To the Editor:
In the past, Kawachi et al have been deservedly cautious in interpreting their surveys of the Nurses Health Study cohort, but they seem to have thrown caution to the wind in their article, “A prospective study of passive smoking and coronary heart disease” (Circulation. 1997;95:2374-2379), as reflected by explosive reports in the news.
In earlier reports by the authors, lack of statistical significance usually prevented conclusions of causality, coupled with a keen concern about confounding and bias. Indeed, this article criticizes other studies that have not adjusted “for the full range of potential confounding factors,” while the authors claim to have adjusted “for a broad range of cardiovascular risk factors”−adjustments that have markedly reduced all crude risks, making most of them not statistically significant. It is not clear, however, what further reductions would have been achieved if the study had also adjusted for other confounders previously identified by the same authors in the same cohort and for which directly applicable data must have been at hand−notably the considerable influence of shift work,1 weight change,2 trans-fatty acid intake,3 and height.4 Possibly, such adjustments might have been inconsequential, but the report is silent.
Although the study is said to represent the prospective experience of >32 000 nurses, a scant 152 cases make it comparable to a modest case-control study subject to the uncertainties of a remote one-time determination of exposure, of misclassification, and of clinical verification of markers−inter alia−as the report acknowledges. The authors mention the negative reports from the American Cancer Society and National Mortality Followback Survey databases5 6 7 but resist warning that a nonsignificant risk from their frail database is inconsistent with the no-risk reports from those very much larger databases.
The authors are justifiably troubled in endorsing passive smoking risks substantially greater than the CHD risk attributed to active smoking by the Surgeon General.8 In attempting to circumvent the problem, however, they appear to have produced a larger one by arguing that the CHD risk of active smoking is understated because smokers are usually compared with nonsmoking control subjects, many of whom are exposed to passive smoking CHD risk. Yet, smokers are also exposed to passive smoking risk at doses and durations far exceeding nonsmoker exposures, which would call for a drastic reduction of the apparent risk attributable to active smoking, assuming the argument and the high passive smoking risks claimed in this report were true.
The article laments the absence of a positive gradient in relation to exposure duration and adduces uncertainties of recall as a reason, but it should be said that the absence of a gradient could very much be real. Also, the authors insist in relating risks to “women,” and although it may be true that all nurses of this cohort are women, not all women are nurses. The authors themselves have documented in almost 100 articles that the women of the Nurses Health Study are quite exceptional and hardly representative of the US female population.
Speculation, of course, is essential and cognoscenti may see through the lines, but feeding such problematic messages to the unwary media challenges responsibility, especially if messages are endorsed by presumed luminaries. Of course, I may be blinded as an occasional consultant to the tobacco industry−still, if there are no answers to the points just raised, perhaps the authors or the Circulation editors might consider releasing appropriate cautionary messages to correct what may have become quite unwarranted public perceptions.
- Copyright © 1998 by American Heart Association
Kawachi I, et al. Circulation.. 1995;92:3178–3182.
Rich-Edwards JW, et al. Am J Epidemiol.. 1995;142:909–917.
Steenland K, et al. Circulation.. 1996;94:622–628.
The Health Consequences of Smoking. Cardiovascular Diseases: A Report of the Surgeon General. Rockville, Md: US Public Health Service, Department of Health and Human Services; 1983.
Dr Adlkofer suggests that selection bias could have accounted for our finding of an association between passive smoking and CHD. This possibility is based on two assumptions: that passive smokers were less likely to visit their doctors and receive a diagnosis of CHD, and that excluding women with a history of CHD at the beginning of each follow-up period led to a higher prevalence of undiagnosed CHD cases in the group of passive smokers. We believe that both assumptions are untenable. The Nurses’ Health Study participants are all registered nurses and have excellent access to health care. For instance, in 1992 when we asked about the participants’ screening behaviors, only 2.4% of the cohort had not had their blood pressure checked during the last 2 years. Moreover, there were no important differences in screening behavior according to passive smoking status. Dr Adlkopfer suggested that selection bias was introduced by our exclusion of women with a history of CHD at the beginning of each follow-up period. As we mentioned in our article (p 2375),R1 we did this to avoid potential misclassification caused by women who alter their exposure to passive smoking after developing a major illness and to assess newly incident disease in a population at risk. By using this methodology, 20 medically confirmed cases of CHD were excluded during the 10-year follow-up period. When we repeated the analysis including these 20 cases, the multivariate adjusted relative risk of total CHD among regular passive smokers compared with women not exposed was 1.73 (95% confidence interval [CI], 1.06 to 2.84) compared with the estimate of 1.91 (95% CI, 1.11 to 3.28) published in our article.R1
Both Dr Adlkopfer and Dr Coggins question the plausibility of the magnitude of the relative risks we obtained by comparing them with smaller relative risk estimates obtained from other studies of active smoking−for example, the American Cancer Society (CPS II) Study, which reported a relative risk of 1.8 for CHD in female smokers. On this issue, Drs Adlkopfer and Coggins were selective in their choice of cited studies; we are aware of just as many cohort studies (eg, the Swedish cohort study by Cederlof et al,R2 the Rancho Bernardo Study,R3 the Finnmark Study,R4 and our own Nurses’ Health StudyR5 ) that found relative risks of CHD of between 2.6 and 3.6 in female current smokers.
