Original article

Scand J Work Environ Health 2017;43(5):396-404    pdf

doi:10.5271/sjweh.3654

Temporal relationships between job strain and low-back pain

by Magnusson Hanson LL, Madsen IEH, Rugulies R, Peristera P, Westerlund H, Descatha A

Objectives Psychosocial working conditions are suggested risk factors for low-back pain, but it is unclear whether these associations are causal. The present study examined whether there are lagged and bidirectional associations between job strain and low-back pain and further controlled for unmeasured time-invariant confounding.

Methods The study was based on four biennial waves of data from the Swedish Longitudinal Occupational Survey of Health (SLOSH), including 3084 men and women. Cross-lagged analyses using structural equation modeling (SEM) were conducted on job strain, a combination of high job demands and low control, and any as well as low-back pain severity (how much any problems affected the respondents life). Analogous SEM (dynamic panel) models with fixed effects were also fitted to remove confounding from time-invariant factors (such as non-observed individual and environmental factors, eg, genetics, childhood conditions, personality).

Results The SEM models indicated bidirectional associations between job strain and any back pain over a 2-year time lag (β=0.21 and 0.19, P<0.05), when adjusting for a range of covariates. Job strain was also associated with an increase in low-back pain severity and vice versa. However, the SEM models with fixed-effects showed no statistically significant lagged relationships between job strain and any or low-back pain severity (β=-0.05 and β=0.00, respectively).

Conclusions This study suggests that associations between job strain and low-back pain with a lag of years may be due to residual confounding by time invariant characteristics. Further studies are, however, needed to elucidate short-term relationships.

The following article refers to this text: 2017;43(5):393-395

Low-back pain remains a major public health concern generating large healthcare costs and productivity losses (1). Low-back pain is the leading cause of years lived with disability globally (2, 3). Over 60–70% of all individuals may experience problems with back pain at some point in their life (4). The work environment in which many people spend a large part of their lives, may influence whether an individual develops disabling musculoskeletal problems, including low-back pain (5). A great deal of research has recognized the importance of physical working conditions for the development of low-back pain (5). People with physically demanding work seem to develop more back problems than others, eg, those with manual labor including heavy lifts, with twisted or bent positions, kneeling or squatting or who are exposed to vibrations (5). A number of studies indicate that also psychosocial factors may be of relevance (57). Studies with a longitudinal design that take earlier symptoms into account indicate causal associations between, eg, high job demands and low job control, job strain (the combination of high job demands and low control), as well as low support at work and low-back pain (6, 7). However, the evidence for psychosocial factors is generally weaker, and it has been questioned whether these associations represent causal associations or are due to bias or confounding (7).

To rule out that unmeasured third variables explain the associations between psychosocial work factors and low-back pain, some scholars recommend cross-lagged analyses based on repeat measures (7). These types of analyses may also shed light on the causal direction of the associations, which remains uncertain (7). Cross-lagged analyses cannot, however, completely rule out the influence of individual differences (8). It is possible that genetic or earlier life experiences can confound these type of associations. Studies have shown that there is a relatively high heritability of disabling low-back pain (9) and that a weak constitution is a risk factor (10).

This study aimed to increase the knowledge about temporal relationship between job strain and low-back pain by examining whether there are lagged and bidirectional associations between the two. In addition, we aimed to rule out confounding from time-invariant characteristics.

Methods

Study population

The study population consisted of participants in the Swedish Longitudinal Occupational Survey of Health (SLOSH) study, a nationally representative longitudinal cohort survey focusing on work life participation, social situation, and health/wellbeing. SLOSH started in 2006 with a first follow-up of participants in the Swedish Work Environment Survey (SWES) 2003 (N=9214), containing individuals from the entire country stratified by county, citizenship, gainfully employed, and 16–64 years of age at the time of enrolment into SWES. About two years later, all eligible SWES participants were followed-up by means of postal self-completion questionnaires, one addressed to people in work, ie. those in gainful employment for ≥30% of full time, and one to people working less or who had left the labor force temporarily or permanently (11). All SWES participants from 2003 who were still alive, with a known address in Sweden, and who had not actively opted out, were asked to complete questionnaires again in 2008, 2010, 2012, 2014, and 2016. In 2008, all participants in SWES 2005 were added to the cohort (raising the number of cohort members to 18 917) to be followed up every second year, see figure 1. The present study is based on those SWES 2003 and 2005 participants who responded and were working in all four waves (2010–2016), in total 3084 individuals. Only data from 2010 and onwards was used since low-back pain was not assessed in the same way in earlier waves. Some characteristics of the 3084 included participants are presented in table 1. The Regional Research Ethics Board in Stockholm approved the study and informed consent was obtained from all participants.

