simple random selection. Given that most data were from males in our population, we also repeated the logistic regression analysis using only the male data while still controlling for age.

We also estimated prediction accuracy statistics by producing 2 × 2 tables, with columns reflecting if the participants experienced a UQI or not, and rows indicating if the participants presented one or more predictors or no predictors from the logistic model developed earlier. We further compared participants presenting two or more or less than two predictors and those presenting all three predictors selected in our regression model to those presenting less than three. The prediction accuracy statistics estimated were sensitivity, specificity, positive and negative likelihood ratios, positive and negative predictive values, and OR (²⁷). Data analysis was conducted using the Statistical Package for Social Sciences 24.

RESULTS

Sample characteristics

A total of 494 CAF personnel volunteers meeting the inclusion criteria provided informed consent and were enrolled in the study including 454 (91.9%) males and 40 (8.1%) females. Five had missing baseline data and were excluded from the analyses. Over the 12-month follow-up, participants were considered lost to follow-up if they missed one UQI monthly survey, which occurred for 64 participants (13%); these participants were excluded from analyses (Fig. 1: participant was posted elsewhere (9.8%), deployed (1.6%), retired (8.2%), opted out (6.6%), or did not respond to reminders (73.8%)). Participants who met the criteria for UQI who did not complete subsequent questionnaires were included in the analyses.

Mean age was 28.6 (±6.8) yr. Most (416 (84.2%)) were junior noncommissioned officers, 40 (8.1%) were senior noncommissioned officers, and 38 (7.7%) were officers. Two hundred and fifteen (43.5%) worked in a combat role compared with 279 (56.5%) support-based.

One hundred and twenty-seven (25.7%) participants were placed on an MEL because of an MSK injury over the 3 to 12 months preceding baseline with durations as follows: <2 wk (n = 49; 38.6%), 2–8 wk (n = 47; 37.0%), 8 wk to 6 months (n = 12, 9.4%), and >6 months (n = 19; 15.0%). There were 290 (58.8%) nonsmokers, 97 (19.7%) who had smoked over 100 cigarettes in their lifetime, and 106 (21.5%) were current smokers.

Participants with missing follow-up data were compared with those with complete follow-up data (see Table S3, Supplemental Digital Content, supplemental tables, https://links.lww.com/MSS/C432) (^36–40). Participants with missing data were significantly younger (25.5 ± 4.2 vs 29.0 ± 7.0 yr, P < 0.01) and included only males, proportionally more junior noncommissioned officers, more combat roles, and more smokers. Those with and without follow-up data did not differ significantly for their previous history of UQI or their status on the dichotomized movement-based predictors UQI, except that proportionately fewer participants without follow-up data had YBT-UQ superolateral worst scores lower than 57.75 cm.

Frequency of upper quadrant injuries

There were 27 participants with UQI over the 12-month follow-up. Of these, there were 9 (33.3%) neck injuries, 14 (51.9%) shoulder injuries, and 8 (29.6%) elbow, wrist, or hand injuries. There were four participants with more than one injury. Only two females had an UQI.

Variables predicting UQI identified during univariate analysis

Table 1 presents the results of the univariate analysis for continuous potential predictor variables. Table 2 presents statistically significant results for dichotomous potential predictor frequencies for the entire group as well as those reporting an UQI (see Table S4, Supplemental Digital Content, supplemental tables, for all results, including subgroup analysis for males only, https://links.lww.com/MSS/C432) (^36–40). Significant results (per P < 0.20 and OR >2.0) included three demographic variables (smoking status (OR, 2.33; P = 0.03), >1 previous UQI (OR, 2.27; χ² = 3.89; P = 0.05), and baseline upper quadrant function ≤90% (OR, 3.34; χ² = 6.04; P = 0.03)), six movement-based test variables (YBT-UQ superolateral worst score ≤57.75 cm (OR, 2.08; χ² = 3.56; P = 0.05), YBT-UQ composite worst score ≤81.1% (OR, 2.66; χ² = 6.33; P = 0.01), failed shoulder clearance (OR, 2.17; χ² = 1.93; P = 0.14), in-line lunge total score <15 (OR, 2.11; χ² = 3.48; P = 0.06), in-line lunge asymmetry >1 (OR, 2.12; χ² = 3.58; P = 0.05), Sorensen time <72.14 (OR, 2.26; χ² = 4.28; P = 0.04)), and three movement-provoked pain variables (SFMA rotation (OR, 45.7; P = 0.06), side bridge (OR, 2.43; P = 0.14), hurdle step (OR, 15.2; P = 0.12)).

