Tobacco Documents Online

THE RELATIONSHIP BETWEEN PASSIVE EXPOSURE TO CIGARETTE SMOKE AND CANCER Jonathan M. Samet, M.D. Associate Professor, Department of Medicine, and the New Mexico Tumor Registry, Cancer Center, University of New Mexico, Albuquerque, New Mexico 87131 INTRODUCTION Causal associations between active cigarette smoking and! cancer of the lung and other sites have been long established on the basis of extensive toxicological, experimental, and epidemiological evidence. Only recently, however, has passive exposure to tobacco smoke been considered as a potential risk factor for lung cancer in nonsmokers. This putative role of passive smoking has become an emotionally charged and highly, controversial subject with potentially important regulatory and, economic implications. Tobacco industry arguments defending the' individual's right to free choice concerning smoking would be severely damaged if passive smoking were shown to cause cancer in nonsmokers. The prevalence of passive smoking in the United States further emphasizes the potential public health consequences of this exposure. Friedman and co-workers (1) questioned 37,881 nonsmoking members of a health maintenance organization concerning passive smoking at home and elsewhere. Overall, 63 percent reported some exposure and 34.5 percent received at least 10 hours per week. Unpublished findings from an ongoing case-control study in New Mexico show that 29 percent of nonsmoking male and 56 percent of nonsmoking female controls have lived with a cigarette smoking spouse. Association between passive smoking and lung cancer derives, biological plausibility from the chemical composition of sidestream smoke, the confirmation of exposure in nonsmokers with biological ~ markers, and the failure to find a threshold for respiratory ~ carcinogenesis in active smokers. Sidestream smoke contains the ~ + ~A ~ _Cb 7~ ! .!.l.,i'f...~19, ' U:e . . ..

same toxic and tumorigenic agents as mainstream smoke; some are present in much higher concentrations because of the burning conditions under which sidestream smoke is generated (2). Investigations with markers of tobacco smoke exposure have convincingly demonstrated that passive smoking results in inhalation and absorption of sidestream smoke components (3). For example, Wald et al. (4) recently reported increased urinary cotinine levels in exposed nonsmokers and a dose-response relationship between urinary concentration and the duration of reported exposure. In Japan, Matsukura and colleagues (5) found that the presence of smokers in the home and in the workplace, and urban residence were associated with increased urinary cotinine levels. Finally, studies of active smoking have uniformly indicated excess lung cancer risks at lower levels of cigarette smoking and none have implied the presence of a threshold (2). This paper will review the epidemiological evidence relevant to the hypothesis that passive smoking causes lung cancer. First, methodological considerations relevant to studying this association will be addressed. Second, the available epidemiological evidence will be reviewed. Finally, the existing data will be assessed against conventional criteria for determining the causality of association - the same criteria, in fact, that were used in the 1964 Surgeon General's Report for evaluating the association between lung cancer and active smoking (6). METHODOLOGICAL ISSUES The association between passive smoking and lung cancer has been approached with conventional hypothesis-testing designs: the case-control and cohort studies (Tables 1 and 2). Each has well characterized advantages and disadvantages (7). The results of both may be affected by misclassification of exposure and confounding by other risk factors, whereas other types of bias uniquely influence each design. The potential for information bias, introduced by the interviewer or the subject, is of particular importance in case-control studies of this hypothesis. Misclassification of exposure refers to the incorrect categorization of actually exposed subjects as nonexposed and of nonexposed as exposed (8). When misclassification occurs randomly in relationship to the selection of a study's subjects, it reduces measures of effect towards unity; if nonrandom, it may increase or decrease effect measures. The questionnaire measures that have been employed in investigations conducted to date may have introduced random misclassification on exposure to cigarette smoke. While gas phase components may also be important for carcinogenesis, the following discussion will primarily consider cigarette smoke particulate. In :;:~

