Work related musculoskeletal disorders and complaints constitute an important health problem in many industrialised countries, as they account for a large number of working days lost and considerable workers compensation and disability payments.^1–3 For a long time, low back pain has been the dominant problem. However, pain from the shoulder and neck region now seems to occur more frequently.^4,5 The prevalence of shoulder-neck symptoms is highest in the 45–65 year age bracket, as well as among women, manual workers, and certain ethnic groups.⁶

However, its aetiology is still incompletely understood. Mechanical exposure at work and psychosocial conditions within and without the workplace, in addition to lifestyle and individual variables (age, previous symptoms, etc) are frequently discussed as causal factors in the literature.^7,8,9,10,11

Shoulder and neck symptoms have been linked to jobs with highly repetitive work, static work, and work above shoulder level.^12–17 However, mechanical exposure explains only part of these complaints. The role of psychosocial factors in the workplace has therefore received increasing attention. On the job pressure, monotonous work, and a high perceived workload have also been associated with musculoskeletal symptoms just as much as working situations characterised by high psychological demands, low decision latitude, and low social support.^1,9,11,14,17

Hence, an aetiological model explaining shoulder and neck symptoms could be based on the assumption of an interaction between mechanical and psychosocial factors at work.^1,9,18 There is, however, a need to clarify the interplay between these risk factors.

A number of shortcomings in previous research into the causes of back and neck pain have been recognised.^1,9,17 Firstly, few studies have separated low back and neck pain, which would seem to be an important distinction because of potentially different epidemiological patterns and assumedly different risk factors.¹⁹ Secondly, studies regarding the role of psychosocial risk factors in these outcomes seldom have adequately controlled their risk estimates with respect to potential confounding from mechanical exposure.¹⁷

Moreover, few studies have been performed on populations having a sufficient variety of both mechanical and psychosocial exposures.^17,20 The presumed high correlation between the two risk factors cited can only be weighed adequately in a large study sample in which diverse job tasks are represented.

Finally, few attempts have been made to quantify the multidimensionality of mechanical exposure.^21–24 Nor have such measures been used in relation to psychosocial ones—particularly not in a prospective study design.

This study has attempted to address all the mentioned issues. Based on a large general population cohort, its intentions are to analyse the importance of both mechanical and psychosocial exposures in the workplace on the incidence of shoulder-neck complaints, and evaluate the possible effect modification between these two exposures.

METHODS

Design and population

The Malmö shoulder and neck study (MSNS) was designed as a prospective questionnaire based study with a follow up period of one year. The cohort is a subset of the larger Malmö diet and cancer study (MDCS) cohort, which consists of all men and women in the city of Malmö (240 000 inhabitants) born between 1926 and 1945—that is, those between 45 to 65 years of age in 1991, when the cohort was defined (n = 53 494).²⁵

The MSNS cohort consists of every participant who completed the baseline examination of the MDCS between February 1992 and December 1994 (n = 14 555), the period when items assessing shoulder and neck pain were included in the questionnaire. People were selected by random from the population register and either received a personal invitation (74.0%), or were recruited by a media campaign (26.0%). Possible selection bias of the MSNS sample was estimated by comparing sociodemographic data from the municipal statistics office regarding the complete age cohort in February 1994. This analysis supported the assumption that the MSNS cohort is reasonably representative of the corresponding general age cohorts in Malmö.²⁶

The baseline questionnaire inquired into musculoskeletal symptoms, marital status, ethnicity, level of education, occupational status, lifestyle factors, mechanical exposures, and psychosocial factors at work. The participants were instructed to answer questions regarding work related exposure, as referred to their present work situation.

A follow up was performed after about one year, when a questionnaire was mailed to all 14 555 participants, of whom 86.6% responded (12 607). The mean follow up time was 403.1 days (standard deviation (SD) 48.9).

