Significant Shortcomings of the U.S. Environmental Protection Agency's Latest Draft Risk Characterization for Dioxin-Like Compounds

Thomas B. Starr,1

TBS Associates, 7500 Rainwater Road, Raleigh, North Carolina 27615

Received June 29, 2001; accepted August 3, 2001


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 DISCUSSION
 REFERENCES
 
The United States Environmental Protection Agency (U.S. EPA) has concluded that 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) is a human carcinogen, and it has stated that the lifetime all-cancer mortality risk attributable solely to the current background body burden of dioxin-like compounds could be as high as 1.3 per 100. U.S. EPA's most current human cancer risk estimate was obtained from a linear dose-response model fit to the data from three epidemiology studies of TCDD-exposed chemical workers. Herein it is shown that the U.S. EPA model fails to provide an adequate fit to that data, whereas an intercept-only model, having no slope whatsoever, is entirely adequate. Although the epidemiology data used by U.S. EPA are consistent with a significant elevation in all-cancer mortality, by about 32%, among TCDD-exposed workers, this elevation should not be attributed to the workers' TCDD exposure.

Key Words: TCDD; U.S. EPA; dioxin; carcinogen; risk assessment.


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 DISCUSSION
 REFERENCES
 
In the most recent U.S. Environmental Protection Agency (U.S. EPA) draft risk assessment documents for dioxin-like compounds (U.S. EPA, 2000b), the agency has characterized 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) as a human carcinogen. Further, U.S. EPA has quantified the lifetime average body burden of TCDD (or a mixture of dioxin-like compounds having the same toxic equivalence [TEQ]) in terms of 1% effective dose (ED01) estimates, i.e., doses that would be expected to increase the lifetime risk of death from all-sites cancer, i.e., all-cancer mortality, by 1%.

U.S. EPA generated both central (ED01) and lower-bound (LED01) estimates using human data from three epidemiology studies of exposed workers (see Table 5–4, pp 148–149 in U.S. EPA, 2000b): a study of a cohort of 1,189 chemical industry workers from Hamburg, Germany (Flesch-Janys et al., 1998Go), a study by the National Institute for Occupational Safety and Health (NIOSH) of a similar cohort of 5,172 workers from 12 U.S. chemical plants (Aylward et al., 1996Go; Fingerhut et al., 1991Go), and a study of a cohort of 243 workers involved in the cleanup and related activities following an uncontrolled release on November 17, 1953, of TCDD from an autoclave being used for trichlorophenol production at a BASF AG plant in Ludwigshafen, Germany (Ott and Zober, 1996Go). Although U.S. EPA has developed another set of estimates from the Kociba et al. (1978) laboratory feeding study of TCDD using Sprague-Dawley rats, the focus herein is on the estimates that the agency obtained from the three above-mentioned epidemiology studies.

U.S. EPA's epidemiology-based ED01 and LED01 estimates were derived with Poisson regression and a linear dose-response model having an intercept constrained to equal unity (U.S. EPA 2000b):

where j is an index of exposure group, Pj is the model-predicted number of cancer deaths, Ej is the number of cancer deaths expected absent any TCDD-related effect, as derived from the mortality experience of an appropriate reference population, Dj is the lifetime average TCDD body burden (ng TCDD/kg body weight), as estimated separately by U.S. EPA, and ß1 is the estimated potency or slope factor per unit lifetime average body burden.

This linear dose-response model, a special case of the more general linear model that has a floating, or variable, intercept (Crump and Allen, 1985Go) was fit separately to the data from each of the three epidemiology studies. In addition, it was fit to a combined data set with observations from all three studies taken together. It is this latter analysis, described by U.S. EPA as a meta-analysis, that yielded the agency's most current upper bound slope factor for estimating human cancer risk based on human data (U.S. EPA 2000b). It is noteworthy that all of U.S. EPA's epidemiology-based ED01 and LED01 estimates exceed the current U.S. background body burden of dioxin-like compounds, estimated by U.S. EPA to be approximately 5 ng TEQ/kg body weight, of which only roughly 10% is attributable to TCDD (U.S. EPA, 2000b), by no more than a factor of about 16. In fact, one estimate, the LED01 for the Flesch-Janys study (1998), is 3.5 ng/kg, which is smaller than the U.S. EPA's current background body burden estimate by a 1.4-fold factor. Thus, U.S. EPA estimates that all-cancer mortality in the United States is increased by anywhere from about 6 per 10,000 to as much as 1.3 per 100 due solely to the current background body burden of dioxin-like compounds.

