Errors in estimates of smoking-related deaths derived from nonsmoker mortality

A number of recent analyses have computed present and future costs associated with a risk by estimating what would happen if the risk were absent. Two sources of bias are associated with this approach: (1) differences in confounding factors between present risk avoiders and risk takers, and (2) the difficulty of selecting an unbiased sample of risk avoiders. A staff memo from the Office of Technology Assessment used this approach to estimate mortality due to smoking. Numbers of deaths and age at death distributions of U.S. smokers and nonsmokers for all causes, all cancers, lung cancers, heart disease, and cerebrovascular lesions are used to assess the accuracy of these estimates. Large errors in the OTA estimates are found. Conditions are discussed that might help reduce errors from this approach.