A PPENDIX (S TUDY V)

Design: The design of study V was consistent with the systematic review methodology outlined by the Cochrane Collaboration (www.cochrane.org). All phases of the review work were developed according to the Cochrane Handbook for Systematic Reviews of Interventions

and the 2009 Updated Method Guidelines for Systematic Reviews in the Cochrane Back Review Group

. We performed three broad comparisons: (1) workplace intervention versus no interventions, (2) workplace intervention versus usual care, and (3) comparison of two or more workplace interventions.

Inclusion/Exclusion criteria: Only RCTs were included, there were no language limitations, and the sole targets were adults who were of working age (18 to 67 years) and were either on the job or were absent (on sick leave, early retirement, or disability pension) but still connected with the workplace through permanent or temporary employment agreements. All sectors, branches, and types of jobs were included. The targeted employees were to have reported neck pain of acute (< 6 weeks), sub-acute (6–

12 weeks), or chronic (≥ 12 weeks) duration. Shoulder pain was included only if it was

described in conjunction with neck pain. The fluctuating nature of neck pain constituted

a challenge when defining the target group for this review, but we solved this problem

by including only studies in which at least 50% of the baseline population had neck

pain. Neck pain due to specific pathological conditions such as fractures, tumours,

infections, inflammatory processes, and ankylosing spondylitis were excluded.

The interventions could be a single strategy, or a combination of strategies, with different intervention programme labels (i.e., modified work, participatory ergonomic, ergonomic workplace visit, RTW interventions, or multidisciplinary ergonomic

interventions). By use of ICF terminology, we defined workplace intervention as: ―any action at the workplace with the aim of preventing health problems and disability, maintaining participation in work and reducing sickness absences, or facilitating early return-to-work. These interventions seek to modify the employees' physical or mental functions, their activity performance, participation challenges or the physical, social or attitudinal environment‖. Studies about clinical and health care interventions conducted outside the workplace were excluded. Also, studies were not included if they concerned exercise

and multidisciplinary biopsychosocial rehabilitation

, because those were covered in other Cochrane reviews.

Harms and other adverse effects were included if they were reported in the studies. The timing of outcome measures was reported according to the descriptions used in the included studies, and they were grouped as being short term (measured closest to four weeks after randomization), intermediate term (measured closest to six months after randomization), or long term (measured one year or longer after randomization)

. Trials were included if they measured at least one of the following outcomes

recommended by the Cochrane Back Review Group

: pain severity or pain prevalence that was self-reported on a visual analogue scale or the NSR scale, or was measured as the proportion of those with pain; absence from work, considered as time on benefits, number of hours or days on sick leave or lost time, proportion of

individuals returning to work, employment status; shift in employment status to working full-time, working part-time, or being on sick leave, disability pension, or early retirement.

Search strategies: Potential trials were identified by computer-aided searches (to July 2009) of these electronic bibliographic databases: CENTRAL (The Cochrane Library 2009, issue 3), MEDLINE, EMBASE, CINAHL, PsychINFO, ISI Web of Science, OTseeker (Occupational Therapy Systematic Review of Evidence), and PEDro (the Physiotherapy Evidence Database). The intervention section of the searches was purposely left open, because of the diversity of terms used to describe workplace interventions. References cited in included trials were also screened, and experts in the field were contacted to obtain additional studies.

Data collection: Before selection, the titles and abstracts (if available) of all identified studies were collected and duplicates were removed. We assessed our interpretation of the inclusion criteria in a pilot study of a sample comprising ten articles, some of which we considered to be definitely eligible, some definitely not eligible, and some

questionable. The inclusion form was revised in this manner. For all articles that had abstracts that appeared to meet our inclusion criteria, or either lacked abstracts or had abstracts upon which a decision could not be made, the full texts were obtained and independently screened by the same two reviewers to determine whether they met our inclusion criteria. Consensus was used to solve disagreements; if disagreements

persisted, a third reviewer was consulted. We dealt with missing data by contacting the

original investigators to request the absent information. Furthermore, any assumptions

concerning methods used to cope with missing data were made explicit, and the potential impact of missing data was addressed.

Data analysis: Initially, two reviewers worked independently to extract data from the included studies and record them on a standardized form. Twelve criteria were used to assess the risk of bias in the included studies

, and each of these was scored ―yes‖,

―no‖, or ―unclear‖. A trial with low risk of bias was defined as, at the least, having met criteria 1 (randomization), 2 (allocation concealment), 5 (outcome assessor blinding), and any three of the remaining nine criteria. Two reviewers independently assessed the risk of bias in a selection of trials and reached consensus on the final results. A third reviewer assessed the risk of bias in all included studies. Only one meta-analysis could be performed due to between-study diversity of interventions, outcomes, outcome measures, type of workers, and follow-up times. The two studies forming the meta-analysis were homogeneous in that they both focused on the body functions. For the outcomes, odds ratios (ORs) were calculated for dichotomous data, and mean differences were computed for continuous data with 95% CIs.

Some of the studies tested a single intervention, whereas others tested a set of interventions. Therefore, a content analysis of the interventions was performed as outlined in the 10 papers included in the review with the objective of delineating the exact content of the intervention. In these efforts, the ICF

was used as a

conceptual framework to help describe the components of the intervention(s) in the included studies. Assessments aimed at determining whether a specific intervention is clinically justified should not be based solely on statistically significant findings. Thus, we attempted to addressed five questions that could help determine the clinical

relevance of the interventions

.

Regardless of whether we had sufficient data to combine the results statistically, we assessed the overall quality of the evidence for our primary outcomes by using an adapted GRADE approach

. The quality of the evidence for a specific outcome was based on the performance against five domains: limitations of the study design, inconsistency, indirectness (inability to generalize), imprecision of results (insufficient or imprecise data), and publication bias across all studies that measured the outcome.

Two review authors worked independently to perform the GRADE analysis. Initially, the quality was good when at least two RCTs with a low risk of bias provided results for the outcome, and it was reduced by one level for each of the subsequent domains that were not met: High quality evidence. At least 75% of RCTs with no limitations of the study design, consistent, direct and precise data and no known or suspected

publication biases. Further research is unlikely to change either the estimate or our confidence in the results. Moderate quality evidence. One of the domains was not met.

Further research is likely to have an important impact on our confidence in the estimate

of effect and might change the estimate. Low quality evidence. Two of the domains

were not met. 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. Three of the domains were not met. We are very uncertain about the

estimate. No evidence. No RCTs were identified that addressed this outcome.