Tobacco Documents Online

se.,ww-.,l r11CAMdazy. vOe. , O PaI H - tlos Dtvlim S.bpr 22<563001 ~ V I~ ~. r_~. _ I~191 I~PI ~ hl~ II~I I~I ®' Il~ il~ Dll' ~WESSES IN RECENT RISK ASSESSMENTS OF ENVIRONMENTAL TOBACCO SMOKE ~ f P.N. LE E & Cuting~ ' J Cadar R ed,SuttonbSwVwy SM2 SDe~'U.K. (Recewed 25 June 1990, Accepted 11 January 1991) Epidemiological evidence of increased Iting cancer riitk in never smokers married to smokers has been used' to estimate annus] deaths from environment.al'tobacec smoke (ETS) exposure. Such estimates are very much higher than those based on dosimettic considerations and misltadingly ignore major weaknesses in the epidemiology. Some authors overestimate total lung cancers occurring in never smokers. There is no scientific basis for extending risk auaessments to include deaths from other causes, from workplace exposure to ETS, and among ex-smokera. Recent rti.k assessments by VGe11L, by Repace and Lowrey, and by Kawachi and' colleagues are given particuli<r attention. WTROnuCTIOIv In 1986 four authorities revi'ewed the evidence on the relationship of environmental tobacco smoke (ETS) and' health (11-4). There was agreement that there was inadequate evidence to determine whether ETS caused heart disease or cancers other than the lung. With regard to lung cancer views were more conflicting. The International Agency for Research on Cancer (1), while noting that "several epidemiological studies have reported an increased' risk of lung cancer in non-smoking spouses of smokers" pointed to "substantial difiticulties" and errors that "could arguably have artefactually depressed or raised estimated risks" so that each study "is compatible either with an increase or with an absence of risk": The Australian National Health and Medical Research Council (2) noted that "the evidence that passive smoking causes lung cancer is strongiy' suggestive" and, although pointing to difficulties in many studies that "preclude a conclusive interpretation", stated that "passive smoking gives rise to some risk of cancer". The US Surgeon General (3) concluded that "involuntary smoking is a cause of lung cancer" but that quantification of the risk for the US population •'is dependent on a number of factors for which only a limited amount of data are currently avaidable". The US National Research Council (4)' noted that a"summary' estimate from epidemiological studies places the 1q.1 increased risk of lung cancer in non-smokers married to smokers compared with non-smokers married to non-smokers at about 34`b" and considered' that, though "to some extent, misclassification bias may have contributed to the results reported i!m the epidemiological literature"; the "bias is not likely to account' for all of the increased risk•'. Although one of the four authorities felt it premature to conclude cause and effect, and two who thought cause and effect could be concludedd felt it could not be quantif ed, there has been an increasing tendency to carry out risk assessments to estimate annual numbers of deaths due to ETS. The purpose of this paper is to underline a number of problems in conducting such risk assessments, and' to comment critically on three that have recently been published. The first, by Wells (5), estimated that annually in the United' States 46,000 deaths per year occurred among non-smokers (i.e. never plus ex-smokers combined)I due to ETS exposure at home and at work. 3,000, were from llung cancer, 1'1,000 from other cancer and 32,000 from heart disease. The second, by Kawachi and colleagues (6), estimated that annually in, New Zealand 2?$' deaths per year occurred among never smokers, 30 from lung cancer and 243 from ischaemic heart disease: 95 deaths were from at home expo.ure and 1178 from at work exposure. The third nsk assessmenti, by Repace and Lowrey (7), was based on a review of nine

n other risk assessments for lung cancer. They noted that "excluding one study whose estimate differs from the mean of the others by two orders of magnitude, the remaining risk assessments are in remarkable agreement. The mean estimate is approximately 5000 ± 2400 non- smokers' lung cancer deaths per year in the US". This paper starts by discussing risk assessment for lung cancer among never smokers based on epidemiological data in relation to spouse smoking, this being the area most intensively studied. Following this problems resulting from extending the risk assessment to cover other diseases are discussed, as are those caused by considering workplace as well as at home ETS exposure. Finally, some other issues are considered. LUNG CANCER L*71dEVFR SMOKERS IN RELATION TO ETS EXP06U'RE FROM TFE SPOUSE An up-to-date review of the evidence (8)' shows there are 27 epidemiolbgical studies of lung cancer (involving nine or more cases) in which risk in never smokers could be related to the smoking status of the spouse (or in five studies to an alternative index of at-home exposure): Eleven studies were conducte& in the US, eceven in Asia and five in Europe, involving a total of 2350 lung cancer cases with relevant data, 90% of these being females. 