Tobacco Documents Online

a marker of relative tobacco smoke exposure. In a large study I conducted in 1985 in the UK (18), I found that the corresponding ratios were about 10 times higher, 0.27% in males and 0.13% in females. Had cotinine been used as the index of exposure in Arundel S_t al's calculations, this suggests that about 160 lung cancer deaths would have been predicted, still a full order of magnitude lower than the EPA estimates. It is doubtful, however, that estimates based on cotinine are valid. In the first place, nicotine itself is not deemed to be a carcinogen. Secondly, even if cotinine is used only as an index of smoke uptake, it suffers from the major problem that while cotinine is a marker of the lung's particulate exposure in active smokers, it is a marker of ETS gas phase exposure in nonsmokers. Comparison of cotinine levels in body fluids of smokers and nonsmokers is therefore misleading. N N d .a A O1 UI 0

-12- 3. Enidemioloeical avnroach There are by now 33 published epidemiological studies of lung cancer for which results relating to ETS exposure have been separately presented for lifelong never smokers. I have recently prepared an up-to-date assessment of the data from these studies [19], drawing partly on an earlier book [20) in which I examined 28 of these studies in detail. The 33 studies I considered included all those 30 considered by EPA, with the addition of recently published studies by Brownson [21j and Stockwell [22], and a study by Kabat [23J for which results were only presented at a conference. It is convenient first to describe how the EPA conducted their risk assessment to reach their estimate of 3,060 lung cancer deaths attributable to ETS. The main steps taken, described in full in sections 5 and 6 of their report, can be summarized as follows: (1) Estimate, for never smoking women, the relative risk of lung cancer associated with marriage to a smoker (or in some studies with living with a smoker) in each study. (2) Adjust the relative risk estimates downward to account for bias caused by a proportion of current and former smokers misrepresenting themselves as never smokers, coupled with the tendency for smokers preferentially to marry smokers. (3) Demonstrate, by combining adjusted relative risk estimates from the relevant studies, that there are statistically significant 4 increases in risk in relation to marriage to/living with a smoker in the 11 US studies, in the 5 Japanese studies, in the 4 Hong Kong studies, and in the 2 Greek studies, though not in

-13- the 4 West European studies or in the 4 Chinese studies, and that the overall evidence indicates an association. (4) Demonstrate that there is a stronger association of lung cancer risk with mhrriage to a heavy smoker (or with marriage to a smoker for a long time) than with marriage to an average smoker. (5) Consider various potential confounding factors (history of lung disease, family history of lung disease, heat sources for cooking or heating, cooking with oil, occupation, dietary factors) and conclude that none explains the association between lung cancer and ETS exposure. (6) Classify studies into four "tiers" by a quality assessment, and show that the associations generally remain statistically significant if attention is restricted to studies considered to be of a better quality. (7) Use the information in (1) to (6) to determine that there is a causal relationship, i.e. that a hazard has been identified. (8) Use an estimate of Z - 1.75 for the relative cotinine level of never smokers married to a smoker and never smokers married to a nonsmoker to adjust US relative risk estimates to a non-exposed baseline. Thus, relative to a never smoking woman unexposed to ETS the risk of a never smoking woman married to a nonsmoker is 1.34 and the risk of a never smoking woman married N tlY to a smoker is 1.59. (N.B. the ratio of risks 1.59/1.34 - 1.19 O N ~ s is the relative risk estimate from the 11 US studies, and the A ~ ratio of excess risks 0.59/0.34 - 1.75 is the Z-factor N N N assumed).

-14- (9) Take an estimate of 9.26, from the American Cancer Society Cancer Prevention Study, for the risk of current and former smokers relative to never smokers, and convert it to an estimate (of 13.8) relative to never smokers unexposed to ETS. (10) Use estimates of the total number of lung cancer deaths in US women in 1985, and estimates of the relative frequency of never smokers married to nonsmokers, never smokers married to smokers, and ever smokers, in conjunction with the relative risks of 1.34, 1.59 and 13.8 to calculate that there are 6,970 lung cancer deaths among never smokers, of which 470 are from ETS exposure from the spouse and 1,030 from other, non-spousal, sources of ETS exposure. (11) Assume estimates of the increased lung cancer risk in never smokers in relation to spousal and non-spousal ETS exposure for women apply equally to men, and calculate that there are 80 deaths from ETS exposure from the spouse and 420 from non-spousal ETS exposure. (12) Assume estimates of the increased lung cancer risk in never smokers in relation to ETS exposure for women apply to former smokers of both sexes who have given up five or more years ago, leading to an estimated further 160 female and 150 male deaths from spousal ETS exposure and 270 female and 480 male deaths from non-spousal ETS exposure. (13) Sum the numbers for men (1,130) and women (1,930) for never (2,000) and former (1,060) smokers, or for spousal (860) and non-spousal (2,200) exposure, to give a total of 3,060 deaths due to ETS, rounded to 3,000.