Table 1Go presents the standardized mortality ratios (SMRs, ratios of observed mortality to that expected absent dose-related effects) for all-site cancers for each of the 12 study/exposure subgroups from the three epidemiology studies, along with the corresponding U.S. EPA-estimated lifetime average TCDD body burdens. For ease of visualization, these data are also depicted in graphical form in Figure 1Go. The range of estimated TCDD body burdens across these exposure subgroups is exceptionally wide, spanning more than three orders of magnitude, from 1.4 ng/kg, about 3-fold higher than U.S. EPA's estimate of the current U.S. background body burden of TCDD alone (~ 0.5 ng/kg [U.S. EPA, 2000b]) up to 2012 ng/kg, and a logarithmic scale was used on the horizontal axis of Figure 1Go to accommodate this large range. Because the range of estimated exposures in the worker cohorts is so great, the combined data set from the three epidemiology studies provides an unusually stringent test of the agency's linear dose-response model, as the predicted incremental risks above background must also range linearly over the same 1400-fold range.


View this table:
[in this window]
[in a new window]
 
TABLE 1 All-Cancer Mortality Standardized Mortality Ratios (SMRs), 95% Confidence Limits, and Fixed and Random Effects Weights Assigned to Each Study/Exposure Group
 


View larger version (22K):
[in this window]
[in a new window]
 
FIG. 1. Standardized mortality ratios (SMRs) and 95% confidence intervals for all-cancer mortality versus estimated lifetime average TCDD body burden for the twelve study/exposure subgroups employed in the initial meta-analysis and subsequent dose-response analyses.

 
When the U.S. EPA Science Advisory Board Dioxin Reassessment Review Subcommittee (DRRS) met to review the agency's newest draft risk characterization (U.S. EPA 2000b) on November 1–2, 2000, a number of criticisms were voiced by various DRRS members and public commenters (U.S. EPA, 2001). Notable among these were concerns that (1) the scientific evidence did not support U.S. EPA's conclusion that TCDD is a human carcinogen; (2) the additional lifetime human cancer mortality risks that U.S. EPA was attributing to current background body burdens of dioxin-like compounds were large and simply not credible; and (3) U.S. EPA had failed to provide sufficient quantitative information regarding the uncertainty inherent in its quantitative risk estimates.

All three of these concerns are addressed in this commentary. First, a formal meta-analysis, including an explicit assessment of heterogeneity among the various study/exposure groups, has been undertaken to assess whether it is appropriate to combine the data from the three epidemiology studies into a single dose-response analysis, as was done by U.S. EPA. Second, the ability of U.S. EPA's linear dose-response model to provide an adequate description of the data with which its slope factor was estimated was assessed using a standard {chi}2goodness-of-fit criterion, and plausible alternative models, both linear and nonlinear, were explored to ascertain whether the fit of U.S. EPA's linear model to the data could be improved upon materially. Third, a sensitivity analysis was undertaken in which the influence of individual data points on the derived ED01 and LED01 estimates was quantified.

Together, these three aspects of a more thorough and comprehensive dose-response assessment exemplify how uncertainty in estimated risks arising from both model and parameter uncertainty can and should be effectively characterized. In addition, the conclusions that follow from the results of this dose-response assessment bear heavily on the question of whether it is appropriate to describe TCDD as a human carcinogen.

Initial Meta-Analysis
A meta-analysis of Poisson distributed data from multiple epidemiologic studies can be thought of as analogous to an analysis of variance (ANOVA) of normally distributed data taken from multiple experimental studies, although the fact that epidemiologic studies are observational rather than experimental in nature can cause significantly greater difficulties (Blair et al., 1995Go; Stroup et al., 2000Go). The lack of control in observational studies of multiple extraneous factors has the potential to undercut any validity of risk estimates generated through the pooling of results from multiple studies.

It is worth commenting in this regard that the combination of data from multiple experimental toxicology studies into a single quantitative analysis of effects is rarely undertaken due to target organ, sex, species, supplier, investigator, laboratory, and experimental protocol differences that are known to potentially confound any comparisons between treatment groups from different experimental studies. There are thus many reasons why it may be inappropriate to develop a pooled estimate of risk from multiple observational studies using meta-analytic methodology (Colditz et al., 1995Go; Greenland, 1994Go).

