We compared the current evidence of cancer risk among individuals exposed to TCDD with the results available at the time of the 1997 IARC Monograph. Since 1997, the strongest evidence for a carcinogenic effect comes from the exposure-response reanalysis of all-cancer mortality among herbicide manufacturing cohort, in which an association was apparent at high doses, or when lagging of exposure was applied (Steenland et al., 1999, 2001
). Supportive evidence comes also from the updated mortality follow-up of the Seveso population (Consonni et al., 2008
). Nevertheless, other results published since 1997—including the main SMR analysis of the US herbicide manufacturing cohort (Steenland et al., 1999
), the updated mortality analysis of one of the largest plants included in the US cohort (Collins et al., 2008a, 2008b), the cancer incidence follow-up of the Seveso population (Pesatori et al., 2009
), the update of the Dutch cohort of herbicide manufacturers (Boers et al., 2010
), and the studies of Vietnam veterans (Akhtar et al., 2004
; Cypel and Kang, 2010
)—do not support an association between TCDD exposure and cancer risk.
Among the four studies with highest TCDD exposure, given the greatest weight in the 1997 IARC review (IARC, 1997
), updated results were reported for the US multicenter cohort (Steenland et al., 1999
) and the Dutch cohort (Boers et al., 2010
). Whereas the US cohort added evidence in favor of an association between TCDD exposure and cancer risk, the Dutch cohort added evidence against it. Although the hypothesis that TCDD is a human carcinogen is plausible based on experimental evidence, in our opinion the weak and contradictory evidence from epidemiologic studies does not support a causal association.
With respect to results for individual neoplasms, the updated mortality analysis of the Seveso population suggested an increased risk of lung cancer (Consonni et al., 2008
), which was not confirmed in the update of any other study. As for NHL, the updates of the mortality analyses of the Seveso population (Consonni et al., 2008
) and New Zealand herbicide sprayers suggested an increased risk, which was not confirmed in the updates of the US multicenter study (Steenland et al., 1999
), the Dutch study (Boers et al., 2010
), or the Ranch Hand study (Akhtar et al., 2004
). An increased mortality from STS was suggested in the updated mortality analysis of the Midland cohort (Collins et al., 2008a
), but not in the analysis of the larger US multicenter cohort (Steenland et al., 1999
) or in the other studies.
Because human and experimental studies on TCDD and cancer risk have continued for more than 40 years, this area of research is heavily charged with political and emotional issues. In such circumstances, specific types of bias might occur, in addition to those affecting epidemiologic studies in general. These include publication, diagnostic and reporting biases. Selective reporting of “positive” results is plausible because of the great concern that TCDD is an environmental carcinogen. As has been shown (Boffetta et al., 2008
), early studies on NHL risk might have been subject to publication bias. To assess publication bias by identifying unpublished studies is difficult. In the case of TCDD and cancer, there is at least one such example: a case-control study of STS and NHL conducted in the 1990s in Ho Chi Min City, Vietnam, on the possible association with Agent Orange exposure during the Vietnam War (Kramarova et al., 1998
). To the best of our knowledge no results have been reported. Furthermore, apart from the findings on breast cancer mentioned above (Engel et al., 2005
), and a brief mention of null results in a study of prostate cancer (Alavanja et al., 2003
), to our knowledge no results on TCDD-contaminated pesticides have been reported from the AHS, the most extensive study of agricultural exposures and cancer conducted to date.
Bias might also arise because several of the cohorts have been subject to intensive medical surveillance, possibly resulting in overdiagnosis as compared to the unexposed populations used for comparison. This potential form of bias would be particularly relevant for neoplasms whose detection is highly dependent on diagnostic intensity (e.g., melanoma, and thyroid and prostate cancer [Adami, 2010]), and for neoplasms suspected to be linked to dioxin exposure, which are subject to diagnostic uncertainties, such as STS and NHL, as has been suggested for pleural mesothelioma in cohorts exposed to asbestos (Siemiatycki and Boffetta, 1998
). However, no direct evidence is available in favor of, or against, this form of bias.
Two of the most informative studies (US herbicide manufacturers and Seveso residents) suggest an association with all cancer only after a latency of 15 or 20 years (Steenland et al., 1999
; Consonni et al., 2008
; Pesatori et al., 2009
). The authors interpret these studies as supporting a causal association between TCDD and cancer risk. TCDD exerts its carcinogenic effect in experimental systems through the Ah receptor, which activates cell proliferation (Safe, 2001
). This mechanism has been invoked to explain an unspecific action on risk of all cancer (Mandal, 2005
; Walker, 2007
). However, a long latency between exposure and cancer is a hallmark of agents acting through DNA damage. Cheng and colleagues (2006)
have proposed to solve the apparent inconsistency by assuming that cell proliferating activity of TCDD should be sustained for a long follow-up. Hence, individuals with less than 10 or 15 years since first exposure would not have acquired a sufficiently long exposure. However, known carcinogens acting via other late-stage mechanisms (e.g., hormones) do not support this notion.
