Objectives To investigate whether psychosocial working conditions predict the development of low back pain (LBP) in female eldercare workers while adjusting for physical workload and depressive symptoms.
Methods We investigated risk for developing LBP for between 1 and 30 days in the past year and developing LBP for more than 30 days in the past year at follow-up in 1537 female eldercare workers with no LBP in the year before baseline. Data were analysed using logistic regression analysis adjusted for sociodemographic variables, health behaviours, physical workload and depressive symptoms at baseline. Follow-up ranged from 18 to 22 months.
Results High emotional demands, high and medium role conflicts and low influence predicted risk for reporting LBP for between 1 and 30 days in the past year at follow-up after adjustment for sociodemographics, health behaviours and physical workload. All associations became statistically non-significant when adjusted for depressive symptoms. Low and medium influence at work and high emotional demands predicted risk of reporting LBP for more than 30 days in the past year at follow-up, after adjustment for sociodemographics, health behaviours and physical workload. For employees with low (OR 4.16; 95% CI 1.36 to 12.75) and medium (OR 3.93; 95% CI 1.37 to 11.22) influence, this risk remained statistically significant after adjustment for depressive symptoms.
Conclusions Most psychosocial working conditions in this study were no longer associated with risk of LBP after adjustment for depressive symptoms. However, low and medium influence at work predicted risk for LBP for more than 30 days after adjustment for both physical workload and depressive symptoms.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
What this paper adds
Numerous studies have shown a prospective association between psychosocial working conditions and risk for the development of musculoskeletal disorders, but only a few studies have investigated this association in healthcare workers.
However, few studies have adjusted for physical workload and negative affect, such as depressive symptoms, rendering the analyses vulnerable to confounding and reporting bias.
This study shows that high emotional demands, high role conflicts and low and medium influence at work predict low back pain after adjusting for physical workload.
This study shows further that high emotional demands and high role conflicts no longer predict low back pain after adjustment for depressive symptoms, although low and medium influence at work predicts low back pain after adjusting for both physical workload and depressive symptoms.
Improving influence at work may contribute towards preventing low back pain among eldercare workers.
Musculoskeletal disorders (MSD) are prevalent in high-income countries1–4 and are associated with costs for individuals, workplaces and societies.5–10 Physical workload and physical exertion have been shown to predict MSD,11–14 but a growing number of studies have also indicated a possible impact of psychosocial working conditions on risk of MSD.8 ,15–17
The aetiology of MSD is considered multifactorial.16 With regard to psychosocial working conditions, two causal pathways have been proposed. First, adverse psychosocial working conditions, for example high role conflicts or low influence at work, may mean that work tasks—including physically strenuous work tasks—cannot be carefully planned and executed, thereby resulting in increased physical workloads and subsequent increased risk of MSD.15 Second, adverse psychosocial working conditions may entail psycho-physiological stress reactions, for example, increased release of cortisol, which may lead to muscle tension or decreased blood supply in the extremities, increasing the risk of MSD.16 ,18–23
Several meta-analyses have been published on the longitudinal association between psychosocial working conditions and risk for MSDs.8 ,16 ,17 In the most recent meta-analysis, Hauke et al examined 54 prospective studies.15 They found that psychosocial working conditions such as low social support, high job demands, low job control, low decision authority and low job satisfaction were associated with risk for pain in the low back, the neck/shoulder and the upper extremities. However, the meta-analysis also revealed important weaknesses in the included studies. First, most studies on psychosocial working conditions and MSD failed to adjust for physical workload, which might be an important confounder. Second, the reviewed studies usually did not adjust for the different types of negative affectivity, such as symptoms of depression at baseline. This is a severe limitation, as baseline depressive symptoms might lead to over-reporting of adverse psychosocial working conditions at baseline24–26 and to an increased risk of MSD,27 ,28 which implies that the observed association between self-reported psychosocial working conditions and MSD may be spurious. Hence, it remains unclear whether the apparent association between psychosocial working conditions and risk of MSD reported in the review by Hauke and colleagues is genuine or due to these two variables, physical workload and depressive symptoms.
