5
Cancer

This chapter reviews the U.S. Environmental Protection Agency (EPA) assessment of the carcinogenicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), commonly referred to as dioxin, other dioxins, and dioxin-like compounds (DLCs), including EPA’s qualitative characterization of their carcinogenicity, the assumption that the dose-response relationship is linear, and the use of animal bioassay and epidemiological data to quantify the dose response. The final section summarizes the committee’s conclusions.1

QUALITATIVE EVALUATION OF CARCINOGENICITY

EPA concludes that dioxin is “carcinogenic to humans” based on the following evidence (Reassessment, Part III, pp. 6-7 to 6-8): evidence from the occupational cohort studies that dioxin exposure increases mortality from cancer aggregated over all sites and from lung cancer “and, perhaps, other sites”; evidence from bioassays of cancer in both sexes of multiple species at multiple sites; and evidence regarding dioxin’s mode of action, including mechanistic evidence that dioxin acts as a tumor promoter via receptor-mediated pathway(s) and the finding that the receptor-mediated pathways that may give rise to cancer in laboratory animals appear to be present and functional in human tissues.

1

The Exposure and Human Health Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds (EPA 2003a, Part I; 2003b, Part II; 2003c, Part III) is collectively referred to as the Reassessment.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment 5 Cancer This chapter reviews the U.S. Environmental Protection Agency (EPA) assessment of the carcinogenicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), commonly referred to as dioxin, other dioxins, and dioxin-like compounds (DLCs), including EPA’s qualitative characterization of their carcinogenicity, the assumption that the dose-response relationship is linear, and the use of animal bioassay and epidemiological data to quantify the dose response. The final section summarizes the committee’s conclusions.1 QUALITATIVE EVALUATION OF CARCINOGENICITY EPA concludes that dioxin is “carcinogenic to humans” based on the following evidence (Reassessment, Part III, pp. 6-7 to 6-8): evidence from the occupational cohort studies that dioxin exposure increases mortality from cancer aggregated over all sites and from lung cancer “and, perhaps, other sites”; evidence from bioassays of cancer in both sexes of multiple species at multiple sites; and evidence regarding dioxin’s mode of action, including mechanistic evidence that dioxin acts as a tumor promoter via receptor-mediated pathway(s) and the finding that the receptor-mediated pathways that may give rise to cancer in laboratory animals appear to be present and functional in human tissues. 1 The Exposure and Human Health Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin (TCDD) and Related Compounds (EPA 2003a, Part I; 2003b, Part II; 2003c, Part III) is collectively referred to as the Reassessment.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment In this chapter, the committee reviews the epidemiological, bioassay, and mode of action evidence and then presents conclusions regarding both qualitative and quantitative measures of carcinogenicity of TCDD, other dioxins, and DLCs. Epidemiological Evidence The epidemiological evidence that provided the basis for EPA’s assessment consists primarily of studies following four cohorts. Of these, the Reassessment reviewed in detail those related to the three cohorts that provided quantitative dose-response estimates linking serum dioxin to cancer mortality (Ott and Zober 1996; Becher et al. 1998; Steenland et al. 2001). The cohorts were quite variable in size and exposure ranges. Ott and Zober (1996) studied a relatively small number of men exposed to an accidental release of dioxin in 1953 (N = 243, 13 cancer deaths). Becher et al. (1998) examined a cohort of 1,189 men employed in pesticide and herbicide production, from which 124 cancer deaths were identified. The third cohort represents a large occupational population originally studied by Fingerhut et al. (1990, 1991), who examined 5,172 male employees in 12 manufacturing facilities. An update on this cohort was provided by Steenland et al. (1999), who applied “job-exposure matrix”2 estimates to 5,132 workers in the original cohort who were followed for 6 more years. The total number of cancer deaths in this cohort was 377. In 2001, Steenland et al. updated this study again on a subcohort of 3,538 workers (with 256 cancer deaths) and used data from 170 members of this cohort for which estimated external exposures and known serum dioxin levels were available to establish a quantitative dose-response assessment. Each study identified a cohort of workers who had been employed in industrial settings in which dioxin was a by-product. These settings included pesticide production (Ott and Zober 1996; Becher et al. 1998) or chemical plants more broadly (Steenland et al. 2001). In each instance, current serum dioxin measurements were available for a subset of workers. Development of exposure estimates for the entire cohort required two extrapolations: from current serum dioxin measurements to historical exposure levels using estimates of serum dioxin half-life, and from workers with current serum dioxin measurements to those without by linking available serum dioxin measurements to job characteristics based on knowledge of the industrial processes. Although these extrapolations decrease the accu- 2 A “job-exposure matrix” refers to an algorithm by which experience in particular jobs are assigned estimated exposure levels. Each job (a row of the matrix) has a corresponding series of exposure levels assigned (columns).

