Workplace interventions to prevent work disability in workers on sick leave

Summary of findings for the main comparison. Workplace interventions compared to Usual care for Workers on sick leave

Workplace interventions compared to Usual care for Workers on sick leave
Patient or population: Workers on sick leave Settings: Workplace Intervention: Workplace interventions Comparison: Usual care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Workplace interventions
Time until first RTW Risk at return to work Follow‐up: 12 months	Study population¹		HR 1.55 (1.2 to 2.01)	608 (5 studies)	⊕⊕⊕⊝ moderate³
	799 per 1000²	917 per 1000 (854 to 960)²
	Low
	500 per 1000²	658 per 1000 (565 to 752)²
	High
	870 per 1000²	958 per 1000 (914 to 983)²
Time until lasting RTW Risk at return to work Follow‐up: 12 months	Study population⁴		HR 1.07 (0.72 to 1.57)	635 (6 studies)	⊕⊝⊝⊝ very low^5,6,7
	834 per 1000²	853 per 1000 (725 to 940)²
	Low
	560 per 1000²	585 per 1000 (446 to 724)²
	High
	920 per 1000²	933 per 1000 (838 to 981)²
Cumulative duration of sickness absence Days. Scale from: 0 to 365. Scale from: 0 to 365. Follow‐up: median 12 months	The mean cumulative duration of sickness absence in the control groups was 165.7 Days⁸	The mean cumulative duration of sickness absence in the intervention groups was 33.33 lower (49.54 to 17.12 lower)		1164 (8 studies)	⊕⊕⊕⊕ high
Recurrences Risk at recurrences of sick leave Follow‐up: 12 months	Study population		HR 0.42 (0.21 to 0.82)	99 (1 study)	⊕⊕⊕⊝ moderate^10,11
	250 per 1000⁹	114 per 1000 (59 to 210)⁹


Functional status Roland disability questionnaire. outcome was measured on different scales in different studies. Scale from: 0 to 24. Follow‐up: median 12 months	The mean functional status ranged across control groups from 5.76 ‐ 81.79	The mean functional status in the intervention groups was 0.33 standard deviations lower (0.58 to 0.08 lower)		628 (6 studies)	⊕⊕⊕⊝ moderate⁶
Depression Depression Anxiety Stress scale, and PHQ‐9 depression scale Follow‐up: 12 months	The mean depression ranged across control groups from 5.9‐24.6	The mean depression in the intervention groups was 0.12 standard deviations lower (0.35 lower to 0.11 higher)		133 (4 studies)	⊕⊝⊝⊝ very low^7,12
Pain Visual Analogue Scale Follow‐up: 12 months	The mean pain ranged across control groups from 3.4‐30	The mean pain in the intervention groups was 0.26 standard deviations lower (0.47 to 0.06 lower)		531 (5 studies)	⊕⊕⊕⊕ high
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: Hazard ratio;
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ Risks ranged from 0.50 to 0.87, therefore lowest and highest risks are presented. ² Risks are presented as the number of people who returned to work. ³ 60% of studies was assigned high or unclear risk of bias ⁴ Risks ranged from 0.56 to 0.92, therefore lowest and highest risks are presented. ⁵ Three of the six studies were assessed as high risk of bias. ⁶ I2 > 50% ⁷ The CI of the pooled effect size is sufficiently wide that the estimate could either support or refute the effectiveness of the intervention. ⁸ Presented as the number of days of sickness absence. ⁹ Risks are presented as the number of people who had a recurrent period of sickness absence. ¹⁰ Not applicable ¹¹ Sparse data, only one study for this outcome. ¹² Three of the four studies were assessed as high risk of bias.

Background

Description of the condition

In recent years, the participation of individuals with chronic health problems in the workforce has become increasingly essential in order to address the decline in labour supply associated with an ageing population (OECD 2010). Many workers with health problems leave the labour market temporarily or permanently, and too few people with reduced work capacity manage to continue working. Work disability may be due to 1) musculoskeletal disorders, such as back pain and upper‐extremity disorders, 2) mental health problems, such as depression and adjustment disorders, and 3) other health conditions, such as cardiovascular diseases or cancer. Work disability is a major public health problem in Western industrialised countries. It has a considerable economic burden for society (Dall 2013; OECD 2010; Veronese 2012; Wilkie 2012). Around 6% of the working‐age population on average relies on disability benefits, leading to public spending on disability benefits of on average of 2% of the gross domestic product across Organisation for Economic Co‐operation and Development countries, rising to 4% to 5% in countries such as Norway, the Netherlands, and Sweden (OECD 2010). Employment rates of people with disability are 40% below the overall level, highlighting the need for efforts to support people with disability to return to work and stay in their jobs. Besides the high costs for society, work disability may have serious consequences for workers (Schandelmaier 2012). Being employed is a valuable societal role and is an important source of income; work disability might therefore lead to poorer quality of life and loss of social identity (Stigmar 2013). Furthermore, work disability may result in permanent exclusion from work (Waddell 2006).

Description of the intervention

Studies indicate that return‐to‐work (RTW) interventions should be carried out at the workplace instead of a rehabilitation centre, for example. It is furthermore important to involve key stakeholders in the RTW process. It has been shown that RTW interventions involving workplace adaptations and stakeholder involvement are more effective on RTW than workplace‐linked interventions such as exercise (Carroll 2010; Franche 2005a; Haugli 2011). In this review, we defined workplace interventions by either changes to the workplace or equipment, changes in work design and organisation, changes in working conditions or work environment, and involvement of (at least) the worker and the supervisor (Anema 2004).

How the intervention might work

Studies indicate that medical interventions solely do not show a positive effect on work‐related outcomes (Anema 2009; Arends 2012; Loisel 2001). If the cause of work disability is associated with the workplace, then a return to an unchanged workplace (with or without appropriate treatment for the disorder) may lead to recurrences in the longer term (Adler 2006; Pichora 2010; Sanderson 2006). By incorporating workplace adaptations, workplace interventions aim to reduce barriers for RTW. Supervisors influence health outcomes of employees, and supervisor support is associated with lower sickness absence (Munir 2012; Skakon 2010). Symptoms are generally not addressed by workplace interventions. However, we do hypothesise that earlier RTW is not associated with more severe symptoms.

Why it is important to do this review

Recent economic developments have led to policy changes. Disability benefit systems now focus more on assessing the remaining work capacity of a person applying for a benefit instead of assessing disability (OECD 2010). Timely RTW is of great benefit for both the sick‐listed workers and their employers, since there is a strong association between increased length of sickness absence and increased risk of future disability pension (Lund 2008). RTW is influenced by various psychosocial factors (Baldwin 1996; Clay 2012; Steenstra 2005; Sullivan 2005; Turner 2007; WHO 2001). It is therefore important to report on the research on individual interventions aimed at reducing workplace barriers to RTW (Nordqvist 2003; Schultz 2007; Young 2005).

This review provides insight into the effectiveness of workplace interventions on improving RTW. There is still uncertainty about the effectiveness of these interventions, especially for workers on sick leave due to mental health problems and other health conditions like cancer and cardiovascular disease.

Objectives

To determine the effectiveness of workplace interventions in preventing work disability among sick‐listed workers, when compared to usual care or clinical interventions.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) only.

Types of participants

We incorporated all studies concerning working‐age adults (18 to 65 years) who were on sick leave. We included studies conducted with full‐ and part‐time workers.

Types of interventions

The Cochrane Work review group has classified workplace interventions as appropriate for disability management (Schonstein 2006). For this review, we used the term 'workplace intervention' for interventions focusing on changes in the workplace or equipment, work design and organisation (including working relationships), working conditions or work environment, and occupational (case) management with active stakeholder involvement of (at least) the worker and the employer (Anema 2004; Franche 2005b). We defined active involvement as face‐to‐face conversations about RTW between (at least) the worker and the supervisor. In this review, a workplace intervention must contain work changes and stakeholder involvement, specifically the employer/supervisor. This definition is a synthesis of the International Ergonomics Association definition of ergonomic interventions and the Waddell et al definition of occupational interventions (Stapleton 2000; Waddell 2001). Changes in the workplace and equipment include changes in the furniture or the materials needed to perform the work. Changes in the work design and organisation include changes in schedules or tasks, training in task performance, and altered working relationships with supervisors and coworkers. Changes in working conditions refer to the financial and contractual arrangements and changes in the work environment concerning noise, lighting, vibration, etc.

We based the definition of a workplace intervention on definitions from studies on musculoskeletal disorders. However, the definition seems suitable for interventions for RTW on workers with mental health problems as well. These interventions comprise at least advice about changes in work processes to facilitate RTW or the preparation of a RTW plan that includes the worker and the supervisor (Blonk 2006; van Oostrom 2008; Vlasveld 2008). We compared identified workplace interventions with either usual care (the care usually offered to a sick‐listed worker, for example guideline‐based care) or clinical interventions (for example graded activity, problem‐solving therapy).

As long as the workplace intervention was a structural part of the intervention (with the intention to offer the workplace intervention to all participants in the intervention group), we did not exclude studies with interventions that included more components than described in the definition of a workplace intervention. Our definition allowed us to include only interventions that were linked closely to the workplace and that focused on work adaptations or the involvement of stakeholders from the work environment. We excluded interventions that were intended to simulate the demands of work in a laboratory setting, without changes to or involvement of the workplace itself in the RTW process.

We also excluded studies if the intervention was:

focused on primary prevention of sickness absence, that is, targeted to healthy workers as opposed to those on sick leave;
not focused on RTW as the main goal;
group‐based (focused on an organisation) rather than individual‐based;
focused on education about ergonomics only, and did not result in work adaptations;
aimed at posture modifications only without RTW as the goal.

Types of outcome measures

Primary outcomes

Sickness absence is operationalised in many ways (Steenstra 2003). However, when studies used different ways of operationalisation, we only analysed the data collected in the following manners.

Time until first RTW: a period of absence from work because of sickness both preceded and followed by a period of at least one day at work (consensus definition, de Vet 2002).
Time until lasting RTW: a period of absence from the first day of sick leave to full RTW in previous or equal work for at least four weeks without dropping out.
Cumulative duration of sickness absence: total days of sick leave during the follow‐up period (resulting from one or more periods of sickness absence).
Recurrences of sickness absence: the number of days until a recurrence; or the frequency and duration of recurrent episodes of sick leave.

Our primary outcome was operationalised in different ways, and we chose to prioritise the outcome time until first RTW, as it is the most commonly used RTW outcome internationally, and hence has greater relevance. The aim of modifying work tasks or processes is to guide sick‐listed workers back to work faster than by applying only care as usual. From a socioeconomic perspective, every day of earlier RTW is beneficial. Since the outcome time until first RTW does not take recurrences into account, we also analysed the outcomes time until lasting RTW (first episode of sick leave), cumulative duration of sickness absence, and recurrences (after the first episode of sick leave). By using these outcomes besides time until first RTW we include the follow‐up of (recurrent) sick leave after first RTW.