The aim of this article is to examine the prospective association between psychosocial working conditions and risk for low back pain (LBP) among female eldercare workers while taking potential confounding by and biases due to physical workload and depressive symptoms into account. Participants in the study were free of pain in the year prior to baseline, and we adjusted the analysis for a comprehensive measure of physical workload and a well-established measure of depressive symptoms. A previous analysis has shown that emotional demands, role conflicts, influence at work and quality of leadership predict long-term sickness absence in the present study population29 and, accordingly, we examined these four indicators of psychosocial working conditions as predictors of LBP in the present study.
We used data from a prospective cohort study of all employees in eldercare services in 36 Danish municipalities. The aim of the cohort study was to investigate work environment, health and well-being among employees in eldercare. Respondents were surveyed using mailed questionnaires. The first round of data collection took place in late 2004 and early 2005 and the second round was conducted in the autumn of 2006. The follow-up period varied from 18 to 22 months. In the first round, we sent questionnaires to 12 744 respondents who were employed in the 36 municipalities. Of these potential respondents, 9949 completed the questionnaire (response rate 78%). One municipality decided to withdraw from the cohort during follow-up, reducing the population to 9847 respondents. Of these respondents, 7864 were still working in eldercare services in the participating municipalities at follow-up and were therefore eligible for participation in the second round survey in 2006. Of the eligible respondents at follow-up, 6307 (80%) responded. An analysis of differences between respondents and non-respondents at follow-up showed no significant differences between respondents and non-respondents with regard to body mass index (BMI), physical workload, smoking behaviour, leisure time physical activity and emotional demands. However, respondents were significantly older (p<0.0001), had longer tenure (p<0.0001), had fewer depressive symptoms (p=0.0481), higher influence (p<0.0001), higher quality of leadership (p=0.0002) and lower role conflicts (p=0.0069) than non-respondents.
To ensure a homogeneous study population, we excluded respondents who were not directly providing care services (N=1153) and male respondents (N=102), yielding a gross study population of 5052 respondents. We further excluded respondents who had any LBP in the 12 months before baseline (N=3440), and respondents with missing values on the measure of LBP at baseline (N=54) and at follow-up (N=21). The analyses of risk for LBP are therefore based on 1537 female care workers who were free of LBP at baseline.
Assessment of LBP
Participants replied to the standardised Nordic questionnaire on musculoskeletal pain symptoms in the low back, as defined by a drawing of this body region.30 The question concerned duration of pain during the last year, with the response categories: 0 days, 1–7 days, 8–30 days, 31–90 days, more than 90 days but not every day, and every day. For the statistical analyses we trichotomised pain into 0 days, 1–30 days and more than 30 days.
Assessment of psychosocial working conditions
We defined and measured four types of psychosocial working conditions using scales from the Copenhagen Psychosocial Questionnaire (COPSOQ)31: (1) emotional demands (four items), sample item: Is your work emotionally demanding? (2) role conflicts (four items), sample item: Are contradictory demands placed on you at work? (3) influence at work (four items), sample item: Do you have a large degree of influence concerning your work? (4) quality of leadership (four items), sample item: To what extent would you say that your immediate superior is good at work planning? Responses are scored on a 5-point Likert scale. The sum scales were rescaled from 0 to 100 points, with high scores indicating high levels of the measured work characteristic. The psychometric properties of these scales were satisfactory and are described elsewhere.32 We divided the scores of each scale into three levels. The first level approximated the lower quartile, the second level the two middle quartiles and the third level the upper quartile of the distribution.
Assessment of covariates
We adjusted the analyses for age, smoking status (current smoker (yes/no)), BMI (four categories: underweight, normal weight, overweight and obese), leisure time physical activity (four categories: predominantly sedentary, slightly active, fairly active and very active), tenure, job group (four categories: social and health care assistants, social and health care helpers, other care staff with no or short-term education and registered nurses/therapists), physical workload and depressive symptoms.