-15- In my recent review of the evidence I concluded that the overall evidence from, the 33 studies indicated a statistically significant relationship between lung cancer risk in never smoking women and marriage to (or living with) a smoker. Of 33 relative risk estimates, unadjusted.for covariates or misclassification of smoking status, 25 were greater than unity (p<0.01) and a "fixed effects" meta-analysis (24] gave an overall relative risk estimate of 1.17 (95% confidence interval (CI) 1.08-1.27). This estimate was only marginally changed (to 1.14, 95% CI 1.05-1_23) if one used, where available, relative risk estimates adjusted for covariates, or if one used "random effects" meta-analysis (to 1.21, 95% CI 1.09-1.36). "Fixed effects" meta-analysis only takes within-study variability into account, but "random effects" meta-analysis also considers between-study variability. An association could also be seen separately in studies in the USA, Europe and Asia, estimates (based on unadjusted relative risks and "fixed effects" meta-analysis) being respectively 1.13 (95% CI 1.00-1.28), 1.40 (95% CI 1.06-1.85), and 1.17 (95• CI 1.03-1.32). 21 studies provided data on risk in relation to extent or duration of smoking by the husband (or cohabitant). Comparing risk in the most heavily exposed group with that in the overall exposed group, the former had a higher risk in 16 studies and a lower risk in only 4, a significant (p<0.05) departure from chance s expectation. Overall the most heavily exposed group had 1.16 times

-16- the risk of the overall exposed group, which suggests that the most heavily exposed group had 1.35 (- 1.16 x 1.17) times the risk of the women not married to a smoker. So far my conclusions were broadly in line with those of the EPA summarized in points (3) and (4) above. However further examination of the data revealed a very large number of flaws in the. EPA's argument which totally overturned their conclusions. These flaws are summarized below: Failure to consider workvlace and childhood ,Qnosure. When evaluating whether an association with ETS exposure exists, it is vital to consider all indices of exposure with adequate data. Although, in 1986, when a number of the major reviews [3-5] were published, there was a worthwhile amount of data only on smoking by the spouse, this is certainly not true now. There are 14 estimates of risk in never smokers in relation to workplace ETS exposure, which, when combined, provide no evidence at all of an association with lung cancer (RR - 1.02,95E CI 0.93-1.12). Similarly there are 14 estimates in relation to childhood ETS exposure from the parents, and again there is no evidence of an association (RR - 0.94, 95% CI 0.84-1.05). There seem to be no strong reasons to believe that smoking by the spouse is a much better marker of ETS exposure than is smoking in the workplace or smoking by the parent in childhood (121. It is therefore grossly biassed to do what the EPA did, namely to conceal from the reader the results for these two ~ N O N a ~ 01 N rn N

-17- alternative indices which show no association at all with lung cancer, and to concentrate solely on the single index, marriage to a smoker, which does show an association_ Eailure to consider histoloeical tvoe of lune cancer. The association of lung cancer with active smoking is much stronger for squamous cell cancer than for adenocarcinoma. If, as EPA assume, ETS is merely a reduced dose of active smoking, one would expect to see effects, if any, for squamous-cell cancer. In fact the evidence regarding histological type of lung cancer is conflicting. There are four studies where the data on spousal smoking seem more consistent with a relationship with squamous (or small cell) carcinoma, four studies where the data seem more consistent with a relationship with adeno (or large cell) carcinoma, one study which found a relationship with both types, and five studies which found no relationship with either type. A major weakness of the EPA report is that it makes no attempt to compare and contrast results for the two major types of lung cancer. Consistency is a criterion that EPA cite for testing causality, but which they do not apply in this context_ Failure to take into account uronerlv the vossibilitv of confoundin¢. There are three fundamental flaws in the EPA's argument. First, they only consider confounding relevant if a single risk factor can be shown to explain the whole association between lung cancer and spousal smoking. This is clearly not sensible. More than one risk factor might confound. Second, when trying to determine whether a factor actually elevates lung cancer risk EPA