Dr Coggins simply reiterates the limitation that we already acknowledged in our article, that measurement of passive smoking was by self-report and at baseline only. Yet, as we discussed in our articleR1 as well as elsewhere,R6 any resulting random misclassification of exposure could only change the relative risk estimates in the direction of the null. If misclassification were nonrandom, the “nonexposed” group in our study is more likely to have included passive smokers than vice versa, because people consistently underestimate their exposure to secondhand smoke.R6 Furthermore, workplace restrictions on smoking became more common during the study period, so that the “exposed” group in our study probably became progressively mixed with women who were no longer exposed. Both types of misclassification push the relative risk estimates in the direction of the null.
Although Dr Coggins questions our use of questionnaire assessment of ETS exposure, he himself places much weight on the study by Layard,R7 which also used questionnaires completed by surrogate respondents. We did not cite Layard’s pooled estimate of relative risks across just three studies because a more thorough meta-analysis incorporating data from 12 studies has been published,R8 which suggests a statistically significant increase in risk of CHD with passive smoking.
Dr Brennan suggests that studying the risks of passive smoking is analogous to making mountains out of molehills. We were unable to follow the logic of his calculations; however, we would point out that according to credible estimates,R9 some 35 000 to 40 000 deaths from coronary disease each year may be attributable to passive smoking.
Dr Gori speculates about how our relative risk estimates might have been influenced by adjusting for other risk factors such as height and shift work. As we stated in our article,R1 we adjusted for a very broad range of factors that we determined a priori to be potential confounders. The reason for not adding shift work to our models was that this exposure was not assessed until the 1988 questionnaire,R10 whereas passive smoking was ascertained in 1982. We nonetheless repeated our analysis from 1988 through 1992, including shift work as a covariate. On the basis of 73 cases of total CHD, the multivariate adjusted relative risk among women regularly exposed to passive smoking compared with those never exposed was 2.12 (95% CI, 0.92 to 4.89). This point estimate is quite similar to the relative risk of 1.91 reported in the full study; the width of the 95% CIs reflects the halving of cases caused by shorter follow-up duration. Lacking a sound rationale as to why other variables such as height might act as important confounders in the relation between passive smoking and coronary disease, we believed that it would be pointless to keep adding more variables to our models.
We were puzzled by why Dr Gori refers to the American Cancer Society CPS-II StudyR11 as a “negative report.” Although two consultants funded by Philip Morris published an analysis of the CPS-II study reporting no association between passive smoking and CHD risk,R12 this report was superseded by a more thorough analysis carried out by the American Cancer Society investigators, which did report a positive association.R11
Finally, Dr Gori states that we “have documented in almost 100 reports that women of the Nurses’ Health Study are quite exceptional and hardly representative of the US female population.” Although this population was selected for study because their training and motivation would provide superior data quality, there is no reason to believe that biological relations based on our findings are not relevant to women in general.
Kawachi I, Colditz GA, Speizer FE, Manson JE, Stampfer MJ, Willett WC, Hennekens CH. A prospective study of passive smoking and coronary heart disease. Circulation. 1997;95:2374–2379.
Cederlof R, Friberg L, Hrubec Z, Lorich U. The Relationship of Smoking and Some Social Covariables to Mortality and Cancer Morbidity: A 10-Year Follow-up in a Probability Sample of 55 000 Swedish Subjects Age 18 to 69. Stockholm, Sweden: Karolinska Institute, Department of Environmental Hygiene; 1975.
Barrett-Connor E, Khaw KT, Wingard DL. A ten-year prospective study of coronary heart disease mortality among Rancho Bernardo women. In: Eaker ED, Packard B, Wenger NK, Clarkson TB, Tyroler HA, eds. Coronary Heart Disease in Women. Bethesda, Md: National Heart, Lung, and Blood Institute, National Institutes of Health; 1987.
Njolstad I, Arnesen E, Lund-Larsen PG. Smoking, serum lipids, blood pressure, and sex differences in myocardial infarction: a 12-year follow-up of the Finnmark Study. Circulation. 1996;93:450–456.
Kawachi I, Colditz GA. Confounding, measurement error, and publication bias in studies of passive smoking. Am J Epidemiol. 1996;144:909–915.
Layard MW. Ischemic heart disease and spousal smoking in the National Mortality Followback Survey. Regul Toxicol Pharmacol. 1995;21:180–183.
Kawachi I, Colditz GA, Stampfer MJ, Willett WC, Manson JE, Speizer FE, Hennekens CH. A prospective study of shift work and risk of coronary heart disease in women. Circulation. 1995;92:3178–3182.
Steenland K, Thun M, Lally C, Heath C Jr. Environmental tobacco smoke and coronary heart disease in the American Cancer Society CPS-II Cohort. Circulation. 1996;94:622–628.
LeVois ME, Layard WW. Publication bias in the environmental tobacco smoke/coronary heart disease epidemiologic literature. Regul Toxicol Pharmacol. 1995;21:184–191.