Tobacco Documents Online

REVIEW 766 COHFIDENTIAT. Subject ref 8b "Women and lung cancer, a comparison of active and passive smokers with nonexposed nonsmokers" G H Miller et al Cancer Detection and Prevention (1994), 18, 421-430 Over the last 20 years or so Miller has produced a number of papers based on his study conducted in Erie County, Pennsylvania, many published~ in rather obscure journals. Early work concluded, inter alia, that men and women have no difference in longevity if they do not smoke, that smoking reduces longevity more in women than in men, that smokers of filter cigarettes have reduced longevity compared with smokers of plain cigarettes, and that passive smoking takes four years off a nonsmoking woman's life. His papers have been notorious for statistical flaws - including misuse of the statistic 'average age of death', failure to use control populations and failure to age standardize. In 1984, in the Western Journal of Medicine (1984, 140, 632-635), Miller claimed that passive smoking doubled risk of cancer, a claim that was highly implausible since active smoking is typically found to be associated with less than a two-fold risk of cancer in the large prospective studies. As I pointed out in my comments at the.time (TAC document TG 1644), there were a large number of major weaknesses in the

paper, which totaLly invalidated it. The study only included five lung cancer deaths in the never smokers and results for lung cancer were not separately presented. Later, in 1990, Miller published a paper in Cancer Detection and Prevention (1990, deaths that were with ETS exposure assumed by Miller 14, 497-503) reporting that the proportion of total to cancer was massively higher in unemployed women due at home (78/412 - 17.0%) and in employed women, all to have ETS exposure at work (108/315 - 34,3%) than in unemployed women with (4/179 - 2.28), no known (or only minimal) ETS exposure at home As I pointed out in Review 558, Miller's results were totally discrepant from the rest of the literature in terms of (a) the strength of the association of ETS with cancer risk, (b) the increase in cancer risk in relation to employment, and (c) the rarity of cancer in the non-exposed group. In my review I pointed out that though his conclusion must be nonsense, it was difficult, unlike his earlier papers, to put a finger on what feature of his study design or analysis had actually produced the ludicrous findings. I suggested that one possible explanation might lie in more intensive interviewing of cancer cases than non-cancer cases, but there was no direct evidence to support this. The present paper is based on.deaths occurring fromnon-accidental causes in Erie County in women aged 30+ in 1972-76 and 1979-84. Based on interviews with surviving relatives, the decedents were classified into three groups:

-3- Smokers : Women who had smoked more than 20 packs of cigarettes during their lifetime. ETS Exposed : Women who had smoked no more than 20 packs of cigarettes Nonsmokers during their lifetime but who had had moderate (20 years). exposure to ETS, from a smoking spouse, additional smokers in the home, smokers in the workplace, or exposure in other environments. ETS nonexposed : Women who had smoked no more than~20 packs of cigarettes Nonsmokers during their lifetime but who had no known exposure or at most a very limited exposure to ETS. Data on cause of death were obtained from Pennsylvania Department of Health, presumably based on death certificates. The main finding of the study can be expressed in the following table: Total Deaths.from Relative Group ea hs Lun¢ cancer Percentaee Percentaee ETS Nonexposed Nonsmokers 528 1 0.2% 1.00 ETS Exposed Nonsmokers 3138 27 0.9% 4.54 Smokers 1747 140 8.0% 42.31 The fact that a much higher proportion of smokers' than nonsmokers' deaths are from, lung cancer is consistent with the literature and tells us nothing new, even if the study were valid. I will concentrate, therefore, on the comparison of the two groups of nonsmokers. There are a number of points that should be borne in mind in considering the higher proportion of deaths from~lung cancer in the ETS exposed group: (i). It is not statistically significant. Table 1 gives an odds ratio of 4.57„ with 95% confidence Limits of 0.67-90.7.

-4- (ii)~ The simple comparison is not age-adjusted. The requirement of 20+ years exposure for the exposed group is likely to mean that the exposed group is older. In Table II, age-adjusted standardized mortality ratios of 17 and 37 are calculated for the nonexposed and exposed groups. This implies that the unadjusted 4.57-fold odds ratio reduces to only 2.18 after adjustment, a difference which would be very far from significant. (The probabilities in Table II relate to comparisons with the overall population.) (iii) There are no apnrooriate controls. Using non-lung cancer cases as controls is only valid if non-lung cancer cases are representative of the living population in respect of smoking and ETS exposure. This is highly unlikely to be true. (iv) There is a large amount of missing data. Data were only collected for about 55% of the decedents. It is quite plausible that there are systematic differences between subjects for whom data were and were not collected. (v) Data collected from relatives may be tt ht Surviving , ~; _ relatives may not always have known the decedent's hiet> : oty of smoking or ETS exposure. Their recall may have been affected by knowledge of the cause from which the subject died. (vi) The index of ETS exposure is not clea ly ggplained. It is stated _ on p424 that 3666 interviews of nonsmoking women were obtained, 528 unex osed and, 3138 d Thi N, p expose . s suggests that all nonsmokers were assigned to one of the two categories. However the actual C ~ description of the categories of exposure seems to divide subjects riJ ~%. ~ into those with 20 years + ETS exposure and those with no known ETS ~ exposure (.or at most very limited ETS exposure) so that some ~

-5- nonsmokers (those with less than 20 years exposure) would appear to be excluded from analysis. Indeed the text refers to two nonsmoking women with lung cancer with less than 20 years exposure, when there is only one nonsmoking women with Lung cancer in the nonexposed group. On p427, however, it is stated that "working wives were placed in the EX (ETS exposed) category"' with no reference to any 20 year minimum. What Miller actually did is therefore rather unclear. (vii) The index is correlated with employment and no adjustment is made in analysis for occupational exPosures. If all working wives are placed in the ETS exposed category, there is an~obvious confounding between employment and ETS'exposure in the analysis. In summary, the paper adds nothing to the evidence on ETS and lung cancer. As the study does not use a proper control group, and as it uses and index of exposure that does not separate out effects of spousal, childhood and workplace ETS exposure, it cannot sensibly be used in TMA's overview of the epidemiological evidence. P N Lee 19.12.94