Musculoskeletal pain and effort-reward imbalance- a systematic review

The search was carried out in the Pubmed, PsycINFO and Embase databases. A secondary search was completed in the reference lists of the articles included and additional reviews [5, 6, 14, 16‐19].

The query was completed in PubMed using the following search syntax:

“effort reward imbalance AND (musculoskeletal* OR shoulder pain OR neck pain OR back pain OR upper extremity pain OR lower extremity pain OR upper limb pain OR lower limb pain OR hip pain)”. The last date on which the search was carried out was 29 August 2012. Duplicates were eliminated following an amended search in the other databases.

The following inclusion criteria were defined for the selection of studies:

Titles and abstracts were screened in line with the inclusion criteria in step one, and then the full text was checked in step two. Screening was performed independently by two reviewers.

To take into account the risk of bias in the individual studies, quality criteria were used to assess the studies’ methodological quality.

Table 1 shows the quality criteria in five categories: study objective, study population, exposure, outcome and analysis. The items were combined differently to allow for the three different study designs. The criteria were adapted from Ariëns et al. [18] and thematically modified as appropriate.

The quality of each item was assessed by two reviewers independently using the grading positive (+), negative (−) or unclear (?). If an item was rated differently, the two reviewers reached a consensus. The inter-rater reliability was moderate (Cohen’s kappa: 0.41). The positive items were added for each study to give a total score. Cross-sectional studies were deemed methodologically good if they scored more than three points (maximum: five points). Case–control studies and prospective cohort studies had to score a minimum of four points to be graded as good (maximum: seven points).

For observational studies, the level of evidence for the association between psychosocial factors defined in the ERI model and musculoskeletal pain was classified as follows (Table 2) [18]:

Results are classified as consistent if at least 75% of the study results were unambiguous. Multiple results were defined as being derived from at least three studies.

As a number of studies yielded several partial results, these were combined to give a main result: if at least half of the partial results were statistically significant in relation to the hypotheses, a significant statistical effect was documented for the association within the study.

By combining the study results with a quality assessment, a level of evidence was allocated to 1) the ERI hypothesis, 2) the OVC hypothesis and 3) the ERI*OVC hypothesis.

The groups of workers examined in the included studies came from different industries. Five studies examined populations in care facilities [21, 23, 25, 26, 39] and four studies researched the association between ERI and musculoskeletal pain in employees of passenger transport companies [27‐29, 36], although the study by Dragano et al. [28] was a reanalysis of Joksimovic et al. [27]. This study was still included because – unlike the paper by Joksimovic et al. – it considered chronification of musculoskeletal pain. In this context, individuals were considered to be suffering from a disorder if they had experienced pain in the previous seven days - in addition to twelve-month prevalence. Lau [22] and Lapointe [37] studied public administration employees. Two studies examined random samples from surveys of the working population [30, 31].

Vineyard workers [32], dental technicians [24], call centre operatives [38], cleaners [33], officers from special police units [34] and bank/insurance staff have also been studied [35].

Seven studies examined associations between ERI and other outcomes: self-related poor health and psychological distress [22, 31], work-related burnout [22], poor general health, lack of vitality, poor psychological well-being [23], gastrointestinal complaints [23, 29], cardiovascular complaints [23], psychosomatic health complaints and exhaustion [21, 30], fatigue, sleep disturbances, common cold, nausea and dizziness [29], arterial hypertension [35], job dissatisfaction [30].

In addition to the psychosocial factors ascertained using the ERI model, ten studies [25‐27, 30‐33, 36, 37, 39] analysed other psychosocial factors. These were based on the demand-control model [10], the Copenhagen Psychosocial Questionnaire (COPSOQ) [41] and the Nursing Work Index – Extended Organisation Questionnaire (NWI-EO Questionnaire) [42].