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment racy of the assessment, they were necessary to provide historical exposure estimates so that there would be a sufficient number of cohort members who could be included in the analysis. In addition to these three cohorts, Part II, section 7.5.4 of the Reassessment describes studies that reported on an occupational cohort of 2,310 workers in two plants that prepared and manufactured phenoxy herbicides in the Netherlands (see Reassessment, Part II, Table 7-21 for a summary of all four studies). Bueno de Mesquita et al. (1993) found no statistically significant increases in cancer mortality among all workers (31 deaths, standardized mortality ratio [SMR] = 107, 95% confidence interval [CI] = 73 to 152) and among a subset of 139 workers involved in a 1963 industrial accident (10 deaths, SMR = 137, 95% CI = 66 to 252). Comparing exposed workers (N = 963) to unexposed workers (N = 1,111), both total cancer mortality (rate ratio, [RR] = 1.7, 95% CI = 0.9 to 3.4) and respiratory cancer mortality (RR = 1.7, 95% CI = 0.5 to 6.3) were nonsignificantly increased. A follow-up study by Hooiveld et al. (1996) reported a statistically significant increase in cancer mortality among workers in one of the two plants (SMR = 146, 95% CI = 109 to 192). No such increase was observed in the other factory. Follow-up analysis by Hooiveld et al. (1998) reported a statistically increased incidence of malignant neoplasms among 140 workers involved in the 1963 industrial accident (SMR = 1.7, 95% CI = 1.1 to 2.7). The incidence of malignant neoplasms was also increased in a larger group of 549 workers (SMR = 1.5, 95% CI = 1.1 to 1.9). A comparison of this group of 549 exposed workers to 482 unexposed workers, also from this cohort, yielded an increased total cancer mortality risk (RR = 4.1, 95% CI = 1.8 to 9.0) and an increased respiratory cancer mortality risk (SMR = 7.5, 95% CI = 1.0 to 56.1). There are three major issues to consider regarding EPA’s review of the epidemiological studies investigating the relationship between dioxin exposure and cancer. First, although EPA identified the cohort studies capable of generating quantitative dose-response information for the dose-response modeling and considered the broader epidemiological literature in the background documents, Part III of the Reassessment did not provide a thorough and systematic analysis of the body of epidemiological evidence from which these three studies were chosen. In particular, although Part II described the complete array of studies, including those by Kogevinas et al. (1997) and Bertazzi et al. (1998), the Reassessment did not analyze site-specific tumors consistently across all studies but rather emphasized the positive findings in each paper without a full discussion of consistency, or lack thereof, across studies. A second issue is EPA’s decision to focus on total cancers instead of specific types of cancer. EPA argues that because dioxin is not genotoxic

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment and is instead presumed to act primarily as a promoter rather than an initiator of cancer, the lack of specificity in tumor type is to be expected. If dioxin promotes cancer through the Ah receptor mechanism, however, then an increased tumor incidence would require expression of the receptor in that tissue. The Ah receptor is expressed in most tissues but to varying degrees. It is uncertain whether the level of expression is an important determinant of tumor promotion. There are also many downstream events from ligand-receptor interaction that are tissue specific and essential for tumor promotion to occur via a receptor-mediated response, and these downstream events differ from tissue to tissue (see also discussion on mode of action later in this chapter). In any case, EPA reasons that, in the face of limited power, increased risk of total cancers (which would reflect the increased incidence across the multiple sites affected by dioxin) is easier to detect than an increased risk of individual cancer types (see Part III, pp. 2-9 to 2-10). This rationale would be valid for a given relative risk (e.g., a doubling of the incidence or mortality). However, a given absolute incremental risk (e.g., an additional 10 cancers due to exposure) would be more readily identified for a specific cancer site than for cancers in the aggregate. The more compelling argument for aggregating across cancer types is the practical one that the results for specific cancers are extremely imprecise in these cohorts of modest size. If, in fact, multiple cancer types all showed a small increment in risk of equal magnitude, there would be greater precision and statistical power for the aggregation. For example, in the case of ionizing radiation, the aggregation across a series of radiosensitive cancers, each with small increases in risk, yields a more statistically precise indication of an increase in cancers related to radiation exposure than do any of the individual cancers. In the case of dioxin, it is not clear that a specific set of cancers is affected that can then be aggregated to enhance statistical power. To evaluate the patterns across cancer sites, the committee examined selected papers from the three cohorts (Ott and Zober 1996; Flesch-Janys et al. 1998; Steenland et al. 1999). This evaluation revealed that only limited information is available regarding numbers of cases at specific sites, hence limiting the opportunity to examine consistency across studies. As noted by others, there is some consistency across studies for respiratory cancers, but there is a general lack of concordance for the other cancer sites reported in more than one study. The degree of replication or lack thereof should not be overstated given the small number of studies and imprecise information on specific cancer sites from all but the Steenland et al. (1999) report. Overall, the committee concurs with the value of conducting analyses of total cancers, given the potential for dioxin to affect multiple types of cancer and the limited precision of risk estimates for individual cancer types. Nonetheless, the potential for effects limited to specific types of

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment cancer, as has been found for other causes, also warrants an analysis of major cancer types (e.g., respiratory cancers), the imprecision notwithstanding. Another concern is the potential role of confounding by lifestyle factors such as smoking and by occupational exposures that co-occur with dioxin. Although smoking is a powerful lung carcinogen, quite capable of generating spurious relative risks on the order of those reported in the epidemiological studies for dioxin of around 1.5, the design of those studies makes its potential role as a confounder unlikely in this case. The key comparisons were not between industrial workers and the general population, which is quite susceptible to confounding by lifestyle factors, but among subsets of workers with different levels of estimated dioxin exposure. It is not likely that smoking histories would differ markedly among men located at different jobs within the industrial plant or in relation to duration of employment. In contrast, there is greater potential for confounding by other workplace agents given that the industrial cohorts had exposure to pesticides and potentially carcinogenic chemicals in addition to dioxin. Although these accompanying workplace hazards likely differed for the three cohorts that contributed to the quantitative risk assessment, confounding could have occurred in each to yield a similar falsely elevated measure of association. The difficulties in isolating the health effects of single agents from the complex mixtures encountered in chemical manufacturing must be recognized. Epidemiological evidence for an association between cancer and exposure to DLCs has been characterized as “inadequate but suggestive” (EPA 1987) and “limited” (IARC 1997). ATSDR (2000) concluded that the epidemiological evidence “taken in totality, indicates a potential cancer causing effect for PCBs.” On the whole, it was the committee’s impression that EPA’s narrative in discussing epidemiological studies in Part III of the Reassessment tended to focus on positive findings without fully considering the strengths and limitations of both positive and negative findings. Part III of the Reassessment would be strengthened if EPA clearly identified specific inclusion criteria for those studies for which quantitative risk estimates were determined. Bioassay Data Several large and well-conducted dioxin-related cancer bioassays (Kociba et al. 1978; NTP 1982a,b; NTP 2004) have reported induction of several types of cancer in both rats and mice. The study in hamsters was confounded by use of dioxane, which is a potential carcinogen, as the delivery vehicle. Table 5-1 summarizes these studies. In all studies in which dioxin elicited an increase in tumors, the increase was site specific. With