Table 1

Characteristics of the study participants responding to the questionnaire for those in paid work 4 time points 2010–2016 (N=3084).

2010 a 2012 2014 2016




N % N % N % N %
Sex
 Men 1293 42
 Women 1791 58
Age
Civil status
 Single 607 20 612 20 615 20 632 21
 Married/cohabiting 2424 80 2453 80 2443 80 2427 79
Occupational status
 Low 863 28 881 29 850 28 836 28
 Intermediate 1395 46 1417 47 1460 48 1446 48
 High 680 22 704 23 691 23 718 24
 Self-employed 91 3 20 1 30 1 31 1
Education
 Max 9 years of school 187 6 187 6 187 6 186 6
 Upper secondary school 1319 43 1311 43 1307 42 1299 42
 University 1578 51 1586 51 1590 52 1599 52
Physical activity
 Physically active 2494 81 2524 83 2474 81 2441 80
 Physically inactive 580 19 535 17 598 19 618 20
Work schedules
 Daytime 2341 77 2366 78 2402 79 2376 79
 Night work 197 7 205 7 194 6 179 6
 Shift work excl. nights 487 16 464 15 431 14 442 15
Physically straining work
 No 2587 84 2619 86 2667 87 2699 88
 Yes 476 16 431 14 388 13 370 12
Job strain
 No 2529 83 2522 82 2520 82 2533 83
 Yes 533 17 558 18 550 18 536 17

a Age range 24–71 years, mean 48 years, standard deviation 8

Figure 1

Total number of respondents/non-respondents to the Swedish Longitudinal Occupational Survey of Health (SLOSH) among the SLOSH sample originally participating in the Swedish Work Environment Surveys (SWES) 2003 and 2005, as well as number of individuals included in the analytic sample.

SJWEH-43-396-g001.tif

Exposure

Both job strain and low-back pain were measured repeatedly in the same way in the four waves. Job strain was measured by self-reports, using the Demand-Control Questionnaire (DCQ) (12), a widely used questionnaire operationalizing the demand-control-support model (13, 14). Four items (working fast, too much effort, enough time, and conflicting demands, Cronbach’s alpha 0.65–0.68) were used to create a score for demands at work and five items (learn new things, high level of skill, require initiative, deciding what to do at work, deciding how to do your work, Cronbach’s alpha 0.66–0.68) to create a score for control at work (15). Median split was used to divide the population according to high and low demands and control, respectively. People with a high demands and at the same time low control were considered exposed to job strain. The remainder formed the reference group.

Outcome

Low-back pain was measured by one item about the presence of back pain in the past three months in combination with an assessment of severity, in line with recommendations for prevalence studies (16). The severity criteria was based on ratings of how much the low-back problems affected the respondent’s life. Four groups were distinguished based on the response options: (i)no low-back pain, (ii) low-back pain, but does not affect my life at all, (iii) low-back pain that affects my life a little, and (iv) low-back pain that affects my life a lot. We also created a dichotomous indicator of any back pain as compared to no low back pain.

Covariates

A number of covariates assessed in the questionnaire in 2010 were included in the models, including sex, age, occupational status, education, civil status, obesity and physical strain at work, since these have been identified as potential confounders (5). Occupational status was coded according to the Swedish socioeconomic classification based on the occupation reported by the respondents, and divided into three categories: (i) low occupational status, including routine and manual occupations; (ii) intermediate occupational status, including non-manual intermediate occupations; and (iii) high occupational status, including higher managerial, administrative and professional occupations). Self-employed formed a separate category. Education was divided into three categories (primary school, upper secondary school, university). Civil status were divided into two categories (married or cohabiting versus single). Reported height and body weight was used to calculate body mass index (BMI) which was divided into obesity (BMI ≥30 kg/m2) versus non-obesity (BMI <30 kg/m2) according to the World Health Organization cut-off criteria. Also physical activity was assessed by one question: “How much do you exercise?”, from which two groups were formed: one active (exercises now and then or regularly) and one inactive group (exercises very little or never). We accounted for physical strain at work through a combination of two questions: whether work entailed heavy lifting or twisted positions at least ¼ of the time). Finally, we included information on current smoking (yes/no) and alcohol consumption (risky: ≥15 drinks/week for women, ≥22 drinks/week for men; or non-risky: <15 drinks/week for women or <22 drinks/week for men).