the United States, cigarette smoking is a major source of indoor respirable particulates and thus a major determinant of variation among individuals in exposure to this pollutant (9-11). Within a' room, concentrations will be determined not only by the strength of sources, such as cigarette smoking, but by building characteristics' and ventilation rate (9). Time-activity patterns further modify the profile of exposure (11). Thus, with regard to domestic exposure, simple descriptions of spouse smoking behavior cannot satisfactorily define gradients of exposure. They can, however, document that exposure to tobacco smoke has occurred. Similar limitations apply to questionnaire derived indices of workplace exposure. With regard to total passive exposure to tobacco smoke, variables that do not include time outside of the home will lead to misclassification. In the population studied by Friedman et al. (1), high proportions of nonsmoking males and females reported exposure outside of the home. Workplace exposure was associated with higher urinary cotinine levels in the recent report from Japan by Matsukura et al. (5). Thus, random misclassification of exposure is likely with questionnaire indices. Studies that have used such measures may be conservative since random misclassification reduces effect measures toward unity. REVIEW OF THE EVIDENCE Evidence concerning passive smoking and lung cancer has been sought indirectly in descriptive data and directly with case-control and cohort studies. Time-trends of lung cancer mortality in nonsmokers have been examined with the rationale that increasing passive smoking should be mirrored by increasing mortality rates. Enstrom (12) calculated lung cancer mortality rates from various nationwide sources for the period 1914-1968 and concluded that a real increase had occurred among males after 1935. In contrast, Garfinkel (13) did not identify time trends in nonsmokers in the Dorn Study of Veterans, 1954 to 1969, or in the American Cancer Society study, 1960 to 1972. In a large autopsy series, Auerbach and colleagues (14) did not find increased abnormalities in the bronchial epithelium of male nonsmokers deceased in 1970-1977 in comparison with those deceased in 1955-1960. While this review emphasizes lung cancer, associations of passive smoking with cancers of other sites or with other diseases would strengthen the evidence concerning passive smoking and lung cancer. An investigation of all cancer deaths in females residing in Western Pennsylvania has been frequently cited as showing an adverse effect of passive smoking (15). Miller interviewed surviving relatives of 537 deceased nonsmoking women concerning the smoking habits of their husbands. A significantly increased relative risk of cancer death was found in the women who were not employed outside of their homes. The large number of potential "".""^'N!~l~~~qqqqf/At~n+1n77ifjNSflfY,f!Mi»f i1+fi7t7~7i7!7p1MilH , ,. , • :,

subjects that were not interviewed and the possibility of information bias detract from this report. Gillis et al. (16) followed 16,171 healthy Scottish individuals, ages 45 to 64 years, over at least a 6 year period. In a preliminary report concerning 8,128 subjects, all-cause mortality was comparable in nonsmoking males with and without domestic tobacco smoke exposure, but was increased by nearly 50 percent in exposed nonsmoking women. A case-control study of 438 cancer cases involving multiple sites and 470 controls showed increased relative risks from exposure during childhood and during adulthood (17). In a 25-year cohort study in Amsterdam, all cause mortality in females was not affected by the husbands' smoking status (18). More relevant is the direct hypothesis-testing evidence provided by case-control and cohort studies. In 1981, two papers were published which reported significantly increased risks of lung cancer in nonsmoking women whose husbands smoked cigarettes (Table 1). Hirayama (19) conducted a prospective cohort study of 91,540 nonsmoking women in Japan. Standardized mortality ratios for lung cancer increased significantly with the amount smoked by the husbands. The findings were unchanged with control of potentially confounding variables and with extension of follow-up from 14 years to 16 years (20). Overall, the relative risk from passive exposure was 1.8 whereas that from active smoking was 3.8. Hirayama has also reported a significantly elevated relative risk (2.94) in nonsmoking men with smoking wives (21). Following its publication, this article received intensive scrutiny and correspondence in the British Medical Journal offered concerns about statistical methodology, about population selection, about uncontrolled confounding by factors such as cooking fuel e=posure. and socioeconomic status, and about the seemingly high relative risk. In his responses, Hirayama satisfactorily rebuffed most of these criticisms; in particular, confounding did not appear to explain the findings though active smoking by reportedly nonsmoking women can not be excluded. In this regard, Hirayama (20) has reported that the findings after 16 years of follow-up are consistent with effects of passive smoking on mortality from emphysema and chronic bronchitis, nasal sinus cancer, and ischemic heart disease. Biologically, these effects seem somewhat less plausible than lung cancer and these new associations raise concern about confounding by unreported active smoking. Hirayama has explained the level of relative risk by the low percentages of women working outside the home in Japan, low divorce rates, small room sizes, and lack of inhibition about smoking in the presence of nonsmokers (21). No data concerning respirable particulate levels in the subjects' homes have been provided, however. Also reported in 1981 were the results of a case-control study in Athens, Greece (22) (Table 1). Female lung cancer cases with a diagnosis other than adenocarcinoma or bronchioloalveolar carcinoma