The operationalisation for time until lasting RTW is based on the Dutch social security system. There are large differences between countries in social security systems and the way sickness absence is registered (de Vet 2002), therefore the cut‐off point for lasting RTW at four weeks is just one example. If studies reported a definition of prespecified time periods for lasting RTW, we included the data from these studies in the time until lasting RTW outcome.

A differentiation between short‐ and long‐term sickness absence is needed (Uegaki 2007). In the past, dichotomous outcomes (returned to work versus not returned to work) were often used for absence caused by sickness. However, use of these measures results in a loss of information about the exact duration of work disability, and the episodic nature of work disability is neglected. Continuous sickness absence outcomes are now more frequently used. This is especially important when an intervention is focused on RTW and when sickness absence is the primary outcome of the study, as in this review. We therefore excluded studies that only reported a dichotomous measure of sickness absence.

Secondary outcomes

Secondary outcomes were:

functional status;
quality of life;
general health;
depression;
pain levels; and
direct and indirect costs of work disability.

These outcomes are likely to be meaningful for workers who are on sick leave, their employers, their care providers (such as treating and occupational physicians), insurers, and the policymakers who are involved in decision‐making.

Search methods for identification of studies

Electronic searches

For this review, we identified studies by searching the following databases:

Cochrane Work Trials Register (31 October 2013)
The Cochrane Central Register of Controlled Trials (CENTRAL) (Cochrane Library) Issue 2, 2015
MEDLINE (PubMed); 2 February 2015
Embase; 2 February 2015
PsycINFO (EBSCO); 2 February 2015

We have presented the search strategies used in Appendix 1. We did not apply any language restrictions.

Searching other resources

We handsearched the reference lists of relevant review articles and eligible studies. We also made use of personal contacts with experts in occupational health to identify studies for inclusion in the review.

Data collection and analysis

The methods of this review followed the Cochrane Handbook for Systematic Reviews of Interventions (Version 5.1.0) (Higgins 2011).

Selection of studies

We stored titles and abstracts (if available) of all identified studies in a new database in Reference Manager. We generated a bibliography that included the title, keywords, and abstract of each reference found after removing duplicate references.

We completed study selection in two steps. In the first step, two review authors (MvV and SvO) screened the titles, keywords, and abstracts of all references retrieved by the literature search to determine if articles met the inclusion criteria. The inclusion criteria were: study design was an RCT, participants were sick‐listed workers, the intervention studied met the definition of a workplace intervention, and sickness absence was measured continuously. In the second step, we retrieved the full‐text article for studies where we were could not decide upon inclusion or exclusion in the first step. We read these full‐text articles and assessed them for inclusion. We used a consensus procedure to resolve disagreements about the inclusion of RCTs, and consulted a third review author (CB) if the disagreement persisted. We documented the criteria for exclusion (design, intervention, population, and outcome) for each study we excluded.

Data extraction and management

Two review authors (MvV and SvO) independently extracted the data onto a predesigned data extraction form. This form included essential study information about participants, interventions, outcome measures, and results. We initially used a small sample of the articles to test whether the form was feasible. We resolved disagreements about the data extraction by consensus between the two review authors, or if the disagreements persisted we consulted a third review author (CB). We contacted the authors of articles that contained insufficient information. We reminded authors who did not respond. We eventually received all the information that we needed that the authors of studies had not reported.

Assessment of risk of bias in included studies

Two review authors (MvV and SvO) independently assessed the risk of bias of the included RCTs. We assessed risk of bias using the Cochrane tool for risk of bias (Higgins 2011). We assessed the blinding of outcome assessment both for sickness‐absence and health outcomes.

We assessed the following criteria:

Random sequence generation (selection bias)
Allocation concealment (selection bias)
Blinding of participants and personnel (performance bias)
Blinding of outcome assessment (detection bias): sickness‐absence outcomes
Blinding of outcome assessment (detection bias): health‐related outcomes
Incomplete outcome data (attrition bias)
Selective reporting (reporting bias)
Other bias

We judged all studies as having high, low, or unclear risk of bias in each of the above domains. We used a consensus method to resolve disagreements, consulting a third review author (CB or HdV) if the disagreements persisted. We identified four key 'Risk of bias' domains for this review:

Random sequence generation
Allocation concealment
Blinding of outcome assessment (sickness‐absence outcomes)
Incomplete outcome data

We considered studies to be at low risk of bias overall when we judged them to be at low risk of bias in all four key domains. We did not consider blinding of participants and personnel to the intervention as a key domain because the context of the workplace does not allow blinding (Schonstein 2003; Tveito 2004). We considered studies to be at high risk of bias overall when we judged them to be at high risk of bias in one or more of the four key domains. We considered studies to be at unclear risk of bias overall when we judged them to be at unclear risk of bias in one of the four key domains and at low risk of bias in the other three key domains.

Measures of treatment effect

We entered the outcome data for each study into the data tables in Review Manager (RevMan) to calculate the treatment effects (RevMan 2014). We used hazard ratios for time‐to event outcomes (time until lasting RTW and time until first RTW). We used mean differences for the continuous outcome cumulative duration of sickness absence, since all studies measured this outcome in days. For the outcomes functional status, pain, and depression, we used standardised mean differences, because studies measured these outcomes with different instruments.

Unit of analysis issues

We included three cluster‐RCTs in this review, and these studies did not account for the design effect (Anema/Steenstra 2007; Loisel 1997a; Noordik 2013a). The Anema/Steenstra 2007 study was a replication of the Loisel 1997a study, and all three studies randomised on the level of the occupational physician. The intracluster correlation coefficient (ICC) is generally low on the level of the treatment provider. Only Anema/Steenstra 2007 reported the ICC, and it was less than 0.01. Since the Anema/Steenstra 2007 study was a replication of the Loisel 1997a study, and all three studies randomised on the level of the occupational physician, we assumed that the ICCs in the Loisel 1997a and Noordik 2013a studies were also considerably small. Because the ICCs were low and did not have a large impact on outcome data (as shown in the Anema/Steenstra 2007 study), we chose not to adjust our analyses for unit of analysis errors.

Dealing with missing data

When data was reported in a format we could not extract from the publication we contacted the primary authors.

Assessment of heterogeneity

We investigated heterogeneity due to differences in populations by performing analyses for specific causes of work disability: musculoskeletal disorders, mental health problems, or other health conditions. We investigated heterogeneity due to difference in applied interventions by performing analyses for workplace interventions offered alone, or in combination with a cognitive behavioural intervention. To determine the presence or absence of heterogeneity, we analysed I². When I² was more than 50%, we considered studies to be heterogenous. We conducted sensitivity analyses by analysing only RCTs we judged to have a low risk of bias.

Assessment of reporting biases

We used a funnel plot to check for publication bias where more than five studies were available for inclusion in the analysis.

We excluded no papers on the basis of language.

Data synthesis

We pooled the data with Review Manager 5.3 software (RevMan 2014). We plotted the results of each RCT as point estimates with corresponding 95% confidence intervals.

Most outcomes regarding sickness absence were time‐to‐event data (time until lasting RTW, time until first RTW, time until recurrence). The Cox proportional hazard model is used to analyse time‐to‐event data. In this approach, workers who do not return to work during the entire follow‐up period are censored to be sure that the total follow‐up period is analysed as sick leave. Cox proportional hazard regression models are used to determine a hazard ratio. We performed log transformations of the hazard ratios. We then combined the study results using the generic inverse‐variance method with the estimates of log hazard ratios and standard errors from the results of Cox proportional hazards regression models. For one study the results of Cox proportional hazard regression models were not available (Feuerstein 2003). We therefore estimated a hazard ratio based on the log rank test (Parmar 1998). For all analyses, we used the random‐effects model because of the heterogeneity in the type of work disability, duration of sickness absence, and the variation in interventions among studies.

Cumulative duration of sickness absence was usually presented as a continuous outcome and, we therefore calculated the mean difference. Functional status, quality of life, pain, and symptoms were continuous outcomes. For these continuous data, we determined standardised mean differences (with 95% confidence interval) to summarise the effect depending on whether or not these outcomes were measured with different scales. We could not combine changes from baseline scores and final post‐intervention scores in the forest plots when using standardised mean differences. We therefore calculated or requested the final post‐intervention scores for all self‐reported outcomes. In an RCT, mean differences based on changes from baseline can usually be assumed to address exactly the same underlying intervention effects as analyses based on final measurements (Higgins 2011). We could not pool quality‐of‐life outcomes due to large conceptual differences in measurement instruments and subscales used. Some studies only measured a couple of subscales of a quality‐of‐life measurement instrument that were related to workers' specific disabilities, while other studies measured an overall quality‐of‐life score.

We have summarised the results on the costs data in Table 1.

Table 1. Cost outcomes

Study	Cost outcomes	Notes
Anema/Steenstra 2007	Total costs: WI: EUR 8993 UC: EUR 9109 Ratio of 1 day: EUR 19	No major difference in costs between work intervention and usual care, but work intervention is associated with larger effects
Arnetz 2003	Total reimbursement from the health insurance system: WI: SEK 57,564 UC: SEK 73,178 Direct cost of WI was SEK 550,000 Total savings SEK 972,900 Benefit‐to‐cost ratio: 6,8	Total reimbursement from the health insurance system significantly lower in work intervention group
Bültmann 2009a	Direct costs: WI: DKK 18,184 UC: DKK 9782 Indirect costs: WI: DKK 153,461 UC: DKK 220,836 Costs per averted absence day (WI vs. UC): DKK 183	In terms of productivity loss, the work intervention seems to be cost saving for society
Busch 2011	Total costs (10‐year follow‐up): BM: SEK 969,077 PT: SEK 1,425,048 CBT: SEK 1,491,298 UC: SEK 1,502,898	There was a decrease in costs per individual in the behavioural medicine rehabilitation group compared to usual care
Lambeek 2010a	Direct costs: WI: GBP 1479 UC: GBP 1262 Indirect costs: WI: GBP 11,686 UC: GBP 17,213	The work intervention for workers sick listed because of low back pain had substantial economic benefits over usual care
Loisel 1997a	1‐year follow‐up. Saved consequence of disease costs against standard care: CAD 604 Cost‐benefit: CAD 220 6‐, 4‐year follow‐up. Saved consequence of disease costs against standard care: CAD 10,697 Cost‐benefit: CAD 16,827 Lower costs in the workplace intervention than in the control group. Significance was not calculated	There was a small number of very costly cases
Tamminga 2013	Total intervention costs per worker in the intervention group: EUR 119 Productivity loss Human Capital Approach: WI: EUR 41,393 UC: EUR 38,968 Productivity loss Friction Costs Approach: WI: EUR 14,030 UC: EUR 13,529 Costs work adjustments: WI: EUR 2975 UC EUR 3025	Costs did not differ statistically between groups
van Oostrom 2010a	Direct costs: WI: EUR 4587 UC: EUR 3560 Indirect costs: WI: EUR 17,842 UC: EUR 16,440	The workplace intervention had no economic benefit compared with usual care
Vlasveld 2012a	Direct costs: WI: EUR 3900 UC: EUR 4600 Indirect costs: WI: EUR 10,110 UC: EUR 11,627	Comparable findings between both groups

BM: behavioural medicine rehabilitation
CBT: cognitive behavioural therapy
PT: physical therapy
UC: usual care
WI: workplace intervention

Quality of the evidence

We assessed the overall quality of the evidence for each outcome using the GRADE approach (Boluyt 2012; GRADE working group, Guyatt 2008a, Guyatt 2008b; Schünemann 2006). Two review authors (MvV and SvO) independently assessed the quality of the evidence. We used GRADEprofiler software (version 3.6). The GRADE approach specifies four levels of quality: high, moderate, low, and very low. The highest quality rating is for randomised trial evidence. The quality rating can be downgraded depending on the presence of the five factors specified below. Every limitation assigned downgrades the quality of the evidence by one or two levels.

