Work-Related Risk Factors for Rotator Cuff Syndrome in a Prospective Study of Manufacturing and Healthcare Workers

From 2002 to 2005, complete, individual on-site assessments were conducted at baseline and annually for up to 2 years to administer questionnaires, conduct biomechanical exposure assessments, videotape all job tasks, and perform clinical assessments (Burt et al., 2013). Our investigators used a questionnaire to collect information on personal characteristics, health history, work history, work environment (including psychosocial factors such as job strain; Hurrell & McLaney, 1988; Karasek et al., 1998; McNair et al., 1971; Radloff, 1977), physical activities outside of work, and musculoskeletal symptoms (neck, hand, wrist, elbow, shoulder; Kuorinka et al., 1987). Trained analysts (e.g., ergonomists, industrial hygienists) conducted biomechanical exposure assessments that included force ratings and determining vibration exposure (yes/no) by job task (Borg, 1982). Each job task, as defined by Bao et al. (2009), was video recorded at 30 frames/s from two angles (17 min for single task jobs and 12 min per task for multi-task jobs). Usually, the cameras were positioned at right angles to each other to allow clear views of the subject’s sagittal and transverse planes. Biomechanical exposures were also reassessed after baseline site visits every 6 months if an individual participant changed job titles, production lines, or departments (Burt et al., 2013).

To diagnose work-related RCS (also called rotator cuff tendinosis, tendinopathy, or disease; subacromial pain syndrome), physical therapists conducted clinical examinations of both arms and hands on all participants at baseline with annual follow-up for up to 2 years. Our case definition for dominant arm RCS case included a combination of (1) shoulder pain during a clinical examination induced by at least one provocative test (Sluiter et al., 2001); and (2) meeting both self-reported shoulder symptom criteria: (a) in the past 12 months, they experienced any shoulder symptoms, and (b) any shoulder pain in the past 7 days (Supplemental Figure 3). Prevalent cases—met all case criteria at baseline—were excluded (censored) from the analyses for this study. Participants who met only criteria #1 or #2 were included in the study.

Univariate and multivariable Cox proportional hazard models were used to analyze the incidence of RCS. The models were stratified within the Cox model by year of follow up (years 1 and 2). Baseline values (year 0) were used for the demographic and psychosocial variables.

Missing exposure variable values were replaced with the first nonmissing value from a previous visit. The final exposure variables used in the analysis represent a lag of 1 year. The final exposure variables were analyzed as continuous and categorical variables (tertiles/thirds). To improve model stability, tertile cut-points were determined using baseline values among cases.

Shoulder pain in the last 7 days at baseline was the only demographic, psychosocial, or biomechanical characteristic excluded as a potential confounder because it is on the causal pathway between exposure and RCS. Based on subject matter knowledge, we included all other demographic/psychosocial factors in this study based on their potential to be a confounder between biomechanical exposures and RCS. Likewise, we considered biomechanical exposures with different primary domains could be confounded by other biomechanical exposures. The selection of confounders for the multivariate models began by selecting potential demographic, psychosocial, and exposure variables associated with time to event in univariate models with p < .20. A form of backward elimination was used to determine the confounders for each model (Harris-Adamson et al., 2015). All potential confounders were entered in a model and were retained if their removal resulted in a change of 10% or more in the regression coefficient of the exposure variable of interest. Potential confounders were tested in order from largest to smallest univariate p value. To avoid overfitting the models for a given exposure variable, exposure variables in the same primary domain (force, repetition, duty cycle, vibration, and posture) were not considered as a confounder. We reduced multivariable model instability attributed to highly skewed distribution of forceful repetition values by using a categorical form of forceful repetition rate (median split). Otherwise, only continuous exposure variables were used in the selection of confounders.

Interactions between the categorical posture variables and continuous, nonposture exposure variables were tested in multivariable models to determine if the effect of a nonposture exposure variable varied by categories of a posture variable. The main effects of one categorical and one continuous variable as well as their interaction were included in a model. We used the final multivariable models for the nonposture exposure variables to analyze interactions.

We used SAS (Release 9.4, SAS Institute, Inc., Cary, North Carolina) to conduct all calculations. The assumptions of the Cox proportional hazard model were tested in univariate models.

We excluded 34 participants from the longitudinal analyses who met the case definition criteria for RCS at baseline, resulting in 451 eligible cohort participants. We lost 58 participants due to lack of health outcome follow-up data, leaving 393 workers in the cohort (Table 1, Supplemental Figure 4). We observed 39 incident cases of RCS in 694 person-years (incidence rate = 5.62 per 100 person-years, 95% CI [4.12, 7.69]). Baseline demographic, medical history, and psychosocial characteristics are presented by case status in Table 1, and descriptive results for baseline biomechanical exposures are presented in Table 2. Compared with noncases, incident cases were older (difference in means = 5.3 years, CI [1.8, 8.8]), were more likely to have a BMI ≥30 kg/m² (X² = 5.1, p = .02), had a higher BMI (difference in means = 2.3 kg/m², CI [0.2, 4.4]), and fewer had completed high school (X² = 6.0, p = .01). Mean years worked at employer for the cohort was 9.9 (SD = 8.1) and was not meaningfully different for cases compared with noncases. Compared with noncases, the 77 participants who had shoulder pain at baseline, but were not classified as a prevalent case, were more likely to become an incident case eventually (p < .001). No significant baseline comparisons were observed for any psychosocial factor measured. Exposures at baseline were not significantly different when stratified by baseline symptom status (data not shown).