were identified at three large hospitals and controls were selected at a hospital for orthopedic disorders. All subjects were interviewed by the same physician and their smoking status and that of their husbands was obtained. Single women were considered as married to nonsmokers and changes in marital status were considered. The final series included 40 nonsmoking cases and 149 nonsmoking controls. A significant trend of increasing risk with presumed extent of passive exposure was present when either the husbands' current or lifetime smoking habits were used for stratification. The findings were unchanged when the series was expanded to 77 cases and 225 controls (23). Less criticism has been published concerning the Greek study than concerning Hirayama's investigation in Japan. As discussed by Kabat and Wynder (24), the attempt to restrict the case series to histologies other than adenocarcinoma appears premature at present. Further, the diagnosis of lung cancer was made without histological or cytological confirmation in 35 percent of the cases. Noncomparability of the case and control series must also be considered when they are ascertained at different institutions; in this context, Trichopoulos et al. did demonstrate comparability of the case and control series for key demographic variables. The possibility of information bias must be raised because case and controls were interviewed by a single physician who may have been aware of the study's hypotheses. Finally, the investigators assessed the statistical significance of their findings with a chi-square for trend in proportions. The assumption that a former smoking husband provided an exposure intermediate between that of a nonsmoker and a current -smoker was not justified by the authors. However, the odds ratio is significantly elevated for the stratum with the highest level of current smoking. The results of another case-control study, published in 1983, also demonstrated a significant association between passive smoking and lung cancer risk (25) (Table 1). Correa et al. obtained information about the smoking habits of the parents and spouses of eight male and 22 female nonsmoking lung cancer cases and of 313 controls. Lung cancer risk increased with the spouses' lifetime cigarette consumption. Maternal smoking was associated with a significantly increased odds ratio in active smokers but not in nonsmokers. On stratification by sex, the increase was statistically significant only in males. The relatively small numbers of subjects in this investigation mandate caution in interpreting its results. However, the overall findings were unchanged, as reported in a recent abstract, when these data were combined with comparable information from two other case-control studies (26). The overall design wa s appropriate but information bias may affect the results of case-control studies that rely on interview for exposure information. The exposure variable, cumulative cigarette consumption, differs from the

measures used by Hirayama (19) and by Trichopoulos et al. (22). It would be useful to reanalyze these data with comparable exposure variables. The results of two other investigations have also been interpreted as showing an increased lung cancer risk associated with passive smoking. In Germany, Knoth et al. (27) accumulated a series of 792 lung cancer cases of which 59 were in females. Thirty-nine of these women had not smoked but 24 of the nonsmokers had lived in households with smokers. Because the investigators did not interview a control series, they relied on census statistics to estimate the anticipated proportion of smoking spouses in the general population. In the age group 50 to 69 years corresponding to the husbands of most patients, the census showed only 22.4 percent currently smoking. In another recent report, Gillis et al. (16) described the results of a cohort study of 16, 171 males and females in Western Scotland (Table 2). Exposure to tobacco smoke in the environment was characterized by four strata: nonsmoker and not domestically exposed, nonsmoker and domestically exposed, smoker and not domestically exposed, and both a smoker and domestically exposed. Mortality rates for lung cancer and for all other cancers were calculated separately for males and females within each stratum. Among males, six lung cancer deaths were observed in nonsmokers; in the control stratum, the annual mortality rate was 4 per 100,000 whereas in the domestically exposed nonsmokers the rate was 13 per 100,000. For males the rates were similar in the two actively smoking groups. In females, with a total of eight deaths from lung cancer in nonsmokers, the variation of mortality rates did not suggest an adverse effect of domestic tobacco smoke exposure. The methodological limitations of these two studies are evident; neither formally tests for association between lung cancer risk and passive smoke exposure. The German report did not involve a comparison series and the appropriateness of substituting census data was not addressed (27). The authors did not formally test for association between passive smoking and lung cancer; in fact, they used their sparse data as a platform for discussing social and political aspects of passive smoking. Interpretation of the Scottish investigation is constrained by the small number of deaths; in this regard, statistical significance testing was not performed (16). The lack of effect of domestic tobacco smoke exposure in females is not consistent with earlier reports (Table 1) but the number of deaths is quite small at present. The results of four other investigations suggest lesser or no effects of passive tobacco smoke exposure (Table 2). Chan et al. (28,29) performed a case-control study in Hong Kong that included 84 nonsmoking female cases with 139 controls. Apparently a single question was asked concerning passive smoking exposure. In a 1979 report, the investigators stated that 40 percent of cases and 47 .Dr W