TABLE 1 - Means and SD for the whole group and for those with and without UQI for the continuous potential predictor variables, with t-test of differences between injured and not injured and ROC analysis results for promising predictors.

Descriptive Variable	Overall, Mean ± SD	With UQI, Mean ± SD	Without UQI, Mean ± SD	T-test (P Value)	Cutoff Indicating Risk of Injury	ROC Curve AUC (P Value)
Age (yr)	28.6 ± 6.8	30.0 ± 8.3	29.0 ± 6.9	−0.71 (0.48)
YBT-UQ medial (cm)
Asymmetry	3.5 ± 2.8	3.9 ± 3.1	3.5 ± 2.8	−0.75 (0.45)
Worst	90.5 ± 8.0	90.5 ± 8.1	90.4 ± 8.0	−0.09 (0.93)
YBT-UQ inferiolateral (cm)
Asymmetry	5.1 ± 4.1	5.5 ± 5.0	5.1 ± 4.0	−0.54 (0.59)
Worst	81.6 ± 9.9	80.4 ± 10.4	81.3 ± 10.0	0.43 (0.67)
YBT-UQ superiolateral (cm)
Asymmetry	3.5 ± 2.8	3.9 ± 3.1	3.5 ± 2.8	−0.75 (0.45)
Worst	61.1 ± 10.0	56.9 ± 11.5	61.0 ± 10.1	2.02 (0.05)	≤57.75	0.60 (0.08)
YBT-UQ composite (%)
Worst	86.3 ± 8.1	83.7 ± 8.8	86.3 ± 8.1	1.59 (0.11)	≤81.10	0.59 (0.13)
YBT-LQ anterior (cm)
Asymmetry	3.3 ± 2.8	3.2 ± 2.1	3.3 ± 2.9	0.11 (0.91)
Worst	63.3 ± 8.2	62.1 ± 8.3	63.2 ± 8.3	0.67 (0.51)
YBT-LQ posteriomedial (cm)
Asymmetry	3.8 ± 3.1	4.0 ± 3.3	3.9 ± 3.2	−0.17 (0.87)
Worst	105.4 ± 9.0	106.4 ± 11.8	105.9 ± 9.0	−0.59 (0.56)
YBT-LQ posteriolateral (cm)
Asymmetry	4.3 ± 3.8	4.7 ± 3.8	4.4 ± 3.9	−0.48 (0.64)
Worst	101.6 ± 9.8	101.6 ± 11.8	101.5 ± 9.7	−0.09 (0.93)
YBT-LQ composite (%)
Worst	100.6 ± 21.9	103.3 ± 36.9	100.5 ± 20.9	−0.63 (0.53)
Side bridge (s)
Asymmetry	14.2 ± 14.3	16.7 ± 15.8	14.1 ± 13.6	−0.95 (0.34)
Worst	71.5 ± 31.3	69.1 ± 32.0	71.4 ± 30.5	0.38 (0.70)
Deep overhead squat	6.7 ± 4.4	7.0 ± 4.9	6.6 ± 4.4	−0.40 (0.69)
Hurdle step
Asymmetry	0.8 ± 1.4	0.9 ± 1.1	0.8 ± 1.4	−0.41 (0.68)
Worst	6.3 ± 2.1	6.2 ± 1.9	6.2 ± 2.1	0.06 (0.95)
Total	13.4 ± 3.6	13.4 ± 3.8	13.3 ± 3.6	−0.09 (0.93)
FMS total score (100 points)	58.1 ± 14.1	56.6 ± 14.5	57.9 ± 14.4	0.48 (0.64)

Variables potentially predictive of UQI are in bold format (P < 0.20). Cutoff values using ROC curves were only calculated if t-test P < 0.20.

TABLE 2 - Statistically significant results from the univariate analyses for dichotomous potential predictor variable frequencies observed for the whole group (overall) and for those with and without UQI, with χ² tests and OR reflecting the association between predictors and UQI event.

