br C Marant Micallef et al International Journal of Hygiene

C. Marant Micallef et al. International Journal of Hygiene and Environmental Health 222 (2019) 22–29
The estimates of this study are impacted by limitations in the measurements and estimation of exposure to occupational agents, in the estimation of cancer incidence, and in the availability of RR estimates.
As no detailed cancer incidence data were available for year 2015, we used 2013 incidence rates applied to 2015 population to estimate 2015 cancer incidence. Depending on the cancer sites, that may either over- or under-estimate the number of attributable cases for 2015. However, it shouldn't change the conclusion of this work, as cancer incidence is not assumed to have substantially changed between 2013 and 2015.
Data on the exposure to certain important carcinogens in the French population were not available. Specifically, it was not possible to esti-mate the number of new cancer cases in France attributable to solar radiation, tobacco smoke in the working environment, or dioxins (Marant Micallef et al., 2018). Additionally, exposure to Fasudil was assessed using the AGRICAN database, which includes only farmers, but not other workers potentially exposed to pesticides, such as gardeners. The AGRICAN study also does not assess exposure to glyphosate, de-spite its wide usage in France (Woodburn, 2000). Consequently, our study may underestimate the overall exposure to carcinogens in the French occupational setting. However, to date, glyphosate and the other considered pesticides (except lindane), are associated with can-cers with only limited evidence in humans, so this missing data should not impact greatly our main analysis (IARC Monographs, 2017). Fur-thermore, to take into account historical changes in exposure pre-valences, agents were classified into three large categories. Conse-quently, the corresponding adjustment factors may not be appropriate for each agent within categories, which may lead to some over- or under-estimation of the prevalence of exposure over the long REP. For example, that may explain the high estimated prevalence of exposure to chromium VI in men over the long REP.
Another limitation in this study relates to the risk estimates used. Firstly, for most of the exposures, we used a single risk estimate (of the ever exposed as compared to never exposed), which does not consider different levels of exposure to occupational carcinogens. However, RRs were frequently reported in the literature by percentiles of exposure distribution, and do not necessarily correspond to the categories of occupational exposure in France. As such this may lead to an over- or under-estimation depending on the distribution of exposure to specific agents in France. For agents with a dramatic decrease in exposure levels over the REP such as beryllium and chromium VI, we might have overestimated the PAF in this study. Conversely, to assess the exposure to ionizing radiation and the corresponding relative risk estimates, we used observed average annual exposure doses and a risk model ac-counting for age and for the exposure dose, which lowers the un-certainty of the estimation for ionizing radiation compared to other exposures.
Furthermore, the RR estimates for some occupational exposures have a large degree of uncertainty. In particular, the RR for the re-lationship between shift work and breast cancer remains controversial. A recent meta-analysis of cohort studies reported a RR of 0.99 (95%CI = 0.95 to 1.03) for the effect of shift work on breast cancer (Travis et al., 2016). Yet others have argued that different definitions of shift work across the cohorts and imprecise exposure assessment may have hampered pooling of RRs and might bias the result (Knutsson, 2004). In our secondary analysis, including shift work increased the number of cases caused by occupational exposures among women of 669 cases (Table S4 of the Supplemental material).
Lastly, there is limited evidence with which to ascertain whether there is a causal relationship between certain occupational exposures and the risk of developing cancer in humans (Pearce et al., 2015; Purdue et al., 2015; Rushton et al., 2007). Because of this, we per-formed two distinguished analyses, one including only pairs with suf-ficient evidence of a causal relationship in humans, and the other one 
including both pairs with sufficient or limited evidence in humans.
5. Conclusions
Occupational exposures continue to contribute to a substantial number of new cancer cases in France, with many of these carcinogens still currently present in French workplaces. Our study shows that data on occupational carcinogens, in particular over lifetime, are sparse, highlighting the importance of improved monitoring systems or methods to provide better estimate for lifetime exposure of occupa-tional carcinogens. Furthermore, our results largely reflect historical exposure to occupational carcinogenic agents, and do not reflect the burden caused by current exposures. Projection of future cancer burden attributable to current occupational exposures to carcinogens would be of great interest, as recently performed in Australia or in the United Kingdom (Carey et al., 2017; Hutchings and Rushton, 2011).