Article Text
Statistics from Altmetric.com
Harris et al1 provide useful new data for evaluating which occupational hazards are most deadly, and how this has changed over time. They have estimated excess deaths for different occupational groups, presumably exposed to specific hazards, via calculation of proportional mortality ratio (PMRs) (observed vs expected deaths) from death certificates, for causes of death strongly associated with occupations. Excess deaths (observed minus expected) over three study periods (1979–1990, 1991–2000 and 2001–2010) are presented (excess/observed deaths is the attributable fraction (AF) among the exposed). For some causes resulting solely from occupational exposures, there are no expected deaths without exposure, so all observed deaths across all occupations are considered excessive (eg, asbestosis, coal miner's pneumoconiosis). Among the more striking findings were the drop in excess deaths in exposed occupations from chronic bronchitis/emphysema, and in deaths from injury and poisoning.
There are a number of limitations to the method, most of which are noted by the authors. Change in annual excess deaths over time conflates improved working conditions/lower exposures, fewer people working in specific occupations (and hence fewer deaths), and declining overall mortality in the population over time. To zero in on improved working conditions/lower exposure, it might be best to emphasise the PMR itself, although clearly excess deaths are an important public health measure.
A large amount of misclassification is inherent in using death certificate occupation as a proxy for exposure. However, while this will bias effect estimates downward for the subset of the occupational group(s) who are actually exposed, it should not bias the PMR for the entire occupational group(s) of interest. Nonetheless, there is also the question of the accuracy of the reported last occupation itself on the death certificate, as well as the use of last occupation itself rather usual occupation (more aetiologically relevant, but not available). The authors note some examples of discrepancies regarding the correspondence between last and usual occupational (see also Luckhaupt et al,2 for US data). Another issue is which occupations are assumed to be at risk of any specific cause of death due to an exposure/hazard. For example, five job groups were thought to be at risk for bronchitis/emphysema (mining/energy manager, glass/ceramic workers, coal miners, molders/electroplaters and other metal manufacturers). These groups accounted for only 2000 of 39 000 deaths (5%) from this cause in 2001–2010; members of other occupations might have also had exposures which put them at risk. There is a presumably trade-off here between specificity and sensitivity.
As noted by the authors, use of the PMR itself to determine excess deaths from a given cause is a limitation because an occupational group may have an excess (or deficit) from all other causes of death which can bias the cause-specific PMR downward (or upward).
Harris et al1 have omitted a number of important outcomes which are known to be associated with occupation, in particular lung cancer, but also others less clearly associated, such as breast cancer (associated with shift work, an International Agency for Research on Cancer probable carcinogen) and coronary heart disease (associated with passive smoking). It would be inherently difficult to pick which occupations are associated with shift work or passive smoking. Regarding lung cancer, the authors argue correctly that lung cancer PMRs by occupation should be adjusted for smoking, which they promise in future work. However, it is not clear why they then included bronchitis/emphysema, also strongly associated with smoking. Furthermore, the methods of indirect adjustment for smoking in PMRs are by now fairly standard, and such adjustments are routinely made in the comparable US data from the National Institute for Occupational Safety and Health.3 ,4
A different approach to estimating the population AF/excess deaths by country has been taken by the Global Burden of Disease (GBD) project,5 and by investigators in specific countries.6 ,7 Here the first step is to derive a rate ratio (RR) for exposure to a given toxic agent in relation to a given disease (eg, asbestos and lung cancer), often via a meta-analysis of studies controlling for confounders, then determine the per cent of the population exposed in the past to that agent as of a given target year, and finally calculate the population AF for deaths in that year. Alternatively, different RRs are assigned to different proportions of a presumably exposed occupational group, reflecting the estimated per cent exposed given levels of the toxin. This method has the advantage of using an RR specific to the toxin, but the disadvantage of requiring an estimate of what per cent of people in a given occupation/industry have been exposed over time (which requires an assumed latent period and occupational turnover rate). The authors note that this second approach has been used in Britain by Slack et al,7 and results for one occupational agent (wood dust/leather dust) and cause of death outcome (sinonasal cancer) are available in both Harris et al1 and Slack et al.7 Applying Slack et al's population AF of 0.22 for deaths due to exposure to wood dust/leather dust, to the ∼40 annual sinonasal cancer deaths in 2001–2010, would result in 8.8 annual attributable deaths in the population, compared to Harris et al's estimated 1.8 annual attributable deaths (see their table 3) in occupations exposed to wood dust/leather dust. This difference might be explained if other workers were exposed outside the four occupational groups chosen by Harris et al. Future further side-by-side comparisons of the two methods for other exposures/outcomes will be instructive.
It is worth noting that death is a less sensitive endpoint than disease; there are many types of occupational surveillance for disease. The GBD method can be used for morbidity as well as mortality. Other types of disease surveillance include registries for usually non-fatal sentinel diseases caused only by occupation (eg, silicosis,8 and for occupational pesticide poisoning.9 One can also conduct cohort studies using national census data on occupation linked to cancer registries.10 The latter method can estimate excess cases of cancer in different occupations, similar Harris et al, but without relying on PMRs. For the moment such occupation/cancer surveillance is limited to a few countries with very good national data, for example, Scandinavia.
Footnotes
Correction notice The paper has been updated since it was published Online First. The abbreviation PMR in the first paragraph was incorrectly expanded and has been corrected.
Competing interests None declared.
Provenance and peer review Not commissioned; internally peer reviewed.