The twelve potential confounders (p < .20) are presented in Table 3 in ascending order by p value. Age in years (HR = 1.05, CI [1.02, 1.08], p < .01), body mass index (BMI, HR = 1.05, CI [1.00, 1.10], p = .04), and TWA forceful repetition rate (HR = 1.06, CI [1.00, 1.13], p = .04) were all significantly associated with increased incidence of RCS. To improve model stability, we controlled for forceful repetition rate using a two-category model with a cut-point at the median, 2.8 repetitions per minute (HR = 1.70, CI [0.87, 3.31], p = .12). Having at least a high school education was associated with a significant decrease in incident RCS (HR = .27, CI [0.10, 0.75], p < .001). Participants who had shoulder pain at baseline, but did not meet the RCS case definition, had an increased risk of developing RCS (HR = 3.91, CI [2.01, 7.63], p < .001). Univariable results are presented in Supplemental Table S1 for demographic and psychosocial covariates where p ≥ .20, and Supplemental Table S2 for biomechanical univariable.

Survival analysis results for adjusted models of the association between biomechanical exposures are presented in Table 4. Five variables were confounders in >69% of models—Site (1, 2, or 3), forceful repetition rate category, supervisor support, age (years), and BMI (kg/m²). For most results by tertile, associations were attenuated toward the null in the highest tertile. We failed to observe any meaningful associations of increased risk of RCS among the single biomechanical exposures by tertile (Table 4) and linear (trend) effects (Supplemental Table 2). It appears that there may have been a difference in risk by tertile for forceful repetition rate (p = .06), where the second tertile was less hazardous than the first or third tertiles. Any risk associated with forceful repetition rate was unlikely to be linear (linear effect: HR = 1.06, CI [0.98, 1.14]). For linear effects, adjusted results did not vary substantially compared with unadjusted HRs (Supplemental Table 2).

Figure 1 presents HRs for interactions between tertiles of upper arm flexion and abduction posture exposures (% time) and analyst rated force (charts A–D), total repetition rate (charts E–H), and forceful repetition rate (charts I–L). When working with the upper arm abducted ≥30° for 12% time to 21% time, each unit increase in total repetition rate and forceful repetition rate was associated with a statistically significant increased risk of incident RCS (total repetition: HR = 1.11, CI [1.04, 1.34]; forceful repetition: HR = 1.18, CI [1.04, 1.34]). Forceful repetition was also associated with increased risk of RCS when working with the upper arm abducted ≥60° for ≥5% time (HR = 1.16, CI [1.04, 1.29]) or flexed ≥45° for ≥29% time (HR = 1.11, CI [1.01, 1.22]). Overall, significant interactions between posture variables and biomechanical variables from other domains were rare; we found no significant increased risk of RCS for interactions between upper arm posture tertiles and force ratings, duty cycle, or vibration exposure measures. Numeric interaction results for all variables are available in Supplemental Tables 3–6.

Positive associations with incident RCS for forceful repetition rate alone approached statistically significance (p < .10), but when combined with medium or high upper arm elevation exposure groups, the risk increased and was statistically significant. In general, as upper arm elevation angles increased, participants could spend less time in those postures (lower tertile cut-points) without increasing their risk of developing RCS for each unit increase in forceful repetition rate. Working with upper abduction ≥60° for as little as 5% time was associated with increased risk when combined with forceful, repetitive work. Although these results were consistent with two cross-sectional studies (Frost et al., 2002; Silverstein et al., 2008), most similar studies included in recent systematic reviews have not included a specific measure of forceful repetition (Dalbøge et al., 2014; Mayer et al., 2012; van der Molen et al., 2017). The Danish JEM lacks a forceful repetition component; their shoulder load variable includes forceful exertion rating, upper-arm elevation above 90° (hours/day), and repetitive work (hours/day). Despite the differences between our forceful repetition variable and the Danish JEM shoulder load variable (includes upper arm elevation >90°), our results for exposure to forceful repetition and interactions with our posture variables seem consistent with the Danish results for combinations of exposures (Dalbøge et al., 2018, 2018; Svendsen et al., 2013). Our results also may support recent work conducted by Gallagher and colleagues, who hypothesize that cumulative damage to musculoskeletal soft tissues can be explained by applying fatigue failure theory to understand cumulative soft tissue damage involved in MSDs (Gallagher & Schall, 2017; Gallagher et al., 2018). If true, forceful repetition could be an important biomechanical risk factor for any MSD, including RCS. More research is needed in this area; however, our results may support their hypothesis (Gallagher & Schall, 2017; Gallagher et al., 2018).

