We presented the development of a JEM cross-classifying 121 job groups with five generic mechanical exposures to the lower extremities. The JEM encompasses the whole labour market in Denmark and provides quantitative exposure measures, except for a minimally exposed job group, which was not included in the JEM. In general, the agreement between the experts’ rankings was fair to moderate and the face validity was found to be high.
The inter-rater agreements were higher than the mean weighted kappa values of between 0.2 and 0.3, which have been previously reported for standing, heavy lifting, and kneeling . This reflected the fact that we did not have the constraints faced by the authors of the just-mentioned study with respect to the grouping of occupational titles, and the fact that their comparisons were based on initial ratings, whereas our comparisons were made after correction of outlying estimates. We estimated that floor-layers were exposed to kneeling/squatting for on average 3.5 hours/day, which is comparable to estimates based on observations and measurements [13–15, 44]. It seems reasonable that no job groups obtained a higher mean for this exposure. We are not aware of other comparable exposure estimates based on observations and measurements. Our first priority was to rank the job groups in a valid way since this is a precondition for exploring exposure-response relationships. The face validity was high, so we think that our quantitative estimates reflected the ranking of true exposures quite well, whereas the absolute values are more questionable.
We grouped occupational titles instead of D-ISCO codes because D-ISCO codes are based on skills required to fulfil tasks and duties of the jobs and thus may not reflect specific exposures. Maybe the agreement could have been improved if we had provided the experts with brief texts describing the work content of the occupational titles represented in the job groups . Such descriptive texts could also make it easier to adapt the matrix for studies of other populations. We refrained from the use of exposure vignettes because our exposure assessment panel included experienced specialists, who knew the tasks of the majority of occupational titles present in the job groups. Another way of obtaining better agreement could be to use 10–15 benchmarks in terms of occupational titles representing specific job groups, which the experts consensus rated before the remaining rating process. In this way the experts could calibrate their estimates to a common scale [17, 28, 45].
We designed the job axis of The Lower Body JEM to contain as homogenous exposure groups as possible . Based on theories of classic and Berkson errors, group-based exposure assessment should be less subject to attenuation bias than individual-based approaches . However, to the extent that we mixed occupational titles with high and low true exposures within the job groups, the observed mean values of the job groups would erroneously seem similar (ultimately, we could have constructed our job groups in such a poor way that all group means were equal, meaning that we would be unable to detect exposure-response relationships). Subsequently, quantitative exposure-response relationships obtained by the JEM may be calibrated by validation studies based on observation or technical measurements of selected exposures for selected occupational titles or job groups.
We classified more than 50% of all occupational titles as minimally exposed. To the extent that these titles were in fact more than zero-exposed, exposure-response relationships based on the JEM would underestimate true associations (if the possibility of a U-shaped relationship is disregarded). The omission of occupational titles judged to entail minimal exposures precludes the use of the JEM to study effects of these exposures, which may be relevant with respect to other outcomes than OA . As a future refinement of the JEM, the large group of minimally exposed occupational titles may be subdivided and provided with exposure estimates. Some of the job groups in the JEM received one or more exposure estimates that were lower than the cut-off points used in the screening process. We kept these estimates in the JEM to reduce the risk of underestimation of associations due to misclassification of exposures that were not minimal.
The use of probability of exposure has been proposed as a means to minimize bias due to misclassification of exposures , and has been used in recent studies of lower body exposure [26, 27]. However, the probability approach may be more meaningful in studies of chemical exposures that occur in specific occupational groups, where some group members are exposed and others are not. For mechanical exposures, the situation is typically different. For instance, standing/walking is widely distributed and does not occur in an on-or-off manner, and exposure to whole-body vibration occurs in few occupations where the majority of the group members are exposed to some extent. Therefore, we found it more informative to provide quantitative estimates of mean exposures.
We did not use different estimates for men and women within the same occupation. Women in heavily exposed jobs may actually be less exposed than their male colleagues, for instance due to gender segregation of tasks within jobs. This would have the effect that women would erroneously seem to be less affected by heavy exposures than men. A perspective for improvement of the JEM could be to provide gender specific estimates for selected groups . However, the Danish labour market is to a large extent gender segregated so that men and women work in different jobs, which means that the practical significance of such an effort may be limited.
The job groups were constructed to have similar exposure profiles across the five exposure variables that we assessed. The relatively large number of job groups means that it will be possible to update specific exposure estimates in The Lower Body JEM as new knowledge is obtained, and other researchers will be able to modify the JEM for use in different study populations. The JEM has already proved useful in a study of the work-relatedness of inguinal hernias  and in our recent case–control study of hip OA . In these studies, the exposure estimates from the Lower Body JEM were used to calculate cumulative exposure measures. The JEM may also prove useful for research into e.g. varicose veins, where prolonged sitting or standing/walking have been suggested as risk factors .
When the JEM is applied for exposure assessment in an epidemiologic study, a high prevalence of job groups with high inter-rater agreement and a low prevalence of job groups with low inter-rater agreement would yield kappa values for the ranking of exposures in the study population, which are larger than calculated for the JEM per se (this situation seems quite likely since large job groups would be particularly well known to the experts). In this situation, the JEM-based exposure assessment must be expected to lead to risk estimates that are closer to the real than suggested by the presented kappa values, provided that high agreement reflects a better estimate of the true exposure. Thus, the influence of agreement between raters on the probability of biased risk estimates is related to the prevalence of the job groups in the study population. It may even be a design option to restrict study populations to job groups with relatively high agreement between raters to counteract biased risk estimates.
Until more accurate and precise methods for exposure assessment have been developed that are feasible for use in large scale population studies of hip and knee OA and other lower extremity disorders, we find it promising to explore the avenue of a JEM approach based on expert ratings of mechanical exposures.