Descriptive Variable	Category	Overall, Frequency (%)	Without UQI, Frequency (%)	With UQI, Frequency (%)	χ2 (P Value)	OR
Smoking status	Nonsmoker (0)	260 (60.6)	249 (58.0)	11 (2.6)	P = 0.03 a	2.33
	Smokers (1 + 2)	169 (39.4)	153 (35.7)	16 (3.7)
>1 previous UQI	No	336 (78.1)	319 (74.2)	17 (4.0)	3.89 (0.05)	2.27
	Yes	94 (21.9)	84 (19.5)	10 (2.3)
Baseline UQ function ≤90%	No	391 (90.9)	370 (86)	21 (4.9)	6.04 (0.03)	3.34
	Yes	39 (9.1)	33 (7.7)	6 (1.4)
YBT-UQ superolateral worst ≤57.75 cm	No	264 (61.5)	252 (58.7)	12 (2.8)	3.56 (0.05)	2.08
	Yes	165 (38.5)	150 (35.0)	15 (3.5)
YBT-UQ composite worst ≤81.10%	No	312 (72.7)	298 (69.5)	14 (3.3)	6.33 (0.01)	2.66
	Yes	117 (27.3)	104 (24.2)	13 (3.0)
Shoulder clearance	Negative	381 (89.4)	359 (84.3)	22 (5.2)	1.93 (0.14)	2.17
	Positive	45 (10.6)	40 (9.4)	5 (1.2)
In-line lunge total score <15	No	320 (74.4)	304 (70.7)	16 (3.7)	3.48 (0.06)	2.11
	Yes	110 (25.6)	99 (23)	11 (2.8)
In-line lunge asymmetry >1	No	250 (58.1)	239 (55.6)	11 (2.6)	3.58 (0.05)	2.12
	Yes	180 (41.9)	164 (38.1)	16 (3.7)
Sorensen time <72.14 b	No	312 (72.7)	297 (69.2)	15 (3.5)	4.28 (0.04)	2.26
	Yes	117 (27.3)	105 (24.5)	12 (2.8)
UQP with SFMA rotation b	No	429 (99.8)	403 (93.7)	26 (6.0)	P = 0.06 a	45.72
	Yes	1 (0.2)	0 (0.0)	1 (0.2)
UQP with side bridge	No	397 (92.3)	374 (87.0)	23 (5.3)	P = 0.14 a	2.43
	Yes	33 (7.7)	29 (6.7)	4 (0.9)
UQP with hurdle step b	No	428 (99.5)	402 (93.5)	26 (6.0)	P = 0.12 a	15.21
	Yes	2 (0.5)	1 (0.2)	1 (0.2)
	Yes	180 (41.9)	164 (38.1)	16 (3.7)

All variables of potential significance (P < 0.20 and OR ≥2.0) are included (see Table S3, Supplementary Digital Content, https://links.lww.com/MSS/C432, for the full list of tested variables) (^36–40).

^aFisher’s exact P value.

^bNot included in the logistic regression model for collinearity or lack of theoretical relevance to upper quadrant injury.

Logistic regression analysis

From the 494 participants, 425 (86%) were included in the regression analysis and 69 (14%) were excluded because of 5 missing baseline data and 64 lost to follow-up. The variables predicting a UQI in the 12-month follow-up period were smoking status (OR, 2.31; 95% confidence interval [CI], 1.03–5.20), baseline upper quadrant function ≤90% (OR, 3.13; 95% CI, 1.15–8.55), and YBT-UQ composite worst score ≤81.1% (OR, 2.94; 95% CI, 1.31–6.61). Using a cutoff probability of 0.1 to classify cases, these three variables accurately predicted 12 (44.4%) of 27 upper quadrant injuries and accurately predicted 344 (85.8%) of the 398 participants would be uninjured (Cox R² = 0.034; Nagekerke = 0.092). The UQI logistic regression prediction model equation was as follows, where numbers in brackets represent the 95% CI and P value estimates from bootstrapping for each β coefficient:

$Z=-{2.064}_{\left[-\mathrm{11.785;}-\mathrm{0.472,}\mathit{p}=0.01\right]}–\left({0.05}_{\left[-\mathrm{0.072;}\mathrm{0.038,}p=0.846\right]}\times \mathrm{age}\right)$$–\left({0.188}_{\left[-\mathrm{0.18.818;}\mathrm{1.070,}p=0.695\right]}\times \text{female}\mathrm{sex}\right)+{1.165}_{\left[-\mathrm{0.361;}\mathrm{2.193,}p=0.011\right]}\times \text{Baseline Upper Quadrant Function}\le 90\mathrm{\%})+\left({1.111}_{\left[\mathrm{0.201;}\mathrm{1.977,}p=0.007\right]}\times \mathrm{YBT}‐\mathrm{UQ}\text{Composite Worst}\le 81.1\mathrm{\%}\right)+\left({0.831}_{\left[-\mathrm{0.47;}\mathrm{1.898,}p=0.052\right]}\times \text{Smoking status}\right).$

The logistic regression analysis using only the male data but still controlling for age found similar predictors, except that smoking status was not retained in the model developed in males only. The P value for smoking status was 0.073 and did not meet the threshold of 0.05 for entry using males only but was included in the model including data from both sexes and controlling for sex (not significant) and age (not significant).