Limitations of the study refer to the 'Risk of bias' assessment of studies.
Consistency refers to the similarity of estimates of treatment effects for the outcome across studies.
Directness refers to the extent to which the participants, interventions, and outcomes in the studies were comparable to those defined in the inclusion criteria of the review.
Precision of the evidence refers the degree of certainty surrounding an effect estimate.
Publication bias refers to the probability of selective publication of studies and outcomes.

The overall quality of the evidence for each outcome was the result of the combination of the assessments in all domains, leading to the following four levels of evidence (Guyatt 2008b):

High‐quality evidence: Further research is very unlikely to change our confidence in the estimate of effect.
Moderate‐quality evidence: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.
Low‐quality evidence: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.
Very low‐quality evidence: Any estimate of effect is very uncertain.

Subgroup analysis and investigation of heterogeneity

We performed subgroup analyses in which we determined the effectiveness of workplace interventions for workers with musculoskeletal disorders, mental health problems, and other health conditions separately. Furthermore, we performed subgroup analyses in which we analysed the effectiveness of workplace interventions only, and the effectiveness of workplace interventions offered in combination with a cognitive behavioural or problem‐solving intervention.

Sensitivity analysis

We performed sensitivity analyses by including only studies we judged to have a low risk of bias.

Results

Description of studies

Results of the search

Literature search and study selection

We ran the searches in CENTRAL, the Cochrane Work Trials Register, EMBASE, and PsycINFO. We have presented the detailed search strategy in Appendix 1. We identified 1350 references from the initial electronic literature search (run up to November 2007), retrieving 30 references for full‐text scrutiny. After further examination we excluded 21 articles. We have presented reasons for exclusion in the Characteristics of excluded studies table. We included 10 articles from six studies in the initial review (Anema/Steenstra 2007; Arnetz 2003; Blonk 2006; Feuerstein 2003; Loisel 1997a; Verbeek 2002a) (see the Characteristics of included studies table).

We ran the updated database search up to 2 February 2015, which yielded at total of 1736 references. After removing double references, the electronic search resulted in 856 hits, from which we selected 21 potentially eligible studies. After further examination of the full articles, we excluded 13 articles (see the Characteristics of excluded studies table). Eight studies were eligible for inclusion, in addition to the six studies already included in the original review (Bültmann 2009a; Busch 2011; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a Vlasveld 2012a) (see the Characteristics of included studies table). Handsearching of the reference lists of relevant review articles and eligible studies and personal contact with experts in the field of occupational health generated four additional potentially eligible studies (Farzanfar 2011a; Rebergen 2009b; Reme 2011b; Vermeulen 2011). However, we excluded these four studies after further examination (see the Characteristics of excluded studies table). We present a PRISMA study flow diagram of the selection process in Figure 1.

Figure 1

Study flow diagram.

Included studies

Participants: types of disorders and duration of work disability

All studies specified a particular condition targeted. Eight studies included workers with musculoskeletal disorders. Of these, four studies included workers with back pain (Anema/Steenstra 2007; Lambeek 2010a; Loisel 1997a; Verbeek 2002a), three studies included workers with a variety of musculoskeletal disorders (Arnetz 2003; Bültmann 2009a; Busch 2011), and one study included workers with upper‐extremity disorders (Feuerstein 2003). Five studies included workers with mental health problems (Blonk 2006; Hees 2012a; Noordik 2013a; van Oostrom 2010a; Vlasveld 2012a), and one study concerned workers with cancer (Tamminga 2013). The duration of work disability before randomisation extended from immediate inclusion of a sick‐listed worker (Blonk 2006; Tamminga 2013), a minimum of 10 days (Verbeek 2002a), two to six weeks (Anema/Steenstra 2007), two to eight weeks (Noordik 2013a; van Oostrom 2010a), at least eight weeks (Hees 2012a), four to 12 weeks (Bültmann 2009a; Loisel 1997a; Vlasveld 2012a), one to six months (Busch 2011), and three to 24 months (Lambeek 2010a). The exact duration of work disability before randomisation was not described in the Arnetz 2003 study. The duration of work disability prior to randomisation was less clear in the Feuerstein 2003 study: inclusion was possible when the work disability claim was accepted within 90 days of filing. In three studies the disability was work related; none of the other studies reported whether the condition for work disability was work related or not (Feuerstein 2003; Hees 2012a; Loisel 1997a). The only reported restrictions with regard to age, sex, and ethnicity of the participants were working age and sufficient understanding of the national language to fill in the questionnaires. Two studies reported inclusion of workers when assigned to light or modified duties (Feuerstein 2003; Loisel 1997a).

Time and setting characteristics

Of the 14 included studies, 12 were conducted in Europe (Anema/Steenstra 2007; Arnetz 2003; Blonk 2006; Bültmann 2009a; Busch 2011; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Verbeek 2002a; Vlasveld 2012a), one in the US (Feuerstein 2003), and one in Canada (Loisel 1997a). Of the 12 Northwestern European studies, nine were conducted in the Netherlands (Anema/Steenstra 2007; Blonk 2006; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Verbeek 2002a; Vlasveld 2012a), two in Sweden (Arnetz 2003; Busch 2011), and one in Denmark (Bültmann 2009a). Participants were working in several economic sectors (manufacturing, health care, office administration, and agriculture) in most studies, except for the Verbeek 2002a study, where all participants worked in hospitals. The duration of the recruitment period ranged from 12 to 34 months and was reported by 12 studies (Anema/Steenstra 2007; Blonk 2006; Bültmann 2009a; Busch 2011; Feuerstein 2003; Hees 2012a; Lambeek 2010a; Loisel 1997a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Vlasveld 2012a).

Work interventions

We have described the content of the interventions (based on published reports and personal contact with authors) in Table 2 and Table 3. Regarding the content of the interventions, 11 studies reported changes to the workplace and equipment, 13 reported changes of work design and organisations, and six and nine studies reported changes to working conditions and work environment, respectively. According to the authors, case management with the worker and employer (supervisor) occurred in 12 studies. The worker, the supervisor or employer, and a professional in occupational health were always involved in the interventions, except for the study concerning adjustment disorders, where no supervisor was involved due to self employment (Blonk 2006). Insurer representatives were involved in three studies (Arnetz 2003; Blonk 2006; Busch 2011a), and union representatives were involved in one study (Loisel 1997a). In general, the number of contacts in the workplace intervention was not described in detail, but ranged from one to 29 contacts. Face‐to‐face contact took place in all studies: often at the workplace, and otherwise at a rehabilitation centre, hospital, psychiatry department, or at home. The Busch 2011 study includes three interventions: all three were compared to usual care. All three interventions met our inclusion criteria, therefore we included all three interventions in this review.

Table 2. Content of the interventions ‐ 1

General characteristics of interventions	Specific characteristics interventions	Anema/Steenstra	Arnetz	Blonk	Bültmann	Busch	Feuerstein	Hees
Applied components definition workplace intervention	Changes workplace or equipment	x	x	x	x	x	x	‐
	Changes work design and organisation including working relationships	x	x	x	x	x	‐	x
	Changes in working conditions	‐	‐	x	‐	x	‐	‐
	Changes to the work environment	x	‐	‐	x	x	x	x
	Case management with worker and employer	x	x	‐	x	x	x	x
Contacts	Number of meetings	3	1	5 to 6	2	?	4 to 5	18
	Duration contact	1 h	?	?	2.5 h	1 h	1 to 2 h	1 to 2 h
Stakeholders involved	Worker	x	x	x	x	x	x	x
	Employer/supervisor	x	x	Self employed	x	x	x	x
	Occupational physician	‐	‐	‐	x	‐	‐	x
	Occupational nurse	x	‐	‐	x	‐	x	‐
	Ergonomist	x	x			x	‐	‐
	Representative of a union	‐	‐	‐	‐	‐	‐	‐
	Representative of an insurer	‐	x	x	‐	x	‐	‐
Type of contact	Face‐to‐face	x	x	x	x	x	x	x
	By phone	‐	‐	‐	‐	x	‐	x
Place of contact	At workplace	x	x	x	x	x	x	x
	Other	‐	‐	‐	‐	Rehabilitation centre	Home and provider office	Psychiatry department
Main treatment provider, work intervention		Ergonomist, occupational nurse	Insurance agency case manager	Labour expert	Social worker	?	Nurse case manager	Occupational therapist
Training treatment provider, work intervention		Yes	?	Yes	?	?	Yes	Yes

A 'x' mark indicates that the study fits the specific intervention characteristic. A '?' mark indicates that it is unclear whether the study fits the specific intervention characteristic.

See: Summary of findings for the main comparison Workplace interventions compared to Usual care for Workers on sick leave

Table 3. Content of the interventions ‐ 2

General characteristics of interventions	Specific characteristics interventions	Lambeek	Loisel	Noordik	Tamminga	van Oostrom	Verbeek	Vlasveld
Applied components definition workplace intervention	Changes workplace or equipment	x	x	‐	‐	x	x	x
	Changes work design and organisation including working relationships	x	x	x	x	x	x	x
	Changes in working conditions	x	x	‐	‐	x	‐	x
	Changes to the work environment	x	x	‐	‐	x	‐	x
	Case management with worker and employer	x	x	x	x	x	‐	x
Contacts	Number of meetings	3 to 29	?	Not prespecified	5	3	? ˜ 3	6 to 12
	Duration contact	?	1 to 3 h	?	25 min	1 h	20 min	?
Stakeholders involved	Worker	x	x	x	x	x	x	x
	Employer/supervisor	x	x	x	x	x	x	x
	Occupational physician	x	x	x	x	‐	x	x
	Occupational nurse	‐	‐	‐	‐	x	‐	‐
	Ergonomist	x	x	‐	‐	‐	‐	‐
	Representative of a union	‐	x	‐	‐	‐	‐	‐
	Representative of an insurer	‐	‐	‐	‐	‐	‐	‐
Type of contact	Face‐to‐face	x	x	x	x	x	x	x
	By phone	‐	x	‐	x	‐	‐	‐
Place of contact	At workplace	x	x	‐	x	x	‐	‐
	Other	‐	‐	Occupational Health Service	Hospital	‐	Occupational Health Service	Occupational Health Service
Main treatment provider, work intervention		Clinical occupational physician	Ergonomist	Occupational physician	Specialised nurse/ social worker	Return‐to‐work coordinator	Occupational physician	Occupational physician care manager
Training treatment provider, work intervention		Yes	?	Yes	Yes	Yes	Yes	Yes

A 'x' mark indicates that the study fits the specific intervention characteristic. A '?' mark indicates that it is unclear whether the study fits the specific intervention characteristic.

