The author of the tobacco industry funded article `The risk of coronary heart disease in nonsmokers exposed to environmental tobacco smoke' argues that previous studies have noted an association because they failed to adjust for confounders, and that the results of three previously unpublished cohort studies (CPS-I, CPS-II and MNFS) should be included (Gross 1998). Unfortunately for the author's hypothesis of no association, studies which did the best job at controlling confounding reported the highest risks (i.e. Garland et al. 1985, RR 2 . 7; Humble et al. 1990, RR 1 . 6; Svendsen et al. 1987, RR 2 . 2). Small increased risks found in other studies may have been understated by a failure to control for negative confounding, speciยฎcally for the eects of alcohol consumption.
The author's attack on the work of Glantz and Parmley (1995) and Wells (1994), who pooled results from various US studies, is completely unconvincing. Wells ranked studies from tier 1 to tier 4 in descending order of quality and arrived at a relative risk estimate for CHD mortality of 1 . 22 (95 per cent CI, 1 . 10ยฑ1 . 40). The only criticism the author can oer is that the estimate was heavily weighted by tier 3 studies. However, when attention is restricted to tier 1 and 2 studies the relative risk estimate was even higher (1 . 79). This estimate is not, as the author suggests, indefensible just because it is higher than expected. This estimate also cannot be ignored solely on the basis of a lack of precision.
The three cohort studies which the author wants included (which, not surprisingly, found only very modest eects) did the poorest jobs in controlling for confounding, adjusting for only age and sex. If the ability to control for confounding is important, then the author should not rely upon the CPS-I, CPS-II and MNFS studies. If it isn't important, the author cannot argue in his abstract that the `myriad . . . of confounders that have not been adequately adjusted . . . makes it foolhardy then to claim an association'.
REFERENCES