completed a questionnaire in 1982 and were then followed up (see U.S. Department of Health and Human Services, 1989). The study found that among current smokers (the vast majority of whom were lifelong smokers consuming, on average, about 20 cigarettes per day), overall death rates were 2 to 3 times higher than for lifelong nonsmokers of the same age and sex (Peto et al., 1992). Indeed, for some diseases strongly associated with smoking, such as lung cancer, death rates were up to 25 times higher for smokers compared with nonsmokers. The disease-specific relative risks derived from this study, together with smoking prevalence data for the United States, have been used by the U.S. Surgeon General to estimate smoking-attributable deaths in the United States by applying the classic attributable risk formula (Levin, 1953).
where P is the proportion of population exposed (smoking prevalence), and RR is the relative (disease-specific) risk of death for smokers and nonsmokers.
Given the size and recency of the ACS study, the question arises of whether its results can somehow be applied, perhaps with suitable scaling, to other populations to estimate smoking-attributable mortality. The problem in doing so is that the ACS study is not even representative of the U.S. population, let alone those of other countries.2 The ACS cohort was biased toward those of higher socioeconomic status and/or adults who were more health conscious than others. The cohort also was undoubtedly contaminated by the "healthy volunteer" effect, whereby those who were already suffering from a serious illness in 1982 were unlikely to participate in the study. There are, in addition, major conceptual problems in applying the ACS results directly to other populations (as has frequently been done), since the hazards of smoking depend very much on past, as well as current, consumption patterns, and on the intensity and distribution of a number of cofactors (e.g., hypertension, hypercholesterolemia). Thus, even if current smoking prevalence (P) were known for a given country, there could be and undoubtedly are sufficiently important differences between smokers in that country and in the ACS cohort to invalidate direct application of the attributable-risk formula.
The methodology suggested here for circumventing these difficulties is to estimate smoking-attributable deaths in populations indirectly by assuming that the current lung cancer rate in a given country reflects adequately the entire smoking history of that country in terms of prevalence, duration, intensity, and relative risk for lung cancer. This methodology is described in detail below. It should be noted here, however, that implicit in this methodology is the basic assumption that lung cancer is essentially unicausal (i.e., smoking) and that other cofactors have a negligible impact. This is clearly not the case for many developing countries, where indoor air pollution is a major cause of lung cancer (espe-