Usual care

The usual‐care conditions were less extensively described in most studies. Despite the fact that studies explored the effectiveness of workplace interventions, only nine studies had a usual‐care condition in an occupational setting: guideline‐based care by the occupational physician (Anema/Steenstra 2007; Lambeek 2007; Noordik 2013a; Tamminga 2013; van Oostrom 2008; Vlasveld 2012a); an eight‐week RTW plan (Arnetz 2003); and usual case management comprising monitoring of the claims process and surveillance of medical treatment (Feuerstein 2003; Bültmann 2009b). In the other studies, usual care consisted of treatment by the attending physician (Busch 2011; Loisel 1997a), the general practitioner (Blonk 2006; Verbeek 2002a), or the psychiatrist (treatment consistent with American Psychiatric Association guidelines) (Hees 2012a).

In two studies, the workplace intervention was followed by a clinical intervention when RTW was not achieved within a predefined period of eight weeks (Anema/Steenstra 2007; Loisel 1997a). This meant that workplace and clinical interventions did not start concurrently. The Blonk 2006 study was a three‐armed trial: a workplace intervention was compared with a clinical intervention (cognitive behavioural therapy) and with usual care.

Outcomes

We selected studies if they reported on the exact duration of work disability. Four studies used self‐reported outcomes with regard to sickness absence (Hees 2012a; Loisel 1997a; Noordik 2013a; Tamminga 2013): the other 10 studies used administrative outcomes. Six studies reported time until lasting RTW, which they defined as duration of sick leave in calendar days until full RTW for at least four weeks without recurrence (Anema/Steenstra 2007; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a). Four studies reported time until first RTW (Blonk 2006; Feuerstein 2003; Loisel 1997a; Verbeek 2002a). Two studies distinguished between time until partial RTW and time until full RTW (Blonk 2006; Noordik 2013a); we decided to focus on time until first full RTW only to avoid introducing any difference from the outcomes of the other studies. One study described the outcome time until first RTW in their published study protocol (Steenstra 2003), but did not include results for this outcome in the published paper about the results (Anema 2007). We therefore requested and obtained the Cox regression output regarding this outcome from the authors. Another study did not publish results on the time until first RTW (Feuerstein 2003). It was clear from papers from the Feuerstein study that they had collected these data, and so we requested the unpublished data to avoid obvious publication bias. Eight studies reported the outcome cumulative duration of sickness absence (Anema/Steenstra 2007; Arnetz 2003; Bültmann 2009a; Busch 2011; Lambeek 2010a; van Oostrom 2010c; Verbeek 2002a; Vlasveld 2012a). Busch et al reported the cumulative duration of sickness absence for the three interventions evaluated. One study reported recurrences of sickness absence in percentages and a hazard ratio (Verbeek 2002a). Six of the included studies used the definition of lasting RTW that corrected for recurrences within four weeks of RTW (Anema/Steenstra 2007; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a). The other studies did not take the sustainability of sickness absence without recurrences into account.

Functional status was measured in the four back pain studies (Anema/Steenstra 2007; Lambeek 2010a; Loisel 1997a; Verbeek 2002a), the study on musculoskeletal disorders (Bültmann 2009b), and the upper‐extremity disorder study (Feuerstein 2003). Of these six studies, three used the Roland‐Morris Disability Questionnaire (Anema/Steenstra 2007; Lambeek 2010a; Verbeek 2002a), two used the Oswestry questionnaire (Bültmann 2009a; Loisel 1997a), and one used the upper‐extremity functional limitations scale (Feuerstein 2003). The studies on mental health problems and cancer reported no outcome on functioning.

Four of the 14 studies reported quality of life and general health as separate outcomes (Busch 2011; Hees 2012a; Tamminga 2013; Verbeek 2002a). To assess these outcomes, the studies used the Nottingham health profile, in Verbeek 2002a, and the 36‐Item Short Form Health Survey (SF‐36) (Busch 2011a; Hees 2012a; Tamminga 2013).

Six studies reported a follow‐up of symptoms. For adjustment disorders, the Blonk 2006 study used the depression anxiety stress scales, and for work‐related upper‐extremity disorders, the Feuerstein 2003 study used a modified version of the carpal tunnel symptom severity scale. The Vlasveld 2012d study reported severity of depressive symptoms (Patient Health Questionnaire), and the van Oostrom 2010c study used the Four‐Dimensional Symptom Questionnaire (4DSQ) to measure stress‐related symptoms. The Hees 2012a study assessed severity of depression with a clinician‐reported measure (Hamilton Rating Scale for Depression), as well as a self‐reported measure (Inventory of Depressive Symptomatology‐Self‐Report (IDS‐SR)). We included the self‐reported measure (IDS‐SR) in the meta‐analysis, as the depression scales the other studies used were also self‐reported. The Noordik 2013a study used the 4DSQ to measure symptoms of depression. The Arnetz 2003 study reported baseline measurements of musculoskeletal symptoms, but conducted no follow‐up of symptoms.

Five studies assessed level of pain or pain intensity (Anema/Steenstra 2007; Bültmann 2009a; Lambeek 2010a; Loisel 1997a; Verbeek 2002a). Of these studies, three used a 10‐point visual analogue scale (Anema/Steenstra 2007; Lambeek 2010a; Verbeek 2002a), one used the McGill Pain Questionnaire (Loisel 1997a), and one used the Örebro Musculoskeletal Pain Screening Questionnaire (Bültmann 2009b).

Nine studies measured direct and indirect costs of work disability, but used different perspectives (Anema/Steenstra 2007; Arnetz 2003; Bültmann 2009a; Busch 2011; Lambeek 2010a; Loisel 1997a; Tamminga 2013; van Oostrom 2010a; Vlasveld 2012d). Seven studies applied the societal perspective for cost analysis. The Loisel 1997a study applied the insurer perspective, and the Arnetz 2003 study did not report the perspective applied. All studies measured the direct intervention costs and indirect costs of sick leave. One study did not measure costs of other treatments (direct medical costs) (Arnetz 2003), while the other studies measured use of other healthcare resources and calculated the accompanying costs. The Hees 2012a study measured direct and indirect costs but has not reported them yet.

Follow‐up

Thirteen studies reported a follow‐up period of 12 months for the sickness absence outcome (Anema/Steenstra 2007; Arnetz 2003; Blonk 2006; Bültmann 2009a; Feuerstein 2003; Lambeek 2010a; Loisel 1997a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Verbeek 2002a; Vlasveld 2012d), and the Busch 2011a study reported follow‐up of three and 10 years. We therefore did not include the Busch 2011a study in the meta‐analysis. For the other outcomes, which were all self‐reported, follow‐up was at: 10 months (Blonk 2006), 16 months (Feuerstein 2003), three years (Busch 2011a), and 12 months (Anema/Steenstra 2007; Bültmann 2009a; Hees 2012a; Lambeek 2010a; Loisel 1997a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Verbeek 2002a; Vlasveld 2012d). One study collected cost data for a mean of 6.4 years (range 5.1 to 7.5 years) (Loisel 1997a).

Short‐term results (less than three months) for the review's outcomes were not reported. Only two studies reported either three‐month, in Verbeek 2002a, or four‐month, in Blonk 2006, results on the self‐reported outcomes (see the Characteristics of included studies table).

Excluded studies

We excluded 37 full‐text articles in total. We excluded 20 studies because the intervention did not fulfil our inclusion criteria. We excluded seven studies because participants were not on sick leave at baseline. We excluded five studies that measured sickness absence in a dichotomous way. We excluded four studies because participants did not have a job at baseline, and we excluded one study that applied the intervention only occasionally.

Risk of bias in included studies

For each study, we assessed the risk of bias by evaluating every study publication we could find. Figure 2 shows the full 'Risk of bias' assessment for each item in every study. Two review authors (MvV and SvO) independently assessed the risk of bias of the RCTs, except for the van Oostrom 2010a study, which MvV and CB assessed.

Figure 2

Risk of bias: review authors' judgements about each methodological quality item for each included study.

Most studies used administrative sickness absence data that was collected without knowledge of group allocation; therefore outcome assessors were blinded for these studies. We classified six studies to have a low risk of bias (Anema/Steenstra 2007; Busch 2011a; Lambeek 2010b; van Oostrom 2010c; Verbeek 2002a; Vlasveld 2012d). We classified three studies to have an unclear risk of bias (Bültmann 2009a; Feuerstein 2003; Loisel 1997a), and five studies to have a high risk of bias (Arnetz 2003; Blonk 2006; Hees 2012a; Noordik 2013a; Tamminga 2013) (see the Characteristics of included studies table).

Risk of bias was different for the secondary outcomes due to their self‐reported nature. The scores on the domain 'blinding of the outcome assessor for health‐related outcomes' changed in all studies, resulting in lower total scores for most studies (Figure 2). As expected, the most prevalent shortcomings were found in the domain 'blinding of participants and personnel', as none of the studies were able to blind participants.

Allocation

Of the 14 studies, 35.7% failed to describe or use appropriate concealment of allocation.

Blinding

Not all studies performed blinding of participants and personnel. Eleven of 14 studies performed blinding of outcome assessment of sickness absence outcomes. Not all studies performed blinding of outcome assessment of health‐related outcomes.

Incomplete outcome data

Of the 14 studies, 21.4% provided incomplete outcome data.

Selective reporting

There was selective reporting bias with 21.4% of the trials.

Other potential sources of bias

Three out of 14 studies did not perform their analyses according to the intention‐to‐treat principle. In four out of 14 studies, the groups were not similar at baseline regarding the most important prognostic factors.

Effects of interventions

1. Workplace interventions compared to usual care

Primary outcomes: Sickness absence

Time until first RTW

The pooled analysis of the outcome time until first RTW showed that workplace interventions were more effective than usual care, with a pooled hazard ratio (HR) of 1.55 (95% confidence interval (CI) 1.20 to 2.01) (Figure 3; Analysis 1.1) (Anema/Steenstra 2007; Blonk 2006; Feuerstein 2003; Loisel 1997a; Verbeek 2002a). This hazard ratio means that after a workplace intervention, people return to work in a more timely fashion and more frequently when compared to the usual‐care group. The quality of this evidence was moderate (Summary of findings table 1; Table 4). We downgraded the quality of the evidence based on the inclusion of studies with high or unclear risk of bias. All studies presented the exact duration of time until first RTW with a median (see Characteristics of included studies table). The difference in median duration of time until first RTW between the workplace intervention group and the usual‐care group ranged from 14 days, in Feuerstein 2003, to 198 days, in Blonk 2006.

Figure 3

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.1 Time until first RTW.

