The association between smoking status and aging cognitive trajectory accounting for bias in attrition.

Abstract

Studies on the relationship between smoking and cognition yield conflicting results, potentially from survival bias. We examined the effect of smoking status on cognitive evolution, accounting for potential bias in attrition. We analyzed 21 216 Black and White participants from the REGARDS cohort study. In order to account for potential differential attrition, a sensitivity analysis compared the association between smoking status and age-sex-specific composite cognitive trajectories using linear mixed models, robust under missing at random assumption, and joint models, robust under the extended missing at random assumptions. Linear mixed model results indicated current smokers had lower overall composite cognitive scores compared to never smokers at baseline (-0.20, 95% confidence interval (CI): -0.23, -0.16) and at 14-years duration in the study (-0.26, 95% CI: -0.33, -0.20), also apparent when comparing current smokers to past smokers, and memory and verbal fluency composite scores. Joint modeling estimates yielded results similar to results from linear mixed models. Joint models are recommended as a potential means to investigate concerns about attrition, particularly in studies of factors associated with both cognition and mortality. This study provides further support that smoking is likely to have negative impacts on cognition.