Abstract

AbstractBackgroundThe proportion of people living with dementia (PLWD) is increasing due to aging global populations. Frailty is a complex, multifactorial, and age‐related state of increased vulnerability commonly co‐occurring with Alzheimer’s disease (AD) and vascular dementia (VaD). Variability in frailty levels of PLWD motivates deeper investigation into dementia’s presenting characteristics. Machine learning techniques applied to large medical datasets can identify characterize disease phenotypes. Therefore, the aim of this study is to use unsupervised machine learning to discover frailty‐related clusters in AD and VaD.MethodWe will use data from the United Kingdom Biobank (UKBB), a large‐scale prospective cohort database which includes more than 500,000 participants aged 40‐69 years at baseline. The data include electronic health records, brain magnetic resonance imaging, polygenic risk scores, and frailty index scores. We will select features for clustering using principal component analysis. Clustering algorithms will include k‐means, affinity propagation, and latent class analysis. After cluster characterization, we will evaluate the association between clusters and prevalent AD (n = 2,634) and VaD (n = 1,664) using logistic regression models. Regression models will be stratified by sex and adjusted for age. Receiver‐operating curve will be plotted, and area‐under the curve will be compared using C‐statistics. This study is supported by the Canadian Consortium on Neurodegeneration in Aging Trainee Synapse Project Funds. Our data application has been approved by UKBB.ResultWe expect that the clustering algorithms will reveal several clusters of similar PLWD and allow evaluations based on shared patterns of regional brain atrophy, the degree of frailty, and polygenic risk. In addition, we also expect that clusters characterized by high frailty levels will be associated with prevalent AD and VaD.ConclusionFrailty is an important factor in the clinical evaluation of PLWD. The use of unsupervised machine learning techniques may reveal distinct high‐risk clusters that are related to prevalent AD and VaD. This work may improve early identification of PLWD and aid in the development of personalized care and therapy for people at risk of AD and VaD.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call