Tobacco Documents Online

DRAFT .Comments on • Law, M.R. et a1.,(1997) ¢ Environmental Tobacco Smoke Exposure and Ischaemic Heart Disease : An Evaluation of the Evidence. ~ BMJ, 1997; 315(7114) :980- Please note - below are some very preliminary comments on the paper appearing in the BMJ, 17th October 1997. It is very difficult to comment in detail as so little detail is given in the paper ! Background : This paper, like the Lung Cancer paper published in the same journal, was also apparently commisioned by the UK Department of Health through their Scientific Committee on Tobacco and Health. Unlike the LC paper, however, TMA were never informed of this work, nor have they been invited to present their response ! Study Obiectives : The objectives were presented as :<< To estimate the risk of ischaemic heart disease caused by exposure to environmental tobacco smoke and to explain why the associated risk is almost half that of smoking 20 cigarettes per day when the exposure is only about 1% that of smoking. • Claims : 1. The paper claims that the epidemiology indicates an excess risk of 23% after adjusting for dietary confounders. 2. That the excess risk for IHD amongst smokers of 20 cigarettes a day has been found to be around 78%. 3. A model is proposed to explain this disproportionately high effect at such low doses. This is based on the claimed effect of cigarette smoke exposure on platelet aggregation. Comments : The paper is based on five sets of analyses as follows • 1) A meta-analysis of 19 studies on non-fatal infarction or death from IHD in never smokers according to whether their spouses were current smokers. The major criticism of this analysis is the blatant ommission of the largest single database on this subject. Law et al cite 19 <~ acceptable )) published studies of risk of IHD in lifelong non-smokers who live with a smoker. Of these, 16 were published studies and 3 cited by others in abstracts or theses. (I note that the latter three are not clearly referenced) : These combined studies include a total of 6,600 IHD events. Analyses of data from the American Cancer Society Prevention Studies, (CPS-i) and the National Mortality Followback Survey, published in two papers in Regulatory Toxicology and Pharmacology in 1995 were ommitted from their analysis. These studies include more cases than the remaining 19 put together, Law BMJ IHD 97.doc/RDE/05.11.97 page 1 of 5

DRAFT (14,981 total fbr CPS-1 alone !). The authors of the BMJ paper justify this exclusion because they show results << statistically inconsistent with the estimate .... from the above analysis of 19 studies ))... and they << took the estimate from the 19 studies as valid and rejected that of Layard and LeVois, since there is not reason to reject an analysis based on 19 independent studies in favour of one from a single group with a vested interest >). They also state that the one data set analysed by Layard and LeVois, and separately by Steenland, shows different results - suggesting that Layard and LeVois are incorrect in their analysis. Again this is unfounded - it would appear, in fact that there is not much difference in the data between the two relevant analyses and that that which there is is due to the different exposure index used, (Steenland used current smokers whereas Layard and LeVois used ever smokers). If we are hoping to keep this debate on a scientific basis then it is totally unjustified to throw out the data based on the largest database available because the authors think that it is tainted ! If they have a problem with the analysis of Layard and LeVois, they should firstly specify the scientific basis for their disagreement, and even better - obtain and analyse the data themselves. The data, after all, comes from the American Cancer Soci.ety - and is one of the databases most widely cited when the issue of active smoking is discussed ! Unfortunately it is impossible to e~traCt from their paper details of their analysis. They do not show the actual data, merely presenting the meta- analysis in a figure where the RRs have been combined into an overall figure for each study. Similarly to the Lung Cancer paper, the authors dismiss Publication bias as a source of bias in their meta-analysis as they claim that 8/19 studies report statistically significant increases in risk and that this is unlikely to be due to chance publication bias. The question of publication bias is not this simple. The ~ publication bias )) theory suggests that studies are more likely to be published if they show a positive effect, almost regardless of the quality of the study. Thus studies which tend to show positive effects which may be due to flaws in the design or upward biases are more likely to be published. Surely the most compelling evidence of serious publication bias in the ETS and HD story is the stubborn refusal to include the data from the ACS etc... ! 2) Analysis of the dose-response curve for smoking and IHD from 5 cohort studies on men. The linear extrapolation model proposed for active smoking seems counter- suggestive. The implication appears to be that based on the dose response observed for <10, 10-15 and >20 cigarettes per day, if extrapolated back to 0 suggests a residual effect. I have not yet had an opportunity to check the source data for figure 2, which are the Hammond studies, the British Drs studies, and the Dorn study, but my recollection from these is that they do not have actual data for smokers of <~ 1 cigarette per day )). The figures here Law BMJ II-ID 97.doc/RDE/05.11.97 page 2 of 5