26 of the 27 studies provide estimates of relative risk in relation to this index of ETS exposure for females; values razp from just under 1.0 to just over 2.0. Five are statistically significantly positive and 20 estiitates are greater than 1.0: Taken as a whole thedata show a positive rei'ationship - the median is.aoout 1.25. Based on 17 of these studies, using fornal rneta-analytic techniques which weighted stufies on sampie size but not on quality of evttbnce, Wells (5) gave an average relative risk of 1144, with 95% confidence limits 1.26 to 1.66. The data for males are more variable, being bswd on 11 studies often with small numbers of Agitfis. Seven relative risks were greater than 1, aoteesignificantly so, with one equall to 1 and three isessthan 1. The median is similar to that for :€earaies. The epidemiologicali evidence has been used for the risk assessments of Wells (5) and liiaeachi et at (6). It has also been used for a ntm,ber of the risk assessments cited by Repace an& Lowrey (7). This is only valid if the epidimiological' evidence itself is sound and not subjw--tv, material bias. In order to investigate this issue, two questions will be addressed; first, "Ils the magnitude of the risk plausible based on what is known about the extent of exposure?" and' ^econd, Are there weaknesses and sources of bias in the epidemiology which could invalidate the approach?' Dosimetric considerations. If lung cancer risk, relative to a non ETS exposed never smoker, is RE in an ETS exposed never smoker and RS in an ever smoker, then the ratio of excess risks X=(RE-1)I(RS-1) is an indicator of the relative effects of ETS exposure alone and of smoking. Since risk associated with smoking is approximately proportional to number of cigarettes smoked, one might expect, were the epidemiology unbiassed, that X would be similar to the ratio of relevant smoke constituents from ETS exposure and from smoking. Table 11 shows,, in rank order, estimates of X based on data for 1'8 studies in females and 7 studies in males. In females, almost half (8/18)', of estimates are 0.2 or greater with the median value 0.14. For males, the results vary more and are base& on many less data points, but the conclusions to be drawn are similar - namely that the epidemiological evidence, if unbiassed, suggests that the extent of exposure from ETS (from spousal smoking) is something like 110-20R6 of that from active smoking. It is clear the ratio of exposure from ETS and' exposure from active smoking is much lower than 10•20°'ro for those smoke constituents that are commonly used as markers. In a large nationally representative study in the Uh (27); mean salivary cotinine levels in non-smokers married to non-smokers, in non-smokers married to smokers and in smokers were respectively 1.22; 3.78 and 331 ng/mli in males and 0.76, 2.21 and 328 ng/ml in females, giving a relative exposure for ETS to active smoking of 0.8% in males and 0.4% in females. Repace and Lowrey (7) give a slightly higher figure, notrng that non-smokers have of "the order of 1% of nicotine uptake of smokers" but it is still an order of magnitude less than the 10-20% one requires to align with the epidemiology. Differences in clearance rates of cotinine reported between non-smokers and smokers are too small to affect this gross discrepancy materially; in any case, since the half-life seems to be longer in non-smokers it would increase the discrepancy (28); not reduce it as Repace and Lowrey (7) claim. 194

TABLE 1 Comparability of relative riska due to ETS exposure (from spouse) and active smoking Study ( r e f) Sex RE' RS" X+ Inoue (9) Female 2.55 4.25 0.48 Geng (10), 2.16 4.18 0.36 Trichopoulos (11Y 2.08 4.37 0M Akiba (12 Y 1.52' 3.24 0:23 Brownson (13), 1.82 4.75 0.22 Koo (14) 1.55 3.56 0.21 Lam 1 (15), 2.01 5.94 0.20 Hole (16), 1.89 5.43 0.20 Lam 2 (17) 1.65 4.97 0.16 Hirayama (18) 1.38 4.12 0.12 Gao (19) 1.19 3.15 0.09 Wu (20) 1.20 3.31 0.09 Correa (21) 2.07 14.10 0.08 Humble (22) 2.34 28.53 0.05 Svensson (23) 1.26 7.17 0.04 Lee (24) 1.03 4.70 0.01 Buffler (25) 0.80 5.91 -0.04 Chan (26) 0:75 3.07 -0.12 Akiba (12) Male 2.10 3.21 0.50 Hirayama (18) 2.34 4.39 0.40 Hole (16) 3.52 15.88 0.17 Humble (22) 4.19 29.36 0.11 Correa (21) 1.97 30.15 0.03 Lee (24) 1.31 12.02 0.03 Buffler (25) 0.51 7.03 -0.08 ' Risk of ETS exposed never smoker relative to non ETS exposed never smoker " Risk of ever smoker relative to non ETS exposed never smoker + Ratio of excess risks, e.g.for first study 0:48 :(2.55-1)/(4.25-1). N.B'. Risks given are unstandardised for age since stand'ardise& estimates were not available in many studies and generally differed little from unstandardised estimates where both were available. Estimates of relative exposure base& on inhaled smoking-related particulates show an even greater discrepancy: Arund'el et al' (29) have estimated that for the US average daily inhale& particulate ETS exposure for all never smokers is 0.62 mg/day for men and 0.2$' mg/day for women as against 387 mg for men and 311 mg for women who currently smoke. Since ETS exposure of exposed non-smokers is about 3 times that of all non-smokers (27); one can~ calculate that the ratio of average exposure for ETS to active smoking is about 0.4% in men and 0.2% in women, similar to an estimate of 0.3% given by Repace and Lowrey (7)'based an their own work. Arundel et al (29) pointed out retention of smoking related particulates is much higher in smokers (80^a) than iin, non-smokers(117c)'. They estimated a relative exposure for ETS to active smoking of around 0.03-0.04% (29)., Using radiotracer techniques, a similar, very low ratio of 0.02% has been estimated based on particulate deposition in the trachea-bronchial region (30). While both ETS and mainstream smoke eontain a wide variety of chemicals, and relative exposure of passive and active smokers will vary quite widely according to which chemical is used as the marker - the factor being higher for vapour phase than for particulate phase compounds (311) - there is certainly strong evidence of a marked discrepancy between the epidemiology and dosimetry. Indeed, since it is commonly believed lung cancer in smokers is associated with depositiom of particulate matter in the lung - the basis of "tar" reduction programmes - the discrepancy seems very large, by two or even three orders of magnitude. One implication is that risk assessments based on dosimetric evidence are likely to give substantially lower estimates than those based on the epidemiological evidence. Another impli!