METHODS

A prospective cohort study is the best observational design for questions of aetiology and consequences. To be included in this review, studies had to be prospective cohort studies with population based samples or samples of working populations dealing with either LBP (that is, any pain in the lower back) or any consequence of LBP (that is, filing injury claim, sick leave, delayed return to work, disability pension). Only full reports written in English were included and not letters or abstracts. Descriptive studies that did not compare exposed workers with a reference population of non-exposed or less exposed workers were discarded. Finally, studies dealing with LBP due to lumbar disc herniations, osteoporosis, cancer, pregnancy, or other specific causes were not included.

Relevant articles from peer reviewed journals were identified by computerised searches in the databases Psychinfo, OSHROM, and Medline for 1990–2002 inclusive. In Psychinfo all articles with the words “back pain” as descriptor were retrieved (n = 798) and abstracts were screened for possible inclusion. In OSHROM abstracts for all articles with index term “low back” in combination with either “disabilities”, “disease”, “disorders”, “injured”, “pain”, “economics”, “epidemiology”, “pathology”, “physiopathology”, “prevention and control”, “psychology”, and “rehabilitation” were screened for inclusion (n = 1316). In Medline, a three step search procedure was performed. First, a search was performed using “low back pain” or “lumbago” or “sciatica” either as MESH term or in the title or abstract (n = 7616). Second, a similar search was performed using “job satisfaction” or “workplace” or “occupational groups” or “occupational exposure” or “employment“ or “occupations” or “workload” or “stress” or “occupational diseases” or “vibration” (n = 218 454). Third, results of the two searches (“hedges”) were combined and titles and abstracts of all remaining articles were screened for possible inclusion (n = 1005). Together, these searches resulted in the inclusion of 38 papers. Finally, relevant references were sought in all included publications and review articles published during the 13 year period, resulting in inclusion of an additional two articles. The computerised searches were assisted by a research librarian.

Abstraction

Using a checklist, each article was abstracted independently by two of the authors (JH, SL) for a list of core items. Completed checklists were subsequently compared and discordances were resolved by discussion until consensus was reached.

Psychosocial variables

Thirty different psychosocial variables were measured in the included studies. For clarity, these were grouped into four categories as follows:

• Perception of work, including the variables job satisfaction, feelings towards work, feelings towards work conditions, enthusiasm for work, enjoyment of work, and low occupational pride.

• Organisational aspects of work, including the variables influence on work conditions, job security, time pressure, conflicting demands, pace, work content, work control, work tempo, quantitative demands, qualitative demands, psychological demands, decision authority/latitude, skill discretion, few possibilities for on the job development, no education at employer’s expense, excessive job demands, and work monotony.

• Social support at work, including the variables recognition and respect, social support, co-worker support, social relations, relationships at work, others listening, external support, and supervisor support.

• Stress at work, including the variables stress, overstrain, job strain, level of distress, total mental exertion.

Quality of studies

No validated, standardised method of assessing the quality of epidemiological studies exists. We therefore constructed our own nine point quality scale based on well recognised requirements for epidemiological studies (table 1). The quality criteria concerned the study sample, the exposure and outcome measurements, and the statistical analysis and reporting. Studies scoring eight or nine points were considered high quality studies.

Assessment of outcomes

Studies were divided into two groups: those dealing with risk factors for low back pain and those dealing with risk factors for any consequence of LBP (filing injury claim, taking time off work, delayed return to work after period of absence due to LBP, etc). In the latter group, we looked for whether the psychosocial variables under investigation affected the outcome in a negative or positive direction.

Assessment of level of evidence and strength of association

Two steps were used in assessment of the association between the psychosocial variables and LBP. First, level of evidence was determined based on the number, quality, and outcome of the studies as follows:

• Strong evidence: Provided by generally consistent findings in multiple (>1) high quality studies.

• Moderate evidence: Provided by generally consistent findings in one high quality study and one or more low quality studies, or in multiple low quality studies.

• Insufficient evidence: Only one study available or inconsistent findings in multiple studies.

This method has previously been used in the assessment of level of evidence of clinical interventions for LBP as well as psychosocial factors as risk factors for LBP.

Further, the strength of the associations was considered according to a method used by Hemingway and Marmot³ as follows:

• No statistically significant positive association: p>0.05 or odds ratio (OR) or relative risk (RR) <1 or 95% CI below or straddling 1.00, that is OR or RR not significantly different from unity.

• Moderate association: OR or RR>1.00⩽2.00.

• Strong association: OR or RR>2.00.

Tables 2 and 3 summarise descriptive items for all studies.

RESULTS

Forty studies fulfilled the inclusion criteria. Eighteen papers reported studies dealing with LBP^10–27 (table 2) and 22 reported studies dealing with consequences of LBP^{14,15,23,28–46} (table 3). Three studies dealt with both pain and consequences and are thus displayed in both table 2 and table 3.^14,15,23 The two papers by Bigos and colleagues^28,29 apparently report on the same study and are only counted as one in the analysis. Ten of the 40 studies scored either eight or nine points out of nine in the quality assessment and were thus labelled as studies of higher quality^{15,16,18,19,23,26,39,41,43,46} (table 4).