DRAFT given must therefore be extrapolations back to zero from the higher dose effects reported. For all five graphs shown on figure 2 there is a large jump from the origin to the first data point at around 10 cigarettes per day. I can think of no justification for such a dramatic change in the dose-response curve at levels below the observable data-points. The authors proposition appears to be that there is an all-or-nothing type of threshold effect taking place, whereby at some point along the dose continuum, between 0 and 10, some biological system becomes saturated and the toxic effect manifests itself virtually immediately. They are suggesting that this takes place below doses provided by exposure to ETS. There is no justification of this from the data itself. The authors appear to be forcing the model to fit their claims that the epidemiological data suggests an increase in risk. This is therefore not the confirmatory evidence that the authors claim, but rather a somewhat extreme model which, to their mind, could explain their interpretation of the epidemiology. Since their meta-analysis is subject to question, this model is equally subject to question. A more reasonable explanation for the leap between the odgin and the first data point in the 5 studies cited, could be that there is a confounding factor in this data and that the exposure in question- is not the sole responsible for the increase in risk. In this latter case, extrapolation would not take you back to the origin, but to some point on the y-axis which represents the background risk for the confounding factor in question. There are also a number of other shapes that the dose response curve could take at lower levels of exposure - none of these alternatives are even discussed by the authors. 3) Estimates of the extent of dietary confounding in the ETS studies. The authors calculate the potential effect of differences in fruit intake between exposed and unexposed subjects in an indirect analysis. They first extract data from various studies showing differences between smokers and non- smokers for consumption based on the proportion of 1 SD decrease in food consumption for various nutrients. Again, I have not had the opportunity to analyse all their source references, but my experience in dietary studies is that they are normally based on quartiles ; (possibly quintiles or tertiles) of consumption, and it is thus straining the data to calculate based on SDs. Secondly, as the authors themselves point out, the data from their source references shows considerable heterogeneity between different communities, different seasons etc.. It is therefore scientifically unsound to come up with a single figure for adjustment and apply it to a single figure for meta-analysis. This does not take into consideration the different contributions this confounding may make to individual studies, nor does it account for the different weightings that each study is given in a meta-analysis. Furthermore, the upper limit suggested in their calculations is in conflict with the data for some of the studies, where confounding was shown to have a larger effect than the 1.06 claimed. The authors fail to discuss the available Law BMJ IHD 97.doc/RDE/05.11.97 page 3 of 5

DRAFT evidence for the albeit limited confounding factors considered in some of the individual study designs. 4) An indirect estimate of the extent of confounding in these latter studies based on data on reversibility of effects aftbr cessation, This estimate is based on the observation that the excess risk for active smoking was almost entirely reversed after cessation in three of the five cohort studies cited. The residual excess risk, given as 6%, was viewed as being an upper limit for the effect of confounding on these studies. This assumes that the only change associated with this reversibility is smoking cessation. They claim that this is justified as these are all based on studies where the smokers would have given up before 1955, << when dietary change wa not widely advocated on health grounds ~>. This is somewhat speculative - even before 1955 it is not unlikely that smoking cessation could be associated with many other behavioural changes. They also claim that recent data would increase the estimate of 6% << to no more than 12% >~ - i.e. a doubling! 5) Estimates of the effect of smoking and ETS exposure on Platelet aggregation and the resulting potential influence on IHD. I am not an authority on the role of platelet aggregation in heart disease and this claim obviously needs careful consideration. However, I note that the authors claim to have analysed the published data from a medline search with a MeSH of <~ Platelet Aggregation >~. VVhen I attempt such a search I get around 2000 hits - I see no evidence in the paper that the authors have analysed such a fountain of literature ! Furthermore, on glancing through some of the more recent papers there does not appear to be any consensus view, as yet, on the role platelet aggregation may play in heart disease. And finally, there appears to be some conflicting data on the effects of smoking exposure (both ETS and active) and platelet aggregation. The authors do not appear to have paid due attention to the complexity of the data. I am still waiting for a more expert analysis on this theory. Once more, one gets the uncomfortable feeling that far from being confirmatory data, the evidence here has been selected and massaged into a form which could fit the authors views on the epidemiology. Conclusions : The authors discussion states that << [e]nvironmental exposure to tobacco smoke is associated with an excess risk of ischaemic heart disease of 30% and is estimated to cause an excess risk of 23% ........ So large an effect from a relatively small exposure, though unlikely on first impression, is supported by a great deal of evidence )>. Firstly, I dispute their premise that the epidemiological data suggests such a 30% increase in risk. If the entire database were considered, including the Law BMJ IHD 97.doc/RDE/05.11.97 page 4 of 5

DRAFT largest and most comprehensive study to date, it is highly unlikely that such a figure would be supported. The ~ great deal >> of evidence which they claim to support this risk is equally doubtful. It seems that a large part of it comes down to speculation, and some rather fancy analyses in order to fit data to their already skewed database. Law BMJ II-[D 97.doc/RD, E/05.11.97 page 5 of 5