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment oral administration, the organ most frequently affected was the liver, reflecting the mode of action of carcinogenicity, as discussed below. Of the 21 DLCs of concern, EPA (Part II, p. 6-30) reported that carcinogenicity bioassays have been conducted on only two pure polychlorinated dibenzo-p-dioxin (PCDD) and 1,2,3,7,8-pentachlorodibenzo-p-dioxin (PeCDD), and a mixture of two congeners (1,2,3,6,7,8- and 1,2,3,7,8,9-hexachlorodibenzo-p-dioxin [HxCDD]). Carcinogenicity bioassays have also been conducted on one polychlorinated dibenzofuran (PCDF) (2,3,4,7,8-pentachlorodibenzofuran [PeCDF]) and one PCB (126; 3,3′,4,4′, 5-pentachlorobiphenyl (PeCB) (Table 5-2). However, the ability of a variety of dioxins other than TCDD and DLCs to enhance the carcinogenicity of known carcinogens (promoter assays) has also been reported for PeCDD, HpCDD, 2,3,7,8-tetrachlorodibenzofuran (TeCDF), PeCDF, and 1,2,3,4,7,8-hexachlorodibenzofuran (HxCDF) (summarized by IARC 1997). Bioassays have also been conducted on mixtures of PCBs, and although they provide some information on the carcinogenicity of components, they do not identify the specific responsible chemical(s). Mode of Action Dioxin does not have structural features that would lead to a reactive electrophile, and it is clearly not DNA reactive, as no DNA binding or adducts were found in rodent tissues (Poland and Glover 1979; Randerath et al. 1988; Turteltaub et al. 1990). Absence of DNA reactivity is supported by negative findings in genetic toxicological assays (IARC 1997). Nevertheless, EPA notes (Part II, p. 6-1) the hypothesis that dioxin might be indirectly genotoxic, either through induction of oxidative stress or by altering the DNA damaging potential of some endogenous compounds, including estrogens. No evidence is available for estrogen-mediated DNA damage resulting from dioxin exposure, but oxidative DNA damage has been documented after 30 weeks administration of dioxin (Tritscher et al. 1996; Wyde et al. 2001). Indirect genotoxicity has been postulated to initiate carcinogenicity, but there is insufficient evidence that dioxin has initiating activity. Dioxin was reported to have weak initiating activity in one study (DiGiovanni et al. 1977) in which it was applied to mouse skin prior to a promoting agent. This finding has not been corroborated, and in contrast to what would be expected from an initiating agent, application of dioxin to mouse skin at a dosage greater than that required for a promoting effect did not induce skin tumors (Poland et al. 1982). Moreover, dioxin has not been specifically tested as an initiator in standard models in rat or mouse liver in which chemicals can be evaluated

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment TABLE 5-1 Dioxin Cancer Bioassays Species/Strain Route and Dose Sex Sites of Tumor Increases Reference Rat/Sprague Dawley Oral in feed 1, 10, 100 µg/kg/day Male Oral cavity Kociba et al. 1978 Female Lung, oral cavity, liver NTP 1982a Rat/Osborne Mendel Gastric instillation 10, 50, 500 µg/kg/week for 104 weeks Male Thyroid   Female Liver   Rat/Sprague-Dawley Gastric instillation 3, 10, 22, 46, or 100 mg/kg, 5 days/week for 104 weeks Female Liver, lung, oral cavity, uterus NTP 2005 Mouse/B6C3F1 Gastric instillation 0.01, 0.05, 0.5 mg/kg/wk for 104 weeks (males) Male Liver NTP 1982a   0.04, 0.2, 2.0 mg/kg/wk for 104 weeks (females) Female Liver, thyroid   Mouse/Swiss Webster Topical application 0.005 µg 3 days/week for 104 weeks Female Skin NTP 1982b

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment Mouse/B6C3 and B6C Intraperitoneal injection 1, 30, 60 µg/kg/week for 5 weeks Male Thymus (both), liver (B6C3 only) DellaPorta et al. 1987 Female Thymus (both), liver (B6C3) Mouse/B6C3 Gastric instillation 2.5, 5.9 µg/kg/week for 52 weeks Male Liver DellaPorta et al. 1987 Female Liver Mouse/Swiss Gastric instillation 0.007, 0.7, 7.0 µg/kg/week for 52 weeks Male Liver Toth et al. 1979 Mouse/TG. AC Topical application for 24 weeks Male Skin papillomas Eastin et al. 1998 Female Skin papillomas Mouse/TP53+/− Gastric instillation 250 µg/kg, 1,000 µg/kg twice weekly for 24 weeks Male None Eastin et al. 1998 Female None Hamster/Syrian Golden Intraperitoneal or subcutaneous injection 50 or 100 µg/kg every 4 weeks Male Skin Rao et al. 1988

