Tobacco Documents Online

A PPt-AJ ~ j X PASSIVE SMOKING MORTALITY A REVIEW AND PRELIMINARY RISK ASSESSMENT A. Judson Wells, Ph.D. Presented at the 79th Annual Meeting Air Pollution Control Association Minneapolis, Minnesota June 24, 1986 Includes minor updatino through September 16,1986

86-80.6 INTRODUCTION In 1981 when Hirayamal and Trichopoulas, et al.2, first published their studies associating passive smoking with lung cancer, and in the years immediately following, there was considerable controversy and skepticism that such an effect could indeed be real. However, as more papers appeared and many of the earlier issues were resolved, there now appears to be a growing concensus among epidemiologists that passive smoking does indeed cause lung cancer, or at least that there is a strong association. 't.awrence Garfinkel, who was one of the most quoted early skeptics, and his coworkers have recently published a paper3 In which they find a statistically significant doubling of lung cancer risks for women married to smokers of 40 or more cigarettes per day compared to women married to nonsmokers. A dose response relationship was confirmed. Also there is now little disagreement that the number of U.S. deaths from lung cancer associated with passive smoking probab l y l i es w i th i n the 500 to 5000 range suggested by Repace and Lowrey's risk assessment4. In 1983 papers started to appear associating passive smoking with deaths from other cancers, chronic bronchitis, emphysema and heart disease. These papers, like the earlier lung cancer papers, have attracted debate, but more recent papers support rather than refute the earlier ones, indicating that a consensus eventually may be reached in this broader area as well. I t i s the purpose of th i s paper to present a summary of the epidemiological literature on passive smoking for four major disease categories, namely, lung cancer, other cancer, emphysema, and heart disease, to discuss differences in disease patterns observed between direct and passive smoking and the probable reasons therefor, and to assess at least in a preliminary way the significance of these findings on expected deaths from passive smoking If the underlying epidemiological results turn out to be correct. Because of the many specialties involved in such a calculation, namely, epidemiology, statistics, population estifiates of passive smokers, aerosol deposition theory, lung structure, chemistry and carcinogenesis, it is not possible In a paper of this scope to give a detailed and sophisticated explanation of each step. Nonetheless, the estimates made Lre believed to be the most probable numbers considering the data now available although it must be realized that the confidence limits are still broad because of the uncertainties that exist at every step. The objective here is to determine the probable future extent of the public health risk from passive smoking if the trend in the epidemiological results continues along the lines it has taken since 1981. METH00§ The epidemiological literature was searched primarily through the 2

86-80.6 publications of the U.S. Government's Office of Smoking and Health5,6 which also provided a printout of all of its documents on passive smoking from 1970 through 1982. In addition many references were received directly from workers In the field. The scope of the Inquiry included all papers that contained original data on adult, nonsmoker mortality or cancer morbidity from passive smoking. All of these papers are iisted'in the subsequent tables or text. From these papers those pertaining to lung cancer, other cancer, emphysema and heart diseRse were selected for analysis and calculation of death rates. " Because some of the papers have rather glaring weaknesses, four criteria were used to admit data to the death rate calculations: 1. Retrospective studies should have controls. 2. Observations should be based on exposure beyond five years. 3. A study should not have serious Internal inconsistencies. 4. Sufficient data should be presented to allow the calculation of a variance. For the eleven studies that met these criteria a relative risk for each disease and sex was estimated (to the extent data were available) by averaging over all exposures Including exposure to ex-smokers, light, medium and heavy smokers. Combined relative risks for each disease category were calculated by a meta- analysis technique which weights the individual relative risks by the Inverse of the variances7. No other. weighting of the accepted studies was attempted. Some of the cancer papers reported morbidity relative risks rather than mortality relative risks. However, an estimate based on published data8 indicates that five year survival rates for both exposed and unexposed cases are similar. Therefore, the incidence ratios were used as good approximations to the mortality ratios. The method used to calculate possible deaths from the combined relative risks in the preliminary risk assessment Is described under that heading. RESULTS Epidemioloav There are eighteen studies now available that bear on adult mortality or cancer morbidity from passive smoking. Luna Cancer - Female. Table Ia lists the papers on female lung cancer. Shown Is the lead author on the paper, the locale of the study, the number of cases, the relative risk for the highest exposure (in most cases 20 or more cigarettes per day smoked by a spouse) and the two-tail p-value for that exposure, a relative risk for all exposures combined, including exposure to ex-smokers if those data are available, and a two-tail p-value for that exposure. A one-tail p-value for a Mantel extension test for trend is shown if available. 3

