Tobacco Documents Online

concludes that ETS is a Group A carcinogen. The EPA~s own guidelines (The Risk Assessment Guidelines of 1986, 51 Fed. B_f,g., 33992; Guidelines for Estimating Exposures, 51 Fed. I~g. 34042; Guidelines for the Health Risk Assessment of Chemical Mixtures, 51 Fed. Egg. 34014) require the following: 1. Chance must be ruled out- Only 5 of the 24 individual studies reviewed by EPA reached statistical significance. Without adjustment for bias and confounding, analysis of a variety of subsets (U. S. studies reviewed by El'A; all U. S. studies to date; all studies reported since the 1986 NRC and Surgeon General's reviews; all EPA studies plus 3 recent studies) produces a non-significant summary odds ratio. (RJR Appendix A). These results indicate that chance cannot be ruled out. 2. Possibility of confounding must be ruled out - Confounders and potential lung cancer risk factors are not adjusted for consistently in the studies. Though the Agency summarily concludes that no correlate of-ETS has been identified as an explanation for the observed associations, several researchers have identified lifestyle characteristics reported to be associated with lung cancer risk also associated with ETS exposure. (Friedman, eta/., 1983; Koo, 1989; Sydney, et aL, 1989). EPA has failed in its burden to rule out the possibility of confounding as an explanation for the observed associations. 3. Bias must be excluded as an explanation for any observed association - Smoking stares misclassification and publication bias are not the only forms of bias that receive inadequate analysis by EPA. The Agency apparently concludes, for example, that bias due to a proxy respondent is negligible. Wynder, 1987, however, concluded that U[r]elatives of a nonsmoking lung cancer patient are more likely to report passive inhalation 8

exposure on the part of a relative than relatives of a control patient .... " Therefore, several sources of bias have not been adequately addressed by the Agency. 4. Strength of association must be evaluated - Even for those studies included by the Agency which show a statistically significant association, the association is weak. "Weak" has been characterized as a relative risk less than 2.0 to 3.0. (Wynder, 1987; Doll, 1985). The EPA itself, in 1989, suggested that relative risks less than 5.0 are considered weak. Observed associations in the ETS lung cancer literature are in the range that strains the limits of the epi.demiologic method. The Agency does_not address squarely the problems inherent in interpreting weak associations. 5. A dose-response relationship must be demonstrated - Not one study reviewed by EPA exhibits a statistically significant dose-response relationship when attention is restricted to exposed subjects. Despite the Agency's statement that plots for trend are consistent with a statistically significant association between ETS and lung cancer, it is apparent that the data are consistent with either the presence or absence of a dose-response. (R JR Figure 2). In fact, many of the studies are consistent with an inverse dose-response relationship. 6. Consistency of the association must be evaluated - Inconsistencies in the studies are minimized by EPA. The fact that 80% of the studies are statistically insignificant at the 5 % level does not demonstrate consistency of association. A striking inconsistency is seen when observed associations are broken down by lung cancer histologic type. (See, e.g., Garfmkel, 1981). The two animal inhalation experiments investigating ETS and lung cancer have found 9

no meaningful histopathological differences between animals exposed to EFS and those which were not exposed. (Halcy, 1987a, 198To; Adlkofcr, 1988). One of the most misleading sections of the Health Assessment is that in which it discusses biological plausibility. The Agency relies on the premise that ETS is chemically similar to mainstream tobacco smoke, an unfounded assumption. Even if one were to accept the unproven assumption that smoking is causally related .to lung cance¢, extrapolation from smoking to ETS exposure is unwarranted: 1. ETS is a qualitatively different chemical mixture than mainstream smoke. The biologic effects of one mixture cannot be predicted based on the presence of constituents found in a second complex mixture. (NRC, 1988 at 3). 2. There are many significant differences between ETS exposure and cigarette smoking. (USPHS, 1986 at 7). The Agency has not conducted a weight-of-evidence analysis of ETS and lung cancer. EPA's review is limited to a statistical manipulation of a subset of the epidemiologic data. No evaluations of the physical-chemical properties of ETS, routes and patterns of exposure, structure-activity relationships, pharmacokinetic properties or animal studies is described. The determination that ETS is a Group A (known human) carcinogen is arbitrary and has no substantial evidentiary basis. EPA concludes that no causal association between ETS and childhood respiratory disorders has been established, yet it further concludes that ETS should be treated as a risk factor for acute respiratory diseases and chronic obstructive pulmonary disorders in infants and young children. The Agency has not performed a weight-of-evidence analysis or any l0

other analysis to justify this conclusion. The Agency must quantify the contribution of bias and confounding acknowledged to be present in these studies OVitorsch, 1990) before it can conclude that the weak and inconsistent observed associations reported in some studies indicate a true association between HIS and childhood respiratory disorders. 11