percent of controls (estimated odds ratio of 0.75) replied affirmatively to a question concerning exposure at home or at work (28). In a 1982 publication; similar findings were reported but the exposure variable was described as related to spouse smoking (29). The conflicting description of this investigation's exposure variable requires clarification. Noncomparability of the case and control series with regard to place of residence and lack of histological or cytological confirmation in 18 percent of cases further limit this investigation. A more recent case-control study from Hong Kong also did not show definite effects of passive smoking. Koo et al. (30) interviewed 200 cases ascertained through Hong Kong health facilities and 200 controls, selected from the general population to match the age, socioeconomic, and geographic distribution of the cases (Table 2). With women not exposed to smoke at home or at work as the reference category, odds ratios for exposure at home and at work were not significantly increased. Nonsmoking cases had fewer hours of total estimated exposure than controls. In contrast to the case-control study in Louisiana (25), an effect of maternal smoking was not found. The most important of the four publications, construed by many as negative, is based on the American Cancer Society's prospective cohort study (13) (Table 2). Between 1959 and 1960, 375,000 female nonsmokers were enrolled and follow-up of mortality lasted through 1972. From this cohort, Garfinkel identified 176,739 nonsmokers whose husbands had never smoked or were current smokers, presumably on enrollment. The standardized mortality ratios for the women with smoking husbands were greater than unity but not significantly. In the smoking-exposed group, there was no evidence for a dose-response relationship. A separate matched analysis, performed to more completely control confounding, provided similar results. The American Cancer Society study should not be characterized as contradictory to the findings of Hirayama (19), Trichopoulos et al. (22), and Correa et al. (25). First, the standardized mortality ratios are above unity for the exposed groups. Second, confidence intervals for the mortality ratios in the American Cancer Society study overlap those reported by Hirayama (19). Third, while each of these investigations employed spouse smoking as the exposure variable, the comparability of dose among the four is uncertain. Repace (31) has suggested that the mortality ratio in the American Cancer Society study has been reduced by misclassification introduced by workplace exposures. His arguments lead to an adjusted mortality ratio of 1.7 for the American Cancer Society cohort. Finally, the use of death certificates to establish diagnosis in the American Cancer Society study probably introduced misclassification of disease status. IMT l.;. . . . ~ .. . ,~, . .. . . , . ,.. ~i ..: ... . . . . . ..... . . . . . .... .~:'r' --- - - -

Recent and preliminary results from a nationwide case-control study also did not demonstrate increased lung cancer risk from domestic exposure to tobacco smoke (24) (Table 2). Kabat and Wynder examined the effects of currently smoking family members and of current exposure at work in 25 nonsmoking male and 53 nonsmoking female cases with equal numbers of controls. For men, the odds ratio for workplace exposure of 2.6 was significantly increased. Current domestic exposure was not significant for males or females. In a smaller subset of cases, adverse effects of spouse smoking ~ were not identified. The authors clearly stated that their results were preliminary and that more data are needed. While the numbers are small, they are equivalent to those in the series reported by Correa et al. (25). CONCLUSIONS In summary, at present, only nine published investigations provide data directly relevant to the hypothesis that passive smoking is a risk factor for lung cancer. Several others offer indirect evidence. This paucity of data contrasts sharply with the literature cited in the 1964 Surgeon General's Report which characterized active cigarette smoking as a cause of lung cancer (6). That report reviewed 29 case-control and seven cohort studies. Their results uniformly and unequivocally demonstrated the association between active smoking and lung cancer. Application of carefully considered criteria for causality to the evidence led to the designation of cigarette smoking as causally related to lung cancer in men. The association was judged on its consistency, strength, specificity, temporal relationship and coherence. The report did not explicitly define "cause" but indicated that the term is generally applied to "... a significant effectual relationship between an agent and an associated disorder or disease in the host". It also acknowledged the multifactorial etiology of lung cancer and did not require a unique relationship between smoking and malignancy. Application of these same criteria to the data for passive smoking highlights their weaknesses. With regard to consistency, the conflicts among the published investigations are immediately evident (Tables 1 and 2). However, because of potential differences in dose among the investigations, it is not certain that each has tested for a common magnitude of effect. Furthermore, given the small numbers of cases in most of the papers, the point estimates of effect are unstable and confidence limits generally overlap from one study to another. In the positive studies, the relative risk estimates have indicated relatively modest effect levels, ranging from about two to three. ~ These values are much lower than those associated with smoking and could more readily be the consequence of bias. active In the ~ face of small and conflicting studies, unidentified sources of bias N o,,•,aqa,:n~N1yNryN11t~N87N1i1~1ifi111~A~l1ljlt~7l~~~~}S~l~ .

