Comparison of Claims‐Based Algorithms for Dementia Identification: A Cross‐Sectional Analysis of the National Health and Aging Trends Study ( NHATS )

BACKGROUND: Accurate identification of Alzheimer's disease and related dementias (ADRD) in administrative data is increasingly important for research, care delivery innovation, and policy applications such as risk adjustment. Despite the widespread use of existing claims-based algorithms to identify ADRD, few studies have compared their use to ongoing longitudinal aging studies. METHODS: In this cross-sectional study, we included 2204 NHATS respondents aged 65 years or older classified as having dementia and enrolled in fee-for-service Medicare in 2019. We determined the performance characteristics of 3-claims-based dementia algorithms using NHATS as a reference standard. RESULTS: Among NHATS respondents classified as having "probable" dementia, sensitivity was generally low (48.8%-70.5%), while specificity was high (91.5%-97.8%) across all 3 algorithms. The Chronic Conditions Warehouse (CCW) algorithm had the highest sensitivity (70.5, 95% CI: 64.3-76.6), followed by Bynum-Standard 3-year (65.8, 95% CI: 59.6-71.9). Bynum-Standard 1-year had the highest PPV (65.6, 95% CI: 56.9-74.4). The Youden Index and F1 score were generally low and clustered closely across the three algorithms. Odds of ADRD misclassification differed across patient characteristics. CONCLUSION: Commonly used claims-based algorithms to identify ADRD had reasonable validity, though each presented important context-dependent tradeoffs and none was uniformly superior. These findings highlight the importance of accurately diagnosing ADRD in clinical settings and of understanding each algorithm's strengths and limitations based on the use case. Limitations include that NHATS is not a "gold standard" for comparison of claims-based algorithms and we only analyzed NHATS respondents in a single year.