Tobacco Documents Online

Critique of the Draft EPA Document Entitled "Health Effects of Passive Smoking: Assessment of Lung Cancer in Adults and Respiratory Disorders in Children" by Joseph L. Fleiss, Ph.D. Professor and Head Division of Biostatistics Columbia University School of Public Health New York, N.Y. and Alan J. Gross, Ph.D. Professor of Biostatistics Medical University of South Carolina Charleston, S.C.

The issue of whether there is an association between exposure of nonsmokers to environmental tobacco smoke (ETS) and the risk of lung cancer is a persistent one. In 1986 the National Research Council and the Surgeon General both published reports purporting to demonstrate such an association. These documents, however, have been shown to be flawed on numerous scientific grounds, and serious questions have been raised with respect to the validity of their conclusions; see, for example, Letzel et al. (1988). Now, four years later, the Environmental Protection Agency has released a draft document dealing with this same issue. We have been asked by The Tobacco Institute to review this draft in light of a recent paper that we co-authored with respect to the statistical technique of meta-analysis, which is the centerpiece of the EPA's lung cancer risk assessment. In our view, the flaws detected in the 1986 documents are carried over in large measure in the EPA's draft. This is especially true of the attempt to use, once again, the tool of meta-analysis to combine the results of epidemiologic studies. The EPA's meta-analysis fails to account for the many biases and confounders that beset the underlying epidemiologic studies, and in any event it inappropriately analyzes the results of studies worldwide when at the very least it should have limited its scope to those studies that relate to the United States. Accordingly, the EPA's meta-analysis constitutes an entirely unconvincing basis for an ETS risk assessment.

2 BIOGRAPHICAL NOTES Joseph L. Fleiss has been Professor and Head of the Division of Biostatistics at the Columbia University School of Public Health since 1975. In addition to his academic appointment at Columbia, he was until 1986 a senior research scientist in biostatistics at the New York State Psychiatric Institute, and from 1976 to the present he has been a consulting biostatistician at the Presbyterian Hospital in New York City. Professor Fleiss has been an officer, member and award recipient of a number of professional societies and the associate editor of several journals. He has served on several expert and review committees for the National Institutes of Health, the Food and Drug Administration, and the American Public Health Association, among others. He has published four books, 18 chapters in books, and nearly 170 journal articles on statistical aspects of medical research, including epidemiologic issues. Alan J. Gross is a professor in the Department of Biometry, Medical University of South Carolina, Charleston, S.C. He received the Ph.D. in statistics from the University of North Carolina in June 1962. Formerly, he was an Associate Professor of Public Health at the University of Massachusetts and a lecturer and research scientist in the School of Public Health at UCLA. He also served in the U.S. Public Health Service as a Senior Assistant Scientist attached to the National Cancer Institute.

- 3 - Professor Gross is the author/co-author of more than 50 scientific articles and two books. He has received research grants from a number of government agencies. Currently he is a member of an NIH study section (Epidemiology and Disease Control, Study Section 2) and an associate editor of Biometrics, a major publication in his field. Each author's curriculum vitae is attached. STANDARDS FOR META-ANALYSIS The issue considered by the EPA draft risk assessment is whether any convincing evidence exists for a relationship between exposure to ETS and the incidence of lung cancer in the nonsmoking population of the United States. To address this question, the EPA draft discusses the relevant epidemiologic studies, which typically involved married women who purportedly never smoked but were married either to a smoker (and were therefore presumably exposed to ETS) or to a nonsmoker (and were therefore presumably not exposed). The methodology employed in the EPA document is a meta-analysis of these epidemiologic studies. In a forthcoming article (1990; presented at the joint 53rd annual meeting of the Institute of Mathematical Statistics and the 2nd Bernoulli Society World Congress, Uppsala, Sweden, August 13-18, 1990 and accepted for publication by the Journal of Clinical Epidemiology), a copy of which is attached for reference, we indicate that meta-analysis, a set of statistical tools for combining and

- 4 - integrating the results of independent studies of a given scientific issue, can be useful if stringent conditions under which such integration is valid are met. The conditions that would need to be met with respect to studies involving ETS exposure and lung cancer include especially the proper adjustment or control, on a study-by-study basis, for the biases and confounding factors frequently present in epidemiologic studies, such as (1) socio-demographic or clinical differences among study populations, (2) misclassification of subjects with regard to case-control 0 status and levels of exposure, (3) factors other than the level of exposure that may affect whether a subject is a case or a control (confounding variables), and (4) the publication bias/file drawer phenomenon whereby studies that fail to show a positive association are not published and thus are not candidates for inclusion in the meta-analysis. It is our conclusion that the EPA draft fails to meet these key criteria. Put differently, a meta-analysis is only as good as the underlying epidemiologic studies that form the basis for the combinational computations that result in an estimate of overall relative risk. Numerous reviewers have called into question the reliability of the individual studies employed by EPA in its meta-analysis, and we need not rehearse those critiques here. Suffice it to say that we fully agree with the succinct conclusions of Dr. Layard (1990) with respect to those studies: "Meta-analysis should not be used unless it can reasonably be concluded that the results of the