Table 4. Grading of the quality of evidence

Comparison	Outcome	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias	GRADE quality
Workplace intervention vs usual care	Time until first RTW	Yes: 60% of studies were assigned high or unclear risk of bias	No: I² < 50%	No	No	Undetected	Moderate
	Time until lasting RTW	Yes: 50% of studies were assigned high risk of bias	Yes: I² > 50%	No	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: majority low risk of bias	No: I² < 50%	No	No	Undetected	High
	Recurrences of sickness absence	No: study with low risk of bias	NA	No	Yes: single study	Undetected	Moderate
	Functional status	No: majority low risk of bias	Yes: I² > 50%	No	No	Undetected	Moderate
	Pain	No: majority low risk of bias	No: I² < 50%	No	No	Undetected	High
	Depression	Yes: > 50% of studies were assigned high risk of bias	No: I² < 50%	No	Yes: wide CI	Undetected	Very low
Workplace intervention vs clinical intervention	Time until first RTW	Yes: study with high risk of bias	NA	No	Yes: single study	Undetected	Very low
Subgroup analyses:
Workplace intervention vs usual care: musculoskeletal disorders	Time until first RTW	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
	Time until lasting RTW	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
Workplace intervention vs usual care: mental health problems	Time until first RTW	Yes: Study with high risk of bias	NA	Yes: PICO deviant (subgroup analysis)	Yes: < 300 workers	Undetected	Very low
	Time until lasting RTW	Yes: >50% of studies was assigned high risk of bias	Yes: I² > 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Low
Workplace intervention only vs usual care	Time until first RTW	Yes: >50% of studies was assigned unclear risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Low
	Time until lasting RTW	Yes: >50% of studies was assigned high risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
Workplace intervention + cognitive behavioural intervention vs usual care	Time until first RTW	Yes: >25% of studies was assigned high risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: < 300 workers	Undetected	Very low
	Time until lasting RTW	Yes: >25% of studies was assigned high risk of bias	Yes: I² > 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate

CI: confidence interval
NA: not applicable

PICO: patients, intervention, control, outcome
RTW: return to work

Subgroup analysis

The subgroups based on diagnosis were not statistically significantly different (P = 0.08). For musculoskeletal disorders, the analysis showed a pooled HR of 1.44 (95% CI 1.15 to 1.82). The quality of this evidence was moderate.

The subgroups based on inclusion of cognitive behavioural elements were not significantly different (P = 0.17). When only a workplace intervention was offered, analysis revealed a HR of 1.35 (95% CI 1.01 to 1.82), with low‐quality evidence. We downgraded the quality of the evidence based on the inclusion of studies with a high risk of bias and indirectness. When a workplace intervention was offered in combination with a cognitive behavioural intervention, the HR was 1.93 (95% CI 1.27 to 2.93). The quality of this evidence was very low, based on the inclusion of studies with a high risk of bias, indirectness, and imprecision.

Sensitivity analysis

An analysis including only studies with a low risk of bias revealed a HR of 1.50 (95% CI 1.18 to 1.92).

Time until lasting RTW

Workplace interventions did not considerably reduce time to lasting RTW compared to usual care (HR 1.07, 95% CI 0.72 to 1.57) (Figure 4; Analysis 1.2), based on six studies (Anema/Steenstra 2007; Hees 2012a; Lambeek 2010a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a). The quality of this evidence was very low, downgraded based on the inclusion of studies with a high risk of bias, heterogeneity between studies, and uncertainty surrounding the effect estimate (Table 4).

Figure 4

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.2 Time until lasting RTW.

Subgroup analysis

The effects in subgroups based on diagnosis were significantly different (P = 0.0009). An analysis of two studies concerning musculoskeletal disorders significantly favoured the workplace intervention with a HR of 1.77 (95% CI 1.37 to 2.29). An analysis of three studies concerning mental health problems showed no difference between treatment conditions (HR 0.79, 95% CI 0.54 to 1.17). We assessed the quality of the analysis of musculoskeletal disorders as moderate, and the quality of the analysis of mental health problems as very low. We downgraded the quality of the evidence for mental health problems based on the inclusion of studies with a high risk of bias and uncertainty surrounding the effect estimate.

The effects in subgroups based on the inclusion of cognitive behavioural elements in the interventions did not differ statistically significantly (P = 0.056). An analysis of three studies offering a workplace intervention only did not show a considerable effect of the intervention on this outcome (HR 0.95, 95% CI 0.74 to 1.21). The quality of this evidence was very low, based on the inclusion of studies with a high risk of bias and uncertainty surrounding the effect estimate. An analysis of three studies offering a workplace intervention in combination with a cognitive behavioural intervention also did not show a considerable effect, with a HR of 1.21 (95% CI 0.56 to 2.62). The quality of this evidence was very low, based on the inclusion of studies with a high risk of bias, heterogeneity between studies, and uncertainty surrounding the effect estimate.

Sensitivity analysis

An analysis including only studies with a low risk of bias revealed a HR of 1.46 (95% CI 0.98 to 2.17).

Cumulative duration of sickness absence, follow‐up 12 months

Eight studies reported the cumulative duration of sickness absence (Anema/Steenstra 2007; Arnetz 2003; Bültmann 2009b; Busch 2011; Lambeek 2010b; van Oostrom 2010c; Verbeek 2002a; Vlasveld 2012c). The pooled analysis showed a non‐significant mean difference (MD) of ‐33.33 days (95% CI ‐49.54 to ‐17.12) between the workplace interventions and usual care (Figure 5; Analysis 1.3), meaning that the workplace intervention group had fewer sickness absence days when compared to the usual‐care group. The quality of this evidence was high (summary of findings Table for the main comparison).

Figure 5

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.3 Cumulative duration of sickness absence.

Subgroup analysis

Subgroups based on diagnosis were not statistically significantly different (P = 0.05). An analysis of studies concerning musculoskeletal disorders showed a significant mean difference in cumulative duration of sickness absence of ‐40.47 days (95% CI ‐55.98 to ‐24.96), favouring the workplace intervention. The quality of this evidence was moderate, due to indirectness. An analysis of studies concerning mental health problems showed a non‐significant MD of ‐8.42 days (95% CI ‐35.99 to 19.16), with low‐quality evidence for this outcome based on indirectness and imprecision.

Subgroups based on inclusion of cognitive behavioural elements in the intervention were not statistically significantly different (P = 0.79). An analysis of studies applying a workplace intervention only showed a MD of ‐31.16 days (95% CI ‐55.87 to ‐6.45). The quality of this evidence was moderate, due to indirectness. We found a significant mean difference for studies applying a workplace intervention in combination with a cognitive behavioural intervention (MD of ‐35.99 days, 95% CI ‐62.21 to ‐9.77), with moderate‐quality evidence, due to indirectness.

Sensitivity analysis

An analysis including only studies with a low risk of bias revealed a MD of ‐27.60 (95% CI ‐69.90 to 14.70).

Cumulative duration of sickness absence, follow‐up 3 to 10 years

Due to a longer follow‐up we did not include the Busch 2011 study in the meta‐analysis. After 10 years of follow‐up, workers in the behavioural medicine rehabilitation intervention had 42.98 fewer sickness absence days per year (P = 0.03). Workers in the physiotherapy or cognitive behavioural therapy groups had comparable sickness absence days with workers receiving usual care.

Recurrences of sick leave

One study reported recurrences of sick leave (Verbeek 2002a), with a recurrence rate of 25% in the usual‐care group and 51% in the workplace intervention group over 12 months, with a corresponding HR of 0.42 (95% CI 0.21 to 0.82). The evidence for this outcome was of moderate quality, since only one study measured this outcome.

Secondary outcomes

Functional status

We found a statistically significant difference in functional status at 12 months' follow‐up. The SMD was ‐0.33 (95% CI ‐0.58 to ‐0.08) between workers on sick leave who received the workplace intervention and those who received usual care (Anema/Steenstra 2007; Bültmann 2009a; Feuerstein 2003; Lambeek 2010a; Loisel 1997a; Verbeek 2002a) (Analysis 1.4). This means that workers in the workplace intervention group had better functional status. The quality assessment showed moderate‐quality evidence from six studies (628 participants) (Table 4), due to heterogeneity between studies.

Sensitivity analysis

An analysis including only studies with a low risk of bias revealed a MD of ‐0.15 (95% CI ‐0.45 to 0.15).

Quality of life and general health

We could not pool the data from four studies on quality of life because the studies used different measurement instruments and different subscales to measure the data (Busch 2011; Hees 2012a; Tamminga 2013; Verbeek 2002a). Hees 2012a measured three subscales of the SF‐36: mental health, role limitations due to emotional problems, and role limitations due to physical problems. The authors found a statistically significant effect between the intervention and usual‐care group on the mental health subscale. Busch 2011 calculated a global score of health‐related quality of life, based on the SF‐36. Their analysis showed that women in the intervention group reported significantly better health than women in the control group. Verbeek 2002a assessed general health perception with six separate subscales of the Nottingham Health Profile. They found no significant differences between the intervention and control group on any subscales. Tamminga 2013 measured quality of life with eight separate subscales of the SF‐36 (no overall score), and a visual analogue scale. Effects measured on quality‐of‐life scales did not differ between groups.

Symptoms

We could not pool symptom‐related data to provide one figure because one study reported the scores for each of the three scales of the Depression Anxiety Stress Scales separately (Blonk 2006). The authors of this study reported scores for depression, anxiety, and stress whilst they found no effect on anxiety and stress. We pooled depression data from the Hees 2012a, Noordik 2013a and Vlasveld 2012a studies. The pooled MD on depression was ‐0.12 (95% CI ‐0.35 to 0.11) (Analysis 1.5). The quality of this evidence was very low, based on the inclusion of studies with a high risk of bias and imprecision. The study on upper‐extremity disorders showed a MD of ‐0.30 (95% CI ‐0.63 to 0.03) for upper‐extremity symptom severity (Feuerstein 2003).

Sensitivity analysis

An analysis including only studies with a low risk of bias for the outcome depression revealed a MD of 0.26 (95% CI ‐0.20 to 0.72).

Pain

We included five studies in the meta‐analysis regarding pain. All of these studies were on musculoskeletal disorders (Anema/Steenstra 2007; Bültmann 2009a; Lambeek 2010a; Loisel 1997a; Verbeek 2002a). Meta‐analysis resulted in a pooled SMD of ‐0.26 (95% CI ‐0.47 to ‐0.06), with workers in the workplace intervention group experiencing less pain (Analysis 1.6). We considered the quality of evidence to be high (Table 4).

Sensitivity analysis

An analysis including only studies with a low risk of bias revealed a MD of ‐0.15 (95% CI ‐0.34 to 0.05).