Low back pain

• Perception of work. Perception of work was examined in relation to LBP in 10 studies,^{10–12,14–19,27} four of which were of high quality^15,16,18,19 (table 2). Of the latter, three reported no association^16,18,19 and in one study a positive moderate association between perception of one’s work and LBP was reported (OR 1.2).¹⁵ Of the five low quality studies, only one reported a positive but strong association (OR 2.00).¹⁴

• Organisational aspects of work. Organisational aspects of work were examined in relation to LBP in nine studies.^{13,17,19–21,23–26} Three studies were of higher quality;^19,23,26 only one of these reported a positive association (OR 2.19).²³ Among the studies of lower quality, only one of six reported a positive association,¹³ however, the OR was not provided. In two studies of lower quality positive associations were reported for certain groups but not for others.^20,21

• Social support at work. Of 11 studies reporting on social support at work in relation to low back pain, only three were of higher quality and all reported no significant positive association.^19,23,26 Among the low quality studies, six reported no significant positive association,^{14,20,21,24,25,27} and in two studies ORs were not provided for significant positive associations.^13,22

• Stress at work. Five studies dealt with stress at work in relation to low back pain,^{14,11,16,23,24,27} of which two were high quality.^16,23 One of these reported a significant and strong positive association (OR 2.68–2.70)¹⁶ and in the other study no significant positive association was found.²³ In all three low quality studies the authors found no significant positive associations.^11,24,27

In accordance with our predetermined criteria for assessment of level of evidence and strength of association, we found moderate evidence for no positive association between perception of work, organisational aspects of work, and social support at work and LBP. We found insufficient evidence for a positive association between stress at work and LBP.

Consequences of LBP

• Perception of work. In 19 studies consequences of LBP were examined in relation to perception of work^{14,15,23,28–36,38–44,46} (table 3). Six of these were high quality studies,^{15,23,39,41,43,46} and three reported significant positive moderate associations (OR 1.20–1.95),^15,41,43 whereas no significant positive association was reported in the other three studies.^23,39,46 Among the 13 studies of lower quality only three reported moderate positive associations (OR 1.53–1.87),^28,29,33,44 while in the remaining 10 studies no association was found between perception of work and consequences of LBP.^{14,30–32,34–36,38,40,42}

• Organisational aspects of work. Nine studies dealt with organisational aspects of work and consequences of LBP^{23,35,37,40,41,43–46} (table 3). Four of these were high quality studies,^23,41,43,46 of which none reported a significant positive association. Of the five low quality studies, two reported significant positive moderate associations (OR 1.40–1.79),^35,44 and in the remaining three studies no significant association was found (table 3).

• Social support at work. Five of nine studies dealing with social support at work in relation to consequences of LBP were of higher quality.^{23,39,41,43,46} Only two of these showed significant positive strong associations (OR 3.40–5.75).^23,46 In the four studies of lower quality, no significant positive associations were found^14,32,35,42 (table 3).

• Stress at work. Only three studies investigated stress at work in relation to consequences of LBP,^23,32,37 of which one was a high quality study.²³ Neither this nor the two low quality studies^32,37 reported significant positive associations between stress at work and any consequences of LBP.

In none of the included studies were work related psychosocial factors found to be protective in relation to LBP or consequences of LBP.

In accordance with our predetermined criteria for assessment of level of evidence and strength of association there was insufficient evidence for an association between perception of work in relation to consequences of LBP. There was strong evidence for no association between organisational aspects of work and moderate evidence for no association between social support at work and stress at work and consequences of LBP.

DISCUSSION

After reviewing and critically assessing 40 prospective studies published between 1990 and 2002, no clear picture of the relation between work related psychosocial factors and LBP emerges. According to the applied quality criteria and cut off values for significance, perception of one’s work—for instance, poor job satisfaction, is not associated with LBP, and associations between perception of work and consequences of LBP could not be determined. Both organisation of work and social support at work from co-workers or superiors showed no association with either low back pain or its consequences in both high and low quality studies. There was insufficient evidence for an association between stress at work and LBP and moderate evidence for no association between stress at work and consequences of LBP.