- 5 - various studies provide comparable estimates of a common quantitative endpoint, in this case, the relative risk of lung cancer among nonsmokers exposed to ETS. Such a conclusion requires that the studies be reasonably comparable with respect to exposure indices, demographic and social characteristics of study populations and disease diagnosis among other factors. In fact, the studies of ETS and lung cancer display substantial diversity in these and other areas. As well as being comparable, each study should be methodologically sound and free from potential biases which could distort the result. .. [T]hose conditions are not met for the studies in question. The small positive associations seen in these studies, to the extent that they are not simply due to chance, may in fact be the result of bias and confounding. If so, a meta-analysis would only serve to provide spurious reinforcement of the invalid results of the individual studies. . . . The weak and inconsistent association seen in the epidemiologic studies of ETS and lung cancer, the fact that bias and confounding cannot be ruled out, and questions about the reliability of the reported results, all indicate that these data do not support a judgment of causal relationship between exposure to ETS and lung cancer." A META-ANALYSIS OF THE U.S. EPIDEMIOLOGIC STUDIES Even if one were to assume, as we do not, that the underlying epidemiologic studies employed in the EPA's risk assessment were individually valid and unbiased, the question remains as to whether one is justified in performing a meta-analysis of all the worldwide studies in order to arrive at a relative risk to be applied to the United States population. We believe that a limitation to United States-based studies would be preferable. As we conclude in our attached paper: "There are many reasons for restricting attention to American studies of whether there is an elevated risk to nonsmokers exposed to

6 ETS relative to nonsmokers not so exposed. One is that this is the population on whom policy decisions will be based, and to whom those decisions will apply. Another is that the summary odds ratios in the individual studies are derived from distributions of smoking amounts and durations, and of brands of cigarettes and other tobacco products, that pertain to populations within the United States, and may thus be expected to be relatively homogeneous. Odds ratios from studies in other countries, on the other hand, are derived from distributions that may differ markedly from those in the U.S. and thus odds ratios may not be relevant to the American experience. Genetic and lifestyle differences between the U.S. population and the populations studied elsewhere (mainly in east Asia) also argue for a meta-analysis only of U.S. studies." Accordingly, if the EPA were to persist in its use of the meta-analysis technique, it should employ only the results of the U.S.-based studies. We performed this exercise in our attached paper, using the case-control studies identified in Tables 3-5 and 3-6 of the EPA draft (referred to therein as BROW, BUFF, CORR, GARF, HUMB, KABA, VARE, and WU) as well as the cohort study of Garfinkel (1981). As these studies are all fully described in both the EPA draft and our paper, only the results of our meta-analysis are presented here. First, the DerSimonian-Laird test for homogeneity yields X2 = 5.46 with df = 8, p. > 0.10 for those studies, indicating relative numerical homogeneity from study to study within the U.S. epidemiologic studies. Proceeding with the method of analysis described in our paper, the overall relative risk computed for the nine American studies is a statistically nonsignificant 1.12 with a 95 percent confidence interval from 0.95 to 1.30. We note that this result accords

7 closely with the calculation by the National Research Council in 1986 of a statistically nonsignificant relative risk of 1.14, with a 95 percent confidence interval from 0.92 to 1.40, for the American studies available at that time. Considering the results of our meta-analysis, we conclude that, based on the available American epidemiologic evidence, there is no convincing scientific basis for rejecting the null hypothesis: The EPA draft risk assessment does not demonstrate that exposure to ETS has been shown to increase the risk of lung cancer in the nonsmoking population in the United States.

8 REFERENCES DerSimonian, R.; Laird, N. (1986) Meta-analysis in clinical trials. Controlled Clin Trials 7:177-188. Fleiss, J.L.; Gross, A.J. (1990) Meta-analysis in epidemiology, with special reference to studies of the association between exposure to environmental tobacco smoke and lung cancer. A critique. (Submitted for publication.) Layard, M. (1990) Environmental tobacco smoke and cancer: The epidemiologic evidence. Chapter 6 in Environmental Tobacco Smoke. Proceedings of the International Symposium at McGill University, Montreal. Letzel, H., Blumner, E. and t)berla, K. (1988). Meta-analysis on passive smoking and lung cancer: Effects on study selection and misclassification of exposure. Environmental Technology Letters 9:491-500. National Research Council (NRC) (1986) Environmental tobacco smoke: Measuring exposures and assessing health effects: National Academy Press, Washington, D.C.