Direct and indirect costs of work disability

We did not pool cost outcomes as we did not consider them comparable across studies (Table 1). Five studies reported no differences in costs (Anema/Steenstra 2007; Loisel 1997a; Tamminga 2013; van Oostrom 2010b; Vlasveld 2012d). The economic evaluation in the Steenstra 2006 study showed that the workplace intervention was more effective than usual care in RTW at slightly higher costs. The study by Loisel 1997b showed a decrease in medical costs in the intervention group compared to usual care (CAD 604 saved), but the intervention was not cost‐effective. Cost‐effectiveness analyses and cost‐utility analyses revealed no differences between the intervention and control group in quality‐adjusted life years(QALYs) or costs in the van Oostrom 2010b study. The cost‐utility analysis in the Vlasveld 2012d study showed that the intervention generated a reduction in both costs and effects compared to usual care, and was therefore not a cost‐effective intervention (Goorden). The Tamminga 2013 study compared mean direct and indirect costs between the intervention and control group. These costs did not differ between groups.

Three studies reported lower costs in the workplace intervention group (Arnetz 2003; Bültmann 2009b; Lambeek 2010c). The economic evaluation of the Arnetz 2003 study showed that the direct cost savings in the intervention group were USD 195 per case, yielding a direct benefit‐to‐cost ratio of 6.8 (Arnetz 2003). The Bültmann 2009b study estimated that the total costs saved in the intervention workers compared to controls were USD 1366 per worker at six months' follow‐up and USD 10,666 per person at 12 months' follow‐up. Lambeek 2010c carried out a cost‐effectiveness analysis from the societal perspective. Total costs in the intervention group were lower than in the usual‐care group. Cost‐effectiveness planes and acceptability curves showed that the intervention was cost effective compared with usual care for RTW and QALYs gained. The cost‐benefit analyses showed that every GBP 1 invested in the intervention would return an estimated GBP 26. Furthermore, the net societal benefit of the intervention compared with usual care was GBP 5744.

Busch 2011a compared three different interventions to a group receiving care as usual. One of these interventions, full‐time behavioural‐medicine rehabilitation, reduced the societal costs of sick leave and disability pension. The total costs in the behavioural‐medicine rehabilitation group after 10 years' follow‐up were SEK 969,077 compared to SEK 1,502,898 in the usual‐care group. The other two interventions, behaviour‐oriented physiotherapy and cognitive behavioural therapy, did not differ in costs when compared to the usual‐care group.

2. Workplace interventions compared to clinical interventions

We could perform comparisons between interventions only when interventions started at the same time. In two studies, the workplace intervention was followed by a clinical intervention if RTW was not achieved within eight weeks (Anema/Steenstra 2007; Loisel 1997a). We could not compare the workplace and clinical interventions because the clinical intervention followed the workplace intervention, and some workers received both. This meant that just one study included a valid comparison of a workplace intervention with a clinical intervention (Blonk 2006).

Sickness absence

We found a HR of 2.65 (95% CI 1.42 to 4.95) (Analysis 2.1) (Blonk 2006), with very low‐quality evidence due to imprecision and high risk of bias of this single study.

Symptoms

The mean differences in anxiety, depression, and stress symptoms were not significantly different at 10 months' follow‐up (Blonk 2006). We regarded this evidence as very low quality.

Publication bias

In the funnel plots for the comparisons for the outcomes time until first RTW, time until lasting RTW, cumulative duration of sickness absence, functional status, and pain, we did not see any indication of publication bias.

Discussion

Summary of main results

We included in this review 14 RCTs that evaluated the effects of workplace interventions on work disability. Eight studies included workers with musculoskeletal disorders, five studies included workers with mental health problems, and one study included workers with cancer. The results of this review show that there is moderate‐quality evidence to support the effect of workplace interventions in reducing time until first RTW and high‐quality evidence to support the effect of workplace interventions in reducing the duration of sickness absence when compared with usual care. Furthermore, we found very low‐quality evidence that workplace interventions lead to lasting RTW, and moderate‐quality evidence that workplace interventions lead to more and faster recurrences of sick leave. Overall, the effectiveness of workplace interventions on work disability showed varying results. Our results show that the effectiveness of workplace interventions differs among workers with specific causes of work disability. Indeed, we found moderate‐quality evidence that workplace interventions reduce sickness absence (sooner first RTW, sooner lasting RTW, and shorter cumulative duration of sickness absence) among workers with musculoskeletal disorders when compared to usual care. We found that workplace interventions reduce pain and improve functional status in workers with musculoskeletal disorders. Taken together, the five studies on mental health problems did not show a considerable effect of the workplace interventions. The study on workers with cancer also did not show a considerable effect of the workplace intervention. We found no considerable evidence that a workplace intervention offered in combination with a cognitive behavioural intervention is more effective than a workplace intervention alone.

Overall completeness and applicability of evidence

There are large cross‐country differences regarding disability policies. For example, it has been shown that eligibility criteria for long‐term work disability compensation explain cross‐country differences in RTW (Anema 2009). Furthermore, the interventions applied to disability claimants in a specific country influence the success of RTW. Work interventions explained cross‐country differences more than purely medical interventions in promoting sustainable RTW (Anema 2009). Given the influence of disability policies on RTW, and the fact that most of the studies included in this review were conducted in the Netherlands, we are cautious about generalising our results to other countries.

We found moderate‐quality evidence to support the effect of workplace interventions for workers with musculoskeletal disorders on the outcomes time until first RTW, time until lasting RTW, and cumulative duration of sickness absence. A minimally clinically important change has not been defined for sickness absence. A consensus paper about definitions for episodes of low back pain proposed that an episode of work absence due to low back pain is defined as a period of work absence due to low back pain, preceded and followed by at least one day at work (de Vet 2002). Furthermore, from a socioeconomic point of view, each day of earlier RTW is important (Ostelo 2005). We found high‐quality evidence to support the effect of workplace interventions on the outcome pain, and moderate‐quality evidence on the outcome functional status. It was not possible to pool cost outcomes (Driessen 2012; Hamberg‐van Reenen 2012). However, the six studies on musculoskeletal disorders that measured costs showed beneficial results from workplace interventions: four studies were cost effective, and two other studies showed more timely RTW at equal cost.

The number of recipients of disability benefit on grounds of mental health conditions is increasing in most Organisation for Economic Co‐operation and Development countries, according to the OECD 2010. This highlights the need for interventions to support workers with mental health problems to prevent work disability. In this review, we found that workplace interventions for workers with mental health problems did not have a considerable effect on sickness absence outcomes. One might argue that the working mechanisms of workplace interventions may be less applicable for workers with mental health problems. For example, the stigma associated with mental health problems can negatively influence involvement of the workplace. This is why other intervention components may be needed to support workers with mental health problems to prevent work disability. A review of previous research on workplace interventions for workers with mental health problems indicated three intervention components that are effective on work outcomes. These components are: facilitation of access to clinical treatment; provision of workplace‐based high‐intensity psychological interventions; and facilitation of navigating through the disability management system (Pomaki 2012). Factor one and three do not involve the workplace, implying that intervention components outside the workplace are effective on work outcomes. Another Cochrane review on interventions to improve occupational health in depressed people showed moderate‐quality evidence that a work‐directed intervention in combination with a clinical intervention reduced sickness absence (Nieuwenhuijsen 2014). This finding was based on three studies, one of which was included in this review (Hees 2012a). The authors of the Nieuwenhuijsen 2014 review used less strict inclusion criteria for a workplace intervention, also including studies in which a workplace intervention was offered in combination with a clinical intervention, while we focused on workplace interventions only. This difference in inclusion criteria may account for the differing conclusions between our review and Nieuwenhuijsen 2014. Furthermore, the reviews of Nieuwenhuijsen 2014 and Pomaki 2012 imply that a combination of a workplace‐based intervention and a clinical psychological intervention might be more effective than a workplace‐based intervention only or a clinical intervention only to improve work outcomes in people with mental health problems such as depression. More research is needed to identify effective interventions or intervention components for this population. The studies by Hees 2012a and Vlasveld 2012a also combined a workplace intervention and a clinical psychological intervention, but failed to find an effect. This shows that it is not clear which intervention components are necessary to support workers with mental health problems to reduce their sickness absence.

The number of studies included in this review has increased from six studies in the original review published in 2009 to 14 studies in this updated review. This suggests that the focus of research on RTW interventions is increasingly on adapting the workplace or formulating a RTW plan instead of employing a clinical intervention in order to reduce symptoms. In this review, we included eight studies that concerned musculoskeletal disorders. RCTs concerning mental health problems or other health problems were conducted less frequently. We added four studies on mental health problems and one study on cancer to this update (Hees 2012a; Noordik 2013a; Tamminga 2013; van Oostrom 2010a; Vlasveld 2012a). It is possible that the smaller number of studies on workplace interventions for mental health problems is due to the lack of recognition of mental health problems by some compensation systems (that is, no compensation is granted in case of sickness absence caused by a mental health problem), difficulties in establishing whether a mental health problem is work related (a prerequisite for receiving compensation in several countries), and stigma concerning mental health problems. Having identified four additional studies on mental health problems since publication of the original review in 2009 highlights the increasing interest in addressing mental health problems to prevent work disability. One possible explanation for this interest is that mental health problems increasingly contribute to sickness absence and long‐term disability. Mental health problems account for about one‐third of all new disability claims, rising to between 40% to 50% in some countries (Cornelius 2011; OECD 2009). For other health conditions, such as cancer or cardiovascular disease, we included only one study in this review (Tamminga 2013). However, the importance of taking workplace factors into account in order to prevent work disability, has been recognized for both mental health problems and other health conditions (Berry 1992; Nachreiner 2007; Sanderson 2006; Tamminga 2010). More RCTs are needed to provide directions for RTW interventions for workers with mental health problems and other health conditions.

Quality of the evidence

In this updated review, we adapted the risk of bias assessment to Cochrane's current tool for risk of bias (Higgins 2011). To determine whether an included study had a high or low overall risk of bias, we set a minimum of four criteria that a study needed to meet in order to be classified as having a low risk of bias. We did not include blinding of participants and personnel in these four key domains. Blinding of participants and personnel is less applicable to the workplace setting, because the workplace often does not allow blinding of participants or personnel, especially if high degrees of worker participation and workplace changes are part of the intervention. Although we did not include blinding of participants in our key domains for the 'Risk of bias' assessment, not blinding participants and care providers could still be a source of information bias in occupational health research.

We assessed selective outcome reporting in our risk of bias assessment, and it is a topic of concern (Chan 2004; Hróbjartsson 2013). In 2008, a review was published about empirical evidence of outcome reporting bias. Results of this review showed that, when comparing trial publications to protocols, 40% to 62% of studies had at least one primary outcome that was changed, introduced, or omitted (Dwan 2008). In our review, we had study protocols available for seven of the 14 studies included in total. For these seven studies, we were able to compare the protocol to the published outcome measures, thereby assessing the risk of selective outcome reporting. No protocols were available for the other seven studies; we therefore tried to assess selective outcome reporting based on the published reports. As we had no study protocols with which to compare the published reports, we cannot be totally sure that selective outcome reporting has not occurred.

