Background
Musculoskeletal disorders are widely recognized as common causes of disability and sick leave [
1‐
3]. Among musculoskeletal disorders, neck and shoulder pain are common, though prevalence estimates tend to differ across studies, primarily due to differences in case definitions. In the general population, estimates of the 12-month prevalence are 2–11% for activity-limiting neck pain [
3] and 5–47% for shoulder pain [
4]. Among workers, 11–14% report activity limitation due to neck pain [
5]. Worldwide, neck pain is the fourth most common reason for years lived with disability [
1] and in Denmark, 16% of days on sick leave in 2015 were caused by neck pain [
6]. Not only does sickness absence imply costs for society [
7]; the potentially detrimental implications to the individual are also well described [
8] as are the association between long-term sick leave and the increased risk of premature withdrawal from the labour market [
9‐
11]. In accordance with the above, sickness absence as a focus of political concern is well established [
7].
Over the past decades, the challenge of rehabilitating sickness absentees with musculoskeletal disorders has been addressed [
12,
13]. Populations suffering from low back pain (LBP) are well represented in the body of literature; studies on sub-acute LBP offer moderate evidence on the positive effect of multidisciplinary rehabilitation in terms of improving disability and reducing sickness absence [
14]. For chronic LBP, it is suggested based on moderate evidence that multidisciplinary rehabilitation is superior to physiotherapy with respect to return to work (RTW), pain and disability and superior to usual care with respect to pain and disability [
12]. A recent review on back, neck and shoulder pain found positive RTW outcomes in studies using a multidisciplinary approach and the assignment of case managers [
15]. The involvement of workplaces has also been proven beneficial [
13‐
17]. In Denmark, the work outcomes of different studies have not been unanimous. Thus, a study from 2009 suggested positive outcomes on RTW and duration of sick leave when applying coordinated, tailored work rehabilitation in workers with musculoskeletal disorders [
18]. In this study [
18] however, only 19% of the participants had neck pain. More recent Danish studies evaluating work outcomes found positive effect of tailored physical activity after 3 months [
19], an effect which was however not maintained at 11 months of follow-up [
20]. Like in the study by Bültmann et al. [
18], these studies included participants with both back, neck and shoulder pain [
19,
20]. So while studies investigating pain and disability in neck and shoulder participants are common, participants with these pain locations often constitute only a minority in studies investigating work outcomes. Regarding shoulder disorders, the work outcomes of a Danish study evaluating physiotherapy exercises and occupational medical assistance are awaited [
21]. In a review on the effect of different treatments for impingement syndrome
1 [
24] only few studies reported RTW as an outcome; neither of these fulfilled the authors’ criteria for “high quality study” and neither of these evaluated the effect of multidisciplinary interventions. Accordingly, how to rehabilitate workers on sick leave with neck and shoulder pain is a question yet to be addressed [
23,
24].
Discussion
Two main findings from this study warrant exploration. One is the lack of difference between a multidisciplinary intervention compared to a brief intervention with respect to RTW, pain and disability in sick-listed workers with neck or shoulder pain. The other is the discouraging fact that less than 60% of the study population returned to work during the first year.
As for the lack of difference between the MDI and the BI; the study conducted by Bültmann et al. [
18] reported a significant improvement in RTW status at 1-year follow-up in a Danish study on sick-listed workers with musculoskeletal disorders. Some notable differences in interventions and study populations may explain why we did not find similar results. The involvement of workplaces was a key element as 45% of participants in the intervention group had roundtable discussions arranged at the workplace in Bültmann’s study. Also, a maximum duration of the intervention equivalent to 3 months was settled on. The mean duration of sick leave prior to the intervention was approximately 6 weeks [
18]. In the present study, only 19 (22%) in the MDI group had roundtable discussions arranged, sick leave was longer and the median duration of the MDI was 4.6 months (IQR 3.3–7.4).
Another possible explanation for the lack of difference between the MDI and the BI groups could be the similarities of the clinical services provided by the rheumatologist and the physiotherapist. The approach to the participants in both groups was based on a non-injury model as inspired by Indahl et al. [
27] and Hagen et al. [
45]. Both Myhre et al. [
46] and Brendbekken et al. [
47] had the same similarities between control and intervention groups. They did not find differences in RTW outcomes either. The reassurance provided by thorough examinations and explanations from two clinicians dedicated to spine disorders should probably not be underestimated—a point which has also previously been stated [
40,
45].
