A Retrospective Study of Administrative Data to Identify High-Need Medicare Beneficiaries at Risk of Dying and Being Hospitalized

BackgroundDeveloping a definition of what constitutes high need among Medicare beneficiaries using administrative data is an important prerequisite to evaluating value-based payment reforms. While various definitions of high need exist, their predictive validity for different patient outcomes in the following year has not been systematically assessed for both fee-for-service (FFS) and Medicare Advantage (MA) beneficiaries.ObjectiveTo develop a definition of high need using administrative data in 2014 and to examine its predictive validity for patient outcomes in 2015 as compared to alternative definitions for both FFS and MA beneficiaries.DesignRetrospective cohort study of national Medicare claims and post-acute assessment data.ParticipantsAll Medicare beneficiaries in 2014 who survived until the end of the year (n=54,717,039).Main MeasuresTwo or more complex conditions, 6 or more chronic conditions, acute or post-acute health services utilization, indicators of frailty, complete dependency in mobility or in any activities of daily living in post-acute care assessments, hospitalization, mortality, days in community, Medicare expenditures.Key ResultsBased on our definition of high-need patients, 13.17% of FFS and 8.85% of MA beneficiaries were identified as high need in 2014. High-need FFS patients had mortality rates 7.1 times higher (16.23% vs. 2.27%) and hospitalization rates 3.4 times higher (40.69 vs. 12.03) in 2015 compared to other beneficiaries. Competing high-need definitions all had good specificity (0.88). Having 3 or more Hierarchical Chronic Conditions yielded a good positive predictive value for hospitalization, at 0.50, but only identified 19.71% of FFS beneficiaries hospitalized and 28.46% of FFS decedents that year as high need, as opposed to 33.92% and 51.98% for the new definition. Results were similar for MA beneficiaries.ConclusionsThe proposed high-need definition has better sensitivity and yields a sample of almost 5 million FFS and 1.5 million MA beneficiaries, facilitating outcome performance comparisons across health systems.