Tobacco Documents Online

REACTIONS TO ENVIRONMENTAL TOBACCO SMOKE:. A Compendium of Technical Information Chapter 4: Environmental Tobacco Smoke and Cancer Prepared by: Joseph L. Fleiss, Ph.D. I , I have been a Professor and Head of the Division of Biostatistics at the Columbia University School of Public Health since 1975. In addition to my academic appointment at Columbia, I was until 1986 a senior research scientist in biostatistics at the New York State Psychiatric Institute, and from 1976 to the present I have been a consulting biostatistician at Presbyterian Hospital in New York City. I have been an officer, member, and award recipient of a number of professional societies and journals, and I have 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. I have published four books, 16 chapters in books, and some 160 journal articles on statistical aspects of medical research,_ including epidemiologic issues. My curriculum vitae is attached. I have been asked to review "Environmental Tobacco Smoke and Cancer," by J.M. Samet, which is Chapter Four of an EPA•draft compendium of technical literature on environmental tobacco smoke.

- 2 - In this chapter, Dr. Samet reviews the epidemiological evidence concern.ing an association-between environmental tobacco smoke (ETS) and lung cancer, and concludes that "involuntary smoking causes lung cancer" (p. 46). In my opinion, such a conclusion is unwarranted, given the almost uniformly poor quality of the epidemiological studies that have been published. There are numerous flaws in these studies; I will limit my comments to those that strike me as most serious from the point of view of a biostatistician. Poor quality in the form of inadequate control has characterized most of these studies from the very beginning, starting with Hirayama's initial study published in 1981. A flood of criticism followed the publication of the article, with responses by Hirayama that Dr. Samet asserts "satisfactorily answered most of these criticisms." I disagree strongly with this assertion. Hirayama never satisfactorily explained, for example, why he controlled in his first analysis for the age of the husband when it was the wife who was the study subject and thus her age that affected her risk for lung cancer. It was in 1984 that Hirayama finally reported the results of analyses that adjusted for the age of the subject herself. The age-adjusted risk for nonsmoking women married to a husband who ever smoked relative to that for nonsmoking women married to a nonsmoker dropped from 1.57 to 1.45, and its two-tailed p-value increased from=

3 0.02 to a barely significant 0.05. The strength and significance of the association were both attenuated-when the correct adjustment was made. The impact of Hirayama's original article would surely have been weaker had the data been properly analyzed. For all of its flaws -- in addition to the absurd control for the age of the husband, Hirayama's study failed to employ actuarial methods in analyzing the data, misstated the statistical significance of the association between exposure to ETS and lung cancer, and combined adenocarcinoma of the lung and bronchial alveolar cancer with o--her lung cancers more strongly linked statistically to cigarette smoking -- the study was prospective and therefore not p:rone to the kinds of systematic errors in ascertaining exposure to ETS that afflict retrospective case-control studies. Dr. ,3amet discusses the effects of misclassification errors on the estimate of relative risk. Even though he points out that nonrandom misclassification error may either exaggerate or reduce the magnitude of the estimate, most of his subsequent analysis of misclassification assumes that misclassification errors occur randomly and with equal probability in caSes and controls. The predictable effect of random misclassification is to underestimate, not overestimate the relative risk. What Ob concerns me, and what should have concerned Dr. Samet more ~ ~ ~ than it did, is the likelihood of nonrandom, systematic error, N W CO Cd

- 4 - r, with the possible consequence that the published relative risks are overestimates-of the true parameter.- I have in mind especially the bias produced in case-control studies by the nonblinded irquiry by the investigator into the patient's and the spouse's smoking history. It is not deliberate misrepresentation or falsification that is of concern so much as the unwitting application of different criteria in evaluating the report of a lung cancer patient or surrogate versus the report of a control. I have long argued that it is as essential for_ validity in a case-control study to blind the investigator as it is in a clinical trial (Fleiss, 1981, p. 206). Nevertheless, the investigators were kept ignorant of the status of the subject as case or control in, apparently, only three of the case-control studies that have been published (Garfinkel, et al., 1985; Akiba, et al., 1986; Brownson, et al., 1987). The nonblinded studies are, in my opinion, seriously and possibly fatally flawed. Similarly flawed are those case-control studies in which a surrogate reporter, someone other than the patient, was relied on for information about the patient's and spouse's smoking history. The Methods sections of the published articles do not always state whether surrogate reports were permitted, but my estimate is that at least half and possibly three quarters of the published case-control studies relied on

- 5 - the recollections of the next of kin when the patient had died. Consider as an example the 1985.study by Garfinkel et al. that Dr. Samet cited as providing supporting evidence for an association between exposure to ETS and lung cancer. The patient herself was the source of information in only 12 percent (16/134) of the cases. In approximately 65 percent of the cases the patient had died and a spouse or child was interviewed for information about the patient. In the remaining 20-25 percent of the cases, someone other than the patient, her spouse or her children supplied the neces-sary information! A body of knowledge exists to the effect that sizable fractions of subjects misreport, sometimes randomly but generally systematically, their lifetime history of smoking and that of their spouses. How much greater must the error rates be when it is the surviving spouse, or a child, or a sibling, or someone not even related, who is the source of information about the patient, or when it is a child, or a sibling, or someone unrelated who is the source for the spouse. In my opinion, it is still not known, nine years after the appearance of the first three articles on the subject, whether exposure to ETS is truly a risk factor for lung cancer. -•Perhaps the epidemiological studies that Dr. Samet mentions as being in progress are sufficiently tightly designed, with appropriate controls, to settle the-issue of

- 6 - whether or not an association exists. The epidemiological studies that have been published thus far, however, are inadequate for the task. *10.l

References Akiba, S., Kato, H., Blot, W.J. (1986). Passive smoking and lung cancer among Japanese women. Cancer Res. 46:4804-4807. Brownson, R.C., Reif, J.S., Keefe, T.J., et al. (1987). Risk factors for adenocarcinoma of the lung. Amer. J. Epidemiol. 125:25-34. Fleiss, J.L. (1981). Statistical Methods for Rates and Proportions. 2nd edition. New York: Wiley. Garfinkel, L., Auerbach, 0., Joubert, L. (1985). Involuntary smoking and lung cancer: A case-control study. J. Natl. Cancer. Inst. 75:462-469. Hirayama, T. (1981). Nonsmoking wives of heavy smokers have a higher risk of lung cancer: A study from Japan. British Med. J. 282:183-185. Hirayama, T. (1984). Lung cancer in Japan: Effects of nutrition and passive smoking. Chapter 14 in Mizell, M. and Correa, P. (eds.). Lung Cancer: Causes and Prevention. Deerfield Beach FL: VCH.