In contrast to this study, other studies have demonstrated that working with extreme upper arm postures (e.g., ≥60°) is a risk factor for shoulder MSDs—especially ≥90° (Bernard, 1997; Dalbøge et al., 2014; Mayer et al., 2012; Seidler et al., 2020; Svendsen et al., 2013). We propose that the lack of positive findings for extreme upper arm postures in this study may be attributed to three factors: (1) Healthy worker survival bias may be a limitation of our study that could account for the unexpected upper arm elevation results; we will discuss this topic in the Limitations section. (2) Compared with most large, Scandinavian registry studies (Dalbøge et al., 2014, 2018, 2020; Svendsen et al., 2013), where exposures are accumulated across a minimum of 5 years up to a person’s working lifetime, our follow-up time was limited to 2 years (Dalbøge et al., 2014, 2018; Svendsen et al., 2013). At low exposure magnitudes, it may take many years of cumulative exposure to increase risk. This is consistent with a recent meta-analysis that calculated risk of developing specific shoulder diseases by calculating cumulative exposures for a number of studies and then conducting a meta-analysis that found an increase in risk after 1000 cumulative hours of work above shoulder level (Seidler et al., 2020). In contrast, recent analyses by Dalbøge et al. (2018, 2020) found that compared with people who worked above shoulder level for no more 2.25 min/day, the risk of developing one of several rotator cuff related shoulder diagnoses was elevated when a person’s cumulative exposure was at least 2.25 min/day for 1 year working above shoulder level (Dalbøge et al., 2020). (3) Due to sparse numbers of cases per tertile, our analysis was sensitive to relatively minor differences in category cut-points. For example, in early analyses, we found a significant association between spending ≥4.3% time in ≥90° flexion; however, after the cut-point was changed to 3.5% time to improve model stability, the association was no longer significant.

A lack of statistical power, our simple method of characterizing shoulder posture, and potential bias due to healthy worker survivor bias (Picciotto et al., 2013) are the three main limitations of this study. (1) As mentioned above, sometimes relatively small changes in the tertile cut-points had a substantial effect on our results due to sparse numbers of cases per group. Despite a relatively large sample size, the number of RCS cases was relatively small. It is possible that some of our results were not statistically significant and unstable due to inadequate power. For example, our power to detect an HR of 2.0 between the middle tertile exposure group compared with the low group, with an α level of .05 was 33% for ≥45° flexion. For each unit change in forceful repetition rate at an α level of .05, our power was 81% to detect an HR of 1.12; for middle tertile exposure group compared with the low group, our power was 81% to detect an HR of 3.0, but only 43% to detect an HR of 2.0 for the same group comparison. Despite this limitation, we found statistically significant, meaningful results for interactions between forceful repetition and three of four posture variables. (2) Our upper arm flexion and abduction measures may have overly simplified shoulder loading. Even using the most sophisticated methods available today (e.g., three-dimensional kinematics using imaging, motion capture, digital human models, multibody kinematics optimization), the three bones, four joints, and 17 muscles help make the shoulder anatomically complex, highly mobile, and challenging to model the joint and muscle loads (Blache et al., 2019; Chopp-Hurley & Dickerson, 2015; Dickerson et al., 2020). Superior humeral head translation (e.g., shoulder shrug), or other scapulothoracic motions were not measured in this study. By assessing two-dimensional upper arm posture angles, we were unable to measure the effects of other shoulder positions on the kinematic loads at the shoulder. Scapulothoracic motions affect rotator cuff muscle demands and, depending on the upper arm elevation angles, can increase or decrease subacromial space (Chopp-Hurley & Dickerson, 2015). Humeral elevation seems to be along the causal pathway to rotator cuff damage (Dickerson et al., 2020). (3) Healthy worker survival bias occurs when healthy workers are lost disproportionately from highly exposed groups due to employment termination (Applebaum et al., 2011; Picciotto et al., 2013; Stayner et al., 2003). For many of our analyses, we observed attenuation or decline of risk for the highest exposure group. This pattern has been observed in other studies of MSDs and is often attributed to healthy worker survival bias (Applebaum et al., 2011; Picciotto et al., 2013; Stayner et al., 2003). Among the 58 participants lost to follow-up, 17 left the study because they left employment, but only one person listed shoulder pain as a reason for leaving employment. Among the 41 other participants who left the study, seven reported shoulder pain at baseline. Also, at baseline, the mean and median time at current job for this cohort were 6.9 and 4.0 years. Although tenure at current employer, current job, and in current occupation were not different between cases and noncases, all values were relatively high. It is possible that these relatively long tenure employees are less vulnerable to developing work-related MSDs that might lead someone to change jobs. When years of tenure met our criteria, we did control for it as a confounder. In Table 4, the footnotes for confounder code g can be used to find the four models that retained the variable.