At the very minimum, it is essential to assess quantitatively the heterogeneity of the study/exposure subgroup-specific risk estimates before giving serious consideration to the development of a pooled risk estimate. It is surprising that U.S. EPA apparently neglected to undertake such an initial heterogeneity assessment. Fortunately, the heterogeneity assessment undertaken herein has revealed a remarkable homogeneity of the different study/exposure subgroup-specific risk estimates, thus providing strong quantitative support for the development of pooled risk estimates.

A fixed and random effects meta-analysis was conducted on the data presented in Table 1Go using the Stata meta command (Stata Release 6, Stata Corporation, College Station TX). Weights assigned by the meta command procedure to each exposure group SMR estimate during the fixed and random effects analyses are also presented in Table 1Go. Results are summarized in Table 2Go. Especially noteworthy is the absence of any significant heterogeneity among the exposure groups, as indicated by the large p-value (0.355) for the heterogeneity test. This is quite surprising, as (1) U.S. EPA has asserted that there is a causal association between TCDD body burden and all-cancer mortality; and (2) as was previously noted, there is a marked difference in the TCDD body burdens that range over more than three orders of magnitude across the exposure groups.


View this table:
[in this window]
[in a new window]
 
TABLE 2 Summary of Meta-Analysis Results: All-cancer Mortality
 
If a causal association between TCDD body burden and all-cancer mortality were truly present, one would expect it to manifest in this initial meta-analysis as between group variability above and beyond simple sampling variability, i.e., heterogeneity, because no adjustment for the large gradient in TCDD exposure across the exposure groups has been undertaken. This is a clear indication that any association between all-cancer mortality and TCDD body burden in these study groups is so weak as to be undetectable in the meta-analysis. The extremely small moment-based estimate of between-studies variance (0.004), consistent with a random effects standard deviation of just 0.06, confirms the remarkable homogeneity of these study/exposure group SMR estimates, despite their marked TCDD exposure differences.

In essence, the only significant source of variability in these data is the sampling variability associated with each exposure group. Each group-specific SMR estimate can thus be characterized as having arisen by random sampling from a single underlying population with a common SMR. Thus, there is no good reason not to combine the data from these three epidemiology studies in a single dose-response analysis.

The pooled SMR estimates of 1.347 (fixed) and 1.343 (random) are virtually identical and significantly greater than unity, indicating that the study/exposure subgroups as a whole have significantly higher all-cancer mortality than their respective comparison populations. As will be demonstrated in the next section, this elevation is not associated with TCDD exposure, so it is most likely attributable to the presence of significant cancer risk factors other than TCDD exposure in these cohorts. The facts that these cohorts had exposure to a number of carcinogens (4-aminobiphenyl, asbestos, tobacco products, for example) in the workplace and possibly elsewhere, and that adjustments for these and other potential risk factors such as lifestyle have not been entirely adequate in any of these studies (Adami et al., unpublished report) need to be kept clearly in mind when interpreting this finding.

TCDD Exposure Modeling
Two of the epidemiologic studies (Flesch-Janys et al., 1998Go; Ott and Zober, 1996Go) relied upon by U.S. EPA only provided exposure ranges for each of their four respective exposure subgroups. To estimate an average exposure level for each of these subgroups, U.S. EPA fit separate lognormal cumulative distribution models to the two studies' subgroup exposure cutpoints, then used the fitted models to calculate an arithmetic average exposure level for each subgroup. Although U.S. EPA presented the lognormal distributions' parameter estimates (a geometric mean and geometric standard deviation) in their latest risk characterization document (U.S. EPA, 2000b), no characterization of parameter uncertainty (e.g., confidence intervals) or model uncertainty (e.g., goodness-of-fit statistics) was provided.

When the U.S. EPA model fits were checked, it was discovered that the agency's model for the BASF study was altogether inadequate (p ~ 0.34 x 10–13). Furthermore, the agency's lognormal model for the Hamburg cohort's cumulative exposure distribution was also inadequate (p ~ 0.02); however, U.S. EPA's parameter estimates for this cohort's exposure distribution were not maximum likelihood estimates. When maximum likelihood parameter estimates were identified and used, the resulting lognormal distribution did provide an adequate fit to the Hamburg study's exposure data (p ~ 0.71).