- cation is that it gives reason to doubt the epi- demtology, and to lbok for sources of bias. 195

Risk assessments based on dasimstry versus those based on epidemiology. Let us consider the situation with regard to the three risk assessment papers which are being studied in detail. A1T three have different approaches. Kawachi et a! takes the epidemiology at face value and do not attempt risk assessment based on dosimetrie evidence (except vide infra to adjust relative risks for at home exposure to those for at work exposure). The discrepancy between the dosimetry and the epidemiology is not evem mentioned'.. Wells (5) also bases his risk assessment on, the epidemiology. However he does note that the mortality observed for passive smoking is "rather high"' relative to the deposited dose of particulate, contrasting relative factors for passive to active smokers of 0.25% for "smoke retention" (Arundel's figures cited above suggest 0.03-0.04%) and' 2.9% for lung cancer (Table 1 suggests 110-2D%): He believes the differences are due to differences in chemistry and' physics between active and passive smoking, and' essentially does not doubt the validity of the epidemiology. Repace and' Lowrey (7) review risk assess- ments based both on dosimetric and epidemi- ologieal' evidence. While this should have revealed major differences between estimates based on the two methods of risk assessment they in fact. claim "remarkable agreement"'. There are many reasons for this erroneous conclusion: i) They rejected the estimate of Arundel et a[ (29), based on retained particulate matter, because iti differs from the mean of the others by two orders of magnitude. i i) They misquote Robins' work in the NRC report (4): They cite his estimates of 2500- 5200 US deaths in lifelong non-smokers per year from passive smoking as being dosimetrically based when in fact they clearly are epidemiologically based. Robins also provides much liower estimates of 45-396 deaths based' on respirable suspended particulates, but Repace an&Lowrey totally ignore these. iii) They quote an early paper by Fong (32), which assumed that the extent of' exposure from ETS was of order 2% to 8~'M that from active smoking, a relative factor far higher than indicated! by the more recent data summarized in the previous section. iv) They amit their own dosimetrically based estimate because it is 'inconsistent with the epidemiology of passive smoking . It is hardly surprising they get "remarkable agreement" if they reject estimates that do not agree! Table 2 presents the various estimates for the studies reviewed by Repace and Lowrey (7). The epidemiologically based estimates are reason- ably consistent and high. The dosimetrically based estimates are much lower. How much lower depends on the smoke constituent used for extrapolation, Weaknesses of the epidemiology. Epidemiology is imprecise. Various sources of bias can produce spurious relative nsks of 2 or even more (38). Since the relative risks seen for ETS exposure are well within this range, and since they seem inconsistent with the dosimetric evidence, it is important to examine the epi- demiolbgical evidence critically. Six potential sources of bias are considered below: Misclassification of di agnosis. Of the 27 epidemiological studies of' ETS and lung cancer, three were prospective and based diagnosis on d'eath certif cates, and only 15 used only (or virtually only) histologically confirmed cases. Faccini,(39) has discussed the dangers of misdiagnosis, particularly of primary breast cancer as lung adenocarcinoma. The magnitude and extent of bias from this source is, however, unclear. Random misdiagnosis would tend to reduce the relative risk„ but differential misdiagnosis might increase it. In theory differential misdiagnosis might occur if a nsk factor for the misdiagnosed disease is correl'ated with ETS exposure, or if knowledge of ETS exposure by the doctor affects diagnostic procedures, but there is no direct evidence of this. N Misclassification of ETS exposure. N None of the studies had any objective measure C4 of ETS exposure, either from ambient air ul measurements in the home or workplace or from ~ measurements of levels of smoke constituents in N body fluids. All information came from C questionnaires. While random miscl'assifi- cation of exposure will~ tend to dil'ute ~ associations, it is possible that in case-control, studies some recall bias might have occurred, with cases overestimating exposure relatively to 196

