Methodological issues in estimating smoking-attributable mortality in the United States.

The authors explored two methodological issues in the estimation of smoking-attributable mortality for the United States. First, age-specific and age-adjusted relative risk, attributable fraction, and smoking-attributable mortality estimates obtained using data from the American Cancer Society's second Cancer Prevention Study (CPS II), a cohort study of 1.2 million participants (1982-1988), were compared with those obtained using a combination of data from the National Mortality Follow-back Survey (NMFS), a representative sample of US decedents in which information was collected from informants (1986), and the National Health Interview Survey (NHIS), a nationally representative household survey (1987). Second, the potential for residual confounding of the disease-specific age-adjusted smoking-attributable mortality estimates was addressed with a model-based approach. The estimated smoking-attributable mortality based on the CPS II for the four most common smoking-related diseases-lung cancer, chronic obstructive pulmonary disease, coronary heart disease, and cerebrovascular disease-was 19% larger than the estimated smoking-attributable mortality based on the NMFS/NHIS, yet the two data sources yielded essentially the same smoking-attributable mortality estimate for lung cancer alone. Further adjustment of smoking-attributable mortality for disease-appropriate confounding factors (education, alcohol intake, hypertension status, and diabetes status) indicated little residual confounding once age was taken into account.