Less than 60% of the participants returned to work during follow-up which is inferior to the results from similar studies describing RTW for more than 70% of their participants [
18,
40,
46], and the modest RTW results warrant exploration.
In the randomized trials by Jensen et al. [
40] and Myhre et al. [
46], a multidisciplinary intervention much similar to the one used in the present study was offered; both reported successful RTW for approximately 70% of their participants. Differences in pain location might be an explanation, as only LBP patients were included in the former [
40] whereas in the latter [
46], both neck and back pain patients were included; however, the distribution of pain locations is not presented. In the above mentioned study by Bültmann [
18], only 12% of the study population had neck pain. Recent studies by Andersen et al. [
19,
20] found promising RTW results of tailored physical activity at 3 month follow-up but these were not maintained at 11 month follow-up; neither the tailored physical activity program nor the pain self management program improved RTW compared to the reference group. The outcome measure in these studies was RTW status (yes/no) and although different from the present four consecutive weeks of RTW [
48], the proportion of participants returning to work was closer to our results than in the studies by Jensen et al. [
40] and Myhre et al. [
46]. A possible explanation could be a larger proportion of the study population suffering from neck and upper extremity pain. However, this information was not provided by Andersen et al.
While involvement of workplaces should be a key element in the process of RTW [
15‐
17,
39,
42], our RTW results were notably poorer compared to the previously published LBP study by Jensen et al., although the rehabilitation programs were very similar [
40]. In contrast to the previously mentioned studies [
18,
40,
46], the present study included only participants with neck and shoulder pain. This may lead to considerations of the possibility of a poorer RTW prognosis for people with neck and shoulder pain in general compared to people with LBP.
Apart from the pain location, the present study population also had baseline characteristics that might have influenced the process of returning to work. At inclusion, the participants were troubled by severe pain intensity and considerable psychosocial impact of their pain (ÖMPQ) (Table
2). Both high pain intensity scores and ÖMPQ scores >90 have been shown to predict future sick leave [
15,
34,
49] and thus may have affected RTW outcomes. At baseline, almost half of the study population had musculoskeletal comorbidity and approximately one-third had ≥3 previous sick leaves. Both factors are known to have negative prognostic value with respect to RTW [
15,
42].
In studies with RTW outcomes similar to ours, explanations may also in part be found in baseline characteristics. Thus, in Andersen et al.’s studies [
19,
20] where approximately 60% returned to work, more than half of the study population had previous sick leave episodes. In the study by Brendbekken et al. [
47], the mean duration of sick leave prior to inclusion was 147 days. Both number of previous sick leaves and current sick leave duration are negative prognostic factors for RTW [
15].
The study had several strengths. One was the randomized design which ensured comparability between the two groups with the exception of a larger proportion of part-time sick-listed participants in the MDI group compared to the BI group. However, this variable was adjusted for. Second, we had 100% follow up on the primary outcome thus eliminating the risk of attrition bias. A third strength of the study was the ITT analysis. The fact that baseline clinical examinations were carried out blindedly before randomization was considered a further strength.
The study also had some limitations. First, given the nature of the interventions, it was not possible to perform all interventions in a blinded manner. A second potential weakness was the recruitment of participants. The GPs received written information about the study with encouragement to refer patients on sick leave due to neck and shoulder pain. They may have referred only high-risk patients because they would consider it more cost-effective to treat low-risk patients in primary care. Whether GPs have had such considerations is unknown. Although the referral pattern was similar to the LBP study [
40] this aspect needs to be taken into account when considering generalizability of the study.
Third, participants with sickness absence lasting 4–16 weeks were included although longer sickness spells constitute an independent risk factor of not returning to work [
15,
39]. An exploratory analysis to test if a more rigid inclusion criterion on sick leave (4–8 weeks) would have yielded different results was performed; this was not the case (data not shown). Fourth, the number of non-responders on the secondary outcomes was substantial (n = 89) introducing a potential risk of selection bias in the assessment of secondary outcomes. Non-response analysis (data not shown) did not show any statistically significant differences between responders and non-responders with respect to intervention groups, RTW or any of the baseline variables. Only the allocation to exercise groups differed between responders and non-responders. This was a difference not suspected to have biased the estimates of the secondary outcomes. Nor do we, to the best of our knowledge, consider the nested RCT [
31] to threaten the estimation of the results in the present study. We base this on the equal distribution of exercise groups between the BI and the MDI groups (Table
2), and the fact that the participants had equal pain improvements following the exercise programmes in the nested RCT [
31].