Physical workload was assessed with the Hollmann index which provides a measure of body postures and weight lifted during the work day.33 Five items describe postures of the trunk (straight upright, slightly inclined, strongly inclined, twisted, and laterally bent). Three items describe the position of the arms (two arms below shoulder height, one arm above shoulder height, and two arms above shoulder height). Five items describe the position of the legs (sitting, standing, squatting, kneeling on one or both knees, and walking or moving). Six items described the lifting of weights. Three items concerned lifting with the trunk upright and three with the trunk inclined at 60°. Each set of lifting items measured lifting of light weights (<7 kg), medium weights (8–30 kg) and heavy weights (>30 kg). All responses were measured on 5-point scales ranging from ‘never’ to ‘very often’. Weight factors, based on the biomechanical model ‘The Dortmunder’,34 were then multiplied by the item scores of the corresponding body postures reported in the questionnaire and then added to an index of physical work load. In summary, this index is a measure of physical work load on the lower lumbar spine over longer periods.33
We assessed depressive symptoms using the Major Depression Inventory (MDI). The MDI consists of 10 items assessing the presence of depressive symptoms (sample item: How large a part of the time in the last 2 weeks have you been feeling low in spirits or sad?). Each item measures the presence of a symptom during the last 2 weeks on a scale ranging from 0 (the symptom has not been present at all) to 5 (the symptom has been present all of the time). In accordance with the MDI guidelines, we built a depressive symptom sum score that ranged from 0 to 50 points.35 ,36
We used logistic regression analysis to examine the prospective association between psychosocial working conditions at baseline and two outcomes: (1) risk of LBP for between 1 and 30 days in the past year at follow-up and (2) risk of LBP for more than 30 days in the past year at follow-up. For both outcomes, we adjusted the analyses in four steps. In model 1, we adjusted for age; in model 2 we adjusted for age, BMI, leisure time physical activity, smoking, tenure and job group; in model 3 we further adjusted for physical workload; and in model 4 we adjusted for depressive symptoms. In all models each of the four psychosocial indicators were entered individually. Because the participants were clustered in work places, observations were not statistically independent.37 Therefore, all analyses were adjusted for random effects at the work place level by using the option ‘repeated subjects’ in the SAS programme. Data were analysed using SAS 9.2 (SAS Institute, Cary, North Carolina, USA).
Table 1 presents descriptive statistics for study variables. It shows that 453 respondents (29.5%) reported LBP for between 1 and 30 days in the past year and 82 respondents (5.3%) reported LBP for more than 30 days in the past year at follow-up.
Table 2 shows the prospective associations between psychosocial working conditions at baseline and risk of LBP for between 1 and 30 days in the past year at follow-up. High emotional demands, high and medium role conflicts and low influence at work predicted risk of LBP for between 1 and 30 days in the past year after adjustment for age, BMI, smoking, leisure time physical activity, tenure and job group (model 2). Further adjustment for physical workload led to attenuation of ORs, but all ORs remained statistically significant (model 3). When we further adjusted for depressive symptoms at baseline, ORs were further attenuated and all associations lost statistical significance (model 4).
Table 3 shows the prospective associations between psychosocial working conditions at baseline and risk of LBP for more than 30 days in the past year at follow-up. High emotional demands, high role conflicts and low and medium influence at work predicted risk of LBP for more than 30 days in the past year after adjustment for the covariates in model 2. After adjustment for physical workload (model 3), ORs were attenuated and statistical significance was lost for role conflicts, whereas emotional demands and influence continued to predict LBP of more than 30 days in the past year (model 3). Further adjustment for depressive symptoms resulted in a loss of statistical significance for emotional demands, whereas low and medium influence at work still predicted risk of LBP for more than 30 days in the past year (model 4).
In this study of female eldercare workers who reported being free of LBP at baseline, psychosocial working conditions predicted LBP after adjustment for physical workload. Further adjustment of baseline depressive symptoms, however, strongly attenuated the ORs and only the association between low and medium influence at work and risk of LBP for more than 30 days in the past year remained statistically significant.
Earlier findings from prospective studies have shown that numerous psychosocial working conditions predict risk of MSD.15 Our results question these earlier findings and suggest that a substantial proportion of these findings may have been due to reporting bias caused by depressive symptoms. According to our results, adjustment for depressive symptoms or other forms of negative affect is crucial when studying the role of self-reported psychosocial working conditions in the aetiology of MSD.