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment TABLE 5-2 TCDD, Other Dioxins, and DLC Cancer Bioassays Congener Bioassay Dioxins   2,3,7,8-TCDD Rat (M,F)/mouse (M,F) 1,2,3,7,8-PeCDD Rat (M,F)/mouse (M,F)/promoter rat (F) 1,2,3,4,7,8-HxCDD No bioassay conducted 1,2,3,6,7,8-HxCDD Combination study 1,2,3,7,8,9-HxCDD Combination study 1,2,3,4,6,7,8-HpCDD Promoter rat (F) 1,2,3,6,7,8- and 1,2,3,7,8,9-HxCDD mix Rat (M,F)/mouse (M,F) OCDD No bioassay conducted Furans   2,3,7,8-TCDF Promoter mouse (F) 1,2,3,7,8-PeCDF No bioassay conducted 2,3,4,7,8-PeCDF Rat (F)/promoter mouse (F) and rat (M) 1,2,3,4,7,8-HxCDF Promoter mouse (F)/rat (M) 1,2,3,6,7,8-HxCDF No bioassay conducted 1,2,3,7,8,9-HxCDF No bioassay conducted 2,3,4,6,7,8-HxCDF No bioassay conducted 1,2,3,4,6,7,8-HpCDF No bioassay conducted 1,2,3,4,7,8,9-HpCDF No bioassay conducted OCDF No bioassay conducted Non-ortho PCBs   3,3′,4,4′-TCB (77)a No bioassay conducted 3,4,4′,5-TCB (81) No bioassay conducted 3,3′,4,4′,5-PeCB (126) Rat (F) 3,3′,4,4′,5,5′-HxCB (169) No Bioassay Conducted Mono-ortho PCBs   2,3,3′,4,4′-PeCB (105) No Bioassay Conducted 2,3,4,4′,5-PeCB (114) No Bioassay Conducted 2,3′,4,4′,5-PeCB (118) No Bioassay Conducted 2′,3,4,4′,5-PeCB (123) No Bioassay Conducted 2,3,3′,4,4′,5-HxCB (156) No Bioassay Conducted 2,3,3′,4,4′,5′-HxCB (157) No Bioassay Conducted 2,3′,4,4′,5,5′-HxCB (167) No Bioassay Conducted 2,3,3′,4,4′,5,5′-HpCB (189) No Bioassay Conducted aInternational Union of Pure and Applied Chemistry numbers in parentheses. Abbreviations: OCDD, octachlorodibenzo-p-dioxin; OCDF, octachlorodibenzofuran; TCB, -tetrachlorobiphenyl. as initiators followed by administration of a promoting substance (Enzmann et al. 1998). Also, in several chronic bioassay studies in which dioxin was administered to female Sprague-Dawley rats for 30 weeks at dosages associated with an increased incidence of liver tumors in carcinogenicity studies, no increase in hepatic preneoplastic lesions indicative of initiation was

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment found (Lucier et al. 1991; Maronpot et al. 1993). Thus, at present, there is no direct experimental evidence that dioxin acts as an initiator in rat liver. A lack of initiating activity would be consistent with an absence of direct genotoxicity (Williams 1992). Nevertheless, some dose-response modeling of data that show a promoting effect of dioxin on rat liver preneoplastic lesions suggested that dioxin also had “a weak” (Moolgavkar and Luebeck, 1995) or “a slight” (Portier et al. 1996) initiating effect. In contrast, analysis of a two-cell clonal growth model reproduced such data without presuming an effect on mutation rates (that is, initiation) (Conolly and Andersen 1997). Resolution of the question of initiating activity of dioxin awaits experimental evidence. Also, the postulated linkage between potential initiating activity and oxidative DNA damage is not established. In an investigation of the mode of action of hepatocarcinogenicity of pentachlorophenol, oxidative DNA damage was not found to produce liver initiation (Umemura et al. 1999). The committee agrees with EPA that TCDD, other dioxins, and DLCs appear to enhance tumor development in female rat liver via tumor promotion. The promoting activity and liver tumor-enhancing activity of dioxin seem to be mediated through activation of the Ah receptor (aromatic hydrocarbon receptor [AHR]), which in turn leads to a variety of changes in gene expression, including notably induction of cytochromes P450 (CYPs) (Whitlock 1989) and genes related to cell proliferation (Puga et al. 1992) (see Figure 5-1). Whether those gene changes mediate the reported oxidative stress is not known. Nevertheless, both CYP induction and oxidative stress could be involved in liver cytotoxicity, which was found in studies that examined this parameter (Maronpot et al. 1993; Viluksela et al. 2000). Cytotoxicity, in turn, elicits regenerative cell proliferation (Williams and Iatropoulos 2002), as reported in several dioxin studies (Lucier et al. 1991). Dioxin-induced changes in gene expression, however, can occur without enhancement of hepatocelluar proliferation (Fox et al. 1993). In fact, increases in cell proliferation have been documented only after 30 weeks of dioxin administration (Lucier et al. 1991). The enhanced cell proliferation arising from either altered gene expression or cytotoxicity or both could be the principal factor leading to promotion of hepatocellular tumors (Busser and Lutz 1987; Whysner and Williams 1996). The sensitivity of female rat liver to dioxin, which apparently does not extend to the mouse, clearly depends on ovarian hormones (Lucier et al. 1991; Wyde et al. 2001). This sensitivity has been ascribed to induction of estradiol metabolizing enzymes (Graham et al. 1988) and is hypothesized to lead either to generation of reactive metabolites of endogenous estrogen or to active oxygen species of estrogens. Oxidative DNA damage has been implicated in liver tumor promotion (Umemura et al. 1999). In contrast to the extensive work on hepa-

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment FIGURE 5-1 Possible mechanism for TCDD hepatocarcinogenicity. tocellular neoplasia, little is known about the pathogenesis of the bile-duct tumors (Table 5-3). Mechanistic issues are discussed in greater detail below in the context of evaluating whether the dose-response relationship is likely to be linear. In any case, the committee agrees with EPA’s general conclusion that there is sufficient evidence from epidemiological studies, animal bioassays, and mode of action studies to support the qualitative conclusion that TCDD, other dioxins, and DLCs are likely to cause cancer in humans with adequate conditions of dose and duration of exposure. Committee’s Perspective on Whether the Scientific Evidence Supports Classification of Dioxin As a Known Human Carcinogen After extensive discussion of EPA’s revised definition of “carcinogenic to humans” and “likely to be carcinogenic to humans” provided in EPA’s

