Tobacco Documents Online

3 (unexposed = 27.2%) than in any of the other three exposed groups which have very similar proportions (= 19.7%, 18.5% and 19.2%). Surely a better test is one based on the simple 2 x 2 table (I+ II vs III+x ETS exposed/unexposed). According to my calculations, this is significant at nearly p = 0.01, refuting the statement made by the authors I cited above. It should also be noted that the study purports to answer the question as to whether diet varies by source of ETS exposure. If so, then the obvious analysis was actually to compare distributions in groups 2, 3 and 4. However, there are no statistical tests of this. What they have done, which is not necessarily the same thing, is to compare each of groups 2, 3 and 4 with the unexposed group 1. A final comment I have on the paper is that ETS exposure has not been validated by cotinine measurements, reliance being placed on subject report. In theory a more sensitive test of diet/ETS relationships may be obtained by quantifying extent (rather than presence) of ETS exposure to diet. In summary, this is not a very impressive paper and should not affect the overall literature, which does show quite clearly that ETS exposure is associated with a poorer diet. PNLee 24.11.99.

~I ~ 3rq 1 REVIEW 1076 CONFIDENTIAL Subject ref 8f "Relation of environmental tobacco smoke to diet and health habits: variations according to the site of exposure" F Curtin et al Journal of Clinical Epidemiology (1999), 52, 1055-1062 This paper describes a study aimed at evaluating whether the association between ETS exposure in never smokers and a less healthy lifestyle varies according to the site of ETS exposure (at home, at work or during leisure). Based on a cross-sectional survey conducted in Geneva in 1993-1996 in women aged 35 to 74 in which participants completed a questionnaire concerning lifestyle factors, reproductive history, classic CVD risk factors, occupational history, smoking behaviour and ETS exposure, and in which weight and height were measured, 914 "never" smokers (defined as smoking less than 100 cigarettes in their lifetime) formed the basis for the statistical analyses presented. These were divided into four groups depending on whether, for each of the three sites, they were exposed to at least one hour per day for at least one year that was still ongoing. Subjects with exposure, as defined, for more than one site, were excluded (surprisingly only as few as 14 women) and this left the following four groups: 1) Unexposed to ETS n = 698 2) Exposed to ETS at home n = 81 3) Exposed to ETS at work n = 83 4) Exposed to ETS during leisure n= 52 The main finding of the study was that though (i) Compared to group I (unexposed), group 3 (exposed at work) showed a wide range of differences consistent with a poorer diet (less fibre, cereals, vegetables, lean meat, iron, beta-carotene, and energy intake) after adjustment for body mass index, social class and

2 age, and though (ii) Group 4 (exposed at leisure) also showed some differences (less cereals, skimmed milk and complex carbohydrates), (iii) Group 2 (exposed at home) did not show any significant differences. The authors also noted that "though women unexposed to ETS belonged more often to classes I and II than women exposed to ETS, the differences were not statistically significant." The main conclusion of the paper is that "confounding factors of the association of ETS and disease vary according to site and populations and therefore should not be invoked as a systematic source of bias in all studies." While I accept the general thesis that the relationship between diet and ETS exposure may vary according to the population studied and the source of exposure (their use of "site" of exposure is rather an odd one), the conclusion that it should not be invoked as a systematic source of bias is rather misleading. Surely the key question of interest concerns whether the overall evidence relating ETS to lung cancer or heart disease is likely to be materially biassed because the great majority of the studies have not taken diet into account as a potential confounder and because, on average over the studies considered, ETS exposure tends to be correlated with a less healthy lifestyle. In any case, it must be borne in mind that the sample sizes studied for the three individual sources of ETS exposure are really quite small, only 83 at most, and much smaller than in a number of the other studies that have reported associations, many of which report an association of diet with spousal or at home exposure. I am also concerned by the quality of the statistical analysis. One problem is that the relationship between social class and ETS exposure appears to have been tested by a chi-square test for the overall association in the 4 x 4 table (four exposure groups x four social class groups), on 9 degrees of freedom. This is a remarkably insensitive test. Looking at the data in Table 1, one can see at a glance that the proportion of higher social class (I + II) is higher in group 1