Data analysis

The associations between job strain and low-back pain, both any low-back pain and low-back pain severity were initially analyzed using cross-lagged models based on structural equation modeling (SEM). The cross-lagged paths estimated the association between job strain and subsequent low-back pain in successive waves and between low-back pain and subsequent job strain, after controlling for the stability of the variables over time and allowing for correlations between the parameters (17). We compared models in which cross-lagged regression coefficients were constrained to be equal across time and models where regression coefficients varied freely over time. As the former showed better fit to the data as compared to the latter (based on chi-square difference tests), henceforth the models were fitted with constraints on the structural paths over time. A robust weighted least squares estimator was used in the main models, which also allows for dichotomous and ordinal variables (18). However, to reduce possible bias introduced by missing information we alternatively used full-information maximum likelihood (FIML) estimation (19, 20). We first tested different models to better understand the temporal relationships between job strain and low-back pain. The first model, used as reference, only included auto regressions (temporal stability effects) between the main variables. The second model included auto regressions plus a path from job strain to low-back pain, while the third model included auto regressions and a path from low-back pain to job strain. Fourth and last, we fitted a model including auto regressions plus paths in both directions. Model fit was mainly assessed by the comparative fit index (CFI), and the root mean square error of approximation (RMSEA) (21). Values of RMSEA <0.05 and CFI ≈1 are assumed to be indicative of a well-fitting model. The adjusted models included sex, age, civil status, occupational status, education, and physical strain at work which have been indicated as confounders of the relationship between job stress and low-back pain (22). In sensitivity analyses, we also controlled for obesity, physical inactivity, smoking and risky alcohol consumption. A multigroup analysis, stratifying for sex, was also conducted to test for differences between men and women.

Additionally, we fitted analogous cross-lagged SEM models with fixed effects by means of dynamic panel models with fixed effects. Fixed effects methods include an alphai term treated as a set of fixed parameters representing all stable characteristics of a person. With this approach, each individual serves as his or her own control and major sources of confounding from time-invariant (fixed) factors (such as non-observed individual and environmental factors, eg, genetics, childhood conditions, and personality traits) are eliminated (23, 24). However, in contrast to standard fixed effects approaches, which use only variation within individuals to estimate the relationships between exposures and outcomes (assuming strict exogeneity), dynamic panel models with fixed effects allow for predetermined (sequentially or weakly exogenous) variables and the dependent variable to affect the predictor variable at a later point in time. Hence, it is possible to simultaneously estimate reciprocal relationships and loss of data due to differencing is avoided. Instead of eliminating all time stable measured and unmeasured factors, this approach uses a fixed-effects latent variable correlated with all time-varying independent variables to adjust for time-invariant characteristics of the individuals. To decrease complexity, separate dynamic panel models with fixed effects were used to estimate the cross-lagged coefficients, which also provides more flexibility in model specification and makes the estimation more robust to misspecification problems (Allison P, causal inference for panel data, available from: statisticalhorizons.com/resources) Reciprocal causation was accommodated by allowing the error term of the independent variable in each equation to correlate with future values of the time dependent predictors (25). As in the cross-lagged SEM models, we also here included contemporaneous effects, and both models with and without covariates were fitted.

All analyses were conducted using the lavaan 5.13 package in R.

Results

The characteristics of the sample are presented in table 1. There was a relatively high mean age of 48 years in 2010, and the majority was married or cohabiting and had a university education. The prevalence of low-back pain and other health conditions including neck/shoulder pain and migraine are presented in table 2. As much as 51% had complaints of low-back pain in 2010, however, only 6% reported low-back pain severe enough to affect their lives a lot. Among those who reported severe low-back pain, over half also reported neck/shoulder pain that affected their lives a lot, and around 25% also reported migraine that affected their lives a lot.

Table 2

Prevalence of low back pain and other types of health problems in 2010–2014.

