Panel Paper: Do Dementia-Care-Licensed Assisted Living Communities Improve Outcomes for Patients with Dementia? An Instrumental-variable Analysis

Friday, November 3, 2017
Acapulco (Hyatt Regency Chicago)

*Names in bold indicate Presenter

Portia Y Cornell1,2, Kali Thomas2 and Momotazur Rahman3, (1)Building 32, (2)Providence Veterans Health Administration Medical Center, (3)Brown University


Assisted living (AL) is growing in popularity as an alternative to nursing homes for individuals with Alzheimer’s disease who cannot live independently at home. We estimated the effect of residing in a dementia-care-licensed (DCL) communities on mortality, nursing-home admission, hospital inpatient days, and emergency department use among individuals with ADRD. The paper contributes to the 2017 APPAM theme, “Measurement Matters,” in the following ways: 1) we leverage a novel methodology to identify Medicare beneficiaries who living in assisted living, creating an unprecedented opportunity to learn about the effects of state assisted-living policy; and 2) utilizing instrumental-variable (IV) analysis to achieve the best possible causal estimates from cross-sectional data.

Study design: We identified Medicare beneficiaries residing in AL by matching the 9-digit zip codes of licensed AL communities with 25 or more beds the addresses of fee-for-service Medicare beneficiaries in six states that issue specialized DCLs (AL, CO, OR, MS, NY, OR, WV) in 2013. Within that sample, we identified a new-admission cohort of 6,761 individuals who resided outside of AL in the two years prior and who had an ADRD diagnosis. We estimated multivariate regression models of the study outcomes on an indicator for whether the community had a DCL license, controlling for residents’ socioeconomic characteristics, chronic conditions, AL community size, market characteristics, and state fixed effects. To address confounding from unobserved resident characteristics correlated with license type, we leveraged distance to the nearest DCL and non-DCL communities as an instrument for the resident choice.

Principal Findings: In the preferred instrumental-variable analysis, we estimated that DCL residents were 10% more likely to die during the study year and had 2.6 additional inpatient days in the hospital than non-DCL residents, controlling for unobserved patient characteristics. In contrast, the naïve multiple-regression models suggested that living in a DCL community was associated with 0.2 fewer ER visits, 0.5 fewer inpatient hospital days, 2.6 fewer days in skilled-nursing care, and no significant difference in mortality compared to living in a non-DCL community.

Conclusions: Residents who actively choose specialized care may have unobserved, systematic differences in their financial resources, preferences, health, and health-care use from non-DCL residents. If distance satisfies the required assumptions for a valid instrument, then the IV estimate is closer to the true causal effect of DCL licensing than multiple regression because it addresses this selection bias. States issue special dementia-care licenses both to ensure a minimum level of staff training on how to care for residents with dementia and basic structural requirements to keep these residents safe. In the presence of scant information about the quality of care in AL communities, licensing may also serve as a signal to individuals with ADRD and their families about staff experience and specialization that go above and beyond the minimum requirements.