All but one of the studies [39] (95%) had a clearly defined objective (no table). Only two studies (11%) achieved a response rate of at least 80% [26, 36]. All of the prospective cohort studies reported a response rate of at least 80% at the time of the follow-up. The only case–control study did not provide a clear definition of cases and controls. Half of the studies which were included (53%) ascertained psychosocial factors using the original ERI questionnaire [22‐25, 27, 28, 32, 34, 35, 37]. The other studies used proxy ERI questionnaires or did not ascertain the intrinsic component of over-commitment. A proxy ERI questionnaire refers to when the dimensions of the ERI model were used and the ERI score calculated, but the items of the dimensions were adapted to a specific work situation. A minority of studies (42%) recorded physical factors and controlled for these in the multivariate analysis [25, 26, 32, 33, 36‐39]. All but one study recorded physical factors by means of self-reported data. The items assessed ergonomic, postural aspects as well as information on physical workload, work intensification and physical factors at home. Gillen et al [39] assessed physical factors with the help of an ergonomic observation. Two studies [28, 34] which only recorded physical factors with the aid of one question (“My job is physically demanding”) did not fulfil this criterion because they were too vague. Other studies neglected to describe their approach in the methodological section but nevertheless presented results on physical factors; these also failed to meet this quality criterion [29, 35]. The quality criterion prospective enrolment was fulfilled by the only case–control study [39]. All of the prospective cohort studies recorded data on the outcome after at least one year. The majority (89%) of the studies presented risk predictors with confidence intervals and controlled for confounding. Using the point system for quality criteria, 14 of the 19 studies were deemed to be of good quality (74%).

According to Siegrist [43], the ratio score of effort (numerator) and reward (denominator) is calculated as follows: ∑ effort/∑ reward * correction factor (correction factor for the difference in the numbers of items of the two scales).

Twelve studies used a dichotomous ERI variable (ERI ratio score values > 1 vs. values ≤ 1) in their calculations. Three of these studies included additional ERI variables in the statistical models, using tertiles, quartiles and the continuous ERI ratio score [25, 33, 38] (Table 3).

The other seven studies which did not incorporate a dichotomous ERI variable (values > 1 vs. values ≤ 1) defined ERI using tertile or quartile limits [21, 23, 26, 28‐30, 36] or only used the continuous ERI ratio score in their computations [39]. In the studies which stipulated an ERI ratio score greater than 1, the frequency of an effort-reward imbalance was between 3% and 66% (Table 3).

A range of 5.6% to 70% was observed for the prevalence or incidence of musculoskeletal pain. Pain was reported in the following parts of the body: neck, shoulders, upper extremities, back, hips and lower extremities (Table 3). Three studies logged pain in only one part of the body [25, 33, 35]. The other 16 papers recorded pain in at least two parts of the body. In five of these studies, the pain reported was separately assessed by area [27, 28, 34, 36, 38], while the other eleven papers collated pain in different parts of the body within a single variable [21‐24, 26, 29‐32, 37, 39]. The following standardised measures were used for assessing musculoskeletal pain: Nordic questionnaire [44], van Korff chronic pain grade [45], SHC – scoring system for subjective health complaint inventory [46], Berger-Schmitt questionnaire [47], Freiburger Beschwerdeliste [48], Wiholm and Arnetz Questionnaire [49], QuickDASH questionnaire [50] and the Roland-Morris scale [51]. In all other cases, the questionnaires were not further specified.

Furthermore, five studies used stratification based on other criteria: Lapointe et al. [37] examined the gender-specific interaction between ERI and posture in workers at Canadian public authorities with computer-based jobs. Toivanen [31] studied working men and women from a Swedish survey and Bernard et al. [32] completed a gender-specific comparison of French vineyard workers. Peter et al. [29] stratified employees at passenger transport companies by different occupational groups: administrative staff, mechanics/skilled manual workers and drivers. Simon et al. [26] stratified caregivers in seven European countries by type of facility: hospitals, care homes and outpatient care.

All 19 studies could be included in the analysis of the ERI hypothesis: 13 studies (Table 4) showed a positive statistically significant association; ten of these were cross-sectional [21‐23, 25, 26, 28, 30, 31, 33, 34], two used a cohort design and one was a case–control study [36, 37, 39]. Six studies found no relationship between ERI and musculoskeletal pain. Of the 14 high quality studies, nine identified an association while five did not. Five studies examined the effect of ERI on workers in the healthcare sector [21, 23, 25, 26, 39] and all of these found a positive association for the ERI hypothesis.

Furthermore, Krause et al. observed negative confounding for physical work-related stress [38]. This means that the association between ERI and musculoskeletal pain is concealed by confounding and only becomes clear after adjustment [52].

Eight cross-sectional studies presented data for reviewing the OVC hypothesis (see Table 4). All of these were high quality. Four studies found an association between OVC and musculoskeletal pain [22, 24, 27, 34]. Lapointe et al. [37] adjusted their analysis for over-commitment, but did not present any risk predictors or level of significance. For this reason, this prospective cohort study could not be taken into account.

Five studies looked into the relationship between ERI and OVC (see Table 4). One high quality cross-sectional study [23] identified an interaction effect, but the other three cross-sectional studies [22, 30, 32] and one cohort study [37] observed no relationship.