Using the data from male and female participants, a predictive model was developed, determining the prediction value of having an increasing number of the three variables from the model above (Table 3). The specificity was highest when all three predictors were present (99.5%; OR, 7.67; 95% CI, 0.67–87.4). Sensitivity was maximized when one or more predictors were present (81.5%; OR, 3.24; 95% CI, 1.20–8.72). Sensitivity and specificity were both reduced when two out of the three predictors were present, but specificity remained acceptable (sensitivity, 44.4%; specificity, 85.6%; OR, 4.83; 95% CI, 2.15–10.84). Overall, 253 members had one or more predictors, 69 had two predictors, and only 3 had three predictors.

TABLE 3 - Sensitivity, specificity, positive and negative likelihood ratios, predictive values, and OR when presenting one, two, or three or more of the predictors of UQI retained from the logistic regression.

No. Predictors	Participants (%)	Sensitivity (%)	Specificity (%)	PLR	NLR	PPV	NPV	OR (95% CI)
≥1	59.1	81.5	42.4	1.41	0.44	0.09	0.97	3.24 (1.20–8.72)
≥2	16.1	44.4	85.6	3.13	0.65	0.17	0.96	4.83 (2.15–10.84)
3	0.7	3.7	99.5	7.43	0.97	0.33	0.94	7.67 (0.67–87.4)

NLR = negative likelihood ratio; NPV = negative predictive value; PLR = positive likelihood ratio; PPV = positive predictive value.

DISCUSSION

Prediction model

Our results support the hypothesis that movement-based clinical tests in combination with demographic characteristics provided a better prediction of military personnel at higher risk of UQI compared with using individual risk factors alone. A prediction model for UQI in a military population was found, combining three modifiable predictors: YBT-UQ composite score ≤81.1%, smoking, and baseline perceived upper quadrant function ≤90%. Participants with two or more predictors were 4.83 (95% CI, 2.15–10.84) times more likely to have an UQI. Teyhen et al. (¹⁵) also developed a model combining three modifiable and nonmodifiable risk factors that was better at predicting MSK overuse injury in the military than individual risk factors alone. A model combining these predictors was able to identify 96% of noninjured soldiers and 31% of injured soldiers correctly (¹⁵). Lehr et al. (²⁸) developed an algorithm incorporating FMS, lower quarter YBT, and injury history that accurately predicted MSK lower extremity injuries in collegiate athletes (relative risk, 3.4; 95% CI, 2–6). Lisman et al. (¹⁷) showed that military personnel with a slow 3-mile run time and a low FMS composite score (≤14) were 4.2 times more likely to sustain an MSK injury. In contrast, our study did not find the FMS to accurately predict UQI, possibly because only two of seven FMS tests (shoulder mobility, trunk stability) are specific to the upper quadrant. To our knowledge, our study is the first to develop a multivariable prediction model focused solely on predicting noncontact UQI in an uninjured population.

Our results are in agreement with Cosio-Lima et al.’s (²⁹) finding that lower composite scores on the YBT-UQ related to higher general training injury risk (P = 0.03) in the military. Our analysis found that individuals scoring ≤81.1% have a higher risk of UQI (OR, 2.66). Such participants may lack flexibility or stability, which could predispose to injury (²⁹). The identification of individuals at risk of UQI could allow for the investigation of a training program to address modifiable risk factors, which may improve movement scores and reduce the risk of injury. For instance, Kiesel et al. (¹⁴) showed that an individualized training program based on the initial FMS performance improved the FMS score and reduced specific movement asymmetry in professional football players. Further work needs to be done to determine if the improvement in movement symmetry then decreases the risk of injury. Encouragingly, other research has demonstrated that targeted training can improve scores on the YBT and reduce injury rates in collegiate athletes (³⁰). Peate et al. (³¹) also showed that a core stability program providing guidance and practice on functional movement and core muscle strength exercises reduced the number of upper extremity injuries (P = 0.03) in active firefighters. A training program to improve upper extremity flexibility and stability and improve scores on the YBT-UQ may therefore lower the UQI risk and should be a focus of future studies.

