Activity profiles of adults aged 50 - 70 years: functional data analysis
Year: 2019
Type: dissertation
Study design: Observational Study
Abstract: Physical activity has a major impact on health. Questionnaires are the most \ncommon method of physical activity assessment. While cost effective, these are \nsubjective and can correlate poorly with actual activity levels. Accelerometers \nhave gained popularity given their accuracy, objectivity and ability to capture \nlarge amounts of data. Simple summary measures such as the total or average \nactivity over the day are often used. However, these fail to exploit the \nlongitudinal nature of the data and do not capture the variation in activity \nlevels throughout the day. This study intends to capitalise on this nature by \nimplementing a functional data analysis approach. Activity data was collected from a cohort of 475 people in Mitchelstown in 2011. The individuals wore wrist worn accelerometers in a free living environment for a week. This data was collapsed into 1 minute epochs and each epoch was then aggregated over the week to get an estimate of daily circadian activity. The discrete wavelet transform was chosen as the smoothing technique to reveal the underlying functional nature of the data. This allows every individual in the cohort to be represented by a smooth activity profile. This study aimed to \nidentify and characterise subgroups within a cohort based on these activity \nprofiles. Functional principal component analysis was applied to these activity profiles in \norder to explore the dominant patterns within the data. Each individual’s \nprofile was approximated by a weighted sum of profiles and these weights were \nthen used to perform a cluster analysis. Five distinct subgroups were identified. \nThese differed from each other in both the magnitude of the activity and the \ntimes at which the activity occured. A more simplified approach, based purely \non the distance between profiles, was also implemented. Two distinct clustering \nmethods identified the exact same 5 subgroups in the cohort. To ensure their \nrobustness, these results were subject to a sensitivity analysis with respect to \nthe epoch length, smoothing technique and number of functional components \nutilised in the clustering. Other studies have clustered accelerometer data in terms of absolute activity volume, as in high or low activity groups. However, they do not place too much value in using the granularity of the data to determine what time of day people \nare active. In addition to the high, moderate and low activity subgroups, our \nanalysis revealed two subgroups which have a propensity to be active in either \nthe morning or evening. It is suggested that these are indicative of an \nindividual’s biological rhythm or chronotype. The Mitchelstown cohort was \nre-screened 5 years later in 2016, which presents an exciting opportunity to \nexamine changes in these profiles over time.
Author: Richard Weedle
Language: English
Topic: Physical Activity and Health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W3042121808