.?~~: . , . . should not be readily dismissed as an explanation for significant but modest elevations of risk. Specificity of association, that is a unique relationship between the factor and the disease, is an irrelevant and unimportant criterion for passive smoking. With regard to the temporal association of passive smoking and lung cancer, the directionality is unquestionably appropriate; exposure precedes the development of the disease. The remaining criterion is the coherence of the association. The biological plausibility of the association between passive smoking and lung cancer has been previously reviewed and this criterion appears to be met. In conclusion, the association between passive smoking and lung cancer does not yet meet criteria applied to active smoking in the 1964 Surgeon General's Report. While confirmation of passive smoking as a risk factor for lung cancer would offer new ammunition against tobacco, the available evidence does not permit definitive judgments. In the face of difficult methodological problems, particularly that of accurately quantifying dose, unimpeachable data will be difficult to obtain. New approaches for studying passive smoking and lung cancer are clearly needed. The problems of dose estimation seem more difficult for lung cancer than for other putative health effects of passive smoking. The relevant exposures may begin at birth and occur under a wide variety of circumstances. Historical reconstruction of exposures by questionnaire may be the only available approach for epidemiological studies. However, further validation of the questionnaire approach is needed with comparisons against biological markers and measured concentrations of tobacco smoke components. The reliability of questionnaire assessment of passive smoke exposure has not been established nor have sources of bias been evaluated. Interviews with next-of-kin may be particularly prone to information bias, almost certainly in the direction of overreporting. In fact, as the public becomes increasingly aware of and sensitized to potential effects of passive smoking, the results of case-control studies will become increasingly difficult to interpret. Unfortunately, the case-control design is the most efficient approach for investigating the relatively small number of lung cancer cases in nonsmokers. Cohort studies, which might offer better exposure data, must involve large numbers of subjects and lengthy follow-up. Investigative approaches which examine outcomes other than lung cancer might provide more immediate answers concerning passive smoking and respiratory tract carcinogenesis. For example, sputum cytology might be evaluated in nonsmokers in relation to passive tobacco smoke exposure. While additional investigations will certainly be performed, the available data may already be satisfactory for both regulation and prevention. For regulatory purposes, the established carcino- genicity of tobacco smoke and the high prevalence of exposure ~sn.~o~•,~uarc".:v~~iUiN7s7~1if{Rn7181i~~

;ai,:3iiiit3~3Ldl~ii~ should be sufficient to prompt action. For prevention, the data on active smoking should be sufficient; smoking prevention and cessation remain the best strategies for minimizing passive exposure. ACKNOWLEDGEMENTS Supported in part by a grant from the National Cancer Institute CA 27187. Dr. Samet is recipient of a Research Career Development Award 5 KO 4 HL00951. The author thanks Dr. A. Judson Wells for his helpful comments and Lee Fernando for preparing the manuscript. REFERENCES 1. Friedman, G. D., Petitti, D. B., and Bawol, R. D. "Prevalence and Correlates of Passive Smoking," Am. J. Public Health 73:401-405 (1983). 2. U.S. Public Health Service. "The Health Consequences of . Smoking. Cancer. A Report of the Surgeon General," (Rockville, Maryland: U.S. Department of Health and Human Services; Public Health Service, 1982). Jarvis, M. J., and Russell, M. A. H. "Measurement and Estimation of Smoke Dosage to Non-smokers from Environmental Tobacco Smoke," Eur. J. Respir. Dis. 65 (Supplement 133):68-75 (1984). 4. Wald, N. J., Boreham, J., Bailey, A., Ritchie, C., Haddow, J. E., and Knight, G. "Urinary Cotinine as Marker of Breathing Other People's Tobacco Smoke (letter)," Lancet 1:230-231 (1984). 5. Matsukura, S., Taminato, T., and Kitano, N., et al. "Effects of Environmental Tobacco Smoke in Urinary Cotinine Excretion in Nonsmokers. Evidence for Passive Smoking," N. Engl. J. Med., 311:828-832 (1984). 6. U.S. Public Health Service. "Smoking and health. Report of . the Advisory Committee to the Surgeon General of the Public Health Service," (Washington, DC: U.S. Department of Health, Education, and Welfare, Public Health Service, Center for Disease Control, PHS Publication No. 1103, 1964). MacMahon, B., and Pugh, T. F. "Epidemiology. Principles and Methods," (Boston: Little, Brown, and Company, 1970). m .I ~ .L1 .I