• Epidemiology
• ergonomics

Research to assess ergonomic interventions that aim to improve the musculoskeletal health of workers is, for good reason, criticised as being of poor quality. Study design can be inadequate with poor or missing control groups, risk of bias and confounders potentially influencing the outcome. Many ergonomic intervention reviews have made this point, from the early review of Westgaard and Winkel1 to the review by Driessen et al2 published in this issue of OEM (see page 277). Researchers have responded to this criticism by upgrading study design, and in some studies aiming for the gold standard of randomised controlled trials (RCTs). Secondary to this development, recent reviews of ergonomic intervention have performed an extensive pruning of studies on the basis of flawed design (eg, Brewer et al3, Burton et al4, Boocock et al5), culminating in the review by Driessen et al2 who only included RCT studies. Inclusion and exclusion criteria follow the Cochrane mould (Higgins and Green6). Literature searches for reviews have typically identified several thousands of potentially relevant studies, but with only a few tens judged to be of sufficient quality for use to further assess the effects of ergonomic interventions.

While the aim of attaining the best study design for ergonomic intervention studies is applauded, the above trend raises (at least) two queries: (1) is the RCT design the best attainable to the exclusion of other study designs and (2) is a review process with exclusion of studies that in the limiting case do not approach the RCT standard reasonable (ignoring for the moment that this is a valid specific objective of a review)? Studies with alternative designs, for example, quasi-experimental and qualitative studies, have been largely ignored in occupational medicine journals, which is criticised (Baker et al7).

The RCT design was developed to guard against bias in the closed circumstances of testing clinical treatment or medication. The use of RCTs in complex environments takes place in circumstances with many opportunities for effect modification, as discussed in the related field of public health interventions (Victora et al8). Testing the effects of ergonomic workplace interventions is exposed to a number of threats typical of dynamic systems responsive to many different influences. Organisational changes, representing psychosocial strain, are common and may easily occur within the typical 1-year observation period. Study setting, operationalised as biomechanical and psychosocial exposure conditions at baseline, is important. The behaviour of intervention stakeholders may critically influence outcome. There will be interaction effects between intervention measures, intervention delivery and workplace setting. These points are illustrated by a kitchen intervention study that did not use an RCT design but included process data (Nielsen et al9). Of the four kitchens in the study, improved worker health was observed in one intervention and one control kitchen, while one intervention and one control kitchen showed no improvements. Process evaluation, generally not available in RCT studies, provided a reasonable post hoc explanation of this result, pinpointing management actions in the delivery of the intervention.

Driessen et al2 aim to review both physical (workplace modifications) and organisational (influencing the temporal pattern of mechanical exposures) ergonomic interventions. Ten studies are included, but only one is classified as an organisational intervention. Their meta-analysis of physical interventions spans the diverse settings of municipal kitchens and office environments. Their conclusion of low to moderate evidence of no effect of ergonomic interventions in improving low back and neck pain of workers is certainly correct on the basis of their evidence. The pertinent question is whether it is fair to extend this conclusion to ergonomic interventions in general and whether the inclusion of studies of other designs will offer a more trustworthy (in this context, more generalisable) result? This parallels in part the three criteria set up by Victora et al8 for assessing public health interventions: probability (best assessed by RCT studies), plausibility (assessment of causality based on considerations of mechanisms and confounder effects) and adequacy (sufficient exposure modification).

Driessen et al2 make the point that their literature base is restricted and the conclusion of no effect of ergonomic interventions cannot be generalised. Furthermore, exposure modification may be small and other risk factors, in particular psychosocial exposures, are not accounted for in all studies. Compliance levels and confounders are mostly unknown. Most studies consider office workers, but also with this limitation there will be heterogeneity by intervention measure, intervention delivery and setting, which is probed to very limited extent. It can be argued a meta-analysis on this basis does not provide a true representation of intervention effectiveness even for office workers.

Can a larger literature base be used to derive a better representation of ergonomic intervention effects? A ‘best evidence’ approach (Slavin10) to reviews of ergonomic interventions seems useful, where studies utilising the strengths of different study designs are combined to provide the best overall assessment of ergonomic intervention effects, considering worker exposures, intervention delivery, setting and other relevant contextual factors. A corollary is that the conclusions of reviews with material emphasising a selection of important study qualities should not be interpreted beyond their stated objectives, with due regard for the (limited) range of evidence serving as the basis for the reviews. The last point is often duly made by the authors, but then forgotten when further referring to the review.