2010 2012 2014 2016




N % N % N % N %
Low-back pain
 No 1477 49 1450 48 1342 44 1388 45
 Yes, but does not affect my life at all 669 22 688 23 739 24 684 22
 Yes, affects my life a little 709 23 739 24 780 25 791 26
 Yes, affects my life a lot 180 6 174 6 204 7 193 6
Neck/shoulder pain
 No 1149 38 1189 39 1068 35 1119 37
 Yes, but does not affect my life at all 796 26 786 26 842 27 876 29
 Yes, affects my life a little 894 29 883 29 965 31 877 29
 Yes, affects my life a lot 214 7 191 6 195 6 191 6
Self-rated health
 Good 2491 82 2507 82 2450 80 2451 80
 Suboptimal 559 18 565 18 608 20 610 20
Major depressive symptoms
 No 2907 96 2943 97 2962 97 2935 96
 Yes 132 4 101 3 98 3 117 4
Cancer
 No 3013 99 3005 99 3309 98 3011 98
 Yes 37 1 45 1 55 2 49 2
Asthma
 No 2838 93 2816 92 2840 93 2833 93
 Yes 212 7 234 8 227 7 227 7
Diabetes
 No 2982 98 2967 97 2966 97 2937 96
 Yes 71 2 86 3 105 3 125 4
Migraine
 No 2615 86 2614 86 2653 87 2638 86
 Yes 434 14 436 14 412 3 424 14

Correlations between the main variables used in the analyses in 2010 are presented in table 3. The distribution of characteristics did not change much over the four waves among the total sample, as also indicated by the stability (autoregressive) coefficients in e Figures 12 in the appendix (www.sjweh.fi/index.php?page=data-repository). Additional analyses showed that a relatively high proportion had stable job strain status over time. All in all, 61% of those working all time points and with valid data on job strain (N=3034) did not experience job strain at any of the time points, 4% experienced job strain all time points, while the remaining 35% changed between no job strain and job strain over time. Among those with valid data on back pain from all time points (N=2958), 21% had no back pain over the period, while 28% experienced any back pain all time points and 49% changed between no back pain and any back pain over the study period.

Table 3

Correlation table based on data from 2010.

SLOSH Job strain Low-back pain Sex Age Civil status Occupational status Education Obesity Physical inactivity Physically straining work
Job strain 0.12 a 0.06 a -0.04 a -0.04 a -0.13 a -0.06 a 0.02 b 0.02 b 0.16 a
Low-back pain 0.03 b 0.00 b -0.01 b -0.14 a -0.11 a 0.06 a 0.05 a 0.16 a
Sex 0.01 b -0.02 b 0.00 b 0.12 a -0.06 a -0.11 a -0.09 a
Age 0.01 b -0.02 b -0.08 a 0.07 a -0.04 a -0.02 b
Civil status 0.06 a 0.02 b -0.04 a -0.02 b 0.01 b
Occupational status 0.53 a -0.10 a -0.02 b -0.37 a
Education -0.16 a -0.10 a -0.28 a
Obesity 0.17 a 0.06 a
Physical inactivity 0.04 a
Physically straining work

a P<0.05.

b Non-significant.

Figure 2

Cross-lagged path coefficients from the SEM model analyzing the bidirectional relationships between job strain and any low-back pain (dichotomous variable) while adjusting for sex, age, civil status, occupational status, education, and physical strain at work. The model was based on 2813 individuals (11 252 observations) with valid information on all variables. w=SLOSH wave, ε=error term, ***P<0.001

SJWEH-43-396-g002.tif

Cross-lagged SEM models

Initial chi-square difference tests showed that the reciprocal model had significantly better fit than models with only auto regressions, auto regression plus paths from job strain to back pain and auto regressions plus paths from low-back pain to job strain. The standardized regression coefficients from the reciprocal models are shown in figures 23, with adjustment for sex, age, civil status, occupational status, education, and physical strain at work. The analyses showed a relatively equal degree of association between job strain and any low back pain two years later [β=0.17, 95% confidence interval (95% CI) 0.13–0.22, P<0.001], and between any low-back pain and job strain (β=0.16, 95% CI 0.12–0.20, P<0.001), also shown in figure 2. The adjustment somewhat attenuated the estimates compared to the unadjusted models (β=0.24, 95% CI 0.20–0.28 and β=0.22, 95% CI 0.18–0.26, respectively). The results also showed significant relationships in both directions between job strain and the 4-category variable indicative of low-back pain severity (figure 3). The results further indicated that job strain was statistically significantly associated with an increase in low-back pain severity. The standardized β coefficients were 0.22, 95% CI 0.18–0.25 (P<0.001) in the unadjusted model and 0.18, 95% CI 0.14–0.21 (P<0.001) when adjusting for sex, age, civil status, occupational status, education, and physically straining work. An increase in low-back pain severity also predicted later job strain (standardized β 0.17, 95% CI 0.14–0.20, P<0.001 in the unadjusted model, and 0.14, 95% CI 0.11–0.17, P<0.001 when fully adjusted). The fit of the unadjusted models, including 2915 individuals, (RMSEA 0.13, CFI 0.71–0.80), as well as those including covariates (based on 2813 individuals) (RMSEA 0.12, CFI 0.72–0.83) was, however, quite poor.

