Identification of Prevalent Dementia in Medicare Advantage Encounters, Compared with Rigorous Cohort Assessments

Rapid growth in Medicare Advantage (MA) enrollment has spurred dementia research using diagnoses in MA encounter data. However, the validity of MA-encounter-based dementia diagnoses is uncertain.

Using four cohort studies at the Rush Alzheimer's Disease Center linked to Medicare data from 2017-19, we validated prevalent dementia indicators in MA encounter data, based on the Bynum-Standard algorithm, against dementia status determined by rigorous cognitive assessment in cohort studies. We replicated analyses among participants in Traditional Medicare (TM).

Of the 508 eligible MA enrollees, mean age was 82.8 (SD = 7.2) years, 81% were female, and 49% were non-Latino White; 62 participants (12%) were classified as having dementia by the cohort assessment. MA-encounter-based dementia indicators performed reasonably well in identifying participants with cohort-assessed dementia: using encounters alone, positive predictive value was 63% (95% CI: 52-75%), negative predictive value was 96% (95% CI: 94-98%), and accuracy was 92% (95% CI: 89-94%). Inclusion of chart reviews did not impact performance. Compared to TM-based indicators during the same period, MA-based indicators yielded similar validity metrics but identified a less functionally impaired group as having dementia. In secondary validation analysis of MCI, sensitivity was low.

As one of the first studies to validate dementia diagnoses in MA encounter data, indicators based on the Bynum-Standard algorithm can be a valid tool for identifying prevalent dementia. Findings support the use of MA-encounter-based dementia indicators for research within MA. However, observed health differences between MA- and TM-identified dementia groups suggest potential for selection bias in MA plans.