controls in an attempt to rationalize their disease. This would probably have been less important for relatively "hard"' questions such as those relating to whether the spouse smoked than for more "soft" questions on extent of exposure. Publication bias. There is strong reason to believe (40) that scientists are less likely to submiti, and journals less likely to accept, papers showing no association than those showing a positive association. If so, published evidence tends to overestimate the true association of a factor with a disease. Since ETS has been the subject of much attention in recent years and since a relati,vely large number of unpublished null studies would be needed to counterbalance the high proportion of studies of spouse smoking and lung cancer showing a positive association, it would seem unlikely non-reporting bias could fully explain the overall positive relative risk. However the fact that the studies showing the highest relative risk are based' on significantly smaller numbers of cases than the studies showing the lowest relative risks (8) is consistent with the notion that small null studies do not get published', and suggests some publication bias exists. TABLE 2 Estimated number of lung cancer deaths occurring in US never smokers from ETS exposure in 1988 (adapted from Repace and Lowrey (7)). Study (ref) Metho& of estimation~ Estimate ' Wald' (33) Epidemiological 5210 Repace & Lowrey (34) Phenomenological " 4310 Robins (4) Epid'emi'ological 4150 Wigle (35) Epidemiological 3650 Kuller (36) Epidemiological 3500 Wells (5), Epidemiological 21,30 Fong (32) Dosimetric - 2% to 8% of effect 1860 Russell (37) Dosimetric - nicotine 710 Repace & Lowrey (34) Dosimetric-respirabiesuspended 490 Robins (4), particulates Dosimetric + 240 Arunde) (29) Dosimetric - retained particulate 40 m atter • As given in (7); rounded, or converted' from estimate for nonsmokers. Dosimetric estimate for .a Robins study added. Based' on comparison of' lung cancer rates in never smoking SDAs (Seventh Day Adventists) and non SDAs (uncorrected for numerous lifestyle factors on which SDAs and non SDAs are known to differ). + Assuming a non-exposed non-smoker inhales the equivalent of 0,01 cigarettes per day. Robins gives 0.0001-0.005 cigarettes per day for the equivalent in terms of respirable suspended particulates. Poor design of some studies. Of the 27 studies which provided information on ETS and lung cancer, 24 were of case-control design. There were clear weaknesses in d'esign in a number of the case-control studies. One study (10) did not even state what the control group was. Four studies (9;, 12, 211, 25) included some patients or d'ecedents with smoking associated diseases in their control group. More seriouslyy there were systematic differences in studyy procedure between cases and controls in a number of studies., In three studies where the case might have been alive or dead (13, 22, 41) the controls were not matched om vital status. Two studies (11„ 15), used cases and controls from different hospitals.. Two studies (17, 23) interviewed cases in hospital and some or all, controls elsewhere. In three studies (13, 21, 22) the proportion of next-of-kin respondents was substantially higher for cases than controls. Although difficult to quantify the effect of such procedural differences it is notable that for females the observed relative risk in the eight studies showing differences was higher (median 1.9) than in the 17 studies where like was being compared with like (median 1.2. p on rank test <0.05). It is also worth noting that three studies (12, 25, 42) obtained a high, proportion of responses from next-of-kin and that in one of'these (42), no association betweeni lung cancer risk and spouse smoking was seen when the P"

subject herself reported the information, but a 3- `old relative risk was seen when the infurmation as obtained from a daughter or a son. Confounding. There were 22 studies in which the index of ETS exposure used was smoking by the husband. One woul'd' have thought that the standard procedure would have been to present an age- adjusted comparison of married never smoking women whose husbands were non-smokers with married never smoking women whose husbands were smokers, and to also present a relative risk adjusted' further for other potentially confounding factors known to affect risk of lung cancer. It was clear this standard procedure was not kept to: About half of these studies included unmarried' women in their non-exposed group so that there was a confounding betweem marital status and! ETS exposure. Three of the 22'studies (11, 15, 43) and also one of the other five (26) did not adjust for age at all while in three others (10, 17, 21), although cases and controls were age-matched' initially, the error was made of failing to age adjust after the never smokers were selected out.. Almost half the studies failed to take into account any other confounding factors and of the :mainder most' looked at only quite a limi2ed number of possible such factors. Those few studies which looked at a reasonable number of confounders were generally those where no significant effect of ETS exposure had been seem anyway._ Koo (44) compared' never smoking women whose husbands did or did not smoke on a wide range of factors and found that those whose husbands did not smoke were "better off in terms of socio-economic status, more conscientious housewives, ate better diets, and had better indices of family cohesiveness". Miscliassification. It is amply documented that active smoking is positively associated with lung cancer