Disparities in undiagnosed and misdiagnosed dementia and the impact on health care utilization and costs - a novel approach using probabilistic dementia classifications

PROJECT SUMMARY/ABSTRACT Alzheimer’s Disease and Alzheimer’s Disease-Related Dementias (AD/ADRD) are the costliest diseases in the U.S. Yet they often go undiagnosed or get diagnosed after substantial delay, leading to undertreatment, delays in care, and underestimation of the total cost of AD/ADRD. Racial and ethnic minority individuals and

underserved rural populations are more likely to be underdiagnosed and undertreated. This project will estimate a novel probabilistic dementia classification for more than 10 million beneficiaries aged 65 and older in Medicare fee-for-service (FFS) and Medicare Advantage (MA). The classification will be

estimated using linked Health and Retirement Study (HRS) and Medicare data, building on an AD/ADRD classification model we developed in prior research that substantially improves the classification accuracy among racial and ethnic groups. We will determine the extent to which racial and ethnic minority groups and

underserved populations are disproportionally affected by AD/ADRD diagnostic errors and delays. Then we will estimate in large Medicare data how AD/ADRD status and diagnoses affect health care utilization and costs, such as emergency department visits, hospitalization, long-term nursing home admissions, Medicare total cost

of care, out-of-pocket costs, and mortality. Medicare claims and encounter data are often used to estimate the health care costs of AD/ADRD because the data cover almost the entire population 65 and older, have detailed records of health care utilization, and have objective claims-based diagnostic information about AD/ADRD. However, the widely used Chronic

Conditions Warehouse algorithm assigns dementia status based on a broad range of International Classification of Diseases diagnosis codes, many of which use cognitive limitations to stand in for dementia, leading to misclassifications and biased cost estimates. We will use a probabilistic correction for these biases.

The project has four aims: (1) Derive a novel probabilistic AD/ADRD classification method using HRS and Medicare data and document the fraction of older adults who live with AD/ADRD without medical diagnoses or receive delayed diagnoses; (2) Estimate the impact of AD/ADRD status and diagnosis on health care

utilization, costs, and mortality; (3) Examine disparities in AD/ADRD status and diagnosis and their cost implications, focusing on differences between racial and ethnic groups and rural and urban populations; and (4) Share knowledge and methods with the Consortium and disseminate findings to stakeholders.

The derived measures of AD/ADRD status and diagnoses will significantly improve research on the economic and health impacts of AD/ADRD and provide more accurate estimates for minority populations. The findings from this work will inform AD/ADRD policymaking by showing potential gains from correcting diagnosis

errors and providing needed services to patients and caregivers.