Workplace interventions to improve work ability : A systematic review and meta-analysis of their effectiveness

Downloaded from www.sjweh.fi on December 09, 2019 This work is licensed under a Creative Commons Attribution 4.0 International License. Print ISSN: 0355-3140 Electronic ISSN: 1795-990X Copyright (c) Scandinavian Journal of Work, Environment & Health Review Scand J Work Environ Health 2018;44(2):134-146 doi:10.5271/sjweh.3685 Workplace interventions to improve work ability: A systematic review and meta-analysis of their effectiveness by Oakman J, Neupane S, Proper KI, Kinsman N, Nygård C-H Affiliation: Centre for Ergonomics and Human Factors, School of Psychology and Public Health, La Trobe University, Melbourne, Australia. [email protected] Refers to the following texts of the Journal: 2002;28(2):85-93 2008;34(1):55-65 2009;35(1):1-5 2012;38(3):282-290 2013;39(5):456-467 2014;40(3):244-251 2015;41(2):153-163 2016;42(1):71-79 2016;42(6):557-560 2016;42(6):490-499 The following articles refer to this text: 2019;45(3):316-317; [online first; 29 November 2019] Key terms: ageing; employment; GRADE; meta-analysis; review; sustainable employment; systematic review; WAI; work ability; work ability index; work ability score; workplace intervention This article in PubMed: www.ncbi.nlm.nih.gov/pubmed/29493713 Additional material Please note that there is additional material available belonging to this article on the Scandinavian Journal of Work, Environment & Health -website. 134 Scand J Work Environ Health 2018, vol 44, no 2 Review Scand J Work Environ Health. 2018;44(2):134–146. doi:10.5271/sjweh.3685 Workplace interventions to improve work ability: A systematic review and meta-analysis of their effectiveness by Jodi Oakman, PhD,1 Subas Neupane, PhD,2 Karin I Proper, PhD,3 Natasha Kinsman, MND,1 Clas-Håkan Nygård, PhD 2 Oakman J, Neupane S, Proper KI, Kinsman N, Nygård C-H. Workplace interventions to improve work ability: A systematic review and meta-analysis of their effectiveness. Scand J Work Environ Health. 2018;44(2):134–146. doi:10.5271/sjweh.3685 Objective Extended working lives due to an ageing population will necessitate the maintenance of work ability across the life course. This systematic review aimed to analyze whether workplace interventions positively impact work ability. Methods We searched Medline, PsycINFO, CINAHL and Embase databases using relevant terms. Work-based interventions were those focused on individuals, the workplace, or multilevel (combination). Work ability – measured using the work ability index (WAI) or the single-item work ability score (WAS) – was the outcome measure. Grading of Recommendations Assessment, Development & Evaluation (GRADE) criteria was used to assess evidence quality, and impact statements were developed to synthesize the results. Meta-analysis was undertaken where appropriate. Results We reviewed 17 randomized control trials (comprising 22 articles). Multilevel interventions (N=5) included changes to work arrangements and liaisons with supervisors, whilst individual-focused interventions (N=12) involved behavior change or exercise programs. We identified only evidence of a moderate quality for either individual or multilevel interventions aiming to improve work ability. The meta-analysis of 13 studies found a small positive significant effect for interventions on work ability [overall pooled mean 0.12, 95% confidence interval (CI) 0.03–0.21] with no heterogeneity for the effect size (Chi2=11.28, P=0.51; I2=0%). Conclusions The meta-analysis showed a small positive effect, suggesting that workplace interventions might improve work ability. However, the quality of the evidence base was only moderate, precluding any firm conclusion. Further high quality studies are require to establish the role of interventions on work ability. Key terms ageing; employment; GRADE; sustainable employment; WAI; work ability index; work ability score. 1 Centre for Ergonomics and Human Factors, School of Psychology and Public Health, La Trobe University, Melbourne, Australia. 2 Faculty of Social Science, Health Sciences, University of Tampere, Tampere, Finland. 