In 1993 Bongers et al asked for more prospective cohort studies dealing with work related psychosocial factors and musculoskeletal disease.⁴ Since then, 40 prospective studies dealing with LBP alone have been published, and this review is, to our knowledge, the first to address both the level of evidence and the strength of association between the two. Hoogendoorn et al assessed the level but not the strength of the evidence and found an effect of low workplace social support and low job satisfaction after reviewing 11 cohort and case-control studies.¹⁹ However, they also noted that slight changes in their rating system would result in a different conclusion.¹⁹ Davis and Heaney, after reviewing 66 articles published before 1999, concluded that studies of better quality appeared to more consistently associate low job satisfaction and job stress with the development of LBP. However, they found it premature to draw causal inferences due to many unresolved methodological issues.² We included and critically reviewed only prospective cohort studies of recent date, and after assessing both the level of evidence and the strength of association found a striking lack of association between work related psychosocial factors on one side and LBP and consequences of LBP on the other. This is in contrast to the a review published by Linton in 2001 in which he concluded that there was strong associations between a number of psychosocial work factors and back pain and that the majority of the studies were of acceptable quality.⁴⁷ In our opinion these conclusions are not justified on the basis of the published paper. First, Linton included only 21 prospective cohort studies indicating unsystematic and incomplete literature retrieval; second, studies not employing prospective designs were mistakenly labelled as prospective; third, he included studies with outcomes different from back pain (sciatica, disc prolapse); fourth, some results cited in the review are not supported by data published in the original papers; and finally, no critical appraisal of the original papers was performed.

New studies show that associations between psychosocial measures and disease outcomes may be spurious. For instance, Macleod and colleagues⁴⁸ found that even though higher self reported psychological stress was associated with symptoms of chest pain, and thus in many cases with a diagnosis of coronary heart disease, most highly stressed patients lacked objective indices of organic heart disease. In fact, self reported stress showed a weak inverse relation to all objective indices of organic heart disease in spite of reported symptoms. Thus persons reporting high stress levels might report more pain even in the absence of organic disease, whereas persons with different personality traits tend to report low stress and no symptoms in spite of objective signs of disease. This same effect may explain the strong association sometimes found between stress at work and LBP, but unfortunately, it is not at present possible to differentiate “organic LBP” from “non-organic LBP”. Also Papageorgiou et al found that dissatisfaction with status and income was not related to employment per se, but rather represented dissatisfaction with life in general.⁴⁹

Accuracy, reliability and reproducibility of measurements are essential features of good research data and uniform, standardised definitions of exposure and disease outcome are necessary if results are to be compared across studies. Unfortunately, due to the subjective nature of LBP, no gold standard definition currently exists. Consequently, the definition of outcomes varied among studies from self reported pain today²⁴ to disability pension due to longstanding LBP,³⁰ which in our opinion, are hardly comparable outcomes. Also the definition of the psychosocial factors varied considerably, and some may even be regarded as intermediate variables between exposure and effect (that is, job satisfaction and stress level). We reported results from the primary studies’ “face value” since the purpose of this review was not to try to disentangle intermediate effects. Finally, data on psychosocial exposures were collected in a wide variety of ways. For instance, in 12 studies the researchers’ own, presumably different questionnaires were used without any reporting of a validation procedure. Further, 20 different validated questionnaires were used in other studies mostly only once, and in five papers the authors did not report how psychosocial data were collected^{10,31,33,39,40} (tables 2 and 3). This lack of agreement and standardisation of both exposure and outcome is methodologically unsound and may give rise to random associations in single studies and account for much of the variation between studies. This tendency towards random associations might be further reinforced by the many statistical tests performed in almost every study reviewed.

According to Karasek, people with jobs characterised by low control over their work and high and conflicting work demands might be at higher risk for disease and less satisfied with their work.⁵⁰ Presumably, a high level of social support may buffer this effect and low social support may amplify it.³ The results of this review lead us to question this model in relation to LBP. First, in all of the studies investigating the variables relating to control and demand, no significant positive relation to LBP was found, and in the higher quality studies a negative or neutral association was found. Second, we found moderate evidence for no association between low social support and both LBP and consequences of LBP.

Alternative models of psychosocial work characteristics and their effect on health are based on a concept of imbalance between the effort at work and the rewards received;⁵¹ that is, prestige or high salaries may cause workers to better tolerate and accept unhealthy environments. This model has to our knowledge not been tested in relation to LBP. Maybe both physical and psychosocial work characteristics affect workers differently depending on factors such as job type, income, ethnicity, country, etc; and perhaps the measurement of, for instance, job satisfaction requires different approaches in different environments.

In our opinion, the completion of many prospective cohort studies such as recommended by Bongers et al in 1993⁴ has not disentangled the relation between work related psychosocial factors and the aetiology and consequences of LBP. The future challenge for researchers will be to develop standardised and valid definitions and operational instruments for reproducible measurements of psychosocial factors internationally and employ these in future studies. Also attention needs to be paid to the biological plausibility of the theories in this field. Until then biological relations remain speculative and strategies for preventing musculoskeletal disease by improving psychosocial work environments will probably be fruitless.

Conclusions

According to recent epidemiological literature we found moderate evidence for no positive association between perception of work, organisational aspects of work, and social support at work and LBP. We found insufficient evidence for an association between stress at work and LBP. Regarding consequences of LBP, there was insufficient evidence for an association between perception of work in relation to consequences of LBP. There was strong evidence for no association between organisational aspects of work and moderate evidence for no association between social support at work and stress at work and consequences of LBP. There were major methodological problems in the majority of studies included in this review and the diversity in methods was considerable. Therefore associations reported may be spurious and should be interpreted with caution.