Figure 3

Cross-lagged path coefficients from the SEM model analyzing the bidirectional relationships between job strain and low-back pain severity (ordinal variable) while including sex, age, civil status, occupational status, education, and physical strain at work. The model was based on 2813 individuals (11 252 observations) with valid information on all variables in the model. w=SLOSH wave, ε=error term, ***P<0.001

SJWEH-43-396-g003.tif

When comparing the model with and without constraints on the structural paths between men and women, the model with constraints fitted the data better according to the chi-square difference test, and RMSEA/CFI. This suggested that there were no clear differences in the estimates between men and women.

We also fitted models based on all 3084 individuals allowing for inclusion of missing information among the included variables and models including all those respondents working ≥30% in 2010 and had responded to any of the following waves (N=7044). Those analyses supported associations of similar magnitude in both directions.

Cross-lagged SEM models with fixed effects

The SEM models with fixed effects showed very good fit to the data according to both CFI and RMSEA. For back pain as a dichotomous variable, the path coefficient from job strain to later back pain was -0.06 (95% CI -0.14–0.02, P=0.70, RMSEA 0.05, CFI 0.98), and the path coefficient from any back pain to job strain was similar (-0.07, 95% CI -0.17–0.04, P=0.20 RMSEA 0.003, CFI 0.99). In the models analyzing back pain severity, the corresponding coefficients were -0.05, 95% CI -0.11–0.02 (P=0.9, RMSEA 0.05 CFI=0.99) and 0.00, 95% CI -0.11–0.11 (P=<0.001, RMSEA=0.06, CFI=0.98), respectively (figure 4a and 4b). All path- and correlation coefficients from the models on job strain and low back severity are further presented in the supplmentary e Figures 3a-b (www.sjweh.fi/index.php?page=data-repository). Inclusion of the same covariates as in the SEM models with fixed effects gave similar estimates, but inclusion of the covariates did not improve the fit of the models. The results were similar when allowing for inclusion of missing values in the analyses (on all 3084 individuals) through full information maximum likelihood estimation with calculations of robust standard errors. Analyses based on the sample responding in any of the waves (N=7044) also showed similar results.

Figure 4a

Cross-lagged path coefficient from the dynamic panel models with fixed effects analyzing the relationships between job strain and low back pain severity (ordinal variable). The model was based on 2915 individuals (11660 observations) with valid information on all variables in the model. w= SLOSH wave, ɛ=error term, α=latent variable to control for all time-invariant confounders, either observed or unobserved.

SJWEH-43-396-g004.tif
Figure 4b

Cross-lagged path coefficient from the dynamic panel model with fixed effects analyzing the relationships between low-back pain severity (ordinal variable) and job strain. The model was based on 1925 individuals (7700 observations) with valid information on all variables in the model. w=SLOSH wave,ε=error term, α=latent variable to control for all time-invariant confounders, either observed or unobserved, ***P<0.001

SJWEH-43-396-g005.tif

Discussion

In this panel study on a sample of the Swedish workforce, we found that bidirectional associations between job strain and low-back pain over a two year time lag disappeared when accounting for all time-invariant individuals characteristics. This indicates that the initial associations may be due to residual confounding from unmeasured characteristics such as genetics, childhood conditions, and stable personality traits.

To our knowledge this is one of the first studies to examine reciprocal relationships between job stressors and back pain (7). A study by Christensen & Knardal (26) also used data from three waves with a 2-year time lag and found a relationship between stable high quantitative demands and both new and persistent/recurrent back pain. They did not examine the relationship between back pain and work factors but acknowledged that there are likely reciprocal relationships. Our initial observations supported this notion. These findings were also similar for men and women, despite that women usually report more musculoskeletal pain (5).