In our review we assessed the quality of the evidence as moderate for time until first RTW and high for cumulative duration of sickness absence, but very low for time until lasting RTW. We downgraded the evidence for the outcome time until lasting RTW because the effectiveness of workplace interventions differed greatly between workers with musculoskeletal disorders and workers with mental health problems and other health conditions. Most included studies on mental health and other health problems reported on the outcome time until lasting RTW, and therefore the analysis of this outcome showed a heterogeneous result. Workplace interventions were effective for workers with musculoskeletal disorders, but not effective or even negative for workers with mental health or other health problems. Furthermore, we assessed 50% of the studies included in the analysis for the outcome time until lasting RTW as having a high risk of bias.

Potential biases in the review process

It is possible that we may have missed relevant unpublished studies. We nevertheless tried to minimise selection bias in our search by screening references of identified trials and systematic reviews, contacting experts in the research field, and using no language restrictions. Careful screening of the identified papers resulted in identification of one study that measured RTW outcomes but which had not published the results (Feuerstein 2003). We contacted the authors and obtained data about the study and the outcomes RTW, symptoms, and functioning. Publishing study protocols is an important step in reducing publication bias (for example not publishing non‐significant results). Trial registers are furthermore a tool to reduce publication bias. We will also screen trial registers in future updates of this review.

The strict inclusion criteria used are likely to have reduced the number of studies included in this review. We included RCTs only since methodologically weaker designs can easily lead to bias. Randomisation is sometimes difficult to perform in the field of occupational health. However randomisation did not seem to be a problem in the individual RTW interventions that we included in this review. It should be recognised that high‐quality prospective studies in workplaces could add relevant information to this review.

This review did not include studies that only reported a dichotomous measure of sickness absence. We excluded five studies that assessed RTW in a dichotomous way only. Of these, two workplace interventions were effective in improving RTW outcomes (Cheng 2007; Netterstrøm 2013); the study by Lindh 1997 showed the workplace intervention to be effective in maintaining work stability after RTW; and two studies did not show a considerable effect of the workplace intervention on RTW outcomes (Haldorsen 2002; Nilsson 1996). We chose not to include studies that only reported dichotomous outcomes for sickness absence, because information may be lost about the exact duration of work disability. Using continuous sickness absence outcomes is especially important when an intervention is targeted on RTW and when sickness absence is the primary outcome of the study.

Workplace interventions aim to hasten RTW and preferably lasting RTW. However, only the Dutch studies reported time until lasting RTW, while almost all studies reported time until first RTW. To use all information in the review optimally, we chose to prioritize time until first RTW.

The results on our secondary outcomes might not completely represent all the available evidence on these outcomes since we based the primary criteria for inclusion of studies on whether a study reported continuous sickness absence as an outcome. Incorporating more studies might have changed the evidence for our secondary outcomes, as we found only moderate‐quality evidence for the outcome functional status. The original version of our review also found limited effects of RTW interventions on health outcomes, as did another review on workplace‐based RTW interventions (Franche 2005a), and the Steenstra 2012 study reported a limited association between functional status and RTW.

Most studies used administrative databases for data on sickness absence, except for three studies that used self‐reported data. Current literature suggests that administrative data and self‐report data reflect different aspects of the RTW trajectory. While administrative data is closely tied to benefit provision and termination, self‐report data is more closely linked to workers' perception of their RTW trajectory and work status (Ferrie 2005; Fleten 2004; Linton 2011; Pole 2006).

Agreements and disagreements with other studies or reviews

Previous studies have investigated the effects of workplace interventions, but the way these studies operationalised interventions and made comparisons differed from our review. Schandelmaier 2012 studied the effectiveness of RTW coordination programs (defined as a direct assessment leading to an individually tailored RTW plan implemented by a RTW coordinator or team that coordinates services and communication among involved stakeholders). This review showed moderate‐quality evidence for the benefit of RTW coordination programs. Other studies compared interventions involving workplace adaptations and stakeholder involvement to workplace‐linked interventions such as ergonomic exercises, and showed a benefit of interventions conducted at the workplace with the involvement of stakeholders (Carroll 2010; Franche 2005a; Haugli 2011). Furthermore it has been shown that purely medical interventions do not have a beneficial effect on work‐related outcomes (Loisel 2001; Nieuwenhuijsen 2014).

Figure 1

Study flow diagram.

Figure 2

Risk of bias: review authors' judgements about each methodological quality item for each included study.

Figure 3

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.1 Time until first RTW.

Figure 4

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.2 Time until lasting RTW.

Figure 5

Forest plot of comparison: 1 Workplace intervention versus usual care, outcome: 1.3 Cumulative duration of sickness absence.

Analysis 1.1

Comparison 1 Workplace intervention versus usual care, Outcome 1 Time until first RTW.

Analysis 1.2

Comparison 1 Workplace intervention versus usual care, Outcome 2 Time until lasting RTW.

Analysis 1.3

Comparison 1 Workplace intervention versus usual care, Outcome 3 Cumulative duration of sickness absence.

Analysis 1.4

Comparison 1 Workplace intervention versus usual care, Outcome 4 Functional status.

Analysis 1.5

Comparison 1 Workplace intervention versus usual care, Outcome 5 Symptoms ‐ Depression.

Analysis 1.6

Comparison 1 Workplace intervention versus usual care, Outcome 6 Pain.

Analysis 1.7

Comparison 1 Workplace intervention versus usual care, Outcome 7 Time until first RTW.

Analysis 1.8

Comparison 1 Workplace intervention versus usual care, Outcome 8 Time until lasting RTW.

Analysis 1.9

Comparison 1 Workplace intervention versus usual care, Outcome 9 Cumulative duration of sickness absence.

Analysis 2.1

Comparison 2 Workplace intervention versus clinical intervention, Outcome 1 Time until first RTW.

Summary of findings for the main comparison. Workplace interventions compared to Usual care for Workers on sick leave

Workplace interventions compared to Usual care for Workers on sick leave
Patient or population: Workers on sick leave Settings: Workplace Intervention: Workplace interventions Comparison: Usual care
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI)	No of Participants (studies)	Quality of the evidence (GRADE)	Comments
	Assumed risk	Corresponding risk
	Usual care	Workplace interventions
Time until first RTW Risk at return to work Follow‐up: 12 months	Study population¹		HR 1.55 (1.2 to 2.01)	608 (5 studies)	⊕⊕⊕⊝ moderate³
	799 per 1000²	917 per 1000 (854 to 960)²
	Low
	500 per 1000²	658 per 1000 (565 to 752)²
	High
	870 per 1000²	958 per 1000 (914 to 983)²
Time until lasting RTW Risk at return to work Follow‐up: 12 months	Study population⁴		HR 1.07 (0.72 to 1.57)	635 (6 studies)	⊕⊝⊝⊝ very low^5,6,7
	834 per 1000²	853 per 1000 (725 to 940)²
	Low
	560 per 1000²	585 per 1000 (446 to 724)²
	High
	920 per 1000²	933 per 1000 (838 to 981)²
Cumulative duration of sickness absence Days. Scale from: 0 to 365. Scale from: 0 to 365. Follow‐up: median 12 months	The mean cumulative duration of sickness absence in the control groups was 165.7 Days⁸	The mean cumulative duration of sickness absence in the intervention groups was 33.33 lower (49.54 to 17.12 lower)		1164 (8 studies)	⊕⊕⊕⊕ high
Recurrences Risk at recurrences of sick leave Follow‐up: 12 months	Study population		HR 0.42 (0.21 to 0.82)	99 (1 study)	⊕⊕⊕⊝ moderate^10,11
	250 per 1000⁹	114 per 1000 (59 to 210)⁹


Functional status Roland disability questionnaire. outcome was measured on different scales in different studies. Scale from: 0 to 24. Follow‐up: median 12 months	The mean functional status ranged across control groups from 5.76 ‐ 81.79	The mean functional status in the intervention groups was 0.33 standard deviations lower (0.58 to 0.08 lower)		628 (6 studies)	⊕⊕⊕⊝ moderate⁶
Depression Depression Anxiety Stress scale, and PHQ‐9 depression scale Follow‐up: 12 months	The mean depression ranged across control groups from 5.9‐24.6	The mean depression in the intervention groups was 0.12 standard deviations lower (0.35 lower to 0.11 higher)		133 (4 studies)	⊕⊝⊝⊝ very low^7,12
Pain Visual Analogue Scale Follow‐up: 12 months	The mean pain ranged across control groups from 3.4‐30	The mean pain in the intervention groups was 0.26 standard deviations lower (0.47 to 0.06 lower)		531 (5 studies)	⊕⊕⊕⊕ high
The basis for the assumed risk* (e.g. the median control group risk across studies) is provided in footnotes. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: Confidence interval; HR: Hazard ratio;
GRADE Working Group grades of evidence High quality: Further research is very unlikely to change our confidence in the estimate of effect. Moderate quality: Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low quality: Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low quality: We are very uncertain about the estimate.
¹ Risks ranged from 0.50 to 0.87, therefore lowest and highest risks are presented. ² Risks are presented as the number of people who returned to work. ³ 60% of studies was assigned high or unclear risk of bias ⁴ Risks ranged from 0.56 to 0.92, therefore lowest and highest risks are presented. ⁵ Three of the six studies were assessed as high risk of bias. ⁶ I2 > 50% ⁷ The CI of the pooled effect size is sufficiently wide that the estimate could either support or refute the effectiveness of the intervention. ⁸ Presented as the number of days of sickness absence. ⁹ Risks are presented as the number of people who had a recurrent period of sickness absence. ¹⁰ Not applicable ¹¹ Sparse data, only one study for this outcome. ¹² Three of the four studies were assessed as high risk of bias.

