Leveraging survey and claims data to identify high-need Medicare beneficiaries in the National Health and Aging Trends Study

Background Multiple algorithms have been developed to identify and characterize the high-need (HN) Medicare population. However, they vary in components and yield different populations, and were developed for varying purposes. We compared the performance of existing survey and claims-based definitions in identifying HN beneficiaries and predicting poor outcomes among a community-dwelling population. Methods A retrospective cohort study using Round 5 (2015) of the National Health and Aging Trends Study (NHATS) linked with Medicare claims. We applied HN definitions from previous studies to our cohort of community-dwelling, fee-for-service beneficiaries (n = 4201) using sampling weights to obtain nationally representative estimates. The Belanger et al. (2019) definition defines HN as individuals with complex conditions, multi-morbidity, acute and post-acute healthcare utilization, dependency in activities of daily living, and frailty. The Hayes et al. (2016) definition defines HN as individuals with 3+ chronic conditions and a functional limitation. We applied each definition to survey and claims data. Outcomes were hospitalization or mortality in the subsequent year. Results The proportion of NHATS respondents classified as HN varied greatly across definitions, ranging from 3.1% using the claims-based Hayes definition to 32.9% using the survey-based Belanger definition. HN respondents had significantly higher mortality and hospitalization rates in 2016. Although all definitions had good specificity, none were able to predict outcomes in the following year with good accuracy. Conclusions While mortality and hospitalization rates were significantly higher among respondents classified as HN, existing claims and survey-based HN definitions were not able to accurately predict future outcomes in a community-dwelling, nationally representative sample measured by the area under the curve.

Automated summary

Different algorithms have been developed to identify and characterize the high-need Medicare population, but existing definitions were not able to accurately predict future outcomes.