The authors reviewed claims-based algorithms. To identify individuals at greater risk of frailty, they developed new algorithms using Medicare fee-for-service claims that were validated using patient assessment data from two types of post-acute care providers: home health agencies and skilled nursing facilities. Finally, they compared the relative performance of the new and existing algorithms at predicting three claims-based outcomes.
- What new and existing algorithms might be used to help stakeholders and researchers identify individuals at greater risk of frailty and functional impairment?
Frailty is a clinical syndrome that is characterized by a constellation of symptoms, including loss of strength, low energy, and weight loss. According to research, the syndrome is associated with negative health outcomes, such as falls, disability, fractures, and increased risk of mortality. Research has also shown that frailty is associated with increased utilization and spending, independent of other medical risk factors. Identifying and quantifying frailty might be an important component of risk-adjustment for value-based payments or might help target specific interventions. Despite its importance, measuring frailty is challenging because of the lack of consistent measurement of frailty-related concepts.
The authors reviewed and refined claims-based algorithms. To identify individuals at greater risk of frailty and functional impairment, they developed new algorithms using Medicare fee-for-service (FFS) claims that were validated using patient assessment data from two types of post-acute care (PAC) providers: home health agencies (HHAs) and skilled nursing facilities (SNFs). Finally, they compared the relative performance of the new and existing algorithms at predicting three claims-based outcomes in a data set representative of all Medicare FFS beneficiaries. Overall, they found that using algorithms previously developed by Kim and colleagues and reported in a 2018 article performed best for most outcomes and subpopulations, although the new algorithms performed slightly better at predicting a nursing home stay in the following year by some metrics, particularly among PAC patients.
- New algorithms were developed and tested on a sample of 35,141,239 Medicare beneficiaries — 18 percent of whom were in the HHA group, 6 percent of whom were in the SNF group, and 76 percent or whom had neither an HHA nor an SNF stay during the study period. In these new algorithms, age and dementia diagnosis were significant predictors of both (1) memory limitation and (2) activity and mobility limitations outcomes.
- In a comparison of the new algorithms with existing algorithms from the 2018 research by Kim and colleagues and from 2015 research by Faurot and colleagues, the Kim model had the best overall performance at predicting claims-based outcomes of interest in a separate sample of Medicare FFS beneficiaries for most metrics and subpopulations tested, although the new algorithms performed slightly better at predicting a nursing home stay in the following year by some metrics, particularly among PAC patients.
- Results did not indicate that differential health care utilization by race/ethnic group or neighborhood socioeconomic status negatively affected model performance.
- Add Kim's claims-based frailty index scores to the Centers for Medicare and Medicaid Services (CMS) Chronic Conditions Data Warehouse to be used by CMS, researchers, and other stakeholders.
- These scores might also be useful to health systems for risk adjustment or for tracking quality of care and utilization for at-risk populations by stratifying measures.
- Researchers might consider including these scores as controls in evaluations of policies or using them to study the potential effect modification of frailty on different interventions.
Table of Contents
Development of the Study Population
Development of Algorithms to Predict Functional Impairment
Validation of the Developed Algorithms
Comparison with Existing Algorithms
Summary and Recommendations
Additional Tables and Figure