J TABLE i a RELATIVE RISKS: LUNG CANCER FROM PASSIVE SMOKING Exposed Highest Exposure All Exposures Mantel'Trend F l Locale Cases R/R 2-Tail p R/R 95% C.L. 1-Tail p ema es Cohort Studies: Hirayama20 Japan 163 1.9 0.002 1.6 1.1- 2.2 0.002 Garfinkelll U. S. 88 1.1 - 1.2 0. 8- 1.6 - Gillis, et al.13 Scotland 6 - - 1.1 0.2-5.6 - Combined Cohort 257 1.34 1.1-1.7 Case Control: Trichopculos, et al10Greece 53 2.6 0.19 2.1 1.2 - 3.6 0.005 Correa, et a1.12 Louisiana 14 3.5 0.017 2.1 0.8 - 5.2 ~ 14 Koo et a1 Hon Kon 64 - - 1.3 0.7-2.3 - w . , Sandler, et a1.15 g g N. Carolina 2 - - inf. Garfinkel, et al.3 U. S. 73 2.0 0.05 1.3 0.8-1.9 0.025 Combined C/C 206 1.54 1. 2- 2.0 Combined Cohort & C/C 463 1.44 1.2 - 1.7 Males Cohort Studies: Hirayama 20 Japan ~ 2.3 0.16 2.2 1.1- 4.8 0.023 Gillis, et a1.13 Scotland 33 07-164 . Combined Cohort 11 2.5 1.2 - 5.2 Case Control: Correa, et al. 12 Louisiana 2 - - 2.0 0.4-10.0 Combined Cohort & C/C 13 2.5 1.3 - 4.7 Ej0LStZ6

TABLE Ib RELATIVE RISKS: LUNG CANCER FROM PASSIVE SMOKING Highest d Exposure E All Exposures Mantel Trend Locale xpose Cases R/R 2-Tail p R/R 95% C.L. 1-Tail p Rejected Studies Case Control: Knoth, et a1.16 W. Germany 24 - - 2.5 1.0- 6.4 (no controls) Chan, et a1.17 Hong Kong 34 - - 0.8 0.4 - 1.3 ~ (current exp. only) Kabat and Wynderl8 Cr Females U.S. 33 - - 0.5 0.2 - 1.2 Males U.S. 20 - - 3.7 1.1-12.7 (current exp. only) Wu, et a1.19 California ? - - 1.2 ? (insufficient data) Data from Dr. Kabat re Kabat and Wynder18 Females Males Exposure Cases Controls Cases Controls Never 20 13 5 12 Work only 17 23 14 10 Home only 7 9 2 2 Work & home 9 8 4 1 Total Exposed n + 2U 13 ViaZscZb

