Dying for a smoke: how much does differential mortality of smokers affect estimated life-course smoking prevalence?

An extensive literature uses reconstructed historical smoking rates by birth-cohort to inform anti-smoking policies. This paper examines whether and how these rates change when one adjusts for differential mortality of smokers and non-smokers.

Using retrospectively reported data from the US (Panel Study of Income Dynamics, 1986, 1999, 2001, 2003, 2005), the UK (British Household Panel Survey, 1999, 2002), and Russia (Russian Longitudinal Monitoring Study, 2000), we generate life-course smoking prevalence rates by age-cohort. With cause-specific death rates from secondary sources and an improved method, we correct for differential mortality, and we test whether adjusted and unadjusted rates statistically differ. With US data (National Health Interview Survey, 1967-2004), we also compare contemporaneously measured smoking prevalence rates with the equivalent rates from retrospective data.

We find that differential mortality matters only for men. For Russian men over age 70 and US and UK men over age 80 unadjusted smoking prevalence understates the true prevalence. The results using retrospective and contemporaneous data are similar.

Differential mortality bias affects our understanding of smoking habits of old cohorts and, therefore, of inter-generational patterns of smoking. Unless one focuses on the young, policy recommendations based on unadjusted smoking rates may be misleading.