Correction of the comparatively small errors in estimated TCDD body burdens that arose from U.S. EPA's inadequate exposure models would not be expected to materially alter any of the conclusions reached herein. These additional errors do, however, prompt one to question the scientific rigor of the agency's risk assessment process.

Dose-Response Modeling
It should be noted that the initial meta-analysis with no adjustment for potential exposure differences provides an omnibus test against unspecified alternatives to the null hypothesis of no subgroup differences that lacks the sensitivity of specific dose-response regression models in detecting exposure-related trends. Thus it should be regarded only as a coarse screening tool, to be supplemented with more specific dose-response modeling, provided no issues are uncovered that would call such modeling of the combined data set into question. As was noted above, the meta-analysis revealed no reason not to combine the data in a single dose-response analysis.

Table 3Go presents results from conducting Poisson regressions of various dose-response models on the study data presented in Table 1Go and depicted graphically in Figure 1Go. The first model to be considered was U.S. EPA's linear model. Note that the standard {chi}2 goodness-of-fit test revealed a highly significant lack of fit (p = 0.0003). Thus, U.S. EPA's linear model with unit intercept provides an altogether inadequate fit to the epidemiologic data. Ironically, this model systematically underpredicts the observed SMRs in the low-exposure subgroups and systematically overpredicts the observed SMRs in the high-exposure subgroups (data not shown).


View this table:
[in this window]
[in a new window]
 
TABLE 3 Goodness-of-Fit Statistics and Estimated Points of Departure for Various Dose-Response Models of All-Cancer Mortality Data from Three Epidemiologic Studies in Relation to TCDD Body Burden
 
Given this marked discrepancy between epidemiologic observations and the U.S. EPA's linear model predictions, it is difficult to justify use of this model to estimate points of departure such as ED01 and LED01, or to conduct extrapolations to other exposure situations. How can a model be trusted to perform well outside the data range in which it was estimated if it performs so poorly within that range? The U.S. EPA linear dose-response model is simply not credible.

Addition of a quadratic term to U.S. EPA's linear model made no material difference in its performance, as is indicated in the second row of Table 3Go. It still provided an inadequate fit to the epidemiologic data (p = 0.0017). It is only when U.S. EPA's constraint of a unit intercept was relaxed that a linear model provided an adequate fit to these data. With a floating intercept, a linear dose-response model provided a perfectly adequate description of the observations (p = 0.313), and addition of a quadratic term to this floating intercept linear model did not materially alter its performance (p = 0.237); even more interesting is the fact that an intercept-only model was also perfectly adequate (p = 0.310).

In the intercept-only model, there is no dependence whatsoever of cancer mortality risk on TCDD exposure. Instead, this model predicts that all of the exposure groups have the same, i.e., constant, elevation of all-cancer mortality, by about 32%, relative to their respective comparison populations. This model prediction is consistent with the presence of significant cancer risk factors other than TCDD exposure in these workplaces, as was noted earlier. Thus, the meta-analysis and the dose-response analyses both indicate that TCDD exposure is not related to all-cancer mortality. Furthermore, the scientific principle of Occam's razor suggests that no model more complicated than the perfectly adequate intercept-only model should be considered plausible.

Sensitivity Analysis
The remaining twelve rows of Table 3Go present results for the floating intercept linear model with individual data points dropped out of the Poisson regressions. The intent of this exercise was to gain insight regarding which, if any, data points individually had an exceptionally large influence on the modeling results as expressed in the associated ED01 and LED01 estimates. Elimination of the highest data point from the BASF cohort (2012 ng/kg) had a fairly strong impact on the estimated ED01, increasing it by a 1.8-fold factor. This is not surprising, as the all-cancer mortality SMR of 2.0 for this subgroup was higher than that for any other study group. Its exclusion left the evidence in support of a linear relationship between all-cancer mortality and TCDD exposure weaker still.

Next in line is the highest data point from the NIOSH study (554.5 ng/kg), whose elimination served to reduce the estimated ED01 by 44%. This result can be understood by noting that the SMR for this group (1.15) was quite a bit lower than one might expect, given their high TCDD body burden, if there were a causal linear association between TCDD exposure and all-cancer mortality. Similarly, elimination of the lowest NIOSH study data point (27.8 ng/kg with an SMR of 1.02) served to increase the estimated ED01 by 53%, presumably because its inclusion forced the predicted SMR in the low end of the exposure range closer to unity than did the other data points in the same range. Its exclusion allowed the model intercept to increase, while the model slope decreased to compensate. Inspection of the remainder of Table 3Go reveals little more; the exclusion of any of the other data points one at a time made little difference in the estimated ED01 or LED01.

