Are Home- and Community-Based Services Cost-Effective?

Description:
In this issue of Medical Care, the authors compare the Medicare and Medicaid costs of care for dually eligible beneficiaries receiving long-term services and supports in nursing homes versus through home- and community-based services (HCBS). They find that HCBS is cheaper, mainly because of lower spending on long-term services and supports. The analysis is careful and includes a number of important robustness checks, which generally support the direction of the main findings. It is an intriguing contribution to the field, especially because, in contrast to some previous studies, they find that HCBS users do not generate significantly higher spending on acute care than do nursing home residents.

How to cite:

Tamara Konetzka R. Are Home- and Community-Based Services Cost-Effective?. Medical Care. 2016; 54 (3): 219-220. Available at: Link

Because the use of HCBS versus a nursing home is a choice, the analysis faces the challenge of selection bias. The authors use propensity score matching to reduce this bias, often the best alternative when one cannot randomize and cannot identify a valid instrumental variable. Specifically, they match beneficiaries on age, sex, race/ethnicity, indicators of functional status and comorbidity, and whether or not the individual lived alone. They show good balance on these variables after matching. The problem with using propensity score matching to reduce selection bias is that one can only match on observable characteristics, a limitation that the authors acknowledge. However, in this case, we know that each pair of matched individuals—similar in demographics, whether they lived alone, and several binary indicators of health status—nonetheless made very different decisions on how to receive their care, one choosing HCBS and one choosing a nursing home. Can we really assume that these decisions were as good as random? Or were they potentially driven by unmeasured health status and the availability of caregivers at home, in which case the selection bias problem has not been solved? The issue of unmeasured health status as a potential problem is underscored by the fact that mortality is higher in the postmatch nursing home group than in the HCBS group, and the magnitude of savings from HCBS decreases by about a third when previous hospitalization is included in the propensity score equation. Would the effect remain, or reverse, with much more refined information on baseline health status?

Owner: