Tobacco Documents Online

expressed as relative risks. A relative risk is the ratio of the disease isscidence rate in the exposed group to the incidence rate in the unexposed group. A relative risk of 1.0 indicates that the observed disease incidence rate in the exposed group is the same as that in the unexposed group. A relative risk above 1.0 indicates that the disease incidence rate is greater in the exposed qzvup, while a relative risk less than 1.0 indicates that the disease incidence rate is smaller in the exposed group. 25. In recognition of tbe complexity of epidemiology and to ensure consistency in its methodology, EPA has adopted certain criteria by which it evaluates epidemiologic data. These criteria are set forth in EPA's own cancer risk assessment guidelines, "Guidelines for Carcinogen Risk Assessment," 51 Fed. .r Reg. 33992 (Sept. 24, 1986) (the "Risk Assessment Guidelines"). The Risk Assessment Guidelines provide that before a conclusion regarding whether an exposure causes a disease (a causal inference) cat, be based upon epidemiologic data, three criteria must be met: (i) the apparent statistical association must be unlikely to be produced by chance; (ii) the possibility of confounding (i.e. the role of other actual or potential factors in the apparent association) must have been considered and ruled out as an explanation for the association; and (iii) there must be no identified bias that could explain the apparent association. 31 FEd. Reg. 33999. In addition, EPA admits in its ETS Risk Assessment that epidemiologic data must also be assessed 11

against other criteria including: the consistency of the data, the stzength or'magnitude of the statistical association, and vhether the data exhibit a dose-response gradient. 26. As is set forth in the following paragraphs, the epidemiologic data relied upon by EPA do not satisfy these criteria and therefore are not sufficient to support a finding that -there is a causal relationship between ETS and lung cancer. According to EPA's own guidelines, such a finding must be made in order to classify a substance as a Group A carcinogen. Indeed the vast majority of the epidemiologic studies relied upon by EPA report no statistically significant overall association between spousal smoking and lung cancer. (a) The Epidc~:_Iocic Studies (i) sbaace 27. The test of statistical significance is crucial to the scientific analysis of epidemiologic data. Without this test, cause nrsd effect conclusions can be erroneously attributed to associations occurring by chance alone, simply due to random sampling variation. Because chance occurrences can never be completely excluded from a study, the likelihood or probability that an observed association could be due to chance is described through the use of tests for statistical significance. Statistical significarce tests can be expressed in terms of a confidence ir,terval which is a numerical range determined by applyirga so-=alled confidence level to the data. A confidence 12

interval calculated with a 95% confidence level, for example, is refert~i to as a 95% confidence interval. An apparent association or relative risk is said not to be statistically significant at the 95% confidence level unless the entire range of the 95% confidence interval for that risk is above or below 1.0. By generally accepted scientific convention, a 95% confidence level is required in epidemiclogic studies to judge an association as statistically significant. 28. EPA analyzed 30 published epidemiologic studies when classifying ETS as a Group A carcinogen. The studies primarily address whether a wow-an's risk of lung cancer may be statistically correlated with whether ber spouse smokes. The _ studies do not measure actual ETS exposure. Instead, they rely on questionnaire responses as to whether a woman's spouse smokes, thereby treating reports of spousal smoking as a surrogate for actual or measured ETS exposure. 29. Df the 30 published ETS studies EPA relied upon, 11 were conducted in the United States. As originally reported, none of the U.S. studies found an overall risk estimate for lung cancer that was statistically significant. 30. Of the remaining studies conducted in seven countries other than the United States, 13 found no overall statistically significant sssociation between spousal smoking and lung cancer as origiaally reported. Put another way, of the 30 studies reviewed by EPA, 24 -- a full 804 -- as originally reported do O Qi GO CD 0