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment TABLE 5-5 EPA Inputs to CSF Estimates Using Epidemiological Dataa Study Function RR(x)b Central Estimate for bb P Value for bb ED01c LED01c Becher et al. 1998 Power: (1 + kx)b b = 0.326 for 0.026 6 NAd   k = 1.7 × 10−4         Additive: 1 + bx b = 1.6 × 10−5 0.031 18.2 NAd   Multiplicative: ebx b = 8.69 × 10−6 0.043 32.2 NAd Steenland et al. 2001 Power: (x/background)b b = 0.097 0.003 1.38e 0.71e   Piecewise linear: ebx b = 1.5 × 10−5 NAf 18.6 11.5 Ott and Zober 1996 ebx b = 5.03 × 10−6 0.05 50.9 25 aED01 values here represent estimates only for males (as is the case in EPA’s Part III, Table 5-4). Female ED01 values are modestly larger because the background cancer rate for females is less than it is for males. For example, the Ott and Zober (1996) ED01 value for females is 62 ng/kg and the corresponding LED01 is 30.5 ng/kg. bSee Part III, Table 5-2. Note that x is exposure expressed in terms of cumulative lipid burden (CLB), ng of dioxin/kg of fat × years. cSee Part III, Table 5-4. The ED01 values are expressed in terms of lifetime average body burden (LABB), ng of dioxin/kg of body weight. dNot available. EPA did not estimate LED01 values (or the corresponding upper bound for ED01), although these values can be calculated, as described below. eEPA reported these values in Part III, Table 5-3. Note that EPA omitted further consideration of the power function for this dataset, stating that “this formula predicts unreasonably high attributable risks at background dioxin levels in the community due to the steep slope of the power curve formula at very low levels” (Part III, p. 5-37). fNot available. Steenland et al. (2001) did not report the P value for this parameter, although EPA reported a value for LED01. ing upper and lower statistical bounds (such as confidence limits) to inform decision makers” (EPA 2005a, p. 3-17). To illustrate the quantitative impact on the range of uncertainty from this one assumption, the committee considered the upper ED01 (UED01) values that correspond to the lower 95% confidence interval on the dose-response relationship. EPA provides the UED01 values for the Steenland et al. (2001) and Ott and Zober (1996) studies (see Table 5-3 in Part III of the Reassessment). As explained below, the committee has calculated the UED01 values for the Becher et al. (1998) study. Together with the ED01 and LED01 values, the UED01 values help to describe the range of plausible ED01 values and hence the uncertainty that attends the CSF estimates due to finite sampling. To estimate the range of plausible ED01 values, the committee assumes that the set of plausible values for the dose-response relationship parameter

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment (b—see column 3 of Table 5-5 in this report) is normally distributed with a mean equal to the parameter’s central estimate (b—see column 3 of Table 5-5) and a standard deviation equal to the estimate’s standard error (bSE).4 When possible, the committee estimated the value of bSE using the P value for the dose-response relationship parameter, as reported in the Reassessment, Part III, Tables 5-2 and 5-3 (shown in column 4 of Table 5-5 of this report). In particular, bm, bSE, and P satisfy the relationship where N−1 is the inverse cumulative normal function. For example, if bm = 0.1 and P = 0.05, then bSE = 0.051. Designating bUED = bm − 1.96bSE, the value of b yielding the UED01, and bLED = bm + 1.96bSE, the value of b yielding the LED01, the UED01 satisfies the relationship RR(bUED, UED01) = RR(bLED, LED01), where the function RR is the dose-response relationship (rate ratio) taking two arguments (the parameter b and a dose). When the P value is not available but both LED01 and ED01 are specified, the committee assumed that bLED − bm = bm − bUED. If necessary, the value of bLED was estimated from the relationship RR(bLED, LED01) = RR(bm, ED01) and UED01 from the relationship RR(bUED, UED01) = RR(bLED, LED01). In the case of the Becher et al. (1998) study, inserting bm into any of the RR formulas along with the ED01 value for that dose-response function yields an RR of approximately 1.09. It was assumed that the UED01 is the dose that yields an RR of 1.09 when inserted into the dose-response function along with bUED. For example, the Becher et al. power function yields an RR of 1.09 if a LABB of 6 ng/kg (CLB = 1,800 ng/kg-year) is used along with the exponent parameter bm = 0.326. In particular, (1 + 0.00017 × 1,800)0.326 = 1.09. The value of bUED is 0.039 and (1 + 0.00017 × 45,300)0.039 = 1.09. That is, CLB = 45,300 ng/kg-year produces the same RR when used with b = bUED. Dividing CLB by 4 × 75 = 300 yields a LABB of 151 ng/kg. The committee assumed that because the LABB of 151 ng/kg also yields an RR of 1.09, this dose is the UED01. Table 5-6 summarizes the ED01, LED01, and UED01 values for the dose-response relationships listed in Table 5-2 of Part III of the Reassessment. The results in Table 5-6 indicate that the set of plausible ED01 values spans at least one or two orders of magnitude for the Becher et al. (1998) study and the Ott and Zober (1996) study. 4 When estimated using a large number of observations, statistical parameters typically have normal error distributions. Of course, it is possible that the error distributions for the bm parameters are not normal and hence not symmetric.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment TABLE 5-6 ED01, LED01, and UED01 Values     LABB (ng/kg) Study Function RR = ED01 LED01 UED01 Becher et al. 1998 Power: (1 + kx) 6 3 150   Additive: 1 + bx 18.2 9.5 200   Multiplicative: ebx 32.2 16 1,000 Steenland et al. 2001 Power: (x/background) 1.38 0.71 8.95   Piecewise linear: ebx 18.6 11.5 49 Ott and Zober 1996 ebx 50.9 25 Infinite NOTES: Bolded values represent the committee’s estimates of UED01 for the Becher et al. (1998) study. These values were estimated by assuming a normal error distribution for b (see column 3 in Table 5-5 of this report). All other values were as reported by EPA in Table 5-3 of Part III of the Reassessment. Consideration of Alternative Points of Departure EPA explains that while a 10% level is generally used as a POD (that is, an ED10 is generally used to estimate the CSF), “where more sensitive data are available, a lower point for linear extrapolation can be used to improve the assessment (e.g., 1% response for dioxin, ED01)” (Part III, p. 5-15). EPA’s cancer guidelines (EPA 2005a, see also Appendix B) state, “Conventional cancer bioassays, with approximately 50 animals per group, generally can support modeling down to an increased incidence of 1-10%; epidemiologic studies, with larger sample sizes, below 1%” (p. 3-17). However, these generalities do not imply that extrapolation down to low levels is justified in all circumstances. EPA’s carcinogen risk assessment guidelines document explains, “Various models commonly used for carcinogens yield similar estimates of the POD at response levels as low as 1%…. Consequently, response levels at or below 10% can often be used as the POD” (EPA 2005a, p. 3-17). The key point here is that a lower response level is justified only if the estimated dose corresponding to this response is insensitive to the functional form (provided the other functional forms fit the data to a comparable degree). The dose-response functions for the epidemiological data identified by EPA suggest this criterion is not satisfied. For example, as detailed in Table 5-3 of Part III of the Reassessment, the ED01 for males in the Steenland et al. (2001) study is 1.38 ng/kg of body burden if the power function is used, more than an order of magnitude less than the ED01 of 18.6 ng/kg calculated using the piecewise linear function. In the Becher et al. (1998) study, the ED01 spans a factor of five, depending on which dose-response function is used.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment Although EPA states that a 1% response above background (corresponding to RR ≈ 1.09) is within the range of observed response for the three occupational cohort studies considered, it is clearly at the low end of the observed range. For example, among the five exposure groups defined in the Becher et al. (1998) study (excluding the comparison group, for which SMR is fixed at 100), the lowest RR is 1.12. Of the six exposure groups in the Steenland et al. (2001) study (excluding the comparison group), one has an RR value below 1.09 (RR = 1.02 for the second lowest exposure group). RR values for the other five groups were 1.26 or greater. For the Ott and Zober (1996) study, RR values for the three comparison groups were 1.2, 1.4, and 2.0. The use of alternative points of departure for the power dose-response relationships would greatly increase the range of plausible CSF values. Table 8-2 (Reassessment, Part II) demonstrates this point. For the Steenland et al. power function, the 95% confidence interval for ED01 spans approximately one order of magnitude. As a result, the CSF calculated with this set of ED01 estimates likewise spans approximately an order of magnitude. In contrast, the 95% confidence interval for ED05 spans approximately three orders of magnitude. Similarly, the ED01 confidence interval derived from the Becher et al. power function spans a factor of approximately 50. The corresponding range for ED05 spans nearly four orders of magnitude. Thus, it is evident that the choice of POD can have a substantial impact on the uncertainty of the final risk estimate, especially if both upper and lower confidence limits are provided. The importance of this assumption is not readily evident in the Reassessment. The transparency of the uncertainty of CSF calculations, and thus risk estimates, would be substantially improved if the document presented CSF ranges and risk estimates calculated from both ED01 and ED05 values to illustrate the importance of this assumption. Consideration of Alternative Dose-Response Functional Forms Because there are so many functional forms from which to choose for the purpose of modeling dose response, EPA should establish criteria for selecting acceptable solutions. For example, there are formal goodness-of-fit tests that can help to identify the best candidates. Note that a higher statistical significance for a positive dose response does not necessarily imply that, using standard statistical criteria, the model adequately fits the data. Evaluating the goodness of fit for the occupational cohort analyses was complicated by EPA’s lack of ready access to the original data. Despite these complications, it is important that EPA provide a cogent set of criteria for determining which functional forms were used. This section identifies four instances in which EPA eliminated from consideration alternative dose-

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment response functional forms without providing adequate justification. EPA should describe the range of ED01 and ED05 values implied by dose-response functions that are statistically consistent with the occupational cohort data and the inclusion criteria established for this assessment. First, EPA eliminated from consideration the power function dose-response relationship calculated from the Steenland et al. (2001) study, explaining only that this relationship “predicts an unrealistic risk for the background exposure” (Reassessment, Part II, p. 8-67) and that it “leads to unreasonably high risks at low exposure levels, based on calculations of the attributable risk that this model would predict from background dioxin levels in the general population” (Reassessment, Part III, p. 5-13). EPA provided no criteria by which it judged the reasonableness of the Steenland et al. power function, nor does EPA provide any further explanation on this point. The Reassessment should provide further scientific rationale for excluding the Steenland et al. power function, or it should be considered as valid as any of the other dose-response relationships. Second, EPA considered only dose-response relationships based on the assumption of no background incremental risk (that is, SMR = 100 at background exposure levels). This assumption is inconsistent with the findings of two analyses identified by EPA (Starr 2001, 2003; Crump et al. 2003) (Part III, p. 5-14) that rejected the assumption on statistical grounds that SMR = 100 at baseline exposure levels. If relaxing this assumption yields an estimate of SMR > 100 at background exposures, the resulting dose-response relationship would tend to be shallower, yielding smaller CSF values. For example, EPA (Part III, p. 5-15) noted that in a pooled analysis of Ott and Zober (1996), Flesch-Janys et al. (1998), and Steenland et al. (2001), fixing SMR = 100 at background exposure levels yielded ED01 = 51 ng/kg-day, whereas dropping this assumption resulted in ED01 = 91 ng/kg. EPA should provide an explanation for assuming SMR = 100 at background exposure levels. Short of doing so, EPA should consider the impact of relaxing this assumption on the estimated value of the ED01. Third, for the piecewise linear dose-response function developed for the Steenland et al. data set, EPA considered only one cut point (40,000 ng/kg × years) (the cut point, or changing point, is the dose at which the slope of the piecewise linear dose-response relationship changes). Although this is the best-fit cut-point estimate and the only relationship of this form reported by the authors, other cut-point values are plausible. (Other cut points would yield dose-response relationships that could not be statistically rejected.) Finally, EPA considered only a subset of the plausible dose-response relationships that could be fit to the data in the Becher et al. (1998) study. Becher et al. considered a family of dose-response relationships of the form RR = (kx + 1)β, where the value of k is chosen arbitrarily. The best-fit value