and also that smokers tend preferentially to marry smokers more often than would be expected by chance. As a result, even if ETS had no effect whatsoever on lung cancer risk, a spurious positive association, between spouse smoking and lung cancer nsk will be seen if a proportion of ever smokers are misclassified as never smokers (2-,). The relationship between the magnitude of this bias and the misclassification ate can be calculated theoretically given the .iegree of between spouse smoking concordance, the observed proportion of ever smokers, the observed proportion of never smokers who are married to smokers, and the observed relative risk in relation to active smoking. Table 3 shows this relationship for four scenarios: US women, US men, Asian women and Asian men. The misclassification bias is much larger where the proportion of smokers is larger, and' where the relative risk in relatiom to active smoking is larger. In order to achieve a bias• of 1.4 for example, one would need less than a 1% mis- classification for US men, about a 2% misclas- sification for Asian men, about a 5% misclassi- fication for US women and about a 30% mis- classifcation for Asian women. Elsewhere (44). I have reviewed in detail the published evidence on the levels of misclassification actually determined in over 100 studies. In studies of self-reported non-smokers under no special: pressure to deny smoking, biochemicali tests suggested that on average around 4% were actually current smokers, with 1 to 2% current regular smokers. I!n additiion to the misclassified current smokers, studies in which subjects were asked questions on multipl!e occasions have shown a somewhat larger number of ax-smokers misclassified as never smokers. The evidence is certainlj consistent with misclassification, bias being of major importance in the US (and' European) studies. However there is virtually no good evidence on misclassification rates in, Asian populations. There has long been, speculation that rates may be particularly high, among women in Japan, where smoking is not considered socially acceptable. A survey of Tokyo University freshwomen (46); among whom 55% of smokers reported that their family did not know they smoked, tends to confirm this. However until cotinine studies are conducted to find out the ttue situation the extent' of bias caused by misclassification in Asian studies will remain unclear. Misclassification also leads to overes- timation of the total number of lung cancerss among never smokers. This is considered below under "other issues": Concllusion. The answerzi to the two questions posed earlier are clear. The epidemiology has indi- cated a magnitude of risk in relation to spouse smoking that is implausibly large compared' with what is known about the extent of ETS exposure involved. There are clear weaknesses and sources of bias in the epidemiology which could invalidate risk assessments based on it. The most important of these are misclassification bias and failure properly to compare like with /"~n 0

like in case-control studies, but failure to properly take confounding variables into account and publication bias are also relevant. All three risk assessments criticised in this document take the epidemiology virtually at face value, with no real discussion at' all of its weaknesses. Thus Kawachi et a!' (6) mentions only' publication bias (and dismisses it), while Wells (5) considers only misclassifeation bias (and' then inadequately corrects for it). Repace and Lowrey (7) do not discuss any sources of bia at all (2hough some of the authors wnose studit they review do so). No reasonable scientir criteria are used to decide what constitutes a val': study before it can be included in a ris assessmenL - studies conducted with complet disregard of basic included as if they designed studies. scientific principles ar were as valid as carefull' TABLE 3 Bias due to misclassification in four scenarios. Scenario % Ever Smoked % ETS Exposed RR for Smoking Misclassification Rate Bias US wocnen 49.0 54.3 6.73 1% 1.06 2% 1.12 5% 1.35 10% 2.02 US men 77.1 38,7 11.83 1% 1.52 2% 2.38 Asiam women 24.5 56.9 2.99 10% 1.07 25% 1.26 40% 1.73 50% 2.82 l Asian men 80.8 6.6 3.48 1% 1.20 2% 1.42 5% 2.36 N'.B. No effect of ETS and between spouse concordance ratio of 3.0 assumed. % ever smoked, % ETS exposed and RR trelative rnsk) for active smoking estimated from those studies providingg relevant data. See (8) for further details. FXTE..''~1DGtVG RLSK ASSESSMENT TO COVER DISFARES OTHER'I'fiAN LUNG CANCER Heart disease. In the risk assessment by Wells (5); heart disease deaths formed 70% of the total. In that by Kawachi et al (6), they formed 89%. As noted above, in 1986 none of the major authorities considered that ETS had been shown to cause heart disease. Evidently Wells and' Kawachi, in assuming that ETS causes heart disease, are jumping to a conclusion that a number of panels of distinguished scientists have not reached. While there are more data now than, in 1986, it remains abundantdv clear that the evidence still does not support this conclusion.. Wells (5), cites data from six published studies (18. 24, 47-50) and' one unpublished study (51): Of these seven studies„ f ve (16, 24, 48, 50; 51) were based on very much small'er number of deaths/cases than the other two (18, 49) so that theyy contribute very little to the overall meta- analysis. While some further small studies have been published since (see 8), none are large. For this reason it is worth taking a detailed look at the two larger studies. The largest of these studies was by H'elsing et al (49h This involved more heart disease deaths among non-smokers than all the other studies combinedl It reported a 24^c increase in heart disease risk in women exposed to ETS, based on 988' deaths, and a 31% increase in men,, based on 370 deaths. Many features of the study and the results render any conclusion that ETS causes heart disease most insecure: iY The companson was of people who lived wtt'h a smoker and of those who did not- with no direct adjustment for the number of people in the household. Clearly the larger the household, the more likely it is to contain a