The access to register data on RTW allowed for 100% follow-up on the primary outcome and the validity of DREAM has previously been demonstrated [
41]. A fifth limitation was that appraisal of register data revealed minor inconsistencies at baseline between self-reported and register-based sick leave status. According to register data, 15 participants did not fulfil the inclusion criteria of sick leave ≥4 weeks. These participants were equally distributed between intervention groups and tentative per protocol analysis excluding these participants did not alter the results (adjusted HR = 0.70. 95% CI 0.44–1.12). It cannot be ruled out that the ITT analysis might introduce a minor degree of non-differentiated information bias. But this does not change the overall estimates of RTW and apart from maintaining the strength of randomization, the ITT analysis also displays high external validity since self-reported sick leave status is the only accessible information on the day of inclusion.
Sixth, the time spent on the MDI warrants consideration. Due to the setup of the study, participants in the MDI group waited 1–2 weeks after randomization before receiving the part of the intervention that differed from the BI group. Meanwhile, time at risk began at the day of randomization for both groups. Remembering the poor prognosis associated with prolonged sick leave [
9‐
11,
15,
39] this was inexpedient but unfortunately unavoidable. Seventh, due to the sample size, there is approximately 30% risk of type 2 errors, i.e. a risk of overlooking an actual difference between the MDI and the BI intervention. We do not, however, consider power problems to explain the lack of difference, but rather characteristics of the population and intervention as described above.
Finally, only a minority of participants in the MDI experienced workplace involvement. In the latest review on workplace interventions, Cullen et al. present strong evidence on the positive work outcomes when applying multi-domain interventions orchestrated from the workplace [
17] and it could be argued that workplace involvement should have been mandatory. As previously described, this was not possible, because the majority of participants preferred to keep their health problems secret to their employers. As described, this discretion regarding health issues is rooted the Danish Health Information Law [
25]. Whether a stronger focus on workplace involvement could have improved the results in the MDI group cannot be ruled out.
On the macro level, the “economic climate” is known to potentially affect sickness absence [
7]. Our choice of outcome measure was constricted to four consecutive weeks of self-support, alternatively four consecutive weeks of holding a job supported by the social system. But since the study was performed during a period of economic recession in Denmark, exploratory analyses were performed allowing for the outcome RTW to be also 4 weeks of unemployment benefits and State Education Fund Grants (both reflecting readiness to return to work). These analyses still did not show significant differences in RTW between the groups but increased the HR in favor of the MDI (data not shown). Rather than interpreting the increased HR as the results of a successful MDI intervention, this merely reflects the termination of employment for some of the MDI participants. The combination of general economic recession and an intervention lasting several weeks may have contributed to the loss of jobs for some of the MDI participants.
In conclusion, no difference was found in RTW rates between the BI and the MDI group. Nor were there any differences in follow-up pain and disability between the groups. We do however assume that the evidence on the effect of multidisciplinary interventions in LBP [
12,
14] and other musculoskeletal disorders [
15,
17] is transferable to neck and shoulder pain. For clinical practice, several studies over the years e.g., [
27,
40,
45‐
47] have suggested efficacy of a brief clinical intervention based on a non-injury approach with a focus of diminishing fear and restoring/maintaining normal daily activities. Add-on of a multidisciplinary intervention including a case manager as in the current study does not seem to improve RTW outcomes. Rather, evidence suggests the necessary involvement of workplaces.
Another implication for clinical practice derives from the above recognition: There is not only a need for efficient RTW interventions but also for increased focus on preventing sickness absence, i.e. how do clinicians identify patients at high risk of sickness absence? Feleus et al. recently published a study identifying three different trajectories for sickness absence (low, intermediate and high risk) in patients presenting in primary care with complaints of the arm, neck and shoulder [
50]. They also identified bio-psycho-social variables associated with these trajectories. For whiplash-associated disorders, a tool predicting both chronic disability and full recovery has been developed [
51,
52]. For neck pain however, current evidence does not support clinical use of neither prognostic nor prescriptive clinical prediction rules [
53].
Better understanding of the prognostic factors and development of clinical prediction rules regarding RTW outcomes in neck and shoulder pain are suggested as future focus areas in research.