These analyses are based on subjects younger than the retirement age of 65 at baseline who had been vocationally active for at least 30 hours per week during the follow up who had not had shoulder or neck pain (defined in the same way as in the cases of this study) for 12 months before the baseline assessment (n  = 4919).

Outcome variables

A modified version of the standardised Nordic questionnaire (SNQ) for the analysis of musculoskeletal symptoms was used to assess shoulder and neck pain at baseline and at follow up.²⁷ The original five point category scale that sought to establish the frequency of the subject’s complaint was modified to focus on the subject’s temporal assessment of occurrences as “never”, “once or twice”, “sometimes”, “often”, or “all the time”. As in the original version of the SNQ, topographical areas were specified through the presentation of drawings indicating the areas referred to in the questionnaire. Separate questions solicited complaints regarding the neck, shoulders, low back, and arms/hands. The five rungs of the SNQ items were dichotomised, and shoulder and neck pain were defined in terms of the two highest rungs, namely, as complaints reportedly experienced “often” or “all the time” during the past 12 months.

Exposure variables

The instrument for assessing the mechanical exposure index (MEI) was constructed on a framework based upon mechanistic theories.²⁴ The MEI is extracted from 24 of 33 questions regarding occupational mechanical exposure, as has been developed and used in the Stockholm MUSIC I study.²⁸ The response scale was revised to one of three points, the verbal quantifiers being “hardly anything/not at all”, “somewhat”, and “a great deal”.¹⁸ Thus, the item scale relates to the subjective perception of the exposure quantity, and does not consider estimation of specific duration or frequency aspects.

The index consists of 11 of 24 items mainly assessing body posture. It was shown to have a much higher validity than job titles. The internal consistency of the index was very high, with a Chronbach’s α of 0.83. Two week test-retest stability expressed as weighted κ value was 0.78; 12 month stability was 0.83.²⁴ The MEI sum thus ranged from 11 to 33. Those with a score of 16 or more were considered to be exposed to high work related mechanical exposure (upper quartile).

Psychosocial work exposure was assessed by an instrument developed by Karasek and Theorell measuring a combination of psychological job demands and job decision latitude.²⁹ Job demands and decision latitude were analysed by means of five and six items, respectively, to which the participants replied on a scale with four rungs. The single items were weighted according to a system suggested by Karasek.²⁹ Those with a total score of 34 or above on the job demands scale (ranging from 13 to 52) were considered to be exposed to high job demands. Those with a score of 56 or less on the decision latitude scale (a range of 18 to 72), were considered to be exposed to low decision latitude. The combination of high psychological demands and low decision latitude was defined as job strain. Job support was determined by five items,³⁰ each carrying a value from zero to four, and thereby yielding a total score ranging from zero to 20. A score of 13 or less was considered “exposed to low job support”.

Age was used as a continuous variable in all the multivariate analyses. Marital status was classified into four groups: married (which included those who cohabited without being married), unmarried, divorced, and widow/widower. The classification of country of origin in this study consisted of a simple dichotomy between people born in Sweden and those born in other countries.

Occupational class was classified according to job title and work tasks, using the manual issued by Statistics Sweden.³¹ Based on the two digit socioeconomic status codes, five groups were constructed. Occupational class group V includes unskilled manual workers; group IV, skilled manual workers; group III, non-manual employees on a low level; group II, non-manual employees on a medium level; and group I comprises non-manual employees on a high level—that is, employees in leading positions and professionals with university degrees. Farmers (very few in this urban population) and entrepreneurs were excluded from the study when this variable was used in the analysis because of their unclear status in relation to other groups in this kind of socioeconomic scale (thus they are included in the category “missing data” in table 1).

Educational level was determined by the self reported total years of education, used in the analyses as a trichotomous variable with the values “nine or less educational years”, “10 to 12 years”, and “more than 12 years”.

