Tobacco Documents Online

,w.:~: ~: .

PASSIVE Sl+IOKING AND HEART DISEASE: A Compendium of Technical•Information CHAPTER 11 PASSIVE S!lORING AND HEART DISEASE EPIDEKIOLOGY, PHYSIOhOGY, AND BIOCHEMISTRY I am Director of the Epidemiology Consulting Group at New York Medical College, and an Adjunct Associate Professor of Community and Preventive Medicine at the same institution. I received the Ph.D. degree in Epidemiology from Yale University in 1979. I was formerly a clinical research scientist at American Home Products and Pharmacia Opthalmics, two pharmaceutical companies, as well as a health care researcher at Blue Cross and Blue Shield of Greater New York. Currently, as Director of the Epidemiology Consulting Group, I have executed epidemiological analyses for numerous drug, chemical, and health companies. My curriculum vitae is attached. I have been asked to review "Passive Smoking and Heart Disease: Epidemiology, Physiology, and Biochemistry," by Stanton A. Glantz and William W. Parmley, which is Chapter 11 of a draft

EPA compendium of technical literature on environmental tobacco smoke (ETS). I will confine my comments to the epidemiology section of the chapter. I. INTRODUCTION In this chapter, Glantz and Parmley review the epidemiological evidence concerning exposure to ETS and heart disease in humans. They cite on page 3 "eleven published" 9 studies. However, they only review and discuss ten studies. Moreover, the study by Martin has not been published, although it was ostensibly presented at the American Public Health Association Conference in October, 1986. The article by Humble is ostensibly in press in the American Journal of Public Health. Neither study is available at this time for review. Therefore it is impossible to determine the validity of Glantz and Parmley's analysis of these papers. The article by He (1989) is available only in Chinese with only the abstract in English, thereby limiting my ability to comment on this study. It is clear, howeveri that the He paper is a retrospective case-control study having only 34 cases, and it is fraught with all of the methodological problems surrounding retrospective case control studies, such as poor recall of ETS exposure and smoking behaviors. I will address the remainder of this paper to the work that has been published on the reported relationship between exposure - 2 -

to ETS and cardiovascular disease. In my view, the studies conducted thus far do not demonstrate that ETS increases the risk of cardiovascular disease. Accordingly, I disagree with many of the conclusions presented by Glantz and Parmley. First of all, the authors' point that "Despite minor changes in methodology, or end points. .. the results of these studies are remarkably consistent," is inaccurate. In fact, the studies vary greatly in their ability to measure ETS exposure adequately, to control for misclassification of smoking status, and to control for numerous important confounding variables. Second, although Glantz and Parmley argue that all of the studies except one report relative risk greater than 1, and imply that this supports a finding of statistical or clinical significance, or even epidemiological proof of causation, I would argue that this is indeed not the case. Third, Glantz and Parmley contend that the conclusions of the Surgeon General (USPHS 1986) and National Academy of Sciences (NAS 1986) are no longer valid.because of studies made available since those reports. As discussed further in this critique, I disagree and believe that the conclusions of the Surgeon General and NAS remain valid. II. EPIDEMIOLOGICAL CRITERIA In epidemiology, it is crucial to distinguish between association and causation. An association merely raises the possibility that causation may exist and suggests that further investigation is warranted. To justify even an inference that - 3 - .;.:..w:.a....~...:.~....:.....;..4.~.,.,._.,.._..~.~...~-.;~. . ,.. .._

epidemiologic data may suggest causation# five criteria must be First, the statistical association shown by the data must be sufficiently strong: There must be a statistically significant increase in the incidence of the disease in the exposed population compared with the non-exposed population, and, for the association to be regarded as meaningful, a relative risk of 2.0 or greater is generally considered necessary. Although there is no precise definition of a"weak" association, relative risks of less than 2.0 are generally considered to be weak. indeed, the noted epidemiologist Jerome Cornfield contended that any relative risk under 3.0 might be considered weak (Wynder 1987). Further, the association should be dose-dependent, that is, higher doses are associated with higher incidence of disease. Second, consistency of the statistical association must exist among the relevant studies. Similar rates of disease must occur at different times and places, under comparable study designs. Third, exposure to the agent under study should have occurred at a reasonable time before the onset of disease, given what is known about how long it takes for a particular disease to develop. ..w..~„..,,,~,.>.........,Y...,.. ....~,~,,...,,~...: . R

Fourth, the statistical association must be specific to the agent under study. Exposure must be shown to be associated with the relevant disease•while controlling for all confounding variables. Fifth, there must be biological plausibility. Under experimental conditions, exposure to the pertinent substance (or similar substances) must be shown to cause biological changes that can lead to the disease in question. Given these five criteria, one must reject Glantz and Parmley's implication that consistency, standing alone, is enough to support a conclusion that an association is causal. I agree with Glantz and Parmley that it is particularly important to control for confounders in epidemiological studies of heart disease. The Framingham Heart Study (Seltzer, 1989), for example, has shown that the following variables are all potential confounders for heart disease: age, sex, cholesterol level, blood pressure, weight, socioeconomic status (including income, occupation and education), and adequate measurement of environmental factors. Glantz and Parmley take the position that, as a whole, the published epidemiological literature adequately controls for these confounding variables. A thorough analysis of this literature does not support their conclusions. Each of the published studies fails to control for one or more important confounding variables, including lifestyle, blood . _~ ~_~~: . ,... . ~. .. •.., ., ..,,-,...._ .