86-80.6 There are eight studies that pass the aforementioned criteria. •Trichopoulos, et al. 10, have a numerical error In the calculation of relative risks. The values for 1-20 and 21+ cigarettes per day should be 1.95 and 2.55 instead of 2.4 and 3.4, respectively. The value in Tabiela-of 2.1 Is a weighted average of exposure to ex-smokers as well as to the two categories of smokers. Koo, et al. 14, report on exposures both at home and at work. The work exposures for nonsmokers average only 2.0 years so these data were rejected for being less than five years. Data for exposures at home and at home plus work are combined here to"*.give the relative risk of 1. 3 shown in Table Ia. The Sandler, et a1.15 , paper was directed largely at total cancer, but they did pick up two lung cancer cases among nonsmokers. These were determined to be female per private communication from Dr. Sandier. Four of the eight lung cancer studies (Trichopoulos, et al.10_, Correa, et al. 12, Koo, et al. 14, and Sandier, et al.15 ) are dise se Incidence studies; Hirayama4, Garfinkelll and Gillis, at al.~3, are mortal ity studies. Garfinkel, et ai.3 , is mixed. Since there is no reason to believe that the very low survival rate from lung cancer is significantly different for nonexposed and exposed cases, the incidence relative risks were used as mortality relative risks. There are four studies that do not pass the criteria, namely, Knoth et al.16, which is a case study with no controls, Chan, et a l. 1~, and Kabat and Mlynder 1g wh i ch are for current exposure only, and Wu, et al. 19, which does not report the number of cases and hence does not ai low the calculation of a variance. The relative risks and significance data for Knoth, Chan and Kabat and Wynder were calculated from data In the papers and from data In a private communication from Dr. Kabat. L uno Cancer - Ma l e. As shown i n Tab l e Ia there are three acceptable passive smoking studies of nonsmoking males with lung cancer. The total of exposed cases is small but the data are consistent. Combined relative risk is 2.5 and p=0.009. Inclusion of the three rejected female studies for which variances can be calculated and the one rejected male study would raise the combined male relative risk for lung cancer to about 2.7 and would lower the female relative risk to about 1.42. These changes would change the overall death date for lung cancer from passive smoking only slightly and In an upward direction. Other Cancer - Female. There are now four studies relating passive smoking to cancer other than lung or to total cancer In females (see Table IIa). This relatively new area is more important in terms of potential number of deaths from passive smoking than is lung cancer because the relative risks are In the same range while the underlying death rates for non-smokers are many times higher. The relative risk shown for Hirayama in 5

TABLE IIa RELATIVE RISKS: CANCER OTHER THAN LUNG FROM PASSIVE SMOKING Highest Exposed Exposure All Exposures Mantel Trend l Locale Cases R/R 2-Tail p R/R 95% C.L. 1-Tail p Fema es Cohort Studies: Hirayama 9 Japan 1879 1.11 0.055 1.08 1.0 - 1.2 0.05 Gillis, et al.13 Scotland 33 - - 1.2 0.6-2.5 - Combined Cohort 1912 1.1 1.0 - 1.2 Case Control: Miller 21 Penna. 66 1.9 1.1 - 3.4 et al.15 Sandler N. Carolina 113 2.0 1.3 - 3.0 rn 0 , Combined C/C 179 2.0 1.4-2.7 Combined Cohort & C/C 2091 1.56 1.3 - 1.9 Males Gillis, et a1.13 Scotland 2 0.6 0.1 - 2.7 Sandler, et a1.15 N. Carolina 5 1.5 0.4-5.7 Combined Cohort & C/C 7 0.9 9t0LSL46

TABLE IIb RELATIVE RISKS FROM PASSIVE SMOKING ISCHEMIC HEART DISEASE, EMPHYSEMA AND CHRONIC BRONCHITIS Highest osed Exposure Ex All Exposures Mantel Trend p Locale Cases R/R 2-Tail p R/R 95% C.L. 1-Tail p Heart Dis. - Female Cohort Studies: Hirayama9 Japan 240 1.3 0.038 1.16 0.9 - 1.4 0.02 Gillis, et a1.13 Scotland 19 - - 3.6 0.9-13.8 Garland, et al.22 California 17 - - 3.5 0.9-13.6 rn Combined 276 1.27 1.0 - 1.6 Heart Dis. - Male Cohort Studies: Gillis, et al.13 Scotland 14 - - 1.3 0.7 - 2.6 Svendsen, et al.23 U.S. (MRFIT) 5 - - 2.1 0.7 - 6.2 Emphysema and Chr. Bronc it s - Female Cohort Study: Hirayama9 Japan 102 1.6 0.08 1.4 0.9 - 2.1 0.05 LTOLSLZ6

