Uncovering selection bias due to enrollment in Medicare claims studies

Abstract Research using administrative claims data often excludes individuals with secondary payors or incomplete coverage. We aimed to determine whether doing so leads to selection bias in longitudinal studies of the Medicare population. Using nationally representative data from the National Health and Aging Trends Study (NHATS), we summarized Medicare enrollment trends from 2011-2019 and measured the association between self-reported sleep medication use and two outcomes: falls and dementia. There were 8,244 Medicare beneficiaries in the NHATS 2011 cohort, and the prevalence of Part C enrollment increased from 29% in 2011 to 42% in 2019. Among those without Part C, the percentage with complete (12 months) Part A, B, and D coverage increased from 45% in 2011 to 60% in 2019. To assess potential coverage-related selection bias, we repeated discrete-time survey-weighted survival analyses after restricting the cohort by common Medicare enrollment criteria (e.g., excluding Part C enrollees). Unadjusted and adjusted associations between sleep medication use and incident dementia or falls were broadly similar across Medicare-defined subsamples and the full cohort. Sleep medication use was not associated with dementia but was positively associated with falls. Sensitivity analyses requiring continuous eligibility showed consistent results. Enrollment in non-traditional Medicare has increased, but in these examples did not lead to significant selection bias.