However, there is a possibility that unmeasured background variables may be confounders of work stressors and strain relationships. In this study, we therefore examined bidirectional relationships at the same time as partialling out all time-invariant individual characteristics. The lack of an association between job strain and low-back pain in the analyses controlling for time-invariant characteristics may indicate that the associations were confounded by unmeasured time-invariant variables such as biological or psychological attributes that could be linked to circumstances earlier in life. A study using a similar approach as ours likewise indicated that the majority of associations between job stressors and mental health were contemporaneous. Only job demands appeared to have an effect on mental health one year later (27). This may also suggest that a shorter time lag would needed to capture the effects of job stressors like job strain even while ruling out major sources of confounding. A number of plausible pathways by which job stress may give rise to back pain have been suggested, eg, enhanced pain perception and increased muscle activity and muscle tension, but more knowledge is needed on mechanisms and the relevant time frame to capture potential effects (28).

This study had a number of additional strengths. By accounting for correlations between job strain and back pain and reciprocal paths the risk of reverse causality is limited in the main analyses. In the SEM models without fixed effects, we also controlled for a number of potential confounders at the first measurement occasion. Both demographic characteristics and biomechanical factors may be of importance to consider and our measure on physical work load was based on multiple items which is preferable to a single item measure (22). A robust version of the diagonally weighted least squares estimation was used which is the recommended approach for obtaining SEM estimates when using categorical data (29). The parameter estimates were also similar for the majority of models when robust standard errors were calculated in conjunction with full information maximum likelihood, an approach that seem to work well for non-normal data (30).

A limitation is that our main analyses is restricted to individuals who responded and were in paid work repeatedly. These people may be selected with regard to working conditions and health/wellbeing. Previous work show that SLOSH participants and especially those who have responded several times are to a higher extent women, older, married/registered partners, and highly educated. Dropout from the study may limit the generalizability of the study to more resilient individuals and have led to an underestimation of the relationship between job strain and low-back pain.

Another limitation of our analyses is that potential time-varying confounding by other factors are not dealt with. It is possible that time-varying factors such as health behaviors and other health problems influence back pain and job strain. However, if health behaviors and other health problems are mediators rather than confounders, adjustment for them would give misleading estimates of total effects.

In this study, job strain was used as a proxy for job stress, but the estimates presented in this study are potentially underestimated by misclassification of exposure. Although the definition of job strain is the most commonly used definition in previous research it should be acknowledged that using median splits is a relatively crude way of operationalizing strain.

A misclassification of the outcome cannot be ruled out, which if non-differential could have contributed to a dilution of the estimates. There is no standard definition of low-back problems. Several expert groups on musculoskeletal problems have, however, recommended that low-back problems should be assessed by a measure that couples pain with a measure of functioning on, eg, work ability or life quality (16, 31), which was done in the present study. Even if we assessed low-back over the past three months, we do not know if low-back pain symptoms had persisted for a 3-month period to be regarded as chronic disabling pain with significant impact on people’s lives and society (32). There is, however, research that supports the validity of retrospective reports of pain status for a 3-month recall period (33).

Finally, our analyses do not clarify the role of job strain as a risk factor or prognostic factor. More research seems warranted also on the role of work stress for first incidence of low-back pain and for progression from acute to chronic pain.

Despite these limitations, the results of this study suggested that lagged relationships between job strain and low-back pain across 2 years may be due to residual confounding by time invariant individual and environmental characteristics. A failure to take into account unobserved time stable characteristics may thus overestimate relationships between psychosocial working conditions and low-back pain. However, it should be acknowledged that the results do not preclude short-term relationships between job strain and low-back pain. To determine if there are bidirectional associations between job strain and back pain future work is needed with more frequent repeat measurements, which may better represent the time lag for cause and effect and that account for time-invariant individual factors.

Acknowledgements

The authors are grateful to the Swedish Research Council for supporting the study, to Statistics Sweden for carrying out data collection, and to the Swedish Research Council, the Swedish Research Council for Health, Working Life and Welfare for supporting the Swedish Occupational Survey of Health. The authors also wish to thank all participants for making the study possible. The funders had no role in the study design, collection, analysis, interpretation of the data, in writing of the report or the decision to submit the article for publication.

