Here is a finding of fault with the methodology. A very serious fault.
Editor—According to Enstrom and Kabat's figures the greater had been a man's cigarette consumption in 1959 the less likely, it seems, was the death of his wife from coronary heart disease.1 However, an age bias existed in those women at the outset. In 1959 their mean age decreased with spousal smoking, such that the wives of men smoking 40 a day were a mean four years younger than wives of men smoking one to 19 a day, probably as a consequence of early death of smoking husbands of similarly aged wives (table 3 on bmj.com).
During the study period mortality from coronary heart disease fell by about 15% every four years.2 The "passive" smokers were therefore predominantly from later cohorts for whom, age for age, mortality from coronary heart disease had fallen significantly in comparison to controls. The same argument applies to never smoking husbands of smoking women who had an average age four to five years lower than controls (table 2 on bmj.com). Adjusting for age alone will not remove this interaction of age and time of observation.
Moreover, the Cox proportional hazard model is critically dependent on assumed proportionality between two survival curves at all points following entry to the study.3 Mortality from coronary heart disease increases almost exponentially for most of adult life and the mortality curves of risk groups for coronary heart disease differ not only in scale but also in doubling time. As such their survival curves cannot be proportional, yet this was not tested.
The effectiveness of age adjustment in this study is questionable, the year of observation should have been taken into account, and the statistical method is potentially unsound. The biological implausibility of the trend in relative risk may well be an expression of systematic bias in the method.
Eugene Milne, deputy medical director