Impact of Model Selection on Estimated Ed01 and Led01 Estimates
Clearly, use of an adequate dose-response model rather than U.S. EPA's inadequate model had a dramatic impact on the estimated ED01. The floating intercept model's ED01 is more than 3-fold higher than U.S. EPA's fixed intercept linear model. Even the lower bound on the ED01, which is known to be particularly insensitive to changes in model specification, is raised by 74% (49 ng/kg) relative to that for U.S. EPA's inadequate model (28.1 ng/kg). Furthermore, if one were to hypothesize that the putative human carcinogenicity of TCDD were strictly a high-dose (i.e., thresholded) phenomenon, then dropping the highest BASF data point from consideration could be justified on biological grounds. If TCDD is in fact a high-dose-only carcinogen, then U.S. EPA's ED01 estimate of 47.1 ng/kg would be increased by more than 5.5-fold to 261 ng/kg if an adequate floating intercept model of only the data below 2012 ng/kg were employed. Finally, it is worth noting that all of the adequate alternative models and truncated data sets that were considered herein yield ED01 and LED01 estimates that are considerably higher than those arising from U.S. EPA's inadequate linear model. This justifies the conclusion that U.S. EPA's ED01 and LED01 estimates are unrealistically low, i.e., U.S. EPA's estimated cancer risks are excessively conservative.


    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 DISCUSSION
 REFERENCES
 
The most remarkable conclusion that can be drawn from the initial meta-analysis and follow-up dose-response analyses described herein is that the epidemiologic data employed by U.S. EPA to develop human cancer risk estimates are incompatible with the agency's linear dose-response model, yet these data are entirely consistent with an intercept-only model, a model that has no slope component whatsoever in relation to estimated TCDD body burden. This conclusion provides strong quantitative support to previous assessments of the epidemiologic evidence as at most limited, and not sufficient, with regard to the potential human carcinogenicity of TCDD (Adami, H.-O., Cole, P., Mandel, J., Pastides, H., Starr, T. B., and Trichopoulos, D., unpublished report dated August 7, 2000, Exponent, Menlo Park, CA; IARC, 1997Go).

Dose response is but one of the so-called Bradford Hill criteria by which the causality of observed associations between exposure and disease can be judged. Others include consistency, specificity, strength of association, coherence, and temporality (Hill, 1966Go). The data from the three epidemiology studies employed by U.S. EPA in estimating potential human cancer risks from exposure to dioxin-like compounds are, in fact, remarkably consistent. Indeed, the initial meta-analysis undertaken herein indicated that sampling variability alone was sufficient to account for the modest differences that are observed among the SMRs for the 12 study/exposure subgroups, even without any adjustment for potential TCDD-related differences. This high level of consistency points toward substantially higher (about 32%) all-cancer mortality in these worker cohorts than would be expected on the basis of the mortality experience of appropriate reference populations. The challenge is to discover the true cause of this significant excess in all-cancer mortality, as it appears to be attributable to factors other than the workers' TCDD body burdens.

Proponents of TCDD's human carcinogenicity have argued that a substance like TCDD that may be acting via a promotional mechanism would not be expected to produce the same specific cancer mortality excesses in different cohorts of workers because their preceding exposures to initiating agents would likely differ. In essence, this is an argument that specificity is not a relevant criterion for cancer promoters, but it is impossible to differentiate this argument from a post hoc rationalization for the absence of consistent excesses in mortality from specific cancers such as soft tissue sarcoma. It is worth noting in this regard that IARC, in its review of the carcinogenicity of dioxin-like compounds, commented that even for ionizing radiation, a clear-cut multisite carcinogen in the atomic bomb survivors, there were clearly elevated risks only for certain specific cancer sites. IARC commented further that "This lack of precedent for a multi-site carcinogen without particular sites predominating means that the epidemiological findings must be treated with caution..." (IARC, 1997Go, p. 337).

