Public Spending on Acute and Long-Term Care for Alzheimer’s & Related Dementias

An estimated four to five million older adults in the United States are living with Alzheimer’s and related dementias (ADRD) (Langa, 2017), a chronic, progressive condition characterized by cognitive decline of sufficient severity to interfere with a person’s ability to carry out daily activities (Alzheimer’s Association, 2017). Despite recent evidence that prevalence rates of ADRD are declining (Langa, 2017), population forecasts indicate significant growth in the absolute number of adults in the U.S. living with dementia (Prince, 2013).

Understanding the magnitude of the medical care and long-term care costs attributable to dementia is important for public and private decision makers, but estimating these costs has been difficult. First, identifying people with ADRD can be difficult in secondary data, since diagnosis can be at different stages of the disease progression or lacking altogether. Second, one must isolate the costs attributable to ADRD among a population that has several co-occurring chronic and acute conditions. Third, our fragmented health system means that many players are responsible for different types of cost; Medicare, Medicaid, and the family all play sizable roles in funding care for individuals with ADRD.

We estimate the public spending on DRD using newly available data from the Health and Retirement Survey matched to Medicare and Medicaid claims data. We identify a retrospective cohort of older adults with ADRD, and perform sensitivity analysis around the definition of dementia onset. We examine Medicare and Medicaid expenditures for the 12 months prior and up to 60 months following a diagnosis of ADRD. In order to isolate the costs attributable to ADRD, we select a comparison group of HRS participants matching on sex, birth year, and HRS entry year. To calculate the marginal effect of ADRD on Medicare expenditures, we use the estimator described by Basu and Manning (2010) for estimating costs under censoring. We estimate costs using a two-part model; the first part estimates the probability of any costs during each month using a logit model, while the second part estimates the magnitude of costs when costs are greater than zero using a generalized linear model with gamma family and power link of 0.95. This estimation is done separately on two samples: (1) months prior to death or censoring, and again (2) for months in which death occurs. An accelerated failure time model based on the lognormal distribution for time is used to estimate each subject’s survival function after accounting for censoring. We use the method of recycled predictions in order to estimate the marginal effects from each of the models.

We estimate the costs attributable to ADRD, paying special attention to who bears these costs. We also examine how the total costs and the burden of costs has shifted over time and over the course of the disease.