Smoking and Cognitive Decline at Older Age: Using Surveillance and Imputation to Deal With Differential Mortality

Background: Smoking is a risk factor for vascular diseases and likely associated with cognitive decline. However, observed associations may be impacted by higher cohort attrition, particularly mortality, among smokers. We incorporated surveillance data between visits using multiple imputation by chained equations (MICE) and assumed the Missing At Random mechanism in missing cognitive function. We assessed the impact of addressing the influence of differential visit attendance driven by mortality on the association of smoking with cognitive decline. Methods: We studied 4,960 participants in Atherosclerosis Risk in Communities (ARIC) Neurocognitive Study, who were free from dementia at baseline (2011-2013) with cognitive batteries at baseline and follow-up (2016-2017, N = 3,078). Cognitive decline was measured as change in standardized global cognitive factor scores. Self-reported smoking was collected at previous visits (1987-1998) and annual follow-up interviews (1998 to 2011) of the ARIC study, and pack years were estimated. We used baseline factor scores, dementia risk factors, and cognitive function screening at annual follow-ups and dementia surveillance information after baseline to impute missing factor scores. Cognitive decline was analyzed using linear regression on both complete data and using MICE. Results: Participants had a mean age of 75.7 (SD: 5.1) at the index visit. 59.6% (2956 of 4960) were female and 18.6% (921 of 4960) were African American. Smoking ≥25 pack-years prior to baseline was associated with moderately lower visit attendance and higher mortality at older age. The association of smoking with cognitive decline was stronger after addressing differential attrition driven by mortality using MICE for smoking ≥25 pack-years, with two-fold increase in the effect size (Table 1). Conclusions: Smoking, particularly of longer duration and intensity, is a risk factor for cognitive decline, with larger effect size when dealing with the differentially higher mortality among heavier smokers.