Identifying clusters of cognitive functioning trajectories in elderly: A comparison of three methodologies

• Activity Number: 175 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Victor Talisa* and Tianxiu Wang and Zhongying Xu and Joyce Chang
• Companies: Department of Biostatistics, University of Pittsburgh and University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: PROC TRAJ ; clustering ; machine learning ; trajectory

Abstract

Dementia is a common disorder characterized by substantial decline in one or more cognitive domains with age. However, some cognitive slowing is typical of normal aging. We analyzed 5-year trajectories of memory and executive functioning (EF) from an ongoing prospective cohort study of adults 65 years or older to determine whether the data support the existence of clusters of trajectories characteristic of normal aging, or a future diagnosis of mild cognitive impairment or dementia. We compared results from 3 statistical approaches: the latent group-based dual trajectories model, the OPTICS algorithm, and the tight clustering k-means procedure (TC). For all analyses, we examined data from 5-year age windows: 65-69, 66-70, ., 85-89 years. Individuals with at least 3 measurements for a given window were included. Both OPTICS and TC used as inputs the coefficients from subject-specific linear regression models of memory or EF on time. The three methods were compared using several metrics, including the total number of resulting clusters, cluster size, degree of cluster separation, and the relationship between clusters and individuals' future Clinical Dementia Rating (CDR) scores.