3 National Institute for Public Health and the Environment, Centre for Nutrition, Prevention and Health Services, Bilthoven, The Netherlands. Correspondence to: Associate Professor Jodi Oakman, Centre for Ergonomics and Human Factors, School of Psychology and Public Health, La Trobe University, Melbourne, Australia. [E-mail: [email protected]] The ageing of the population will necessitate longer working lives for many workers. Official retirement ages are being raised in many industrialized countries as a strategy to encourage workers to delay their exit from paid employment ( 1 ). Remaining employed requires workers to have the ability to meet the inherent requirements of their work, which can be challenging for those with injuries, chronic conditions, or normal ageing processes, particularly in jobs with physical demands ( 2, 3 ). Good health and engagement with work ( 4, 5 ) are key aspects of maintaining employment; an increased focus on sustainable employability is required to assist workers to remain employed for longer ( 6 ), consistent with the need for extended participation in the paid workforce. Sustainable employment requires a good match between a person and their working environment. A person–environment (PE) fit model proposes that attitudes, behaviors and other individual-level outcomes result not only from the person or their environment but rather from the relationship and interactions between the two ( 7–9 ). The concept of work ability, defined as the degree Scand J Work Environ Health 2018, vol 44, no 2 135 Oakman et al to which an employee is mentally and physically capable of performing their current role, and a balance between a person’s resources (eg, health, competence, and attitudes) and work demands (environment and community, organization of the work) ( 10, 11 ) is of relevance in the conceptualization of PE fit. Sustainable employability requires maintenance and promotion of work ability across all ages to prevent decline and potential early departure from the workforce. Work ability has been measured using different methods. A widely-used measure is the work ability index (WAI), which comprises a set of measures to determine a person’s current work ability ( 10, 11 ). The work ability concept and the WAI was developed in Finland in the 1980s and 90s with municipal employees in their midlife followed in a longitudinal study ( 10, 11 ). Factors related to the management, ergonomics, and lifestyles explained both a decline and an improvement in work ability during ageing. In other studies, poor work ability has been associated with older age, obesity, high mental work demands, lack of autonomy, poor physical work environment, and high physical work load ( 12 ). Individuals with poor work ability have an increased risk of early retirement ( 13–15 ), long-term sickness absence and work disability ( 16 ) as well as decreased functional ability and higher mortality in old age ( 17 ). Workplace interventions to address issues of PE fit, which include work ability are complex and thus challenging to implement and appropriately measure the effectiveness of. According to macro-ergonomics theory, which focuses on the interactions between organizations and systems, work productivity is improved when individuals are well matched to the inherent requirements of their work, which results in better outcomes at an individual and organizational level ( 9 ). This is consistent with the requirements for sustainable employment, which requires cognizance of the relevant individual and organizational determinants of work ability. Appropriately targeted interventions are required to ensure maintenance of work ability and prevention of decline and to take into account the complex multifactorial relationships between employees and the work environment. Whilst interventions to improve work ability have been undertaken in a range of populations, for example nurses, cleaners and construction workers ( 18–20 ), a comprehensive understanding of whether these initiatives are most effective when targeted at individuals or the organizations in which they work is lacking ( 21 ). The purpose of this review was to assess systematically the available evidence on the effectiveness of work-based interventions on the work ability of employees. In addition, it aimed to examine whether effectiveness is different for individualor organizational-focused interventions. Methods Search strategy To identify relevant studies of workplace interventions to promote work ability, an electronic literature search of the following databases was undertaken: Medline, PsycINFO, CINAHL, and Embase. The search was limited to English language, between January 2000 – August 2016 to capture articles relating to the contemporary work environment. Bibliographies of included papers were searched, and a cited reference search was undertaken using Web of Science. A sample search strategy for Medline is provided in the online appendix, table S1 ( www. sjweh.fi/show_abstract.php?abstract_id=3685 ). Selection