Low and medium influence at work, however, predicted LBP with ORs in the magnitude of 4, even after adjustment for depressive symptoms and the other covariates. Hence, there was an apparent effect of influence at work on risk of LBP that is unlikely to be explained by reporting bias. Confounding by socioeconomic position also appears unlikely, as we examined a homogenous study population where all participants worked in eldercare, and where we adjusted for the remaining relatively small differences in occupational status.
Finally, many studies investigating associations between work-related psychosocial phenomena and risk for MSD have used measures of job satisfaction and job stress as predictors.15 However, such factors are less direct measures of psychosocial working conditions but rather reflect psychological reactions towards the psychosocial work environment. In the present study we focused on more direct indicators of psychosocial working conditions instead of psychological reactions. This was motivated by the consideration that our results would help identify those psychosocial factors that can be altered by changes in work organisation and the results of the present study are, therefore, more directly applicable in terms of preventing LBP among eldercare workers.
Mechanism linking influence at work to LBP
Low influence at work may affect risk of LBP through at least two pathways. First, employees with high levels of influence at work may to a greater extent be able to plan the execution of physically strenuous work tasks than employees with low influence, resulting in reduced physical workload and reduced physical strain. Second, a low level of influence at work may be a psychosocial stressor, resulting in psycho-physiological stress responses such as increased muscle tension or decreased blood supply in the extremities, and subsequent risk of MSD.16 ,18–22
We adjusted the analyses for potential confounding by physical workload and for reporting bias due to depressive symptoms. Whereas both adjustments attenuated the effect estimates, adjusting for depressive symptoms resulted in greater attenuation than adjusting for physical workload. Depressive symptoms might cause participants to over-report both adverse psychosocial working conditions and level of LBP, causing differential misclassification of the exposure and consequently a spurious association between psychosocial working conditions and LBP. Over-reporting of adverse psychosocial working conditions may be due to a tendency of depressed participants to perceive the environment, including the work environment, more negatively than their non-depressed colleagues.24 ,38 We acknowledge, though, that adjusting for physical workload and depressive symptoms are not without problems. Exposure to physical workload might not be a confounder, but an effect modifier, that is, the effect of psychosocial working conditions on risk of LBP theoretically might be expected to be stronger or weaker in the presence of high physical workload. However, it was not possible to investigate effect modification in the present study because of sample size restrictions. With regard to depressive symptoms, it can be argued that they are not a source of reporting bias, but that they are instead an intermediate step in the pathway linking adverse psychosocial working conditions to risk of LBP. It has been suggested that adverse psychosocial working conditions increase the risk of depressive symptoms39 and that depressive symptoms increase the risk of MSD.27 ,28 If that is the case, then adjustment for depressive symptoms would be inappropriate when studying the association between psychosocial working conditions and risk of MSD, as it, falsely, might lead to the conclusion that psychosocial working conditions were not associated with risk for LBP. In this case, the ORs reported in model 3 would be more appropriate than the ORs in model 4. Unfortunately, though, we cannot determine whether depressive symptoms are a confounder or a mediator in our data. We recommend, therefore, that further studies carry out more in-depth investigation of the role of depressive symptoms in the relationship between psychosocial working conditions and risk of MSD.
When estimating the association between an exposure and an outcome, the calculation of ORs typically yields higher risk estimates than prevalence ratios when the prevalence of the outcome increases. This issue is particularly important when the outcome has a prevalence of more than 10%.40 In our study, LBP for between 1 and 30 days had a prevalence rate of 29.5% at follow-up. Therefore, we repeated the analyses on this outcome using a log-binomial model to estimate prevalence ratios. We estimated log-binominal models only for the analyses in models 1, 2 and 3 (table 2), and were unable to estimate prevalence ratios for model 4. As described by Deddens and Petersen,40 in many situations statistical software fail to converge when estimating prevalence ratios as the maximum likelihood estimate may lie on the boundary of the parameter space. Such non-convergence is increasingly likely when more variables are added to the model. Although the ORs provided higher risk estimates for LBP for between 1 and 30 days than the prevalence ratios, statistical significance did not differ in the two types of analysis (ORs vs prevalence ratios).