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment of β depends on the value of k selected. The relative plausibility of different values of k can be determined by comparing likelihood function values. Finally, holding β = 1 yields a linear function where the value of k is uncertain. Becher et al. reported that k = 0.00017 maximizes the likelihood function and that, for this value of k, the corresponding value of β = 0.326. Holding β = 1 yields k = 0.000016. However, Figure 1 of Becher et al. indicates that the likelihood function is relatively insensitive to the value of k selected. Hence, other dose-response relationships are plausible. Estimating the ED01 values corresponding to these alternative dose-response relationships would require further primary analysis of the data. Uncertainty Associated with Estimation of Historical Exposures The assumed half-life for dioxin in humans plays a major role in the back-extrapolation of dioxin lipid concentrations to the estimation of peak body burdens in occupational cohorts. The Reassessment states, “Using published first-order back-calculation procedures, the relatively small difference (<10-100-fold) in body burden between exposed and controls in the dioxin epidemiology studies makes exposure characterization in the studies a particularly serious issue” (Part III, p. 5-7). The high exposures in the occupational cohorts suggest a high likelihood of enzyme induction during the period of occupational exposure that may have led to a reduction of the half-life to less than the assumed value of 7.1 years. Aylward et al. (2005) discussed the issue of half-lives and the impact of this parameter on risk estimates. EPA’s Reassessment compared the impact of using either a half-life of 4 years or the default of 7.1 years on the back-extrapolation estimate. EPA reported that using a 4-year half-life increases the peak body burden and the area under the curve (AUC) by 4.6-fold and 3.8-fold, respectively. This difference would have increased the estimated ED01 values by the same amount and hence decreased the CSF estimates, resulting in a lower risk estimate. Given the potential importance of this issue, the committee finds the following statement by EPA surprising: “This bounding exercise suggests that impacts on back-calculated peak and AUC values may become significant if the models predict prolonged periods with half-lives of less than 4 years” (Part III, p. 5-8). Because the impact of the half-life used for back-extrapolation depends on the back-extrapolation duration required in any particular study, EPA should have estimated the impact of using the 4-year alternative value for each of the main epidemiological studies separately. EPA should also consider the issues raised by Aylward et al. (2005). Overall, the Reassessment does not provide sufficient quantification of the impacts of these choices, and the committee believes these decisions influence the estimated dose-response relationships.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment Overall Uncertainty Table 5-4 (Reassessment, Part III) summarizes EPA’s all-site cancer ED01 values. For the three occupational cohort studies, these values span less than an order of magnitude (6 to 50.9 ng/kg). The corresponding CSF values range from 5.7 × 10−4 to 5.1 × 10−3 (pg/kg-day)−1. On the basis of this range, EPA concludes that “A slope factor estimate of approximately 1 × 10−3 per pg of dioxin per kg of body weight per day represents EPA’s most current upper bound slope factor for estimating human cancer risk based on human data” (Part III, p. 5-28). The animal-based ED01 values listed in EPA’s Table 5-4 range from 22 to 30.9 ng/kg, leading to the conclusion that “A slope factor of 1.4 × 10−3 per pg dioxin/kg body weight/ day represents EPA’s most current upper bound slope factor for estimating human cancer risk based on animal data” (Part III, p. 5-28). Whereas a CSF of approximately 1 × 10−3 per pg/kg-day (equivalently, an ED01 of approximately 30 ng/kg LABB) lies within the range of plausible values, this discussion has focused on the relative magnitude of the range of plausible values. Consideration of the set of all plausible parameter values (that is parameter values within the 95% confidence interval) for the dose-response functions considered by EPA considerably widened the range of values estimated from the Becher et al. and Ott and Zober studies (see Table 5-6). The CSF values (risk per pg/kg-day) can be calculated by first converting the ED01 expressed as ng/kg LABB to an ED01 expressed as a daily intake (ng/kg-day) using EPA equation 5-1 (Part III, p. 5-18) and then dividing this intake into a risk of 0.01. For the Becher et al. (1998) study, the resulting CSF values range from 3.0 × 10−5 to 1.0 × 10−3 per pg/kg-day, more than two orders of magnitude. The 95% confidence interval for the CSF calculated from the Ott and Zober study (1996) has a lower bound of zero and an upper bound of 1.2 × 10−3. Only the Steenland et al. (2001) study retains an ED01 range (CSF range) with a span confined to less than two orders of magnitude (CSF 6.1 × 10−4 to 3.0 × 10−2). Figure 5-2 compares the range of plausible CSF values identified by EPA with the range of plausible values consistent with the dose-response parameter 95% confidence intervals. Consideration of alternative points of departure can greatly inflate the confidence interval for the power function dose-response relationships. Using an ED05 (rather than an ED01) broadens the confidence interval for the function to more than three orders of magnitude. Consideration of alternative dose-response relationship forms could further broaden the range of plausible CSF values, although primary analysis of the data would be required to quantify the impact. The Reassessment could develop a distribution for the CSF by assigning some probability to different options for each of the assumptions discussed here (McKone and Bogen 1992; Evans et al.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment FIGURE 5-2 Range of plausible CSF values: consideration of parameter confidence intervals only. Solid blocks are central estimate values. For EPA, this value (1,000) corresponds to EPA’s stated central estimate of dioxin’s CSF (1 × 10−3 per pg/kg-day). For the Becher et al. (1998) study, the central estimate in the figure corresponds to the average of the three ED01 values that EPA reports in Part III, Table 5-4. For the Ott and Zober (1996) and Steenland et al. (2001) studies, the central estimate corresponds to the individual ED01 values listed in EPA, Part III, Table 5-4. 1994a,b). In any case, a more thorough consideration of plausible alternative values for key assumptions is needed to portray accurately to risk managers the magnitude of uncertainty that underlies the quantitative risk estimates derived from epidemiological studies. CONCLUSIONS AND RECOMMENDATIONS Qualitative Weight-of-Evidence Carcinogen Classification The committee concluded that the classification of dioxin as “carcinogenic to humans” versus “likely to be carcinogenic to humans” depends greatly on the definition and interpretation of the specific criteria used for classification, with the explicit recognition that the true weight of evidence lies on a continuum with no bright line that easily distinguishes between these two categories. The committee agreed that, although the weight of epidemiological evidence that dioxin is a human carcinogen is not strong, the human data available from occupational cohorts are consistent with a