of studies Inclusion criteria. Inclusion and exclusion criteria covered participants and the interventions. Only studies reporting on currently employed workers were included. In relation to the intervention, studies were included if they were connected to the workplace or a component of the intervention occurred at the workplace. Interventions could include modifications to: (i) the physical work environment, (ii) the work routine, (iii) work hours, and (iv) exercise or lifestyle change programs. Comparators could include current practice or other interventions. Only randomized control trials (RCT) were included in the review to improve the quality of the studies reviewed. Three reviewers (SN, JO and NK) independently assessed studies for inclusion. Initial selection of studies was based on title and abstract. In cases of disagreement between reviewers, the fulltext of studies was accessed and consensus reached. The three reviewers assessed the fulltext of all studies selected for potential inclusion and, where consensus was not reached, a third or fourth reviewer were consulted to resolve any differences (KP/CHN). Outcome measure Work ability, using the WAI, was the outcome measurement used in this review ( 10, 11 ). The complete WAI includes a range of questions relating to the physical and mental demands in relation to their work, diagnosed diseases, work limitations due to disease, sick leave, work ability prognosis and psychological resources. The WAI score range is 7–49 ( 10, 11 ). Reliability and validity of the WAI have been previously reported ( 22, 23 ). One single item in the WAI, the work ability score (WAS), was also used; participants are asked to compare their current work ability with their lifetime best and rate this from 0–10. Previous studies have reported the WAS to be a reasonable alternative to WAI ( 15 ). 136 Scand J Work Environ Health 2018, vol 44, no 2 Review of workplace interventions on work ability Data management A customized form was used to extract data from the relevant studies. Study characteristics were extracted and summarized including: study design, country where intervention was implemented, participant details, type of intervention, and results. Meta-analytic approach A meta-analysis was conducted using Review Manager 5.3 developed by the Cochrane Community. The outcome variable, work ability, was measured and reported as a continuous variable in all included studies. However, only the studies reporting mean, standard deviation (SD) or standard error (SE) and sample size for both intervention and control group at the last round of follow-up, or the studies reporting the change in mean and SD values from baseline to follow-up, are included in meta-analysis. The pooled standard mean difference (intervention minus control) and their 95% confidence intervals (CI) is reported as an overall synthesized measure of effect size using random effect models. Random effect models are used because the data from a series of studies where the effect size is assumed to vary between studies ( 24 ). Heterogeneity among studies was assessed by Chi2 test indicating heterogeneity when P<0.05. Moreover, I2 values are reported to describe the variability among studies (0–100%), where increasing values shows increasing heterogeneity. We first synthesized the effect size from all studies included in the meta-analysis, grouped by WAI and WAS. We also analyzed individualand multi-level-focused interventions, grouped by WAI and WAS, to determine the effect of the individual and multi-level-focused workplace interventions on work ability. Sensitivity analysis was undertaken among studies that used WAI as the outcome measure to calculate the pooled effect size among studies using multilevel interventions versus current practice, individual-focused interventions versus other interventions and individualfocused interventions versus current practice. The findings are presented as forest plots. Assessment of risk of bias Individual studies were assessed for risk of bias using a domain-based evaluation as recommended by the Cochrane Handbook ( 25 ). Three reviewers (SN, JO, NK) independently assessed the studies, with any differences resolved by consensus. Six areas of bias were assessed: selection, performance, detection, attrition, reporting, and "other". Each area of bias included several assessment domains. Due to the nature of workplace interventions, which does not allow for blinding, the criteria related to blinding of participants and providers (domains within performance bias) was not assessed, consistent with