In this study we investigated the association between psychosocial working conditions and risk for LBP for shorter or longer periods. It could be argued, though, that pain duration indicates little about the severity of the experienced pain. In the study questionnaire we also asked respondents to assess the intensity of their LBP. Additional analyses showed that with increasing pain duration, the likelihood of respondents reporting higher levels of pain intensity increased (Spearman correlation=0.47, p<0.0001, N=1537). Thus, it appears that duration of pain also approximates severity of pain to a considerable extent.⇓
Strengths and weaknesses
The strengths of the study are the prospective design, the differentiation between durations of LBP, and the use of established measurements for the assessment of both psychosocial working conditions and LBP. Unlike most other studies, we carried out a comprehensive assessment of the physical work environment and we adjusted for reporting bias by using an established instrument for measuring depressive symptoms.
These strengths have to be balanced against several limitations. First, all variables were measured by self-report, yielding measurements vulnerable to recall bias. This may be particularly relevant with regard to our measures of LBP, as we asked the respondents to remember how many days in the past year they had experienced pain. Imprecision of recall would increase randomness in the responses, which is most likely to result in an underestimation of the ORs and a widening of the CIs. Second, we only assessed the predictors at a single point in time, which implies that the present study yields little information on the effects of changes in psychosocial working conditions on risk for LBP among eldercare workers. Third, the present study had a follow-up period that varied between 18 and 22 months and at follow-up the respondents were asked to provide information on LBP in the past 12 months. This means that we have no information on cases of LBP in the remainder of the follow-up period. Accordingly, the analyses provide information on the association between psychosocial working conditions and risk for LBP in the longer term but not in the short term. Fourth, as we do not have any information on the LBP status of respondents prior to 1 year before follow-up, we do not know whether we actually are measuring recurring cases of LPB or whether the cases we observe are the first experiences of LBP among the respondents. However, by excluding respondents who had LBP in the 12 months before baseline, we investigated risk for LBP in a population that had not had LBP for a considerable period prior to its entrance into the present study. Fifth, non-respondents reported lower influence, lower quality of leadership and higher role conflicts. Thus, non-response was associated with three of the predictors at baseline. We do not know if non-response was also associated with onset of LBP at follow-up. If this was the case, for example because non-responders did not respond because of sickness absence due to LBP, then non-response might have caused underestimation of the true association between working conditions and risk of LBP in this study. Finally, we acknowledge that by adjusting for depressive symptoms we did not capture all types of negative affectivity. Although the MDI measures a broad range of negative affects, for example, low mood, sadness, tiredness, guilt and nervousness, it does not measure affects such as hostility or irritability that are included in negative affectivity scales, such as the PANAS-X scale by Watson et al.41 Thus, we cannot rule out that unmeasured aspects of negative affectivity may have biased our results. On the other hand, the use of a depressive symptom scale instead of a negative affectivity scale may also have advantages considering that depressive symptoms are currently intensively discussed as a major source of bias in occupational epidemiology.24 ,42 Moreover, depressive symptoms strongly predicted onset of low back and neck pain in a Canadian population-based study.28 This indicates that depressive symptoms may be an important factor in the aetiology of LBP and therefore should be considered as a confounder in studies on psychosocial factors and LBP.
As the health and eldercare services in most high-income countries are facing care staff shortages in the coming decades,43 multifaceted initiatives for preventing MSD may be of particular importance. Our study results suggest that initiatives to prevent MSD should also include improvements in psychosocial working conditions, particularly regarding employee influence in the planning and execution of their work tasks.
After adjustment for both physical workload and depressive symptoms, most associations between psychosocial working conditions and LBP lost statistical significance in this study of female eldercare workers. Low and medium influence at work, however, remained a statistically significant predictor of LBP for more than 30 days in the past year in all analyses. Improving influence at work might contribute to the prevention of LBP in female eldercare workers.
Contributors TC contributed to the data collection, conceived and designed the study, performed data analysis, wrote up the first draft of the study and revised the study. LLA conceived and designed the study, performed data analysis, and contributed to revising the study. AH and AFBJ conceived and designed the study, and contributed to revising the study. BA contributed to the data collection, conceived and designed the study, and contributed to revising the study. RR conceived and designed the study, contributed to the first draft of the study, and oversaw the revision of the study.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.