86-80.6 Table IIa of 1.08 is obtained by combinfng his values for higher and lower exposures. The Mtller'1 result is for total cancer. bur interest here is in a relative risk for cancer other than lung, not total cancer. However, according to the calculation described later, less than 3% of Miller's total cancer cases should be lung cancer, and the relative risks for lung and total cancer are similar. Therefore, his total cancer relative risk Is a good approximation to the relative risk for cancer other than lung. The paper by Sandier, et al.t5 , is also directed at total cancer. Here the number of lung cancer cases is known to be two. Therefore, again the total cancer relative risk 14 a good approximation to other cancer relative risk. Other Cancer - Male. The data for cancer other than lung for males (See Table IIa) are much scarcer 13hrn for females. The results from the two available papers • are conflicting, neither result Is statistically significant nor is the combined relative risk. Therefore. a value of 1.0, indicating no association, Is assumed for the death rate caiculations until more data become available. Heart Disease - Female. There are now three papers associating passive smoking with heart disease among females (see Table IIl). Htrayama's paper9 contains data for ischemic heart disease in women by smoking habit of the husbands for two levels of exposure. The relative risk for all exposures of 1.16 Is a weighted average of the two exposure levels. Gillis, et al. 13, report data for myocardial Infarction (M1-410) and for other ischemic heart disease (1H0-411-414). These have been combined to yield the ischemic heart disease relative risk of 3.6 shown In Table IIb. Garland, et al. 22, report data for nonsmoking women who were married either to nonsmokers, ex-smokers, or current smokers. The overal t relative risk of 3.5 In Table IIb is ca l cu l ated from a we i ghted average of thei r age adjusted mortality rates for exposure to ex-smokers and smokers. Heart Disease - Male. There are two papers associating passive smoking with heart disease In males. One, shown In Table IIb is Gillis, et al. 13 Their relative risk of 1.3, although not statistically significant, is reinforced by Svendsen, et a1.23 , who found a relative risk of 2.12 (p=0.19) for the MRFIT cohort of relatively high risk individuals. Since the two papers concur and there are no negative or neutral results, the Gillis, et ai. result, 1.3, Is used as the passive smoking relative risk for male Ischemic heart disease In the general population until better data become available. ' Emphvsema and Chronic Bronchitis. Hlrayama9 has the only data on these diseases, and they are for females only. As shown in TableIIb,a relative risk of 1.4 is obtained by combining his results for high and low exposures. 7

86-80.6 Specific Cancer Sites Other than lun9. Table III shows results on passive smoking risks for specific cancer sites other than lung. It Is interesting that the sites other than lung that are normally associated with direct smoking are absent, with the possible exception of cervix. The data on breast cancer are very preliminary. Hirayama 20 reported that risk elevation for all cancer sites becomes non-significant when lung, nasal sinus, brain and breast are excluded. The Sandier, at ai. 15, relative risk is of borderline tatistical significance. It is interesting that Hirayama 2~ found stomach cancer, with 635 cases, to be specifically not associated with passive smoking. Table III. Epidemioloaical studies on aassive Specific cancer sites other than lun9. Cancer Site Nasal Sinus Brain Breast: Endocrine Cervix: Stomach Investigators Locale Hirayama 20 Hirayama 20 Japan Japan H i rayama Sand l er, Sandier, 20 et et a l. al. 15 15 Japan N. Carolina N. Carolina Buckley, et al.24 Brown, et al. 25 Sandier, et al.15 Hlrayama 20 England Canada N. Carolina Japan a Some statistical significance claimed by authors. se R1R 23 2.1a 31 4.5a - >1.0 32 2.0 13 4.4a 27 4.fia 30 3.7a 56 2.1a 635 1.0 All Causes of Death. There are five studies that contain data on passive smoking and all causes of death. The relative risks are diluted by a large number of deaths that are not related to passive smoking at all so the results are scattered and lower. The Miller study26 In 1978 is the pioneer study on mortality from passive smoking. He observed a lowering of life expectancy from 78.8 years to 74.7 years for 601 non-smoking wives whose husbands smoked compared with those whose husbands did not smoke. Gillis, et al. 13, observed a relative risk of 1.5 (81 cases, p=0.17) for f ema l es and 1.0 (28 cases ) for ma l es . Gar l and, et a l. 22 , observed a relative risk of 1.05 (57 cases) for females. Svendsen, et al. 23, observed a relative risk of 2.0 (11 cases, p= 0. 0 7) f or ma l es i n the MRF I T cohort. The on l y negat i ve relative risk comes from Vandenbroucke, et al. 27, In Holland (0.8 for females; 207 cases, p=0.12). However, this study Is flawed in that non-exposed non-smokers had a death rate 15% higher than direct smokers so it would not pass criterion number three that a study should not have a serious inconsistency. Having reviewed the mortality literatures on passive smoking, what does it ali mean? First, there is quite a bit of It. Second, taken as a whole there is growing evidence of an association 8