As it has been claimed that musculoskeletal complaints could be a part of a generalised pain syndrome,³² we constructed a variable termed “pain from other regions”. People who reported pain from the lower back or from arms/hands at the baseline assessment were, by the definition stated above, classified as having “pain from other regions”. This variable was used in the multivariate analysis to account for possible confounding from a generalised pain syndrome, which could be expected to have different causes than specific shoulder and neck pain.

Statistical methods

The associations between mechanical and psychosocial exposures and a one year cumulative incidence of shoulder and neck pain was calculated as an odds ratio (OR) and a 95% confidence interval (CI) was applied. Adjustments for potential confounders were made by logistic regression analysis. Educational level was chosen as the marker of socioeconomic status in the multivariate models, as it permitted the use of more people, and because occupational class was highly correlated to mechanical exposure, thus creating the risk of colinearity problems in the model. In this study, effect modification was defined as a departure from additivity, because, according to Rothman, this phenomenon should be evaluated on an additive scale.^33,34

RESULTS

Table 1 describes the demographic and socioeconomic characteristics of the cohort. In tables 2 and 3, neck and shoulder pain were analysed separately to assess their similarity in relation to demographic and socioeconomic variables. It can be noted in table 2 that the cumulative incidence of neck pain seems to peak after the age of 50 in vocationally active men, while in vocationally active women it seems to reach its highest prevalence before age 50. The risk of developing neck pain during the follow up period was significantly higher among women born outside of Sweden (age adjusted OR  = 1.81, 95% CI: 1.15, 2.92). It was also noted that occupational class was not a predictor for neck pain in women, while having a skilled manual occupation was significantly associated with a higher cumulative incidence among men (OR  = 2.04, 95% CI: 1.27, 3.29, using high level non-manual employees as the reference). Moreover, being a widower was associated with a significantly increased risk.

Concerning educational level, there was a statistically significant higher risk for neck pain in the group of men with the least education. The age adjusted OR was 2.31 (95% CI: 1.48, 3.61). Pain from other musculoskeletal regions was a strong predictor in both sexes, supporting the notion of a generalised pain syndrome, with age adjusted ORs of 3.71 (95% CI: 2.62, 5.25) for men and 2.73 (95% CI: 1.47, 3.77) for women.

Basically the same pattern (except regarding country of origin) was found for shoulder pain (table 3), which justified the merging of the two pain categories in the further analyses.

In tables 4 (A) and (B) it can be noted that mechanical exposure was the strongest predictor for shoulder and neck pain. Statistically significant higher age adjusted ORs were found for both men and women (OR  = 2.17, 95% CI: 1.65, 2.85 and OR  = 1.59, 95% CI: 1.22, 2.06, respectively). Adjusting the ORs concerning mechanical exposure for job strain, too, produced a very marginal change.

High psychological job demands, low job decision latitude, and low job support were not associated with increased risk for getting neck and shoulder pain during follow up, either in men or in women. However, the combination of high psychosocial demands and low decision latitude (that is, job strain) showed a statistically significant risk in women, although not in men (age adjusted OR  = 1.73, 95% CI: 1.29, 2.31 and OR  = 1.22, 95% CI: 0.84, 1.79, respectively). Adjusting further for mechanical exposure reduced these risk estimates only slightly: the ORs remained statistically significant regarding job strain among women. Further adjustments for the factors previously mentioned, as well as for marital status, country of origin, and educational level, only marginally reduced the ORs regarding mechanical exposure and the work related psychosocial factors cited; all statistically significant associations remained. In the final model, the ORs were adjusted for pain in other regions as well, slightly decreasing them, and the risk estimate regarding high mechanical exposure among women lost its statistical significance.