moker, so any risk factors related to househol'd' size could contribute to the association. ii ) The study was not a properly conducted prospective study, in that data were only collected on whether a given subject had or had not died in W'ashingtom County over the 12-year period. Differences in smoking habits and disease status between those who left the county and those who did not may have caused substantial bias. iii) There was no dose-response relationship in the exposed groups. Indeed, in men the risks (relative to the non-exposed) were somewhat lower with increasing exposure score. iv) Adjustment for effects of age, marital status, years of school and quality of housing used a procedure that was unclear and which had a huge effect. Thus in women, the passive smoke exposed' group had a crude heart disease death rate 34~"p lower than the non- exposed group. After adjustment it was 24% higher. Such a large effect of adjustment makes one wonder just how contingent the reported results were on the exact list of confounding variables ineluded, the statistical, technique used for adjustment, and the accuracy with which the confounding variables were measured. v)~ A whole range of factors have been rei'ated to heart disease. Among major factors not considered in the stud~ were hypertension and aholesteroi level.. While it' is difficult to determine the relative importance of the features listed above, it is clear that one must have very consid'erable reser- vations about the results from this study. The Japanese prospective study of Hirayama (18) is superficially very good, being very large, having a long follow-up period and being apparently reasonably representative. However, following detailed scrutiny given to his study following the 1981 paper (52) which really brought ETS to public attention, a number of authors have identified various weaknesses (53, 54, 55). His questionnaire was extremely short and crude by modern standards, severely limiting the number of risk factors studied and the depth to which they could be investigated. The population was only interviewed once with no changes in habits -ecorded in 16 years. The mortality of his .degedly representative population is too low to reconcile satisfactorily with nati,onal' rates,, indicating that tracing of deaths was incomplete,, with deficits varying by age and marital status (53). His statisticaI presentation is inadequate in a number of ways: the methods used were not appropriate for analysie of long-term cohort studies;, rates for heart disease in women were age adjusted' to their husband's age rather than their own age; and some basic mistakes in analysis were made. One error, noted in 1981 (54), resulted in enormous inflation of the significance of the lung cancer associatiom A second, noted more recently (55), concerned the total inconsistency of results for heart disease reported in 1981 and 1984, and was only resolved by Hirayama (56) admitting his earlier data were in error. A number of approaches have been made to Hirayama to release his data for independent verification of his findings by more appropriate statistical methods, but Hirayarna has always refused to release his data„ which only casts more doubt on his findings. While his findings show a 16% increased risk of heart disease in never smoking women, married to smokers which is marginally significant when a dose-rel'ated trend test is used, it is difficult to place much faith inn his findhngs. Al2hough it has been demonstrated above that the risk assessment for heart disease essentially rests on the results from two studies, both of which seem unreliable, a number of other general points can be made. First, there are a very large number of risk factors for heart disease. It is evident that adjustment for these factors in the studies has always been incom- plete, and oftem seriously incomplete. Second, the extent of the association seen in some of these studies, which in some cases is close to that reported in relation to active smoking, is implausibly high when viewed against the extent of the association seen in relation to active smoking. Third, there is a major danger of publication bias. It is notable that the literature is still relatively sparse despite the numerous ongoing studies of heart disease and the fact that heart disease in a non-smoker is probably 50 times or so more common than lung cancer in a non-smoker. Any prospective study that has reported on lung cancer c1'earl'y could have done so for heart disease. The fact that the American Cancer Society million person study, which reported for liung cancer (57), has not reported any results om the relationship of heart disease to ETS can reasonably be read as implying no relationship was found in that study. If this is in fact true, and' its results were published, the picture from the meta-analysis would change dramatically since the study would' involve so many deaths from heart disease in non-smokers. ,)nn

