LAUSR

W ith Medicare’s nearly $690 billion dollars of spending in 2021 (1), getting payment right is a highstakes business. Risk adjustment, the process of assigning a score to the health status of individuals to predict health care costs, is a central component used to determine payment in value-based models of care and Medicare Advantage. The same measures are also used by health systems to segment their population into distinct groups based on the type of care they need. Typically, morbidity, measured by number and types of diseases, is used for risk measures. Because most chronic diseases are age related, older adults, in whom multimorbidity is common, tend to be flagged as high risk. But in addition to chronic disease, age is associated with the onset of functional impairments and frailty. And people with functional impairments or frailty use more health care and have higher medical costs than those without these characteristics, independent of disease (2). Unlike the case of disease, our electronic health records (EHRs) and diagnostic codes for billing do not adequately capture functional impairment or frailty, which are therefore not historically included in risk adjustments. Algorithms using claims data to capture frailty have been developed, but concern remains that the limitations of claims data underestimate the influence of function and frailty on health care spending (3). In their article, Ensrud and colleagues demonstrate that the addition of functional status and frailty measured through patient reports and objective criteria to the claims-based measures of morbidity and frailty identifies large group differences in spending attributable to functional status and frailty (4). Their insight is important because it demonstrates just how much spending can be “missed” when using claims-based measures alone. The question is: how much value in the real world of delivering care to older adults would be added by using both patient-reported and claims-based measures in risk assessment? When the enriched functional status or frailty measures were included in the statistical models, Ensrud and colleagues first found that functional impairments were associated with an additional $2354 to $11 720 per year in Medicare spending and the presence of frailty was associated with an additional $6172 to $8532 per year in Medicare spending. The strong association in this study indicates that the enriched measures are better at grouping people by risk for high spending than previous methods and may therefore allow for better targeting of specialized models of care to prevent hospitalizations and adverse outcomes (5, 6). Keep in mind, however, that health care utilization in this study represents clinical patterns of care in fee-for-service Medicare from 1 to 2 decades ago (due to the research cohorts used); whether the same relationship exists in contemporary health care delivery remains to be shown. In addition, although functional status was associated with spending in both men and women, frailty does not reach significance in men, a finding consistent with prior research (7). Lastly, the usefulness would be further enhanced if the measures could be shown to be associated with persistently high costs, as opposed to the high costs that accrue in the last months of life (8), given the very high mortality with greater functional impairment noted by Ensrud and colleagues. Although the strength of association is useful for population segmentation, the ability to predict costs is needed when one is determining how much to pay providers for their care of a defined population (as in Medicare Advantage) or determining benchmarks or penalties (as in value-based payment models) (6). Ensrud and colleagues show that accuracy of prediction was statistically improved by using the enriched functional status and frailty measures in women, but not in men, which raises concern about the broad usefulness of the method. The accuracy (that is, percentage of people classified correctly into the top spending quartile vs. the bottom 3 spending quartiles) increased from 73.9% to 75.4% (or 1.5% more) for women and 75.6% to 76.1% (or 0.5% more) for men. These differences in accuracy are small and the costs of collecting the requisite function and frailty measures are likely high. Those costs would include embedding collection of function and frailty measures into the routine clinical workflow and incorporating them as structured data elements in EHRs for easy extraction. Doing so across all health systems participating in value-based payment orMedicare Advantage programs would entail high costs in labor and infrastructure changes, and increase the burden of measurement. It is unclear whether these marginal improvements in predictive accuracy could by themselves justify the systemic changes that would be required to obtain them. On the other hand, incorporation of enriched functional measures could be profoundly impactful when one considers combined Medicare and Medicaid spending. Older adults who have functional impairments and frailty not only have high medical spending, but are also at risk for requiring long-term care services paid by Medicaid among dually enrolled beneficiaries (8). Opportunities to integrate across these sectors are now evolving with the expansion of special needs plans (SNPs); SNPs are Medicare Advantage products that target people who have chronic and disabling conditions, are dually eligible, or who meet criteria for institutional care. Ensrud and colleagues' study likely understates potential costs of functional impairment to the federal government because of its focus on only Medicare spending in community-dwelling elders. Incorporation of enhanced frailty measures into EHRs, while costly, could potentially allow greater return on investment when one considers the potential for financial incentives that encourage more targeted management of functional impairment to prevent need for long-term care. But perhaps the most impactful use of enriched measures of functional status and frailty is not for the purpose of