pressure, serum cholesterol, obesity and socioeconomic status. Furthermore, none of the studies provides an adequate measurement of ETS exposure. Additionally, none of the studies demonstrated a specificity of association between ETS exposure and cardiovascular disease. All of the studies suffer from one or more serious methodological problems, including small sample size and possible misclassification of spousal smoking status.. These confounding variables and methodological problems also preclude. any demonstration of consistency of association among the III. PUBLISHIED STUDIES There are seven published epidemiological studies examining exposure to ETS and cardiovascular disease: five are prospective studies and two are retrospective case-control studies. The designs, findings and methodological problems are summarized in Table One. To support my conclusions, a complete analysis-of each study follows. To begin with the prospective studies, I note first that Hirayama's work has several major methodological problems. First, there is potential misclassification of smokers and non- smokers. Many of the wives who stated they were non-smokers may in fact have been ex-smokers or even current smokers, and thus 6 - ~.~v.~.:v., ~.... ~.u _ _... - - .~. .~. ._ ....~.~.... ._ ~_

likely to have had or continue to have direct (as opposed to indirect) exposure to cigarette smoke. Second, Hirayama's study included a disproportionate number of women of lower socioeconomic status. In Japan, these women live in much closer proximity to their cooking quarters and may have more exposure to charcoal or kerosene stoves than women of higher socioeconomic status. This exposure has been associated with lung cancer in women in Hong Kong. Women in Japan of higher socioeconomic status live farthex away from their kitchens and are more likely to use electric burners. The Hirayama study failed to control for these confounding variables, which may well be associated with ischemic heart disease. Third, there is the problem of misclassification of dose response. Ex-smoking husbands were lumped with current cigarette smokers of 1-19 cigarettes/day. Because ex-smokers are very different in their cigarette exposure rates and lifestyles than smokers of 1-19 cigarettes/day, this factor could skew the data. Fourth, Hirayama examined only the exposure of the wife in the context of the husband's cigarette smoking behavior. No attempt was made to quantify any exposure to ETS outside of the home, such as in the workplace. Fifth, the population studied by Hirayama was not representative of Japanese society but only of an agriculturally - 7 - ,«+w...w.:+ww...*...e,~....~....v..... ~w... :,.....,.......,.~,+_ . ~..~-~.-- .«,._..,w...

based population, which is not typical for Japan. In addition, six prefectures were chosen to participate in.the study based on the fact that they appear to have had the best conditiohs for collecting data. Hence, random sampling was not used. Sixth, the Hirayama study did not control for other risk factors associated with cardiovascular disease, such as systolic blood pressure and plasma cholesterol. Although the Hirayama study offers a],arge prospective cohort to examine the relationship between presumed exposure to ETS and ischemic heart disease, one cannot draw definitive conclusions because of the methodological problems that I have mentioned. B. Garland The next study, by Garland, also exhibits important methodological difficulties. First, Garland later reported a corrected relative risk of 2.7 (not 14.9 as reported in his 1985 publication). The p value is still less than .10 and not statistically significant. Second, after 10 years of follow-up in this study, only 19 deaths from ischemic heart disease occurred.l This small sample size is compounded by the fact that 15 of the 19 deaths occurred in non-smoking women married to husbands who had stopped smoking . .~ ..~.n....»c.~..a-...,~.,... ,.. .~

at entry. As the study did not ascertain the number of cigarettes formerly smoked per day by'the ex=smoking husbands, it is not possible to measure any sustained effects;of ETS from this former smoking group. Without more detailed characterizations of these women's exposure to ETS, it is difficult to show an association between ETS and ischemic heart disease. The Garland study does attempt to control for certain important cardiovascular confounders,'such as obesity, blood pressure, and cholesterol. The small sample size and the lack of adequate measurement of ETS from a former cigarette smoking group, however, render the results at best suggestive and certainly not definitive. Next, the work by Hole and Gillis also has several methodological problems. First, it does not have sufficient statistical power to demonstrate an association between ETS and ischemic heart disease. The sample size is too small. Second, the relative risk of 2.01 detected for ischemic heart disease for non-smokers compared with controls is too similar to the relative risk of 2.27 for active smokers compared with controls to make sense. The explanation for this is not clear, but may be due to small sample size as well. 9 _ ........... ,..d. ... ~ ..~..,...,.. ..~,, ,~.::,.-. ....... .~,. .... .,.~ . ., _ ..~