I. The Meta-Analvsis Is Fundamentally Flawed And Provides No Basis For Derivin~ ETS Lun~ Cancer Mortality Projections The t~rm "meta-analysis" was coined to describe statistical methods that were developed by social scientists and others to summarize overall quantitative trends a~'oss studies of a particular topic (Glass, 1976). Meta-analysis was developed to systematically and quantitatively evaluate and integrate results from sets of studies. Ac~:ess to raw data is not required because its results are the unit of statistical analysis. The technique was developed to complement, not supplant, the traditional narrative review and study-I~lying approaches. The Agencymeta-analyzed 22 epidemiologic studie~ of ETS and lung cancer using an extended Mantel-Haenszel procedure. The overall estimate of relative risk for nineteen case- control and throe cohort studies was calculated to be 1.41 (95% CI -- 1.26, 1.57). Adjustment for smoldng stares misclassification reduced the overall relative risk to 1.28 (95% CI -- 1.12, 1.45). EPA's modification for "background exposure" to ETS elevated the summary relative risk to 1.48 (95% CI -- 1.21, 1.87). EPA relies on the metaoanalysis for two purposes: (1) to increase the studies' power to determine whether the observed associations overall are ascribable to chance alone, and (2) to obtain an overall measure of the observed association's magnitude for use in deriving mortality projections. The meta- analysis presented in the Health Assessment is invalid for either purpose. The Agency has not justified its use of meta-analysis, the selection of studies to be combined or the assumptions used in deriving mortality projections. Meta-analytic techniques encompass a variety of statistical methods. Regardless of the method used, at least three conditions must be satisfied for the results to be reliable: 1. Combined studies must be comparable in terms of design, conduct and 12

analysis. They must be investigations of the same dependent and independent variables. Noncomparable studies cannot be aggregated. m Individual studies must be evaluated for quality prior to inclusion in the meta- analysis. Low-quality studies must be excluded. Subject to (1) and (2) above, studies to be combined must represent all or a representative sample of all studies of the hypothesis. (Glass eta/., 1981; Hunter et a~, 1982; Hedges and Olkin, 1985; Wolf, 1986; and Hunter and Schmidt, 1990). The meta-analysis described in the Health Assessment violates all three conditions. Application of meta-analytic techniques to epidemiologic studies raises additional issues. The National Research Council Committee on the Epidemiology of Air Pollution in 1985 warned that the application of meta-analysis to epidemiologic studies must be done cautiously: Although meta-analysis is seductively simple, it contains serious perils when applied to most epidemiologic studies, and its quantitative nature can mask serious flaws in data. In essence, meta-analysis assumes that the results of studies can themselves be treated as random variables with predictable distributions. That assumption might be reasonable for experiments repeated under very similar conditions, but it is rarely so for epidemiologic studies, in which extraneous factors are harder to control and nonrandom errors dominate the random ones. (NRC, 1985, at 218). The Health Assessment contains no discussion of the appropriateness of applying meta-analytic techniques on the ETS and lung cancer literature. The combined studies are not comparable; poor quality studies were included; and, the combined studies are not representative of all studies of spousal smoking and lung cancer. Meta-analysis should not be used unless it can be concluded that the studies of a topic are reasonably comparable (Mann, 1990, at 478). The NRC Committee on the Epidemiology 13

of Air Pollution cautioned that combining noncomparable studies may produce misleading results: The pooling of observations from independent studies to increase sample size is questionable in environmental epidemiology, because it ignores the differences between studies altogether. Numeric combination of results from different studies still has only a small role in epidemiology. (NRC, 1985, at 218, 219). EPA justifies combining the studies by testing for statistical homogeneity of the relative risk estimates presented in the case-control studies.- (The Health Assessment at 3-12). This test does not demonstrate comparability in a relevant sense. Due to the relatively small sample sizes, the 95 % confidence intervals for the relative risk estimates overlap even though the point estimates vary greatly. The absence of power in the individual studies - not similarity between the studies - produces the statistical homogeneity. Statistical similarity among the relative risk estimates does not demonstrate comparability. The Agency notes several differences between the case-control studies but concludes that "[s]tudy differences do not invalidate statistically testing the hypothesis that exposure to ETS is unrelated to lung cancer occurrence." (The Health Assessment at 3-12). No rationale for this statement is offered. The Agency has assumed rather than demonstrated comparability. Demonstrating comparability requires examining the design, conduct and analysis of individual studies. Studies combined by the Agency in the Health Assessment meta-analysis are noncomparable on several bases. The studies draw subjects from disparate cultures. Populations in Greece and Japan differ from U. S. subjects in life style and perhaps even genetic susceptibility. Geographic and ethnic differences could influence host factors and 14

