Aging
Examining 20-year income volatility and 10-year memory decline in a longitudinal synthetic cohort. Katrina Kezios* Katrina Kezios Scott Zimmerman Peter Buto M. Maria Glymour Adina Zeki Al Hazzouri
Income volatility from young to middle adulthood predicts worse midlife cognition, but its impact on cognitive decline is unknown. Examining this association in a single cohort is difficult: few studies span early-adulthood to later-life, and those that do lack detailed, repeated, and prospective data on finances and cognition. To fill this gap, we created a synthetic cohort by linking two US national cohorts of baby boomers: the National Longitudinal Survey of Youth 1979 (NLSY79) and the mid-Baby Boomer Health and Retirement Study (HRS) subcohort. In NLSY79, we defined income volatility as the number of income drops >25% between successive surveys from 1990-2010 (0 vs. 1, 2, ≥3). In the HRS, we examined memory function every 2 years from 2010-2020 via a composite score that incorporated direct and proxy assessments. To create the synthetic cohort, each HRS participant was matched to and assigned the income volatility history of their 20 most similar NLSY79 counterparts based on the following matching variables measured in 2010 (when age~50 in each cohort) and harmonized: family income, memory score, education, age, sex, race/ethnicity, parental education, employer-provided health insurance, age first married, marital status, wealth, self-rated health, and chronic health conditions. Each of the 20 pairs for a given HRS participant was redistributed into 1 of 20 analytic datasets. In each dataset, we used confounder-adjusted linear mixed models to estimate the effect of income drops on 10-year memory decline, pooling results across datasets using Rubin’s Rules. In the synthetic cohort, experiencing more income drops was associated with faster memory decline in a dose-response fashion (β1_v_0_drops=-0.0015, 95% CI: -0.005, 0.002; β2_v_0_drops=-0.0035, 95% CI: -0.007, -0.0003; β3_v_0_drops=-0.0065, 95% CI: -0.009, -0.004). The synthetic cohort approach allowed evaluation of a novel question, but its validity depends on strong assumptions about matching across cohorts.