The respiratory tract is the only specific cancer site that has shown modestly elevated mortality across most of the retrospective cohort mortality studies of TCDD (IARC, 1997Go). However, smoking is such a strong risk factor for respiratory tract cancer mortality, as well as other specific smoking-related diseases such as myocardial infarction and chronic respiratory disease (see Smith and Lopipero [2001] and Bertazzi et al. [2001a,b] for related discussion), that anything less than virtually perfect control for smoking intensity and duration could give rise to the modest elevations in respiratory tract cancer mortality that have been reported. Furthermore, although the results are not presented herein, a similar initial meta-analysis and subsequent dose-response modeling of respiratory tract cancer mortality data from the three epidemiology studies used by U.S. EPA have revealed essentially the same pattern as was observed with all-cancer mortality: a modest (33%) and homogeneous elevation that was clearly not associated with the workers' TCDD body burdens. Unfortunately, the observed number of deaths from respiratory tract cancer (totaling 106) was too small to provide the {chi}2 goodness-of-fit test with sufficient sensitivity to reject U.S. EPA's linear model with unit intercept, despite its having the largest {chi}2 statistic of all of the dose-response models considered.

The apparent strength of association between all-cancer mortality and TCDD exposure as represented in U.S. EPA's linear model was markedly weakened when a floating intercept was introduced. The linear model's slope factor, expressed per unit of TCDD body burden, dropped more than 3-fold from the U.S. EPA model's statistically significant value of 0.9 x 10–3 per ng/kg (95% confidence interval = 0.3 x 10–3, 1.6 x 10–3) to a statistically insignificant 0.3 x 10–3 per ng/kg (–0.3 x 10–3, 0.9 x 10–3) for the floating intercept model. The strength of association as determined using U.S. EPA's linear model is clearly inflated by the constraint on its intercept parameter to equal unity.

Coherence is essentially the same as believability (Monson, 1990Go), and it includes the concept of a mechanism of action that would serve to explain the observed association. U.S. EPA has proposed multiple mechanisms of action for the carcinogenicity of TCDD; indeed, their newest risk characterization document is replete with speculative hypotheses regarding how cancer might arise subsequent to TCDD exposure. However, none of these proposals has yet been confirmed by rigorous scientific experimentation, and U.S. EPA admits this: "Clearly, a causal link between the Ah receptor-regulated gene expression (regardless of the mechanism of gene modulation) and any of the demonstrated toxic effects has not been established." (U.S. EPA 2000a). In fact, it is not yet certain that occupancy of the Ah receptor is a necessary prerequisite for the clear carcinogenicity of TCDD in highly exposed laboratory animals, although the hypothesis of Ah receptor occupancy necessity but insufficiency is one that enjoys widespread acceptance in the scientific community.

Finally, temporality, the concept that cause must precede effect, is virtually assured in the three epidemiology studies relied upon by U.S. EPA, as all of the workers included in these studies almost certainly were exposed to TCDD at some point(s) prior to their deaths from cancer or other causes.

In summary, which the exception of temporality, the Bradford Hill criteria discussed above do not lend support to U.S. EPA's proposed classification of TCDD as a human carcinogen. The study results do demonstrate consistency, but they are consistent with the null hypothesis of no dose-response relationship between cancer mortality and the workers' TCDD body burdens. Indeed, given the extraordinarily wide range of exposure levels that were experienced in the three worker cohorts, there is a remarkable absence of dose-response. The results from the analyses presented herein demonstrate clearly that U.S. EPA's most current risk estimates for human cancer mortality are seriously flawed and excessively conservative.


    NOTES
 
1 To whom correspondence should be addressed. Fax: (919) 876-0201. E-mail: tbstarr{at}mindspring.com. Back


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 DISCUSSION
 REFERENCES
 
Aylward, L. L., Hays, S. M., Karch, N. J., and Paustenbach, D. J. (1996). Relative susceptibility of animals and humans to the cancer hazard posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin using internal measures of dose. Environ. Sci. Technol. 30, 3534–3543.[ISI]

Bertazzi, P. A., Consonni, D., Bachetti, S., Rubagatti, M., Baccarelli, A., Zocchetti, C., and Pesatori, A. C. (2001a). Health effects of dioxin exposure: A 20-year mortality study. Am. J. Epidemiol. 153, 1031–1044.[Abstract/Free Full Text]

Bertazzi, P. A., Consonni, D., Bachetti, S., Rubagatti, M., Baccarelli, A., Zocchetti, C., and Pesatori, A. C. (2001b). Bertazzi et al. respond to Smith and Lopipero. Am. J. Epidemiol. 153, 1048–1049.[Free Full Text]