Summary of findings for the main comparison. Workplace interventions compared to Usual care for Workers on sick leave

Table 1. Cost outcomes

Study	Cost outcomes	Notes
Anema/Steenstra 2007	Total costs: WI: EUR 8993 UC: EUR 9109 Ratio of 1 day: EUR 19	No major difference in costs between work intervention and usual care, but work intervention is associated with larger effects
Arnetz 2003	Total reimbursement from the health insurance system: WI: SEK 57,564 UC: SEK 73,178 Direct cost of WI was SEK 550,000 Total savings SEK 972,900 Benefit‐to‐cost ratio: 6,8	Total reimbursement from the health insurance system significantly lower in work intervention group
Bültmann 2009a	Direct costs: WI: DKK 18,184 UC: DKK 9782 Indirect costs: WI: DKK 153,461 UC: DKK 220,836 Costs per averted absence day (WI vs. UC): DKK 183	In terms of productivity loss, the work intervention seems to be cost saving for society
Busch 2011	Total costs (10‐year follow‐up): BM: SEK 969,077 PT: SEK 1,425,048 CBT: SEK 1,491,298 UC: SEK 1,502,898	There was a decrease in costs per individual in the behavioural medicine rehabilitation group compared to usual care
Lambeek 2010a	Direct costs: WI: GBP 1479 UC: GBP 1262 Indirect costs: WI: GBP 11,686 UC: GBP 17,213	The work intervention for workers sick listed because of low back pain had substantial economic benefits over usual care
Loisel 1997a	1‐year follow‐up. Saved consequence of disease costs against standard care: CAD 604 Cost‐benefit: CAD 220 6‐, 4‐year follow‐up. Saved consequence of disease costs against standard care: CAD 10,697 Cost‐benefit: CAD 16,827 Lower costs in the workplace intervention than in the control group. Significance was not calculated	There was a small number of very costly cases
Tamminga 2013	Total intervention costs per worker in the intervention group: EUR 119 Productivity loss Human Capital Approach: WI: EUR 41,393 UC: EUR 38,968 Productivity loss Friction Costs Approach: WI: EUR 14,030 UC: EUR 13,529 Costs work adjustments: WI: EUR 2975 UC EUR 3025	Costs did not differ statistically between groups
van Oostrom 2010a	Direct costs: WI: EUR 4587 UC: EUR 3560 Indirect costs: WI: EUR 17,842 UC: EUR 16,440	The workplace intervention had no economic benefit compared with usual care
Vlasveld 2012a	Direct costs: WI: EUR 3900 UC: EUR 4600 Indirect costs: WI: EUR 10,110 UC: EUR 11,627	Comparable findings between both groups
BM: behavioural medicine rehabilitation CBT: cognitive behavioural therapy PT: physical therapy UC: usual care WI: workplace intervention

Table 1. Cost outcomes

Table 2. Content of the interventions ‐ 1

General characteristics of interventions	Specific characteristics interventions	Anema/Steenstra	Arnetz	Blonk	Bültmann	Busch	Feuerstein	Hees
Applied components definition workplace intervention	Changes workplace or equipment	x	x	x	x	x	x	‐
	Changes work design and organisation including working relationships	x	x	x	x	x	‐	x
	Changes in working conditions	‐	‐	x	‐	x	‐	‐
	Changes to the work environment	x	‐	‐	x	x	x	x
	Case management with worker and employer	x	x	‐	x	x	x	x
Contacts	Number of meetings	3	1	5 to 6	2	?	4 to 5	18
	Duration contact	1 h	?	?	2.5 h	1 h	1 to 2 h	1 to 2 h
Stakeholders involved	Worker	x	x	x	x	x	x	x
	Employer/supervisor	x	x	Self employed	x	x	x	x
	Occupational physician	‐	‐	‐	x	‐	‐	x
	Occupational nurse	x	‐	‐	x	‐	x	‐
	Ergonomist	x	x			x	‐	‐
	Representative of a union	‐	‐	‐	‐	‐	‐	‐
	Representative of an insurer	‐	x	x	‐	x	‐	‐
Type of contact	Face‐to‐face	x	x	x	x	x	x	x
	By phone	‐	‐	‐	‐	x	‐	x
Place of contact	At workplace	x	x	x	x	x	x	x
	Other	‐	‐	‐	‐	Rehabilitation centre	Home and provider office	Psychiatry department
Main treatment provider, work intervention		Ergonomist, occupational nurse	Insurance agency case manager	Labour expert	Social worker	?	Nurse case manager	Occupational therapist
Training treatment provider, work intervention		Yes	?	Yes	?	?	Yes	Yes
A 'x' mark indicates that the study fits the specific intervention characteristic. A '?' mark indicates that it is unclear whether the study fits the specific intervention characteristic.

Table 2. Content of the interventions ‐ 1

Table 3. Content of the interventions ‐ 2

General characteristics of interventions	Specific characteristics interventions	Lambeek	Loisel	Noordik	Tamminga	van Oostrom	Verbeek	Vlasveld
Applied components definition workplace intervention	Changes workplace or equipment	x	x	‐	‐	x	x	x
	Changes work design and organisation including working relationships	x	x	x	x	x	x	x
	Changes in working conditions	x	x	‐	‐	x	‐	x
	Changes to the work environment	x	x	‐	‐	x	‐	x
	Case management with worker and employer	x	x	x	x	x	‐	x
Contacts	Number of meetings	3 to 29	?	Not prespecified	5	3	? ˜ 3	6 to 12
	Duration contact	?	1 to 3 h	?	25 min	1 h	20 min	?
Stakeholders involved	Worker	x	x	x	x	x	x	x
	Employer/supervisor	x	x	x	x	x	x	x
	Occupational physician	x	x	x	x	‐	x	x
	Occupational nurse	‐	‐	‐	‐	x	‐	‐
	Ergonomist	x	x	‐	‐	‐	‐	‐
	Representative of a union	‐	x	‐	‐	‐	‐	‐
	Representative of an insurer	‐	‐	‐	‐	‐	‐	‐
Type of contact	Face‐to‐face	x	x	x	x	x	x	x
	By phone	‐	x	‐	x	‐	‐	‐
Place of contact	At workplace	x	x	‐	x	x	‐	‐
	Other	‐	‐	Occupational Health Service	Hospital	‐	Occupational Health Service	Occupational Health Service
Main treatment provider, work intervention		Clinical occupational physician	Ergonomist	Occupational physician	Specialised nurse/ social worker	Return‐to‐work coordinator	Occupational physician	Occupational physician care manager
Training treatment provider, work intervention		Yes	?	Yes	Yes	Yes	Yes	Yes
A 'x' mark indicates that the study fits the specific intervention characteristic. A '?' mark indicates that it is unclear whether the study fits the specific intervention characteristic.

Table 3. Content of the interventions ‐ 2

Table 4. Grading of the quality of evidence

Comparison	Outcome	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias	GRADE quality
Workplace intervention vs usual care	Time until first RTW	Yes: 60% of studies were assigned high or unclear risk of bias	No: I² < 50%	No	No	Undetected	Moderate
	Time until lasting RTW	Yes: 50% of studies were assigned high risk of bias	Yes: I² > 50%	No	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: majority low risk of bias	No: I² < 50%	No	No	Undetected	High
	Recurrences of sickness absence	No: study with low risk of bias	NA	No	Yes: single study	Undetected	Moderate
	Functional status	No: majority low risk of bias	Yes: I² > 50%	No	No	Undetected	Moderate
	Pain	No: majority low risk of bias	No: I² < 50%	No	No	Undetected	High
	Depression	Yes: > 50% of studies were assigned high risk of bias	No: I² < 50%	No	Yes: wide CI	Undetected	Very low
Workplace intervention vs clinical intervention	Time until first RTW	Yes: study with high risk of bias	NA	No	Yes: single study	Undetected	Very low
Subgroup analyses:
Workplace intervention vs usual care: musculoskeletal disorders	Time until first RTW	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
	Time until lasting RTW	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
Workplace intervention vs usual care: mental health problems	Time until first RTW	Yes: Study with high risk of bias	NA	Yes: PICO deviant (subgroup analysis)	Yes: < 300 workers	Undetected	Very low
	Time until lasting RTW	Yes: >50% of studies was assigned high risk of bias	Yes: I² > 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Low
Workplace intervention only vs usual care	Time until first RTW	Yes: >50% of studies was assigned unclear risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Low
	Time until lasting RTW	Yes: >50% of studies was assigned high risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
Workplace intervention + cognitive behavioural intervention vs usual care	Time until first RTW	Yes: >25% of studies was assigned high risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	Yes: < 300 workers	Undetected	Very low
	Time until lasting RTW	Yes: >25% of studies was assigned high risk of bias	Yes: I² > 50%	Yes: PICO deviant (subgroup analysis)	Yes: wide CI	Undetected	Very low
	Cumulative duration of sickness absence	No: Majority low risk of bias	No: I² < 50%	Yes: PICO deviant (subgroup analysis)	No	Undetected	Moderate
CI: confidence interval NA: not applicable PICO: patients, intervention, control, outcome RTW: return to work

Table 4. Grading of the quality of evidence

Comparison 1. Workplace intervention versus usual care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Time until first RTW Show forest plot	5	612	Hazard Ratio (Random, 95% CI)	1.55 [1.20, 2.01]

1.1 Musculoskeletal disorders	4	543	Hazard Ratio (Random, 95% CI)	1.44 [1.15, 1.82]
1.2 Mental health problems	1	69	Hazard Ratio (Random, 95% CI)	2.64 [1.41, 4.95]
2 Time until lasting RTW Show forest plot	6	885	Hazard Ratio (Random, 95% CI)	1.07 [0.72, 1.57]

2.1 Musculoskeletal disorders	2	330	Hazard Ratio (Random, 95% CI)	1.77 [1.37, 2.29]
2.2 Mental health problems	3	422	Hazard Ratio (Random, 95% CI)	0.79 [0.54, 1.17]
2.3 Cancer	1	133	Hazard Ratio (Random, 95% CI)	0.88 [0.53, 1.47]
3 Cumulative duration of sickness absence Show forest plot	7	950	Mean Difference (IV, Random, 95% CI)	‐33.33 [‐49.54, ‐17.12]

3.1 Musculoskeletal disorders	5	679	Mean Difference (IV, Random, 95% CI)	‐40.47 [‐55.98, ‐24.96]
3.2 Mental health problems	2	271	Mean Difference (IV, Random, 95% CI)	‐8.42 [‐35.99, 19.16]
4 Functional status Show forest plot	6	628	Std. Mean Difference (IV, Random, 95% CI)	‐0.33 [‐0.58, ‐0.08]

5 Symptoms ‐ Depression Show forest plot	4	410	Std. Mean Difference (IV, Random, 95% CI)	‐0.12 [‐0.35, 0.11]

6 Pain Show forest plot	5	531	Std. Mean Difference (IV, Random, 95% CI)	‐0.26 [‐0.47, ‐0.06]

7 Time until first RTW Show forest plot	5	612	Hazard Ratio (Random, 95% CI)	1.55 [1.20, 2.01]

7.1 WI only	3	347	Hazard Ratio (Random, 95% CI)	1.35 [1.01, 1.82]
7.2 WI and cognitive behavioral intervention	2	265	Hazard Ratio (Random, 95% CI)	1.93 [1.27, 2.93]
8 Time until lasting RTW Show forest plot	6	885	Hazard Ratio (Random, 95% CI)	1.07 [0.72, 1.57]

8.1 WI only	3	395	Hazard Ratio (Random, 95% CI)	0.95 [0.74, 1.21]
8.2 WI and cognitive behavioral intervention	3	490	Hazard Ratio (Random, 95% CI)	1.21 [0.56, 2.62]
9 Cumulative duration of sickness absence Show forest plot	7	950	Mean Difference (IV, Random, 95% CI)	‐33.33 [‐49.54, ‐17.12]

9.1 WI only	4	494	Mean Difference (IV, Random, 95% CI)	‐31.16 [‐55.87, ‐6.45]
9.2 WI and cognitive behavioral intervention	3	456	Mean Difference (IV, Random, 95% CI)	‐35.99 [‐62.21, ‐9.77]

Comparison 1. Workplace intervention versus usual care

Comparison 2. Workplace intervention versus clinical intervention

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Time until first RTW Show forest plot	1		Hazard Ratio (Random, 95% CI)	Subtotals only

Comparison 2. Workplace intervention versus clinical intervention