Defying the Circadian Rhythm: Clustering Participant Telemetry in the UK Biobank Data

TL;DR

This study develops a scalable clustering method for wrist-worn accelerometer data from UK Biobank, revealing how physical activity impacts survival in individuals with disrupted circadian rhythms.

Contribution

It introduces a standardized, low-dimensional feature and a matrix variate mixture model approach for efficient clustering of large-scale telemetry data.

Findings

Distinct survival outcomes identified for different clusters

Physical activity shows protective effects on circadian disrupted individuals

Methodology enables scalable analysis of large biobank telemetry data

Abstract

The UK Biobank dataset follows over 500,000 volunteers and contains a diverse set of information related to societal outcomes. Among this vast collection, a large quantity of telemetry collected from wrist-worn accelerometers provides a snapshot of participant activity. Using this data, a population of shift workers, subjected to disrupted circadian rhythms, is analysed using a mixture model-based approach to yield protective effects from physical activity on survival outcomes. In this paper, we develop a scalable, standardized, and unique methodology that efficiently clusters a vast quantity of participant telemetry. By building upon the work of Doherty et al. (2017), we introduce a standardized, low-dimensional feature for clustering purposes. Participants are clustered using a matrix variate mixture model-based approach. Once clustered, survival analysis is performed to demonstrate distinct lifetime outcomes for individuals within each cluster. In summary, we process, cluster, and analyse a subset of UK Biobank participants to show the protective effects from physical activity on circadian disrupted individuals.