approaches used by others ( 26 ), leaving a total of ten domains. Each domain was assessed as high, low or unclear risk. The risk of bias associated with intention to treat analysis was assessed as high if >20% loss to follow-up occurred and with no intention to treat analysis ( 27 ). In cases where information was not available or a trial protocol not published or registered, the corresponding risk of bias domain was assessed as unclear. Grading the level of evidence The quality of evidence for the work ability outcome was assessed using the Grading of Recommendations Assessment, Development & Evaluation (GRADE) tool described in the Cochrane Handbook ( 28, 29 ). Evidence quality was assessed in relation to six criteria: study design, risk of bias or study limitations, consistency of results, directness, precision, and publication bias ( 30 ). Two authors (JO and SN) undertook the GRADE process with consensus reached by discussion. Using the GRADE system, the study design for each included study prescribes the starting level of evidence and, following further assessment, can be downor upgraded. An overall level of evidence was evaluated for each outcome as follows: (i) high quality - further research is very unlikely to change our confidence in the estimate of effect or accuracy; (ii) moderate quality - further research is likely to have an important impact on our confidence in the estimate of effect or accuracy and may change the estimate; (iii) low quality - further research is very likely to have an important impact on our confidence in the estimate of effect or accuracy and is likely to change the estimate; (iv) very low quality - any estimate of effect or accuracy is very uncertain. A statement of evidence quality (an impact statement) was then developed, which took into account the level of evidence and the likely impact on the particular outcome. Impact statements used in this review are based on standard qualitative statements developed by Glenton and colleagues ( 31 ). Results Selection of studies A total of 4737 references were retrieved following a search of electronic databases (Medline, PsycINFO, CINAHL, Embase). After removing 2517 duplicates, 2220 unique references remained. When assessed on title and abstract, 53 references were selected for potential inclusion ( figure 1 ). A bibliographic search of included Scand J Work Environ Health 2018, vol 44, no 2 137 Oakman et al articles and a citation search via Web of Science identified four additional references. Following fulltext examination of the 58 potential articles published between 2001 and 2016, 36 were excluded on the basis they were either not an RCT (17 articles) or WAI/WAS not used as an outcome measure (19 articles). Study characteristics Of the 17 studies (22 articles), 10 compared interventions with current practice and 7 compared interventions with other interventions ( table 1 ). Using a macro ergonomics approach as outlined previously, interventions were categorized as either taking an individual, workplace, or multilevel (individual and workplace) focus. Individual-focused interventions (12 studies) included exercise programs and education of individuals on healthy behaviors or coping strategies; however, no changes to the workplace were made. Workplace interventions were targeted at making changes to the workplace such as working hours or schedules, the physical work environment or job design, reduction of workplace hazards (see supplementary table S2, www. sjweh.fi/show_abstract.php?abstract_id=3685 ). Of the studies, 5 were classified as multilevel and 0 were considered only workplace focused (table 1). Comparators to the intervention were classified as "current practice" or "other interventions". Current practice was defined as the situation when no other intervention was instigated by the researchers; ie, participants continued all other activities in which they were usually involved. Other interventions were those assessed by the research group as a comparator to the intervention being tested, such as education or ergonomic training. This might have been a less intense version of the intervention being studied. Meta-analysis A total of 13 studies (8 with WAI and 5 with WAS) were analyzed in the meta-analysis. The pooled estimates of the effect of intervention on work ability were calculated Screening Included Eligibility Identification Records after duplicates removed (N = 2225) Full-text articles assessed for eligibility (n = 58) Studies included in qualitative synthesis (N = 17) (22 articles) Studies included in quantitative analysis (N = 13) (15 articles) Records excluded (N = 2167) Full-text articles excluded, with reasons (N =36) 12 = not an RCT 19 = WAI/WAS not used as an outcome measure 4= study protocol only Records identified through database searching (N = 4737) Additional records identified through other sources (N = 5) Records screened (N = 2225) Figure 1. Selection of studies: PRISMA flowchart. 