We looked at the value of simply counting the number of predictors identified in our regression model to predict future UQI in order to identify a cut point that could quickly identify candidates to investigate the effect of a prevention program. When counting predictors, it is important to note, however, that specificity (ability to rule in those at risk to get future UQI with this combination of predictors) does go up with each additional predictor, but that sensitivity (ability to rule out the risk of a future injury in those without the given combination of predictors) drops dramatically as the predictor count goes up. Therefore, combining a number of predictors as an additive score may not be a valuable mechanism for predicting who will get injured. Instead, we noted that participants with two of the three predictors (69 of 425 in our analysis) had a greater risk of UQI with a high specificity (85.6%) and may benefit from a preventive training program. In contrast, when one or more predictors were present, the model lacked specificity (42.4%) and a cost/benefit analysis may not justify implementing a prevention strategy for the higher number of members identified. Participants in our study with none of the three predictors in our model were unlikely to experience a future UQI. Identification of these individuals not requiring further intervention would allow prevention resources to be allocated to those at higher risk. Thus, the clinical implications of this analysis approach are that prevention efforts could be investigated targeting those with two or more of the predictors, and those presenting with none of the predictors would have a very low risk of future UQI. Those with one predictor represent a group where the likelihood of future UQI is less clear.

It was not surprising that self-reported perceived baseline upper quadrant function of ≤90% was a predictor. Previous or recurrent injury, and employment limitations due to injury are known risk factors for MSK injury in the military (^15,17). These factors and/or low confidence in upper quadrant abilities would be reflected in a lower baseline function rating and would logically predispose an individual to future UQI; however, further studies are needed to address the validity of this assumption.

As expected, smoking was included in the prediction model. Smoking is associated with poor functional outcomes regardless of the severity or nature of injury (³²). Smoking has been associated with neck pain, shoulder dysfunction, and rotator cuff tears (³³). Our results concur with Teyhen et al., who found that current smoking (OR, 6.7; 95% CI, 1.7–26.4) and having smoked at least 100 cigarettes (OR, 2.0; 95% CI, 0.9–4.3) increase the risk of overuse MSK injury in US army rangers (¹⁵). In our study, smoking was defined as current regular smoking or having smoked at least 100 cigarettes (OR, 2.33) to include those who do not classify themselves as smokers but who smoke irregularly. Smoking is a modifiable risk factor with many potential behavioral and pharmacological strategies for smoking cessation. The exact mechanism of the association between smoking and future UQI needs further investigation.

Study limitations

There was a low frequency of participants with UQI over the 12-month period (n = 27). The identification of injuries was based solely on self-reported surveys and may have resulted in response or recall bias. We attempted to avoid response bias by specifically defining what constituted an injury in simple language. We attempted to avoid recall bias by surveying for injury at monthly intervals. The anonymized self-reported method of identifying injuries (rather than surveying of medical records) was chosen intentionally to encourage reporting without fear of workplace restrictions. Only 79.8% of CAF regular forces personnel with repetitive strain injuries seek medical care (³). Carragee and Cohen (³⁴) showed that military personnel often deny low back pain on medical records (97%), despite there being a high incidence of low back pain detected using short-interval questionnaires. Although there was no verification from healthcare providers or medical records that the perceived injury or pain was, in fact, MSK in nature, at least 50% of upper limb pain is due to common MSK conditions (³⁵) and therefore was most likely related to an MSK condition in this healthy population.

Our proposed prediction model should be validated in a new cohort. Nevertheless, estimates obtained using bootstrapping were very similar to those obtained without, offering some validation evidence using an internal validation approach. Three or less participants reported an UQI and experienced upper quadrant pain during SFMA rotation or hurdle step tests, which resulted in large OR values (45.7 and 15.2, respectively) with wide variability. The same data were used to determine the cut points using ROC curves and to test the cut points in the prediction model. One is more likely to predict injury by chance with this approach than when using cut points determined with different data (¹¹). We also chose to dichotomize our outcome variables for clinical ease of use and interpretation; however, a model with continuous predictors could provide further insight and could be an avenue of future work. Furthermore, although our model controlled for sex, only two females had an UQI so results presented are most applicable to a male military cohort, and further investigation into sex differences is warranted.

Future research should focus on identifying individuals at risk and employing risk mitigation strategies to determine if they can reduce UQI rates. A survey showed that 40.8% of Canadian regular forces personnel believe that incorporating injury prevention in their training would positively impact their health (³). Military personnel identified at risk of UQI could have increased motivation to incorporate injury prevention interventions into training.

CONCLUSIONS

We developed a novel multivariable predictive model for UQI including baseline predictors of YBT-UQ composite score ≤81.1%, upper quadrant function ≤90%, and smoking status that can identify military personnel at risk of future UQI. Participants with two or more predictors have a greater risk of UQI with a high specificity (85.6%). This combination produced a specific model that may be valuable for informing a screening strategy that can be done in the field with minimal resources to predict an important number of disabling UQI. Utilizing the predictive model to identify military personnel at risk for UQI could facilitate the development of an injury prevention program to address the identified deficits, reduce the risk of UQI, and thereby reduce military costs and improve combat readiness.