Disparities in the Diagnosis of Alzheimer's Disease Among African American and European American Veterans

Seattle Inst For Biomedical/Clinical Res, Seattle WA

Investigators

Linked publications & trials

Abstract

Summary/Abstract Alzheimer's disease (AD) is a progressive neurodegenerative disorder that currently affects more than 5 million Americans, a number that is expected to triple by 2050. The National Plan to Address AD has articulated a vision for the development of specific treatments for AD by 2025, and it is critical that minorities and Veterans be included within the scope of this projected research progress. However, the diagnosis of AD is complicated by the lack of confirmatory diagnostic tests and the presence of other comorbid processes, including cerebrovascular disease, PTSD, and traumatic brain injury. These issues have contributed to a significant underdiagnosis of AD among Veterans, particularly among African Americans (AAs). Considering diagnosis of AD in an ethnic minority context means paying particular attention to AAs, who have estimated dementia prevalence and incidence rates that are about two times higher than the rates in European Americans (EAs). There is also evidence of racial disparity in dementia care, though the magnitude of this disparity is unknown. To better understand the magnitude of the underdiagnosis of AD, this proposal will leverage the big data resources of VHA, where ~0.6 million AA Veterans who are age 65+ receive care?this is one of the largest samples of older AAs with associated electronic health records (EHRs) in the U.S. Given that it is infeasible to conduct manual record reviews or in-person evaluations in every Veteran served by VHA with undiagnosed AD, Specific Aim 1 will develop an AD-status model by applying novel weakly supervised machine-learning methods to the EHRs of 22,000 AA and EA Veterans who are age 65 or older. These weakly supervised methods are novel in that they will permit data mining within a large number of EHRs, including both structured and unstructured data, without initial human annotation or labeling. By developing an AD-status model in this way, the proposed work will avoid the potential limitations of past natural language processing studies of AD, which used predetermined search terms to generate models and were not sufficiently accurate. Specific Aim 2 will conduct medical record reviews and telephone assessments, the ?silver standard? for diagnosing AD, in 600 randomly selected Veterans and their caregivers. This combination of machine learning and follow-up assessment, which has not been previously utilized in AAs or AD, will enable the validation and refining of the AD-status model so that it can predict the likelihood that AA Veterans are true AD cases. In this way, the proposed interdisciplinary team of dementia specialists, bioinformaticists, and biostatisticians will likely demonstrate the rapid and efficient detection of complex and costly disorders like AD. This has the potential to lead to earlier and more accurate diagnoses of AD, which can then improve Veteran access to medical and psychosocial supports and delay institutionalizations. Thus, future investigations of short- and long-term biomedical, pharmacological or system-level AD interventions will be feasible in AAs.