not support the Agency's classification of ETS as a Group A carcirogen. 31. In an attempt to make otherwise non-statistically significant study results appear statistically significant, EPA "reanalyzed" the 30 epidemiologic studies. Specifically, EPA lowered the threshold for achieving statistical significance by lowering the standard 954 confidence level employed by all but three of the authors of the 30 studies to an unorthodox 90% confidence level. In altering the original analyses of the authors of the studies, EPA doubled the possibility that any statistically significant association is simply a random and meaningless event. Its use of the lower 90% confidence level contrasts with its use of the generally accepted 95` ..vel in the 1990 draft of the ETS Risk Assessment and its routine use of studies employing the standard level in other carcinogen risk assessments. 32. Even after zeanalyzing the studies using the lower confidence level, EPA failed to obtain a statistically significant overall association from over two-thirds of the studies. Of the 11 U.5. studies, only one yielded a statistically significant overall association after reanalysis. 33. Other data which EPA refused to account for in its analysis also demomstrate that any observed association between ETS and lung cancer may be due to chance. Nine of the 30 epidemiologic studies of female lung cancer and spousal smoking 14

also gathered data on male lung cancer risk and spousal smoking and, zf these, vnly cne, a Japanese study by Hirayama that is methodologically flawed on other grounds, is statistically significant even at the lower confidence level. Twelve of fourteen reported studies that have examined workplace ETS exposure and lung cancer risk have reported no statistically significant association at the lower confidence level. Similarly, 12 of 13 studies that have examined childhood ETS exposure and adult lung cancer risk fail to show a statistically significant association at that level. 34. The large number of studies, both those considered by EPA and those EPA refused to consider, which report no overall statistically significant association between ETS and lung cancer, overwhelm the few studies that EPA claims demonstrate an association and suggest that those few results may simply be the result of chance. Wben this evidence is coupled with EPA's choice of a confidence level which doubles the risk of an association due to chance being labeled statistically significant, chance becomes the likely explanation for any statistical association claimed by EPA between ETS and lung cancer. (ii) Coafoundiac 35. Confounding of data exists when an association between a disease and in exposure to one agent can be explained, in whole or in part, by an exposure to a second agent that is associated 15

with.batb (a) the disease and (b) the first exposure. The presence of such potential confounding factors undermines the reliability of epidemiologic data because of the difficulty in disentangling one potential risk factor from another. Because human disease causation, and especially chronic disease causation, is an extraordinarily complex issue, epidemioiogic data must be scrutinizsd closely for confounding before it can be relied upon to identify potential risks. 36. Sound science and the Risk Assessment Guidelines require that EPA may not infer causation unless "(t)he possibility of confounding has been considered and ruled = as explaining the associ'-W`ion." 51 Fed. Reg. 33999 (emphasis added). Here again, EPA violated accepted scientific standards and its own guidelines. 37. The scientific literature identifies the following elements, among others, as potential confounders: diet, personal r-edical history, family health history, lifestyle choices, occupational factors, and environmental factors. Some of these potential confounders are of such magnitude that they are large enough to account completely for any apparent association between spousal smoking and lung cancer elaimed by EPA. 38. EPA acknowledged the existence of some potential confounding factors but ignored the Risk Assessment Guidelines' requirement to rule out confounding as an explanation for an association prior to basing a causal inference on epidemiologic 16

data. Instead, EPA deemed the criterion satisfied largely because 3t claimed to be unable to identify any single confounding factor that in itself would explain the apparent association between ETS and lung cancer which appeared in EPA's analysis. EPA's methodology falls well short of the Guidelines' requirement to "rul[e3 out" the possibility of confounding. (iii) JW2 39. Bias in statistics refers to any trend in the design, collection, analysis, interpretation or publication of statistical data that causes or may tend to cause a systematic distortion of the true nature of the relationship studied. 40. Both the Risk Assessment Guidelines and accepted epidemiologic principles reguire that bias must be excluded as an explanation for an observed association before it can be concluded that a statistically significant association exists. Various sources of bias, including publication bias and respondent bias, could explain any association claimed by EPA between ETS and lung cancer. 41. EPA recognized only one source of bias in its Risk Assessment -- the tendency of smokers to misrepresent themselves as nonsmokers ("smoking status misclassification bias") -- and chose to adjust for it by using an unpublished scientific model that ccmtniri Tramerous mathematical and conceptual inconsistencies, including assumptions based on nonrepresentative data. Z t'EPA bad used representative data, EPA's own analysis ~ 17 tp - O V! Cn Cr .:a