138 Scand J Work Environ Health 2018, vol 44, no 2 Review of workplace interventions on work ability and presented in two groups according to the scale used to measure work ability ( figure 2 ). A synthesis provided a positive effect of intervention (overall pooled mean difference 0.12, 95% CI 0.03–0.21) with no heterogeneity for the effect size (Chi2=11.28, P=0.51; I2=0%). The pooled standard mean difference between intervention and control for the group of 8 studies that used WAI was 0.11 (95% CI -0.02–0.24) with low heterogeneity (Chi2=9.36, P=0.23; I2=25%). The pooled effect size from 5 studies using the WAS was 0.17 (95% CI 0.01–0.33) with no significant heterogeneity problems identified (Chi2 = 1.39, P=0.85; I2=0%). Figure 3 shows individual-focused interventions grouped by either WAI or WAS. A small positive significant overall pooled effect was found (overall pooled mean 0.12, 95% CI 0.01–0.22) among individualfocused interventions studies with no heterogeneity for the effect size (Chi2 = 8.68, P=0.37; I2=8%). The pooled effect size from six studies using the WAI and three studies using WAS was positive but not statistically significant with no major heterogeneity problems identified. Figure 4 shows multi-level focused interventions. The overall pooled effect of such interventions (four studies in total) was positive but not statistically signifiTable 1. Summary of studies. [+ = positive effect; - = negative effect.] First author/ country Population Intervention (I) and comparison (C) Follow-up Conclusion on effect Addley et al, 2014, Northern Ireland Government department employees, N=180 I: Lifestyle & physical assessment plus Health Coaching) and access to web based tools such as online personal trainer I: Lifestyle & physical assessment only C: current practice 12 months No effect Blangsted et al, 2008, Denmark Public administration authority employee. N=549 I: specific resistance training I: all round physical exercise C: Introduced to intervention via presentation of similar projects that had improved work conditions and occupational health. 12 months No effect De Boer, 2004, Netherlands International manufacturing company employees N=116 I: Occupational health program by occupational physician, liaison with employer and health practitioners. C: current practice that included consultation with occupational physician on request. 6 months, 2 years + Flannery et al, 2012, USA Nurse assistant residential aged care employees N=39 I: Worksite Heart Health Improvement Project C: education session only 3, 6 months + Jakobsen et al, 2015, Denmark Female healthcare workers N=200 I: Work or home based physical exercise for 10-weeks I: Home based physical exercise for 10-weeks 10 weeks + Koolhaas et al, 2010 & 2015, The Netherlands Hospital and university employees N=125 I: Individual workplace assessment, action plan and support training for supervisors C: regular annual appraisal Baseline, 3 & 12 months - Nurminen et al, 2002, Finland Female laundry workers N=260 I: worksite exercise training, feedback on physical capacity tests and advice on self-directed leisure time physical activity. C: Feedback on physical capacity tests and advice on self-directed leisure time physical activity. 3, 8, 12, 15 months No effect Perkio-Makela, 2001, Finland Female farmers N=126 I1: Physical exercise & ergonomics work techniques training group sessions I2: as above plus lifestyle and risk management education C: basic occupational health care services 1, 3, 6 years No effect Sundstrup et al, 2014, Denmark Slaughterhouse employees N=66 I: strength training for shoulder/arm/hand C: ergonomic training provided as per standard worksite prescription 10 weeks +ve effect Viester et al, 2015, The Netherlands Construction workers N=314 I: On-site life style coaching program C: current practice 6 and 12 months No effect Barene et al, 2014, Norway Hospital employees N=107 I1: Soccer sessions I2: Zumba sessions C: current practice 12 and 40 weeks No effect Coole, 2012, United Kingdom Employees with low-back pain referred to rehabilitation service N=51 I: Individual work support, liaising with health practitioners/employer. C: Group multi-rehab focused on self-management including education, physical conditioning, cognitive behavior approach 6 months No effect Gram et al, 2012, Denmark Construction workers N=67 I. 12 week exercise program during work hours 3x20 mins per week C. no exercise training but received 1 hour lecture on general health promotion. 