The authors declare no conflicts of interest.

REFERENCES

1 

Maher, C, Underwood, M, & Buchbinder, R. (2017, Feb 18). Non-specific low back pain. Lancet, 389(10070), 736-47, https://doi.org/10.1016/S0140-6736(16)30970-9.

2 

Collaborators, GBDRF, Forouzanfar, MH, Alexander, L, Anderson, HR, Bachman, VF, Biryukov, S, et al. (2015). Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks in 188 countries 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet, 386(10010), 2287-323, https://doi.org/10.1016/S0140-6736(15)00128-2.

3 

Hoy, D, March, L, Brooks, P, Blyth, F, Woolf, A, Bain, C, et al. (2014). The global burden of low back pain: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis, 73(6), 968-74, https://doi.org/10.1136/annrheumdis-2013-204428.

4 

Hoy, D, Brooks, P, Blyth, F, & Buchbinder, R. (2010). The Epidemiology of low back pain. Best Pract Res Clin Rheumatol, 24(6), 769-81, https://doi.org/10.1016/j.berh.2010.10.002.

5 

SBU. (2014). [The importance of the work evironment for back problems. A systematic review]. Stockholm: Statens beredning för medicinsk utvärdering (SBU) [Swedish Council on Health Technology Assessment]. Arbetsmiljöns betydelse för ryggproblem. En systematisk litteraturöversikt.

6 

Hauke, A, Flintrop, J, Brun, E, & Rugulies, R. (2011). The impact of work-related psychosocial stressors on the onset of muskuloskeletal disorders in specific body regions: A review and meta-analysis of 54 longitudinal studies. Work & Stress, 25(3), 243-56, https://doi.org/10.1080/02678373.2011.614069.

7 

Lang, J, Ochsmann, E, Kraus, T, & Lang, JW. (2012). Psychosocial work stressors as antecedents of musculoskeletal problems: a systematic review and meta-analysis of stability-adjusted longitudinal studies. Soc Sci Med, 75(7), 1163-74, https://doi.org/10.1016/j.socscimed.2012.04.015.

8 

Hamaker, EL, Kuiper, RM, & Grasman, RP. (2015). A critique of the cross-lagged panel model. Psychol Methods, 20(1), 102-16, https://doi.org/10.1037/a0038889.

9 

Nielsen, CS, Knudsen, GP, & Steingrimsdottir, OA. (2012). Twin studies of pain. Clin Genet, 82(4), 331-40, https://doi.org/10.1111/j.1399-0004.2012.01938.x.

10 

Leboeuf-Yde, C. (2004). Back pain--individual and genetic factors. J Electromyogr Kinesiol, 14(1), 129-33, https://doi.org/10.1016/j.jelekin.2003.09.019.

11 

Magnusson Hanson, LL, Theorell, T, Oxenstierna, G, Hyde, M, & Westerlund, H. (2008). Demand, control and social climate as predictors of emotional exhaustion symptoms in working Swedish men and women. Scand J Public Health, 36(7), 737-43, Epub 2008/08/08.

12 

Theorell, T, Perski, A, Akerstedt, T, Sigala, F, Ahlberg-Hulten, G, Svensson, J, et al. (1988). Changes in job strain in relation to changes in physiological state. A longitudinal study. Scand J Work Environ Health, 14(3), 189-96, https://doi.org/10.5271/sjweh.1932.

13 

Sanne, B, Torp, S, Mykletun, A, & Dahl, AA. (2005). The Swedish Demand-Control-Support Questionnaire (DCSQ): factor structure, item analyses, and internal consistency in a large population. Scand J Public Health, 33(3), 166-74, https://doi.org/10.1080/14034940410019217.

14 

Fransson, EI, Nyberg, ST, Heikkila, K, Alfredsson, L, Bacquer de, D, Batty, GD, et al. (2012). Comparison of alternative versions of the job demand-control scales in 17 European cohort studies: the IPD-Work consortium. BMC Public Health, 12, 62, https://doi.org/10.1186/1471-2458-12-62.

15 

Chungkham, HS, Ingre, M, Karasek, R, Westerlund, H, & Theorell, T. (2013). PLoS One (Vol. 8). Factor Structure and Longitudinal Measurement Invariance of the Demand Control Support Model: An Evidence from the Swedish Longitudinal Occupational Survey of Health (SLOSH), 8, p. e70541, https://doi.org/10.1371/journal.pone.0070541.