A more fundamental challenge in the evaluation of TCDD as a human carcinogen is the emphasis on risk of all cancer rather than specific neoplasms or a specific subset of neoplasms. As already explained in a previous review (Cole et al., 2003
), we have not found convincing evidence for a central role of the Ah receptor in dioxin-related carcinogenesis. Hence, the ubiquitous presence of the Ah receptor should not guide the interpretation of the epidemiologic evidence we have summarized above. Indeed, a non-organ-specific carcinogenicity of TCDD would represent a unique feature in cancer epidemiology. Known nongenotoxic carcinogenic agents, including those acting through pathways present in all organs and tissues (e.g., overweight/obesity [World Cancer Research Fund, 2007
]), target only one or a few specific organs.
An increase in all-cancer risk has not been clearly shown in animal carcinogenicity tests. The reader is referred to previous reviews (IARC, 1997
; US EPA, 2003
; Knerr, 2006
) and results of recent studies (NTP, 2006
). TCDD causes specific tumors in mice (hepatocellular carcinoma, skin tumors, and lymphoma in multiple studies, as well as thyroid and lung tumors in single studies), rats (hepatocellular carcinoma, cholangiocarcinoma, and lung and oral cavity tumors in multiple studies, as well as tumors of the thyroid, skin, and uterus in single studies), and hamster (skin tumors in multiple studies). TCDD is therefore classified as an experimental carcinogen (IARC, 1997
; US EPA, 2003
; Baan et al., 2009
); however, these evaluations are based on the evidence of increased incidence of groups of specific tumors (and in particular those commonly occurring in mice and rats), and not on all-cancer incidence in experimental animals.
Consistency of results is a key criterion for assessing causality. Shifting the emphasis from specific cancers to all cancer allows for the possibility that different types of cancer-specific biases affecting various studies (e.g., residual confounding by different risk factors) would generate an apparently consistent increase in all-cancer risk. Protection from bias should therefore be subject to more stringent scrutiny when evaluating an epidemiological hypothesis that some risk factor causes all cancers.
In addition, the hypothesis of an increase in all-cancer risk would not dispense with the requirement that cancer-specific results have to be consistent across studies, especially in the case of more common cancers, such as lung cancer, for which random fluctuation can hardly be invoked as cause for the lack of consistency. For none of the specific neoplasms is there a consistent pattern showing an increased risk in populations exposed to TCDD.
In some of the epidemiologic studies, an exposure-response is suggested, in the absence of an overall increase in cancer risk. In these circumstances, the apparent trend arises from a comparison of exposure categories with SMRs distributed below and above the null value of 1.0. One example is lung cancer mortality in the cohort of US chemical workers: the overall SMR was 1.06 (95% CI 0.88,1.26), and the p
value of test for trend in an internal analysis by cumulative exposure score (log-transformed and lagged 15 years) was .03 (Steenland et al., 1999
). In this analysis, the SMR of three of the seven exposure categories was below 1.0.
A further problem in the interpretation of the epidemiologic studies of TCDD exposure and cancer risk arises from the fact that mortality was used as the outcome measure in several of them. In addition to possible differential misclassification of causes of death between exposed and unexposed subjects, there are other issues, including confounding by different quality of treatment among exposed and unexposed due to different access to health care. Another problem in population-based studies such as the cohort of Seveso residents is that mortality analysis is not incidence based. Hence, there is a contribution to the mortality from cases that were already prevalent at the time of exposure. Conversely, when occupational cohorts are compared with the general population, a healthy worker effect is quite likely due to a lower prevalence of existing cancer among those who are occupationally active.
In conclusion, the carcinogenicity of TCDD may be plausible on the basis of animal experiments conducted at high doses, but the epidemiological evidence falls far short from conclusively demonstrating such a relationship in humans. In the case of complex data such as the epidemiologic studies on TCDD exposure and cancer risk, it is important to consider all the evidence, and not just selected components that might support one particular hypothesis. The use of mechanistic data to link experimental and human results is justified when there is a specific and strong correspondence between the different systems such as presence of the same DNA damages and mutations in the same cancer-related genes in both human tumors, experimental animals and in vitro systems, as in the case of benzo[a]pyrene, alterations in TP53
gene, and lung cancer (IARC, 2010
). The exercise becomes questionable, however, when weaker data are used and inconsistencies are ignored. Furthermore, TCDD is a likely example of how epidemiologic studies in a controversial area might be particularly susceptible to multiple types of bias. These considerations support our conclusion that the epidemiological evidence of carcinogenicity of TCDD in humans is not “sufficient” and remains “limited.”