overall health patterns. It is inappropriate to combine studies from disparate cultures with those conducted in the United States. The Agency notes that "a real difference in risk in the populations studied" may account for the different observations produced by the Japanese and American cohort studies by Hirayama and Garfinkel respectively. (The Health Assessment a/ 3-38). At 3-39 of the Health Assessment, EPA suggests that exposure related factors may vary between the two cultures. In fact, the risks observed in the United States studies are statistically different from those observed in the Asian studies. As demonstrated in P.JR Appendix A to these comments, the summary estimate for-risks observed in the U. S. is 1.08 and is not statistically significant (95% CI ffi 0.68, 1.73).3 Differences between the Asian and U. S. studies are unlikely to be attributable to chance alone. Combination of the Asian studies produces a relative risk of 1.37 that is statistically different from the U. S. summary risk (p ffi 0.022).4 This inconsistency undermines EPA's conclusion that ETS exposure causes lung cancer. Instead, it suggests that some life-style factor in the foreign studies is responsible. It demonstrates also that the Asian data should not be used for projecting risk to the U. S. population. The studies also are noncomparable in a variety of methodologic respects. RJ-R's positioa is that a meta-analysis of spousal smoking and lung cancer studies is invalid for the reasons described in these comments. P-JR believes further that the EPA has applied meta-analytic techniques to the spousal smoking and lung cancer literature in a selective manner that is designed to maximize the summary risk estimate. R JR retained George Howard of the Wake Fore~ Unive~ty Bownum Gray School of Medicine to assist it in evaluating ~tatistically the meta-analysis presented in the Health Assessment. His report and a copy of Dr. Howard's curriculum vitae are Appendix A to Note, however, that there is great uncertainty and disagreement even amcmg the Asian studies. Gao a a/., in a study funded and co-authored by the NCI controlled for many confmmders not controlled for in other Asian studies and observed a relative risk of 0.9 (95% CI = 0.6, 1.4) among nonsmoking Chinese women married to smokers. This raises serious questions about the munmary risk achieved by combining other, less well controlled, Asian studies. 15

Combination of case control and cohort studies is inappropriate. When studies with different durations of observation are combined, bias is introduced. If the effect is time dependent, or the number of individuals at risk changes markedly, conventional approaches are likely to produce distorted estimates of the true relative risk. In addition, studies conducted at different time periods may be dissimilar, even though of comparable duration. For example, it is known that levels of air pollution have changed over recent years, presumably affecting any studies of respiratory health. The combined studies evaluated varying histological lung cancer types. For instance, Brownson et al., 1987, restricted their analysis to adenocarcinoma while Trichopoulos, 1981, specifically excluded cases of adenocarcinoma from his study. The Agency has not justified combining studies with such fundamental design characteristic differences. As shown in RYR Table I, the sources of controls in combined studies differ greatly, and include: atomic bomb survivors (Akiba, 1986), bone or colon cancer cases (Brownson, 1987), orthopaedic patients (Chan, 1979, Lam, 1987, Trichopoulos, 1981), colon or rectal cancer patients (Garfinkel, 1985), general population controls (Buffler, 1984, Svensson, 1989), licensed drivers (Varela, 1987), matched neighborhood controls (Wu, 1983), and hospital patients with "non-smoking associated diseases" (Correa, 1983, Kabat, 1984). Moreover, factors used in matching controls in the case control studies varied widely. (RJR Table II). Less than half of the studies match for race, hospital or area of residence. Only two (Geng et aL, 1988; Lee, 1981) matched for marital status. Four (Garfinkel, 1985; Kabat, 1984; Lee, 1986; Varela, 1987) matched for smoking status. The index of exposure varied across the studies in a manner that precludes their 16

Study TABLE I SOURCE OF CONTROLS ISour~ of Co~trol~ Atomic bomb marvivo~ v~th disense~ other ~ c~nce~" BROW Coloa and bone marrow patient~ BUFF Popeh~oa and deu~lent mntrol~ CORR Patients with no--rooking associated diseases GAO Neighborhood GARF(Co) Proqx~tive w.~dy GARF ~ of ~olon or re~mm GENG Not given GILL(Co) Prospective study HIRA(Co) Prospective study HUMB Random telephone sample plus Medicare participants INOU Cerebrovascular decedents KABA Patients with nonsmoking associated diseases KOO District controls LAMT Orthopaedic patients LAMW Neighborhood LEEPatients with nonsmoking usso~iated diseases PERS Women from general population SVEN Population controls TRICH Orthopaedic controls from different hospitals VARE Li~z~xl driven WU I Neighborhood aggregation. (PJR Table III). In general, a woman was considered exposed if her husband was a smoker. However, the basis for determining whether the husband was a smoker varied. In Inoue, 1989, a smoker at home was defined as someone who smoked five or more cigarettes per day. If a husband smoked fewer than four cigarettes per day, his wife was considered unexposed. In contrast, Akiba, 1986, Garfmkel, 1985, Geng, 1988, Humble, 17