Cancer other than the lung. Kawachi et al (6) did not' include deaths from cancers other than the lung in their risk assessments, but W'ells (5) did, although he only made estimates for females since he considered data for males to be too sparse. In fact, there is by now rather more evidence available than Wells considered„ and the picture is completely unconvincing as to the effect of ETS exposure. Of 10 studies providing some evidence, six give no real indication of an effect of ETS. These includ'ed' two moderate sized case-control studies of bladder cancer (58, 59) which both gave relative risks close to unity, a case-control atudy of cervix cancer (60) which found' no association with spouse smoking after controlling for smoking by the femal!e subject, and a prospective study (47) which found a non-significant relative risk of 1.201far cancers other than the lung based on 43' deaths. Another study showing no effect wass the case-control study of Miller (61) from which an age-adjusted relative risk of 0:97 for lung cancer in reiation to husband's smoking history could be cal'culated', It is interesting to note that Miller, while presenting data by age, did not age- stand'ardise, and gave a relative risk of 1.40, while Wells (5), though he did age-standardise, unaccountably used data for unemployed rather than all women, giving a non-significant, relative risk of 1.25. The largest study showing no~ effect was the Washington County study on which the Helstng heart disease results (49)' were based! A later paper (62) reported that relative risks for all cancer for living with a smoker were 1.01 in~ males, based om 115 deaths, and 1.00! in females, based on 501 deaths. Turning now to~ the four studies that provided at least some suggestion of an effect, the smallest was thaU by Reynolds et al (63). This prospective study found no association betweem smoking by the spouse and risk of cancer in men, not giving detailed results. In women, a positive association was found, but this was only of marginal significance (p=0,035), and the relative risk of 1.68 had quite wide confidence limits, being based on. only 71 cancer deaths, only five of which were considered to be smoking, related. In a large case-control study of cervix cancer in Utah (64), a significant positive trend in risk was noted' in relation to various indices of passive smoking exposure. There were many weaknesses in this study, incl!udi'ng f'ailure to adjust for religion (42`'c of cases an& 58`1 of controls were titormons); large and differential non-response rates, m)sclassiifacation of smoking status, and failure to adjust adeq,uatek for sexual' habits. A crude relative risk of 14.8: in relation to ETS exposure for three or marn hours per day dropped to 2.96 after adjustment f'o• the reported number of sexual partners of tho woman. As number of sexual partners is only ar inaccurately measured surrogate of the trut sexually related cause of cervix cancer presumably, a sexually transmitted' infection, th( adjustment will be incomplete and the excess relative risk in relation to ETS may be wholl% spuri.ous representing a residual confoundlnE effect of sexual habits (65). The other two studies reporting a positive association were both~ cited by Wells (5) and were the major contributors to his risk assessment for cancers other than the lung. The study by Sandler et a! (66) for which Wells cites a relative risk o: 2.0 based on 231 cases of cancer other than the lung, use& a mixture of friends or acquaintances of patients and people randomly selected bN . systematic telephone sampling as controls, a very questionable procedure. Response rates also varied substantially between cases and controls. The unconvincing nature of the findings was heightened by study of the results for individual cancer sites where large effects were claimed for ETS for a number of cancers (breast, th)Toid. l,eukaemiaAymphoma) that have little or no relationship, to smoking. The largest study is that by Hirayama (18, 52. 67). Wells (5) cites a relative nsk of 1.11 (95`h confidence limits 1.0-1.2) based on 2505 deaths from, cancer other than the lung. This is unconvincing for a number of reasons. First. most of the comments made about this study when considering the heart disease results apply. Second, the relative risk is only at best of marginal significance (trend p = 0.05 on a one- tailed test). Third, the association with spouse smoking arises mainly because of elevated risks of brain and breast cancer, cancers that are not smoking related. The overall evidence for cancer other than the lung is clearly remarkably unconvincing in demonstrating any effect of ETS exposure. Where any association is reported' it is generally for cancer sites not affected by active smoking.. Wells (5) has great (and un)ustified) faith in, the epidemiology, claming "these differences inn mortality effects are probably renl." Because it is certainly true (though as yet unquantified) that smokers have higher ETS exposure than nonsmokers it is a prtori very difficultit to see how an association with any disease could be observed only in response to ETS exposure, a ;J-/ I no ,.