-18- restrict attention virtually completely to evidence from the ETS/lung cancer studies themselves, ignoring abundant relevant data from other sources- Third, they ignore the growing evidence [19,25] that ETS exposure is associated with exposure to dietary and other risk factors, and fail to reach the appropriate conclusion that a nonsmoker married to (or living with) a smoker is generally more exposed to other risk factors than is a nonsmoker with no smoker in the household. In a study I conducted recently with my colleagues Alison Thornton and John Fry [25], I identified 33 lifestyle "risk factors", i.e. factors generally perceived to be associated with adverse health consequences, not necessarily with increased lung cancer risk. Of the 33 factors, 27 showed a highly significantly (p<0.001) increased prevalence in smokers of 20+ cigarettes a day compared to never smokers, and only 2 a decreased prevalence. 14 of these factors were also significantly (pG0.01) increased in never smokers with a smoker in the household, and on were significantly decreased. The factors included low fresh fruit and vegetable consumption, high fried food consumption, working in an occupation with a possible cancer risk, low social class, poor eduction, and high alcohol consumption, all factors linked to an increased lung cancer risk. Clearly the question is not whether confounding exists, but what magnitude of bias it causes. Analyses presented elsewhere [19.25] suggest that confounding could explain a material part of the reported association of lung cancer with spousal smoking. It should be noted that the extent to which confounding variables were taken into account in the epidemiological studies of spousal smoking on which the EPA's estimate was based was very limited. Thus over

-19- half the studies where spouse smoking was the index of exposure failed to restrict attention to married subjects, thereby producing a serious confounding between potential effects of ETS exposure and potential effects of marital status (and its correlates). Furthermore, although many studies reported having recorded numerous risk factors, few took any account of these in analysis. Thus, 31 of the 33 studies did not adjust for dietary factors, one of these being a study [22] which actually reported a striking relationship between diet and lung cancer among never smokers in another paper [26]! F_a_ilure to correct fully for bias due to misclassification of active mok n. Adjustment for misclassification is a complex issue involving a number of assumptions that are not easily justified, and variables that are not precisely known. In the EPA report, Wells estimates the bias to be negligible, only increasing the overall relative risk estimate by a factor of about 1.02. However, as discussed in detail elsewhere [27], there are two major reasons why this analysis may understate the effect of bias. One reason lies in the error of applying a misclassification rate, estimated from data virtually all of which comes from North American, European and Australian populations, to results from lung cancer studies conducted in Asia. The fact is that in some countries, such as Japan, smoking by -women is considered socially unacceptable. Consequently misclassification rates are likely to be much higher there. The second reason lies in underestimating the extent of misclassification in the countries for which data are available. N w 0 N ~ ~ O> N UI ~

-20- Elsewhere [19) I present the results of analyses adjusting the meta-analysis relative risk for the US studies for various plausible values of the two key parameters, the misclassification rate and the concordance ratio (the measure of the extent of the association between the smoking habits of husband and wife). Using what I regarded as the most plausible astimates (2.58 misclassification, concordance ratio of 3.0) reduced an unadjusted relative risk of 1.13 (95% CI 1.00-1.28) to 0.96 (95% CI 0.84-1.09). Even assuming only a 1.0% misclassification rate reduces the relative risk estimate to 1_06 (95% CI 0.94-1.20), halving the estimated excess risk and making it become non-significant. Failure to address oublication bias. It is well known that any meta-analysis should consider the possibilities of bias due to failure to publish null studies, but EPA do not consider this issue at all. In fact there is some evidence of this, with a tendency for relative risk estimates to decrease with increasing sample size. Thus, in the 11 studies with less than 50 lung cancer cases, the relative risk estimate was 1.41 (95% CI 1.02-1.93); in the 12 studies with 50 to 100 cases it was 1.33 (95% CI 1.12-1.58); and in the 10 studies with more than 100 cases it was 1.10 (95% CI 1.00-1.21). Failure to test for effects of study weaknesses. lteta-analysis is conventionally used to combine results from similarly designed randomized controlled trials conducted in different populations. It is much more open to question when, as here, it is applied to