12 weeks No effect Jorgensen 2011, Denmark Female cleaners from hospitals, cleaning companies & large businesses N=294 I1: Physical coordination training I2: Cognitive behavioural training C: health check of 1 hr duration at baseline 12 months No effect Muller et al, 2016, Germany Registered nurses N=70 I: Selection, optimization and compensation (SOC) training at work. C: Received SOC training one year later 12 months No effect Oude Hengel et al, 2011 & 2013, The Netherlands Construction workers N=293 I: Physical therapy advice and cognitive training 3, 6, 12 months No effect Von Thiele et al, 2015, Sweden Hospital employees N=202 I: integration of new health promotion system with continuous improvement system C: employees not working in the new system Baseline, 12 and 24 months No effect Scand J Work Environ Health 2018, vol 44, no 2 139 Oakman et al † Flannery et al, 2012 study was excluded because of the different scale (0-80) used to measure work ability index. Nurminen et al, 2002 study was excluded because of no detail data on work ability index provided at the last round of follow-up. ‡ Hengel et al, 2011 study was excluded because of different scale used to measure work ability score (0-20). Barene et al, 2014 study was excluded because of no detail data on work ability score provided at the last round of follow-up. Figure 2. Standard mean difference of work ability index † and work ability score ‡ and their 95% confidence interval (CI) between intervention and control group at the last round of follow-up. Figure 3. Standard mean difference (intervention vs control) and their 95% CI for work ability index and work ability score among studies with individually focused interventions. 140 Scand J Work Environ Health 2018, vol 44, no 2 Review of workplace interventions on work ability cant effect (overall pooled mean difference 0.15, 95% CI -0.04–0.33). Sensitivity analysis of those studies examining individual-focused versus other interventions and individual-focused interventions versus current practice using WAI as an outcome are presented separately (see supplementary figures S1 and S2, www.sjweh.fi/show_ abstract.php?abstract_id=3685 ). Individual-focused interventions with WAI as an outcome (three studies in the meta-analysis), showed no statistically significant pooled effect (mean 0.25, 95% CI -0.09–0.59) and a substantial level of heterogeneity (Chi2=7.26, P=0.03; I2=72%). No statistically significant pooled mean effect of individual-focused interventions versus current practice (0.03, 95% CI -0.16–0.23) was found. The pooled effect size of the studies that used multilevel interventions versus current practice with WAI as an outcome (two studies in the meta-analysis), found no statistically significant effect of the intervention (mean 0.06, 95% CI -0.20–0.31) with no heterogeneity (Chi2 =0.52, P=0.47; I2=0%) among studies (supplementary figure S3, www.sjweh.fi/show_abstract.php?abstract_id=3685 ). Risk of bias analysis Risk of bias assessments for all studies are presented in table 2 . Risk of bias was assessed as low, high, or unclear for all study types. Grading the evidence and synthesis Table 3 provides detail on evidence quality as assessed by GRADE. Four "summary of findings" subtables were generated on the basis of the focus of the intervention (individual/multilevel focus) and comparison groups (current practice/other intervention). The table also provides the impact statement for the outcome, based on the evidence quality. The GRADE approach assesses the overall body of evidence rather than the individual study; differences may result between a single study and an outcome effect for groups of studies. Effect of individual-focused interventions on work ability compared to current practice Six studies investigated the effect of individual-focused interventions compared to current practice. Current practice involved consultation with health physicians ( 32 ), health checks ( 19 ), and no changes to current practices ( 18, 33–35 ). Individual-focused interventions included health assessments and advice on behavior change ( 33, 34 ); three studies used exercise programs as the intervention ( 18, 19, 32 ) and one study involved training nurses in the use of selection, optimization and compensation to develop active coping skills ( 35 ). Three