would nct have resulted in a statistically significant association between ETS and lung cancer. (iv) atreactb 42. Strength refers to the magnitude of an apparent association. Associations of less than 3.0 are generally considered in the scientific community and by EPA to be weak and equivocaZ. Associations under 2.0 are considered to be extremely weak and are far more likely to be an artifact produced by chance, bias or confounding than are stronger associations. The weaker an association, the less reliable it is for evaluating a potential causal relationship. EPA admits as much in stating in its own Risk Assessment Guidelines that 'spidemiologic studies are inherently capable of detecting only comparatively large increases in risk." 51 Fed. Reg. 33996. 43. Of the 30 epideaiological studies analyzed by EPA in connection with its classification of ETS as a Group A carcinogen, 80$ did not report an overall statistically significant association of any magnitude between ETS and lung cancer. Even without consideration of the requirements of statistical significance, all of the studies reported overall relative risks under 3.0 and 21 reported overall relative risks under 2.0. Even when the 11 U.S. studies were pooled by EPA the adjusted relative risk estimate it constructed was only 1.19. 44. Dr. Morton Lippmann, chairman of the committee of the EPA's Science Advisory Board ("SAB") which reviewed the draft ETS 18

Risk Assessment, acknowledged the weakness of the association found by•EPA when he noted to reporters at a press conference called to publicize the draft that the risk of ETS was "probably much less than you took to get here through Washington traffic." 45. In other contexts, EPA has concluded that relative risks greater than the risks claimed by EPA for ETS were insufficisat to classify a potential carcinogen as a Group A carcinogen. For example, in a draft report on electromagnetic fields, the EPA concluded that, "(t]he association between cancer occurrence and exposure to either ELF or RF fields is not strong enough to constitute a proven causal relationship, largely because the relative risks in the published reports have seldom exceeded 3.0 in both childhood residential exposures and in occupational situations." U.S. EPA, Office of Health and Environmental Assessment, Evaluation of the Potential Carcino,genieity•mf E2ectronacnetic Fields, EPA/600/6-90/005A, Workshop Reviev-flraft, June, 1990. The 1.19 relative risk reported for ETS by EPA is less than one-tenth of the 3.0 relative risk found "not strong enough" by the SAB with regard to electromagnetic fields. 46. Because any statistical associations between ETS and lung cancer relied upon by EPA are so weak, it is very likely that they are produced by chance, bias, or confounding. 19 ~ I C7

(v) pose-Reiooase 47. a dose-response relationship mQans that as the extent of exposure, and hence dose, increases, the observed incidence of disease also increases. 48. EPA's Risk Assessment Guidelines note the importance of finding a dose-response relationship when trying to determine if an observad association is causal. 51 Fed. Reg. 33999. EPA attempted to satisfy this criterion by employing another unorthodox approach: the substitution of a test for "trend" for a true dose-response test. A trend test does not measure dose- response and is not an accepted substitute for a dose-response analysis. 49. None of the 30 epidemiologic studies relied upon by EPA shows a statistically significant dose-response relationship. 50. Several studies actually show a reverse dose-response, higher reported levels of spousal smoking are associated with an apparent decreased risk of lung cancer. One study reported a reverse dose-response that was statistically significant. (vi) consisteacv 51. EPA states in the ETS Risk Assessment that the presence of a consistent association across several independent studies in different populations may be evidence of a causal relationship. ConveTSely, a lack of consistency across studies undermines the reliability of any apparent associations and argues against 20