Send letter to journal:
Re: Estimating the numbers of work-related musculoskeletal disorders

We are grateful to John Hodgson for his comments on our paper. It was not our intention to investigate the overall accuracy of Labour Force Survey statistics. Rather, we aimed to assess the potential for error from one specific source, namely the impossibility of meaningfully attributing a disorder to work when: a) there are no special clinical features that distinguish between “occupational” and “non-occupational” cases; and b) the relative risk associated with occupational exposure is not so high that all cases in exposed persons can reasonably be attributed to the exposure. In these circumstances, nobody can meaningfully classify an individual case as “occupational in origin” or “work-related”. However, it is possible to estimate the population burden of illness or disease that is attributable to work by calculations based on estimates of relative risk and of the prevalence of exposure in the population under consideration.

Our findings illustrate the inconsistent relation of self-reported attribution to calculated attributable numbers, and raise the possibility that the degree of inconsistency could vary over time. For this reason, time trends in counts of self-reported occupational illness should be treated with some scepticism.

Keith Palmer David Coggon

Send letter to journal:
Re: Estimating the numbers of work related musculoskeletal disorders

Dear Sir

I would like to respond to some of the issues raised by Palmer et al's recent paper "How common is repetitive strain injury?" and the associated editorial by Fred Gerr on the surveillance of work related musculoskeletal disorders.

The central issue is that of the reliability with which individuals can attribute their musculoskeletal symptoms or conditions to work. The question would be easier to resolve if there was a "gold standard" test against which such attributions could be tested. As Fred Gerr points out, this is not the case: our choice is between a number of imperfect methods.

Palmer et al claim that self reports will systematically overestimate the numbers of work-related cases, and seem to imply that there is sufficient error in such reports that they should not play an important role in the formulation of policy.

In respect of the over attribution claim, it is important to realise that Palmer et al's findings do not imply that the estimates HSE has drawn from the LFS for (self-reported) work-related upper limb disorders are exaggerated. On the contrary, their study implies that the prevalence (by their preferred attributable fraction measure) of work-related arm pain over a 12 month period was 6.5% (14% times 46%), while HSE's LFS measure gives an estimate of 0.9%.

In any survey the context in which questions are asked, and precise wording, can make substantial differences to the proportions of respondents who respond positively. The fact that a version of HSE's LFS question performed very differently in the context of the MRC research than it does in the LFS is not surprising. The MRC version of the HSE LFS question produced, in their study, a prevalence of work-related arm pain of 25% (46% times 54%) compared to 0.9% from the LFS. It is not surprising that this kind of question met in the context of a survey of "aches and pains in the community" will elicit a different response than a similar question in the LFS where the central topic is jobs and related matters.

What Palmer et al's study shows is that if you take an "attributable fraction" estimate due to specifically identified work activities, you get a different, and in this case lower, overall estimate than can arise from self reports. What the research cannot show is whether this difference is due to fundamentally mistaken attribution in the self reports, or from respondents applying different cut-offs for work-relatedness along a basically valid scale. The MRC research could not address this question because the attributable fraction approach only gives you an overall estimate, and does not identify individual cases as work-related or not.

If people's responses to questions about attribution were purely driven by their psychological makeup, with no connection to the true sources of their condition then, of course, such measures would carry no information relevant to the understanding and prevention of genuinely work -related musculoskeletal harm. However this would be an extreme, and somewhat implausible, claim. It is more likely that individuals have effectively different thresholds for identifying what seems to them worth describing as "work-related". No doubt, individual psychology plays a role in determining how these thresholds are set, but that is by no means the same thing as saying that the scale they are working on is essentially flawed.

The measurement of work-related ill-health is not straightforward. HSE responds to these difficulties by using a range of sources, including self reports in the LFS, but also using medical surveillance, compensation data and death certificates. In any measure it is important to be consistent from year to year so as to have the best possible chance of detecting the underlying trends. It is these trends, particularly over the medium and long term, which are important, rather than the precise levels.

We also keep our statistical methods under review and are currently working on analyses to improve our understanding of the recent rise in the self-reported work-related illness measure from the LFS. We will also be organising a workshop on the measurement of work-related ill-health to consider this and a range of related issues later this year.

Yours sincerely

John Hodgson Head of Statistics Health and Safety Executive