view endorsed by IARC (1): Wells argues competing risks might be the explanation, effects of ETS exposure on such cancers as brain, sndocrine glands, lymphoma, and breast only occurring in people with a particular susceptibility, and that people with this susceptibility; if they smoke, die first from lung cancer or other smoking-related cancers. This seems a remarkably unattractive and implau~ sible hypothesis, for which there is no supportive evidence. Mortality patterns for lung cancer in terms of age, d'ose and duration of smoking are well described by models involving no component for variation in susceptibility at all, the response arising from random variation. Of course susceptibility might in fact vary to some extent (68, 69), but hardly so much that any, effect in active smokers would be ruled out. The simpler hypothesis that any relationship seen betweem ETS and cancer of sites other than lung is due to chance or bias seems more plausible. FxTf."r'DNG RISK • ccEcSMZNT TO COVFT2 ETS F:XR)6URE FROM THE WORKPLACE Wells (5): took account of ETS exposure outside the home in two ways in his risk assessment. First, he estimated the proportion exposed by adding the proportions of never smokers living with ever smokers (taken from the controls of the US based epid'emiological studies) to the proportions of alll nonsmokers who did not live with a smoker but who where still exposed at home or at work (takenifrom Friedman (70)). Second, he adjusted relative risk estimates upwards, except in Greece„ Japan and Hong Kong, by assuming that nanexposed nonsmokers were actually exposed to 1/3 the extent of the exposed nonsmokers. Essentially he assumed that exposure outside the home had the same effect as exposure from the spouse. Kawachi et al (6) estimated the proportion of people exposed at home and at work from surveys. From the relative risk in relation to home exposure, 1.3, they multiplied' the excess relative risk- 0.3,, by a factor, 4.0, based on Repace and Lowrey s estimate (34) of the relative extent of exposure Go~ the particulate phase of ambient' tobacco smoke at work (11.82 mg/day) to at home (0.45 mglday); thus estimating relative nsk of lung cancer in relationi to work exposure, 2.2. They commented that 'this estimate is consistent: with the rel'ative risk of 3.3' (95'' confidence interval 1.0•1'0.5) for never smokers exposed to passive smoking at work reported by Kabat and~ Wynder (71) in one of the few studies that has distinguished exposure at work from, exposure at home. However, we have adopte& the more conservative estimate of 2.2". It is surprising that neither Wells (:5)1 nor Kawachi et al (6) seem to have actually taken into account the total epidemiological evidence on lung cancer in relation to workplace exposure. Had they done so (see Table 4) they would have found that overall it gives no indication of a positive association at all, with only four out of elievem relative risk estimates greater than 1.0 and only the single estimate (Kabat 1- males); selectively cited by Kawachi et al (6) even close to being significantly positive. The upper confid'ence limiti for seven of the eleven estimates is less than the estimate of 2:2' used in their risk assessment. Most lung cancer cases occur at an age after people have retired. W}ule Wells (5) adjusts the exposed fraction down with increasing age, Kawachi et al (6) make no such adjustment, assuming that their unjustifiably high relative risk of 2.2 in relation to workplace exposure operates at age 80 as at age 40: The estimates by Kawachi et al (6) of risk due to workplace exposure from risk due to at home exposure are in any case methodologically un- sound. Even assuming (and these are very big as- sumptions); that meta-analysis gives unbiassed estimates, that risk is linearly related to extent of exposure to smoke constituents, and that the estimates of relative exposure at work and at home are valid, the equation they used is totally incorrect. The formula only makes sense for a companson of those exposed at work and not elsewhere with those exposed at home and' not elsewhere. If at home and at work exposure are positively correlated (as is likely) double counting of deaths arises. In the extreme situation where everyone is exposed to both or to neither source, their method for estimating deaths due to at home exposure yields an answer appropriate for both exposures combined'. Using their procedure, which would then multiply up deaths due to ETS by five„might lead to there beingg more deaths due to ETS than actually occur in, all: The validity of the factor of 4 for relative exposure at' work to at' home is anyway verv dubious. A recent large survey in London (74): found little difference between particulate matter levels measured in the home an& at work. Indeed where smoking took place, the level! at work was 1= than at home. 202