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment modest positive association between relatively high body burdens of dioxin and increased mortality from all cancers. Positive animal studies and mechanistic data provide additional support for classification of dioxin as a human carcinogen. However, the committee was split on whether the weight of evidence met all the necessary criteria described in the cancer guidelines (EPA 2005a, see also Appendix B) for classification of dioxin as “carcinogenic to humans.” EPA should summarize its rationale for concluding that dioxin satisfies the criteria set out in the most recent cancer guidelines (EPA 2005a, see also Appendix B) for designation as either “carcinogenic to humans” or “likely to be carcinogenic to humans.” The committee agreed that other DLCs are most appropriately classified as “likely to be carcinogenic to humans.” Should EPA continue to classify dioxin as “carcinogenic to humans,” more justification will be required to rationalize why a mixture containing dioxin would not also meet the classification of “carcinogenic to humans.” If EPA continues to designate dioxin as “carcinogenic to humans,” it should explain whether this conclusion reflects a finding that there is a strong association between dioxin exposure and human cancer or between dioxin exposure and a key precursor event of dioxin’s mode of action (presumably AHR binding). If EPA’s finding reflects the latter association, EPA should explain why that end point (e.g., AHR binding) represents a “key precursor event.” The committee considers the distinction between these two categories to be based more on semantics than on science and recommends that EPA spend its energies and resources more carefully delineating the assumptions used in quantitative risk estimates for TCDD, other dioxins, and DLCs derived from human and animal studies. Quantitative Risk Estimation of Cancer Potency The committee concludes that there is an adequate scientific basis to support the hypothesis that the shape of the relationship between dioxin dose and cancer risk is sublinear at low doses, perhaps reflecting responses indistinguishable from background risk at doses below which dose-response data are available, including evidence that (1) TCDD, other dioxins, and DLCs are not genotoxic; (2) dioxin acts through receptor mediation, and receptor-mediated carcinogens tend to exhibit sublinear dose-response relationships; (3) dioxin-induced liver tumors are secondary to hepatotoxicity and enhanced rates of cell proliferation; (4) bioassay results suggest sublinearity (Hill coefficient central estimates substantially greater than 1); and (5) epidemiological results do not help to distinguish between zero and nonzero responses at the low-dose end of the dose-response curve. Accordingly, a risk assessment can be conducted without resorting to default assumptions.

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment To the extent that EPA favors using default assumptions for regulating dioxin as though it were a linear carcinogen, such a conclusion should be supported with scientific evidence. (For example, EPA could explore whether background exposures raise the population to the linear portion of the dose-response relationship.) Alternatively, the decision to use the linear dose-response relationship could be made as a part of risk management, although the risk assessment should provide the scientific strengths and weaknesses for both linear and nonlinear approaches. EPA should adhere to the division between risk assessment, which is a scientific activity, and risk management, which takes into account other considerations, as described by the National Academy of Sciences more than two decades ago (NRC 1983). EPA has not adequately justified use of the 1% excess risk level as the POD for the analysis of either the epidemiological or animal bioassay data. Although demonstrating that the POD is within the range of the data is necessary, it is not sufficient to justify its use. Other conditions, such as demonstrating that the POD is relatively insensitive to functional form (as noted in EPA’s cancer guidelines), must also be satisfied. EPA should acknowledge the larger extrapolation from justifiable POD values down to environmentally relevant doses that would be necessitated by use of a higher-response-level POD. Regarding EPA’s review of the animal bioassay data, the committee recommends that EPA establish clear criteria for including different data sets. The reliance on one site from one gender of one species, as reported by a single study, does not adequately represent the full range of data available. The committee recommends that EPA consider the full range of data, including the new NTP animal bioassay study on TCDD for quantitative dose-response assessment. Characterization of Uncertainty Surrounding Cancer Risk Estimates EPA should characterize more completely the uncertainty associated with risk estimates inferred from the epidemiological data by (1) taking into account the full range of EDxx values statistically consistent with the data (not just the central and lower estimates), (2) considering alternative PODs, (3) considering alternative dose-response functional forms consistent with the data, and (4) considering uncertainty associated with the half-life estimates of dioxin in humans for the purpose of back-extrapolating exposures in occupational cohort studies. The committee recognizes that explicit characterization of uncertainty could result in an especially wide range of risk estimates. Narrowing consideration to a subset of those estimates could be made as part of the risk management task. For example, a “health protective” (conservative)

OCR for page 108
Health Risks from Dioxin and Related Compounds: Evaluation of the EPA Reassessment estimate could be used to support an imperative to protect public health. Alternatively, if the goal is to compare that risk with other risk management priorities, countervailing risks, or the economic costs of risk mitigation, a central or arithmetic mean value could be used. Finally, to address uncertainty associated with specification of the dose-response relationship functional form below the POD (that is, linear vs. nonlinear), EPA could choose to use a margin of exposure approach in place of estimating population risk. These options are the purview of risk management rather than risk assessment. On the whole, it was the committee’s impression that EPA’s narrative in discussing epidemiological studies in Part III of the Reassessment tended to focus on positive findings without fully considering the strengths and limitations of both positive and negative findings. Part III of the Reassessment would be strengthened if EPA clearly identified specific inclusion criteria for those studies for which quantitative risk estimates were determined.