Blair, A., Burg, J., Foran, J., Gibb, H., Greenland, S. Morris, R. Raabe, G., Savitz, D. Teta, J., Wartenberg, D. et al. (1995). Guidelines for application of meta-analysis in environmental epidemiology. Regul. Toxicol. Pharmacol. 22, 189–197.[ISI][Medline]

Colditz, G. A., Burdick, E., and Mosteller, F. (1995). Heterogeneity in meta-analysis of data from epidemiologic studies: A commentary. Am. J. Epidemiol. 142, 371–382.[ISI][Medline]

Crump, K. S., and Allen, B. C. (1985). Methods for quantitative risk assessment using occupational studies. Am. Stat. 39, 442–450.[ISI]

Fingerhut, M., Halperin, W. E., Marlow, D. A. Piacitelli, L. A., Houchar, P. A., Sweeney, M. H., Greife, A. L., Dill, P. A., Steenland, K., and Suruda, A. J. (1991). Cancer mortality in workers exposed to 2,3,7,8-tetrachlorodibenzo-p-dioxin. N. Eng. J. Med. 324, 212–218.[Abstract]

Flesch-Janys, D., Steindorf, K., Gurn, P., and Becher, H. (1998). Estimation of the cumulated exposure to polychlorinated dibenzo-p-dioxins /furans and standardized mortality ratio analysis of cancer mortality by dose in an occupationally exposed cohort. Environ. Health Perspect. 106,(suppl. 2), 665–662.[ISI][Medline]

Greenland, S. (1994). Invited commentary: A critical look at some popular meta-analytic methods. Am. J. Epidemiol. 140, 290–301.[Abstract]

Hill, A. B. (1966). Principles of Medical Statistics. Ch xxiv. Oxford University Press, New York.

IARC (1997). IARC monographs on the evaluation of the carcinogenic risks of chemicals to humans. Vol. 69. Polychlorinated dibenzo-para-dioxins and polychlorinated dibenzofurans. International Agency for Research on Cancer. Lyon, France.

Kociba, R. J., Keyes, D. G., Beyer, J. E., Carreon, R. M., Wade, C. E., Dittenber, D. A., Kalnins, R. P., Frauson, L. E., Park, C. N., Barnard, S. D., Hummel, R. A., and Humiston, C. G. (1978). Results of a two-year chronic toxicity and oncogenicity study of 2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicol. Appl. Pharmacol. 46, 279–303.[ISI][Medline]

Monson, R. R. (1990). Occupational Epidemiology, 2nd ed.. CRC Press, Inc., Boca Raton, Florida.

Ott, M. G., and Zober, A. (1996). Cause specific mortality and cancer incidence among employees exposed to 2,3,7,8-TCDD after a 1953 reactor accident. Occu.p Environ. Med. 53, 606–612.[Abstract]

Smith, A. H., and Lopipero, P. (2001). Invited commentary: How do the Seveso findings affect conclusions concerning TCDD as a human carcinogen? Am. J. Epidemiol. 153, 1045–47.[Free Full Text]

Stroup, D. F., Berlin, J. A., Morton, S. C., Olkin, I. Williamson, G. D., Rennie, D., Moher, D., Becker, B. J., Sipe, T. A., and Thacker, S. B. (2000). Meta-analysis of observational studies in epidemiology: a proposal for reporting. JAMA 283, 2008–2012.[Abstract/Free Full Text]

U.S. EPA. 2000a. Exposure and human health reassessment of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and related compounds Part II: Chapter 2. Mechanism(s) of Action. Draft Final. EPA/600/P-00/001Be.

U.S. EPA. 2000b. Exposure and human health reassessment of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and related compounds Part III: Integrated summary and risk characterization for 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and related compounds. September 2000 SAB Review Draft. EPA/600/P-00/001Bg.

U.S. EPA. 2001. Dioxin reassessment – An SAB review of the Office of Research and Development's reassessment of dioxin: Review of the revised sections (Dose Response Modeling, Integrated Summary, Risk Characterization, and Toxicity Equivalence Factors) of the EPA's reassessment of dioxin by the Dioxin Reassessment Review Subcommittee of the EPA Science Advisory Board (SAB). May 2001. EPA-SAB-EC-01–006.