16 

Dionne, CE, Dunn, KM, Croft, PR, Nachemson, AL, Buchbinder, R, Walker, BF, et al. (2008). A consensus approach toward the standardization of back pain definitions for use in prevalence studies. Spine (Phila Pa 1976), 33(1), 95-103, https://doi.org/10.1097/BRS.0b013e31815e7f94.

17 

Finkel, SE. (1995). Causal analysis with panel data. Sage: Thousand Oaks. https://doi.org/10.4135/9781412983594.

18 

Muthén, B. (1984). A general structural equation model with dichotomous, ordered categorical, and continuous latent variable indicators. Psychometrika, 49, 115-32, https://doi.org/10.1007/BF02294210.

19 

Shafer, J.L, & Graham, JW. (2002). Missing data: our view of the state of the art. Psychol Methods, 7, 147-77.

20 

Allison, PD. (2003). Missing data techniques for structural equation modeling. J Abnorm Psychol, 112(4), 545-57, https://doi.org/10.1037/0021-843X.112.4.545.

21 

Hu, LT, & Bentler, PM. (1999). Cutoff Criteria for Fit Indexes in Covariance Structure Analysis: Conventional Criteria Versus New Alternatives. Structural Equation Modeling-a Multidisciplinary Journal, 6(1), 1-55, https://doi.org/10.1080/10705519909540118.

22 

Davis, KG, & Heaney, CA. (2000). The relationship between psychosocial work characteristics and low back pain: underlying methodological issues. Clin Biomech (Bristol, Avon), 15(6), 389-406, https://doi.org/10.1016/S0268-0033(99)00101-1.

23 

Allison, PD. (2005). Fixed Effects Regression Methods for Longitudinal Data Using SAS. Cary, NC: SAS Institute.

24 

Gunasekara, FI, Richardson, K, Carter, K, & Blakely, T. (2014). Fixed effects analysis of repeated measures data. Int J Epidemiol, 43(1), 264-9, https://doi.org/10.1093/ije/dyt221.

25 

Wooldridge, JM. (2001). Econometric Analysis of Cross Section and Panel Data. Cambridge, MA: MIT Press.

26 

Christensen, JO, & Knardahl, S. (2012). Work and back pain: a prospective study of psychological, social and mechanical predictors of back pain severity. Eur J Pain, 16(6), 921-33, https://doi.org/10.1002/j.1532-2149.2011.00091.x.

27 

Milner, A, Aitken, Z, Kavanagh, A, LaMontagne, AD, & Petrie, D. (2016). Persistent and contemporaneous effects of job stressors on mental health: a study testing multiple analytic approaches across 13 waves of annually collected cohort data. Occup Environ Med, 73(11), 787-93, https://doi.org/10.1136/oemed-2016-103762.

28 

Lundberg, U. (2015). Work conditions and back pain problems. Stress Health, 31(1), 1-4, https://doi.org/10.1002/smi.2633.

29 

Li, CH. (2016). The performance of ML, DWLS, and ULS estimation with robust corrections in structural equation models with ordinal variables. Psychol Methods, 21(3), 369-87, https://doi.org/10.1037/met0000093.

30 

Rhemtulla, M, Brosseau-Liard, PE, & Savalei, V. (2012). When Can Categorical Variables Be Treated as Continuous? A Comparison of Robust Continuous and Categorical SEM Estimation Methods Under Suboptimal Conditions. Psychological Methods, 17(3), 354-73, https://doi.org/10.1037/a0029315.

31 

Hagberg, M, Violante, FS, Bonfiglioli, R, Descatha, A, Gold, J, Evanoff, B, et al. (2012). Prevention of musculoskeletal disorders in workers: classification and health surveillance - statements of the Scientific Committee on Musculoskeletal Disorders of the International Commission on Occupational Health. BMC Musculoskelet Disord, 13, 109, https://doi.org/10.1186/1471-2474-13-109.

32 

Blyth, FM, Van Der Windt, DA, & Croft, PR. (2015). Chronic Disabling Pain: A Significant Public Health Problem. Am J Prev Med, 49(1), 98-101, https://doi.org/10.1016/j.amepre.2015.01.008.

33 

Turk, DC, & Melzack, R. (2011). Handbook of pain assessment (3rd ed). NY, US: Guilford Press.


Additional material