Abstract
BACKGROUND: Cancer and dementia are becoming increasingly common co-occurring conditions among older adults. Yet, the influence of participant cognitive status on the validity of self-reported data among older adults in population-based cohorts is unknown. We thus compared self-reported cancer diagnoses in the U.S. Health and Retirement Study (HRS) against claims from linked Medicare records to ascertain the validity of self-reported diagnoses by participant cognitive and proxy interview status. METHODS: Using data from HRS participants aged ≥67 who had at least 90% continuous enrollment in fee-for-service Medicare, we examined the validity of self-reported first incident cancer diagnoses from biennial HRS interviews against diagnostic claim records in linked Medicare data (reference standard) for interviews from 2000 to 2016. Cognitive status was classified as normal, cognitive impairment no dementia (CIND), or dementia using the Langa-Weir method. We calculated the sensitivity, specificity, and κ for cancer diagnosis. RESULTS: Of the 8 280 included participants, 23.6% had cognitive impairment without dementia (CIND) or dementia, and 10.7% had a proxy respondent due to an impairment. Self-reports of first incident cancer diagnoses for participants with normal cognition had 70.2% sensitivity and 99.8% specificity (κ = 0.79). Sensitivity declined substantially with cognitive impairment and proxy response (56.7% for CIND, 53.0% for dementia, 60.0% for proxy respondents), indicating poor validity for study participants with CIND, dementia, or a proxy respondent. CONCLUSIONS: Self-reported cancer diagnoses in the U.S. HRS have poor validity for participants with cognitive impairment, dementia, or a proxy respondent. Population-based cancer research among older adults will be strengthened with linkage to Medicare claims.