Our evaluation of the relationship between AN exposure and lung cancer mortality using external comparisons based on plant-specific regional mortality rates produced results similar to those we reported in our reanalysis of the first NCI AN cohort study follow-up (25). Namely, we observed statistically significant deficits in lung cancer deaths among unexposed workers in contrast to SMRs at or near the null value among the most highly exposed workers. This indicates that the statistically significant 1.43-fold lung cancer excess among workers in the highest quintile of cumulative AN exposure (lagged 10 years) reported by NCI using internal rate comparisons (HRs) was fundamentally a result of comparing a null value to a statistically significant deficit (21). In this second reanalysis we observed a similar but more pronounced pattern for pneumonitis mortality in which a statistically significant 4.73-fold excess stemmed from comparing a 6% deficit in deaths among the highest exposed to a large, statistically significant deficit among the unexposed.
As we discussed in our first reanalysis (25), several possible explanations exist for the large differences in the lung cancer relative risks in this study population when internal or external comparison rates are used. First, internal comparisons produce more valid results because selection bias stemming from the healthy worker effect can reduce the putative effect of high exposure to acrylonitrile when external comparison rates are used. However, the NCI cohort has now been followed for several decades and much of any healthy worker effect present in the first update has attenuated. Also, the selection for workers who are healthy at time of hire is usually more relevant for chronic cardiovascular and nonmalignant respiratory diseases than lung cancer, which has a relatively sudden onset, short survival time and high case-fatality rate(24). Second, external comparisons produce more valid results because the unexposed group has a different underlying lung cancer risk than the exposed group. The inordinately low SMRs for lung cancer among unexposed workers overall are puzzling given that we used regional standard population rates. As regional rates can help adjust for the social, cultural and economic factors related to diseases such as lung cancer, and even help to adjust for geographic variability in tobacco use, it is difficult to postulate what non-occupational factors may have had such a profound influence on the lung cancer mortality experience of the unexposed workers. Supplemental Table 12 compares the distribution of the major demographic factors of unexposed and exposed workers in the NCI cohort. Factors possibly related to AN exposure (e.g., sex, age at hire, year of hire, wage class) do not differ markedly between unexposed and exposed workers and our adjustments for these factors in internal relative risk models for lung cancer likely did not lead to any residual confounding.
Third, given the large number of lung cancer deaths and overall robustness of the NCI study, chance or under-ascertainment of deaths are unlikely explanations for the low SMRs among unexposed workers. Finally, the possibility remains that some heretofore unknown selection factors for low lung cancer incidence were operating on members of this cohort, or that some type of protective effect for lung cancer arose from a particular exposure or combination of exposures encountered at the study plants. Some of these explanations also apply to our findings for pneumonitis although the heterogenous nature of this cause of death category and their uncertain etiologies further complicates their interpretation. Consequently, the underlying reason for the inordinately low SMRs among unexposed workers remains unknown.
In the NCI study, investigators have attempted to adjust lung cancer risk estimates for confounding by smoking via a nested case-cohort study of lung cancer based on a 10% random sample of the cohort. Blair et al. (5) initially reported an exceedingly low RR for lung cancer among ever smokers compared to never smokers of 3.6 (95%CI = 1.6–8.2) suggesting that smoking status was most likely misclassified among subjects. This observation was confirmed later by Cunningham in 2005 (26). The Blair et al. finding contrasts starkly with the corresponding HR of 19.1 (95%CI = 5.3–68.9) reported by Koutros et al. (21) in the updated cohort based on imputed smoking data for the new lung cancer cases. While the more recent estimate is in line with well-known relative risks for smoking and lung cancer, it is not clear why this estimate increased so dramatically from the first update and was reported with much less precision, especially considering that 646 new lung cancer cases that were added to the case-cohort study in the recent update. Koutros et al. (21)also limited smoking adjustment to lung cancer HRs among workers in the highest cumulative AN exposure category, thus precluding an evaluation of how smoking adjustment impacted AN exposure-response relationships.
In contrast, our application of the Richardson indirect method enabled adjustment of lung cancer risks for potential confounding by smoking at the cohort level and for workers in all categories of AN exposure. Also, unlike the NCI study, this method enabled additional adjustment of lung cancer risks for asbestos exposure and smoking adjustment of bladder cancer risks. Consequently, our cohort level adjustment of lung cancer RRs for confounding by smoking and asbestos yielded for the total cohort and within the eight study plants (smoking adjustment only) mostly decreased RRs and much less evidence of a positive association with cumulative AN exposure than reported by NCI.
In the current reanalysis, we recognized markedly lower mesothelioma SMRs based on regional vs. U.S. death rates as well as uniquely higher lung cancer SMRs among unexposed and exposed workers from Plant 4 and noted that the local area includes the Newport News Shipyards where many asbestos-containing materials were used historically. Shipyard work is associated with elevated lung cancer and mesothelioma risks, and some Plant 4 workers (and persons in the local general population) may have been employed for some time in the yards. Despite the elevated lung cancer rates among Plant 4 workers, we observed no mesothelioma deaths among Plant 4 workers (the 21 mesothelioma deaths occurred in Plants 1, 3, 5, 6, 7, 8)
While our Richardson lung cancer adjustment for asbestos likely accounted for at least some of any shipyard related asbestos exposures that may have occurred among Plant 4 workers, residual confounding from asbestos exposures unique to Plant 4 may remain in the cohort. Thus, we repeated key mortality comparisons omitting or isolating workers from Plant 4. This sensitivity analysis of the remaining cohort revealed decreased lung cancer mortality risks and even less evidence of an AN exposure-response relationship, particularly in RR models adjusted for both smoking and asbestos exposure. Thus, while Plant 4 comprised only 13.3% of the total cohort, it had a relatively large impact on the overall findings for lung cancer that were not recognized in the NCI study (5, 21).
As expected from the small relative risk for smoking and bladder cancer, our application of the Richardson method to bladder cancer revealed little evidence of positive confounding by smoking and similar lack of a positive relationship with cumulative AN exposure as reported by NCI (21). In an independent cohort study of Plant 3 workers, (14) reported a statistically significant excess risk for bladder cancer based on four observed deaths. An expanded Plant 3 study to investigate the bladder cancer excess (15)and a further expansion and extended follow-up of the cohort (16) found that the bladder cancer excess decreased to a not statistically significant level. As noted by Koutros and colleagues, because of high survival rates, bladder cancer risks are best evaluated in studies that include both incidence cases and deaths.
Our cohort level adjustments for smoking based on internal mortality comparisons were corroborated by our smoking adjustment of regional rate-based SMRs for lung and bladder cancer among workers in the highest cumulative AN exposure category. This analysis revealed considerably higher smoking rates among workers compared with plant-specific standard state populations indicating that unadjusted lung and bladder cancer SMRs in the NCI study were heavily and positively confounded by smoking.
The results of our reanalysis of the 2011 update of the NCI study continue to reflect the lack of clear and consistent evidence of an association between AN exposure and mortality from lung cancer both across earlier studies and within the current NCI study. In the latter case, we observed considerable inconsistencies in results when using external vs. internal mortality comparisons and when considering potential confounding by smoking and/or asbestos and/or the impact of Plant 4. Given that consistent evidence of elevated risks and exposure-response relationships across and within studies are requisites to establish a causal association, the absence of such overall evidence argues against a causal association for AN and lung cancer.