Integrating Raw Accelerometry Data and Time-Use Data for Activity Type Recognition

KeyNAKO-955

Project leadMichael Stein

Approval date20.08.2024

Published date16.10.2024

SummaryIn the NAKO study, hip-worn accelerometry data are available from approximately 70,000 participants, representing 20 TB of data. Our group was responsible for performing the quality assurance of these data (NAKO-647), conducted on the high-performance computing cluster at the University of Regensburg. As part of that project, we derived a comprehensive dataset of movement at 15-minute-, hourly, daily, and recording-level resolutions, now available for use by the broader research community [1]. In addition to our work on the accelerometry data, our group conducted quality assurance of the computer-based 24h physical activity recall (cpar24) data (NAKO-71 [2], NAKO-418 [3]), which involved extensive data cleaning. The cpar24 is a time-use survey that assessed physical activity along with contextual information over 24 hours in about 60,000 participants. In the next phase of our work (the current application), we aim to derive additional accelerometer variables in the subpopulation of participants who provided both accelerometry and cpar24 data. Specifically, we propose to use contextual and temporal data from the cpar24 to develop algorithms that distinguish individual types of activities from accelerometer data. Linking accelerometer data to cpar24 data will produce variables on specific activity types, such as lying down, walking, sitting, standing, and potentially more complex activity types like cycling, sitting in a vehicle, or engaging in occupational physical activities. We will also examine whether physical activity based on the combination of raw accelerometry and cpar24 contextual data, and circadian activity rhythms, are associated with health. Thus, the current proposal entails initially developing novel accelerometry variables, followed by an assessmenmt of their content validity. This will be achieved by verfying whether the newly derived variables correspond with baseline health indicators in both the expected direction and magnitude. By doing so, we aim to further enrich the resources available to NAKO researchers interested in conducting detailed analyses of physical activity.

Keywords-

InstitutionsUniversität Regensburg, Leibniz Institut für Präventionsforschung und Epidemiologie - BIPS, Deutsches Krebsforschungszentrum, Studienzentrum Berlin-Mitte (Charité), Leibniz-Institut für Präventionsforschung und Epidemiologie BIPS

Go back