Physical activity is an important determinant of morbidity and mortality. In recent years, 3D accelerometry has become the state of the art in assessing physical activity to provide so-called "objective" measurement points that are free of any information distortion.

However, most approaches to analyze the data are far from fully exploiting the existing potential. The goal of this seed capital project is therefore to secure funding for the development of open source software for the analysis of accelerometer data. By using novel approaches to machine learning that go beyond the state of the art of available software packages, it will be of high relevance for public health researchers. In order to achieve this goal, we will cooperate in an interdisciplinary way with partners of the Laboratory for Machine Learning and Data Analysis (MaD) at the University of Erlangen-Nuremberg. We will conduct a literature review and organize an expert workshop to collect results and methods relevant to public health. In addition, we will conduct small experiments using available data from an AEQUIPA subproject called OUTDOOR ACTIVE and newly generated data. Finally, we will write a proposal to the German Research Foundation (DFG) to secure further funding.

Detailed description

Physical activity (PA) has a major impact on overall health, morbidity and mortality. According to the World Health Organization, about 4% of deaths worldwide are due to lack of exercise. In the past, PA was evaluated by means of questionnaires. More recently, measurements using accelerometers have become easier as a cost-effective alternative that provides unbiased data. However, the data are usually processed using very simplified approaches that a) waste a large part of the information collected and b) make assumptions about the data that are often not justified. For example, the data is downloaded from the device either as a raw signal or in so-called activity counts (AC). In the past, AC was used for small storage capacities and has become the de facto standard. The algorithms that calculate this AC from the raw signal vary depending on the manufacturer and often remain elusive. In addition, PA is then often categorized into intensity levels based on official PA recommendations. However, the limits for these categories are set arbitrarily and the AC limits usually remain unstandardized. In addition, many researches only perform simple regression analyses and do not use the full range of statistical possibilities.

The planned project will use the expertise of computer scientists and their state-of-the-art approaches to signal processing, pattern recognition and machine learning to feed into the field of lifestyle epidemiology and its implications for the development of physical activity programmes. Recent breakthroughs in machine learning for time series data could be exploited by adapting them to data analysis for physical activity assessment. By bringing the disciplines together, the solutions and open access software offered by the future project will have a lasting impact on the digital public health sector.

In a first step, evidence for design decisions from a public health perspective was collected and compared with the technical possibilities. To ensure the relevance and validity of the third-party funded project proposal, expert knowledge is actively sought inside and outside the LSC DiPH. A workshop open to all members of the LSC is held to discuss the results of the activities. To further prove the concept, small experiments will be conducted and data from practical experience will be analyzed. More than 1800 data sets of accelerometer data collected within the AEQUIPA subproject OUTDOOR ACTIVE can be used for this step. The data will be enriched by health-related longitudinal data.

The main goals are the development of open source software for the analysis of accelerometer data with high relevance for public health scientists, going beyond the state of the art of available software packages, and the submission of an application to the German Research Foundation for an interdisciplinary third-party funded project to develop innovative methods for the analysis of accelerometer data.

Contact
Fabian Flaßkamp
Universität Bremen
Institute for Public Health Nursing Research
AG Epidemiologie des demographischen Wandels
Raum: A1060
Grazer Straße 2a
28359 Bremen
Telefon: +49 421 218-68872
E-Mail: flasskamp@uni-bremen.de