studies used the full WAI as the outcome measure and three studies used the single-item WAS measure ( table 1 ). The overall quality of the evidence was moderate. The principal reason for downgrading of the evidence was poor compliance in four of the six studies. Of these four studies, three involved exercise programs ( 18, 19, 32 ) and one involved lifestyle coaching ( 33 ). None of the studies reported a significiant effect on work ability, either as WAS or WAI. Figure 4. Standard mean difference (intervention vs control) and their 95% CI for work ability index and work ability score among studies with multi-level focused interventions. Scand J Work Environ Health 2018, vol 44, no 2 141 Oakman et al Effect of individual-focused interventions on work ability compared to other interventions Six studies represented by seven articles (table S2) compared individual-focused to other interventions. Interventions were either based on exercise ( 36–40 ) or lifestyle education of employees ( 41 ). Other interventions included a passive comparator such as information ( 36–38, 40, 41 ) and in one study training and education was provided as part of a hazard prevention system ( 39 ). Five studies used the full WAI and one used the single-item WAS. Only one exercisebased intervention reported a significant positive impact on work ability ( 39 ). Overall evidence quality was moderate. Again, the principal reason for the downgrading of the evidence was low compliance in four of the six studies. Of the four studies with low compliance, three were exercisebased ( 36–38 ) and one was based on physical activity education ( 41 ). Effect of multilevel-focused workplace interventions on work ability compared to current practice Four studies represented by seven articles compared multilevel focused interventions to current practice ( 20, 42–47 ). Interventions in the four studies were multifaceted and included several components: empowerment training, use of a tool to design rest breaks, an education booklet, supervisor training, and the implementation of a system to integrate health and work through the use of a continuous improvement process. Work ability was measured using the full WAI in three studies ( 20, 42, 44 ) and the single-item WAS in one study ( 43 ). Evidence quality was moderate, and downgrading was primarily due to low or uncertainty about compliance across all four studies reviewed ( 20, 42–44 ). Effect of multilevel-focused workplace interventions on work ability compared to other interventions One study compared a multilevel-focused approach to other interventions ( 48 ). The intervention covered eight vocational sessions along with group rehabilitation. The other intervention Table 2. Summary of risk of bias (-) = low risk, (?) = unclear, (+) = high risk. [ITT=intention to treat] Study Selection bias Detection bias Attrition bias Reporting bias Performance bias ITT analysis performed Random sequence generation Allocation concealment Group similarity at baseline for t a he most important prognostic factors Blinding of outcome assessors Timing of outcome assessments comparable in all groups Outcome data complete Free of selective outcome reporting a Cointerventions avoided b Compliance acceptable in all groups Addley et al, 2014 - ? - - - - - ? - - Barene et al, 2014 - - - - - + - ? + - Blangsted, 2008 - - - - - - - ? + - Coole, 2012 + ? - - - - - ? + - De Boer, 2004 ? ? - - - - - ? + - Flannery et al, 2012 - - - - - + - ? + - Gram et al, 2012 - - - - - - - - + - Jacobson, 2014 - - - - - - - ? + - Jorgensen, 2011 - - - - - - - ? + - Koolhaas, 2010 & 2015 ? - - - - - - ? ? + Muller, 2016 - - - - - - - ? - - Nurminen, 2002 - - - - - - - - - - Oude Hengel et al, 2011 & 2013 - - - - - - - ? + - Perkio-Makela, 2001 ? ? - - - + - ? + + Sundstrup et al, 2014 - - - - - - - ? ? - Von Thiele et al, 2015 - - - - - ? - - ? ? Viester, 2015 - - - - - - - ? + ? a Selective outcome reporting assessed as unclear if the protocol was not available (ISRCTN register checked and authors contacted if studies not registered). A possible limitation of this assessment – when assessing outcome reporting for studies with multiple articles in languages other than English, some articles may not have been detected in literature search (which was restricted to English) and therefore outcomes assessed as being not reported (resulting in overestimation of the risk). b Co-intervention defined as treatment/activity that would not normally be experienced in standard practice – if studies did not explicitly state co-interventions were avoided, an unclear rating was applied.