Title: Searching for clusters that are replicable and clinically meaningful with applications to sleep health
Authors: Meredith Wallace - University of Pittsburgh (United States) [presenting]
Abstract: Applied statisticians tasked with clustering face a sometimes-insurmountable challenge: revealing solutions that are replicable and clinically meaningful. This task becomes even more daunting as high-dimensional, multi-modal data become the norm rather than the exception, and with the realization that variables are often generated from underlying biological processes that follow skewed or non-normal distributions. This is particularly true in sleep research, where one sleeps, health can be captured across multiple dimensions and modes of measurement. Moreover, many sleep characteristics are highly skewed, even in seemingly homogeneous subgroups. We will delve into statistical challenges and potential solutions related to identifying replicable and clinically meaningful clusters in sleep health data. This will include a discussion on how to rigorously perform variable and model selection in a way that reduces spurious findings, cutting-edge methods for evaluating cluster stability and separation with non-normal distributions, establishing clinical utility of identified clusters, and the importance of replicating findings externally. We will demonstrate these challenges and potential solutions in the context of my ongoing National Institute of Aging grant, which aims to identify sleep health phenotypes and link them to prospective health outcomes in a large multi-cohort sample of community dwelling older adults.