The results of the analysis regarding effect modification between mechanical exposure and job strain are shown in table 5 (A) and (B). In the absence of mechanical exposure, job strain did not seem to infer a statistically significant risk of neck and shoulder pain in men or in women (age adjusted OR  = 0.96, 95% CI: 0.49, 1.88 and OR  = 1.33, 95% CI: 0.87, 2.05, respectively). In contrast, the effect of mechanical exposure in the absence of job strain showed a statistically significant impact in men, although not in women (age adjusted OR  = 2.14, 95% CI: 1.59, 2.88 and OR  = 1.33, 95% CI: 0.97, 1.83, respectively). Simultaneous exposure to both risk factors clearly increased the risk for shoulder and neck complaints in both men and women (age adjusted OR  = 2.25, 95% CI: 1.41, 3.59 and OR  = 2.60, 95% CI: 1.78, 3.79, respectively).

Adjusting these estimates for age, marital status, country of origin, and educational level slightly reduced the ORs, but did not change any case of statistical significance. Adjusting also for pain from other regions further decreased the ORs. Among men, the estimate regarding simultaneous mechanical exposure and job strain was just beneath the threshold for statistical significance, but not among women (OR  = 1.64, 95% CI: 0.99, 2.73 and OR  = 2 .01, 95% CI: 1.34, 3.01, respectively).

When synergy index was calculated by Rothman’s method using the fully adjusted ORs, evidence for a synergistic mechanism was found for women, but not for men (synergy indices estimated at 2.20 and 1.02, respectively).

DISCUSSION

We found that mechanical exposure (for men) and job strain (for women)—in other words, the combination of high job demands and low job decision latitude—were the factors most strongly associated with a higher risk for developing shoulder and neck pain during the one year follow up period.

We also found evidence for a synergistic effect of these two factors in heightening the risk of developing shoulder and neck pain among women, but not among men.

Our results may have been biased by selection, misclassification, and confounding. It is probable that the most vulnerable people had left their jobs and thus were excluded from the cohort in this study—either because they were no longer pursuing a physically demanding vocation, or because they were already ill at the time of the baseline assessment. This would bias the risk estimates towards the null.

Furthermore, it is known that people who are ill have less of a tendency to participate in studies. A higher non-participation rate among people complaining of shoulder and neck pain during the follow up study could also lead to an underestimation of the true association between mechanical and psychosocial exposures and the incidence of shoulder and neck pain. In actuality, the high participation rate in our follow up study (86.6%) would render this possibility a negligible consideration.

Bias attributable to dependent misclassification should not influence the results of this prospective study, as the exposures were determined at baseline, and case status at follow up. However, another possibility of misclassification could be present, namely, between mechanical and psychosocial exposures. It is possible that exposure to one of these factors might affect the person’s assessment of the other. However, the correlation coefficient between these two exposures were rather moderate (r = 0.15, not shown in tables), which ought to exclude the possibility that the results were very much influenced by this type of bias.

Another bias of importance could be confounding. One potential confounder could be age, but adjustment for this factor in the multivariate analysis only resulted in a very marginal change in the ORs.

The most important confounder to account for was job strain regarding the estimated impact of mechanical exposure, and mechanical exposure regarding the impact of job strain. However, including these variables in the model only decreased the age adjusted ORs slightly, without changing their statistical significance. It can, thus, be concluded that the impact of mechanical exposure and job strain on shoulder and neck pain exhibits little or no confounding by each others’ effect.

Further potential confounding factors were marital status and country of origin. Once again, inclusion of these variables in the multivariate models hardly changed the ORs. Occupational status was not included in the multivariate analysis because of its close association with both mechanical and the psychosocial exposures. For this reason, the inclusion of occupational status in the multivariate model would most probably result in an over-adjusted model, or problems of multicolinearity. Educational level was therefore chosen as the variable denoting socioeconomic status in the confounding analysis. But even when this variable was added to the multivariate model, its effect on the risk estimate was moderate.

The calculated synergy index value was 1.02 for men and 2.20 for women, showing a synergistic relation between mechanical exposure and job among women, but not among men.

