Machine learning to distinguish HAND from Alzheimer's disease in HIV over age 60

Investigator: Victor Valcour, MD, PhD
Sponsor: NIH National Institute of Mental Health

Location(s): United States


The proposed work will examine how clinicians can determine if older patients with HIV are also suffering from early Alzheimer's disease (AD). With patients now living into age ranges where both diseases become possible, guidance on how to distinguish impairment due to HIV from the early stages of AD is critically needed.

The CDC estimated that one-quarter of Americans living with HIV were over the age of 55 in 2012. By next year, they will be over age 60, entering into the age demographic where Alzheimer's disease (AD) becomes a distinct differential for clinicians. Because up to one-half of people living with HIV experience cognitive impairment from HIV or related factors along, the likelihood for delayed diagnosis of early AD is substantial. Differentiating HIV-associated Neurocognitive Disorder (HAND) from the Mild Cognitive Impairment stage of AD (MCI-AD) is one of the most pressing issues in geriatric neuroHIV. Current HAND nosology does not provide guidance on this issue. Published work suggests the likelihood for distinct phenotypes that would facilitate diagnostic sorting with commonly available inputs from neuropsychological testing and structural imaging. In this application, we will use a new approach that leverages computational machine learning with inputs from structural imaging, neuropsychological testing, motor examination and affective/behavioral assessments to determine the factors that most accurately discriminate HAND from MCI-AD. Our preliminary examinations using this novel technique demonstrate a likelihood that this approach will provide diagnostic sorting that exceeds 90% accuracy. We will examine tightly characterized phenotypes using HIV tests to exclude HAND and PET amyloid scanning to exclude AD among 75 HIV+/amyloid marker negative participants with HAND to 50 HIV-negative/amyloid+ cases with MCI (MCI-AD group), all age, sex and disease severity matched and all over age 60, the population of interest due to dual risk. Our methodology will iterate the most distinctive aspects of each disease's phenotype to inform sorting and subsequently, guidelines. We will validate the identified inputs that most clearly contribute to the algorithm though clinical correlations and through the ability of the determined clusters (e.g. diagnostic group) to predict the meaningful outcomes of disease progression. The long-term goal of this work is to inform clinical guidelines, thus, the modalities examined are readily available in clinical care. This work will also extend our understanding of neuropathology in older HIV patients and may identify factors that shift paradigms because our novel approach does not rely on a priori assumptions to inform neuropsychological abnormalities and brain structural alterations linked to HAND in older age. In an exploratory aim, we extend this examination of HAND neuropathogenesis with the added examination of diffusion tensor imaging (DTI) and a monocyte associated inflammatory marker, soluable CD163 (sCD163), two measures tightly linked to HAND in published work among virally suppressed patients in the current era.