A propensity analysis of cigarette smoking and mortality with consideration of the effects of alcohol

Although it is well established that cigarette smoking causes excess mortality, the extent of the increased risk has been challenged because self-selection biases and confounding factors may not have been adequately accounted for in prior studies. We therefore performed a propensity analysis on a population-based cohort. A logistic regression model was used to generate a propensity score for current smoking in 6,099 adults (mean age 46 years, 54% men, 36% current smokers) participating in the National Heart Lung and Blood Institute’s (NHLBI) Lipid Research Clinic Prevalence Study. During 12 years of follow-up, 513 subjects (8%) died. After adjusting for age, current smoking was strongly associated with death (compared with never and former smokers, relative risk [RR] 2.69, 95% confidence interval [CI] 1.98 to 0.64, p <0.0001 and RR 1.79, 95% CI 1.26 to 2.55, p = 0.001, respectively). After adjusting for a propensity score based on 27 covariates and the covariates themselves, current smoking remained strongly and independently predictive of excessive death risk in smokers compared with never and former smokers (adjusted RR 2.96, 95% CI 2.16 to 4.05, p <0.0001 and adjusted RR 1.87, 95% CI 1.31 to 2.67, p = 0.0006, respectively). Although smokers were more likely to also drink alcohol, an interaction was noted, whereby, after adjustment for propensity score and other covariates, current smoking was associated with a moderately strong increase in mortality among drinkers (adjusted RR 2.00, 95% CI 1.42 to 2.82, p <0.0001), but was also associated with a markedly increased death risk among nondrinkers (adjusted RR 4.74, 95% CI 3.24 to 6.92, p <0.0001). The independent association of smoking with death even after a rigorous propensity analysis argues that it is highly unlikely that the link between smoking and mortality is materially biased or confounded.