A number of previously published studies concerning shoulder and neck pain have included both mechanical and psychosocial exposures. Most of these have been performed within occupationally homogenous groups, such as newspaper workers,^35,36 transit operators,¹⁷ forestry workers,³⁷ carpenters,³⁸ car workers,³⁹ homecare workers,⁴⁰ aluminium smelters,⁴¹ students,⁴² medical secretaries,⁴³ and female nursing staff.⁴⁴ However, there are a few studies that have used groups drawn from the general population.^13,45–52

Most of the aforementioned studies were cross sectional,^{17,35–41,43,44,46–48,51} and one was a case-referent study.¹³ In these investigations, the effect of mechanical exposure was controlled for psychosocial exposure, and vice versa. However, only two studies assessed possible effect modification between the two types of exposure.^36,40 Both of these were cross sectional, and were based on occupationally homogenous samples (newspaper workers and female homecare workers).

Most of the studies mentioned used a kind of demand/control instrument to assess work related psychosocial exposure,^{17,35–38,40–42,44,45,47,51} while a great variety of measures were used for assessing mechanical workload. These ranged from time spent in a certain occupation (for example, Krause et al¹⁷) to observational assessments of workload and postures (for example, Kamwendo et al⁴³). A wide array of instruments was also used in these studies to determine shoulder and neck pain. In some instances, a version of the standardised Nordic questionnaire was used.^13,37,47

In conclusion, there seems to be a lack of previous prospective studies that have been (1) undertaken on large, general, population based samples (or at least samples representing a wide variety of occupations/work tasks), and that (2) use well recognised instruments for assessing mechanical and psychosocial exposure. Such standards are required to optimally address the question of whether one or both of these exposures can be convincingly linked to shoulder and neck pain. The preceding caveat may especially hold true if the objective is a valid analysis of effect modification.⁵³

In a recently published review article concerning neck pain, in which 22 cross sectional, two prospective, and one case-referent study were evaluated, the authors concluded that awkward work postures could be linked to neck disorders with a reasonable degree of certainty.⁴⁹ This coincides with our findings that mechanical exposure (as assessed by an index primarily based on awkward work postures) is associated with an increased risk of developing shoulder and neck pain during the follow up period, independent of psychosocial exposure.

In the case of women, we also found a statistically significant association between work related psychosocial factors such as job strain and the heightened risk of developing shoulder and neck pain during the follow up period, independent of mechanical exposure. Similarly, in some cross sectional studies, high psychosocial job demands were found to be associated with shoulder and neck pain—once again, independent of mechanical exposure.^17,35,37,47 Several studies found low decision latitude associated with this outcome—again, independent of mechanical exposure.^36,38,40,41 Other studies, however, failed to find such an association.^37,47 One prospective study, however, did find high decision latitude associated with an increased risk of shoulder/neck pain.⁴²

Two studies found the combination of high job demands and low decision latitude (that is, job strain) associated with shoulder neck pain, independent of mechanical exposure,^44,47 agreeing with our findings concerning women. One of these studies comprised only women,⁴⁴ while the other consisted of a general population sample.⁴⁷

It seems like effect modification between mechanical exposure and such psychosocial factors as job strain have been rather neglected in previous studies. We could only find two cross sectional studies—both with rather small, occupationally homogenous samples—addressing this issue.^36,40 The aforementioned studies both found support for a possible synergistic effect between mechanical exposure and psychosocial factors. One of them concluded that high job demands, low decision latitude, and low job support all acted synergistically vis-à-vis mechanical exposure regarding the risk for shoulder and neck disorders.³⁶ The other study found that low decision latitude acted synergistically regarding exposure to awkward body positions vis-à-vis the cited outcome.⁴⁰ It is noteworthy that one of these studies was solely comprised of female subjects.⁴⁰ Once more, this is basically in line with the results of our analysis of effect modification, namely, that there is statistically significant support for the existence of a synergistic relation between mechanical exposure and job strain in women (but not in men) when it comes to the risk of developing shoulder and neck pain.