Aging
Cross-walking proxy reports and direct episodic memory assessments to reduce selective attrition in dementia studies Eleanor Hayes-Larson* Eleanor Hayes-Larson Hayes-Larson Hayes-Larson Hayes-Larson University of Southern California
In dementia research, bias due to selective attrition can arise when cognitively impaired participants are unable to complete cognitive assessments. Proxy respondents can provide alternative cognitive assessments, but proxy-rated cognition is not directly comparable to respondent-based memory test scores. Using data from the 2016 Health and Retirement Study (HRS) Harmonized Cognitive Assessment Protocol (HCAP), we developed and validated a cross-walking approach to convert proxy-rated cognition scores to directly-assessed episodic memory scores in the HRS. Among 3,029 HCAP participants with both direct (immediate and delayed 10-word recall) and proxy cognitive assessments (IQCODE), we cross-validated prediction models to estimate total (sum of immediate and delayed) word recall scores from proxy-reported IQCODE scores, evaluating 16 model specifications (including linear or spline IQCODE, respondent age, proxy relationship, and interactions) and selecting the best-performing model based on RMSE. We applied the final model to construct longitudinal memory trajectories for the HRS cohort followed 2000-2022using generalized estimating equations (GEE), restricted to individuals aged ≥50 who were self-respondents in 2000 (N=15,456). We used respondent word recall scores when available, and predicted scores based on the final model when only proxy cognitive assessments were available. For comparison, we implemented inverse probability weighting (IPW) to account for differential likelihood of proxy vs self-response over time. GEE models (Figure) confirmed that analyses of only respondent direct assessments underestimated the rate of memory decline over time due to selective attrition; cross-walking and IPW produced similar and steeper trajectories. The cross-walking approach retains cognitively impaired participants using a simple conversion formula, offering a flexible alternative to weighting methods that can be complex to implement longitudinally.
