Aging
Assessing the Impact of Differential Alcohol Consumption Operationalization and Covariate Adjustment on Cognitive Outcomes in KHANDLE: A Multiverse Analysis David Kapaon* David Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Kapaon Department of Epidemiology, Boston University, School of Public Health
Research indicates heavy drinking harms older adults’ cognition. However, evidence on low-to-moderate drinking is mixed, in part due to variability in operationalizing alcohol consumption. We used multiverse analysis to assess alternative measures of alcohol consumption and differential covariate adjustment on cognition.
In the Kaiser Healthy Aging and Different Life Experiences Study, executive function (EF) and verbal episodic memory (VM) scores were z-standardized to wave 1. Each wave, participants reported the number of days/week drinking in the past 3 months (frequency: 0; 1-6; 7) and drinks/day when drinking (volume: 0; 1; 2; 3; 4+). We derived a continuous measure of alcohol consumption (drinks/week) and four sex-specific measures using definitions from US/Canadian federal agencies. Using parametric g-computation, we looped through all exposure-covariate combinations and ran linear mixed models with random intercepts (10,240 models). All models adjusted for sex, race/ethnicity, and age (linear; linear & quadratic). Covariates in the multiverse included sociodemographic, anthropometric, psychosocial and behavioral factors, interview mode, and enrollment cohort. For each model, we estimated contrasts in cognitive levels and declines between the observed world and a counterfactual world where all respondents abstained from drinking.
75% of the sample reported no alcohol consumption. Models with the CDC’s drinking definitions showed the greatest heterogeneity in estimated effects, ranging from moderately worse outcomes (EF level = -0.12 SD; 95% CI: [-0.21, -0.02]) to small improvements (VM level = 0.05 SD; [-0.05, 0.15]) for abstention across models. No models showed substantial differences in average rates of EF or VM decline.
Abstention achieved small population impacts on cognition, likely due to high proportions of non-drinkers. Multiverse analysis revealed effects differed in sign depending on alcohol consumption operationalization and covariate adjustment.
