Abstract
Physical inactivity is an urgent public health concern, being strongly implicated as a risk factor for many chronic diseases including certain types of cancer. It is not possible to understand or alter physical activity and inactivity levels without first being able to accurately measure these behaviors. Objective assessment of physical activity and inactivity using wearable accelerometer-based monitors has overcome the problems and limitations associated with more commonly used self-report tools. These wearable sensors are ideal for collecting information about physical activity (PA) and sedentary behavior (SB). They can be worn for extended periods of time, impose minimal inconvenience to the participant and researcher, are relatively inexpensive and can produce detailed accounts of PA and SB that are relevant to health (e.g. estimates of energy expenditure, time in moderate-to-vigorous physical activity, sedentary time) These PA and SB metrics can then be used to comprehensively characterize dose-response relationships. This presentation will highlight recent advances in objective monitoring of physical activity and sedentary behavior. The talk will describe recent work that uses machine learning tools for processing activity monitor data to classify physical activity intensity and estimate energy expenditure. Additionally, the assessment of sedentary behavior, which has been shown to be an independent risk factor for disease, will be discussed.
The approach we use to assess physical activity and sedentary behavior with wearable sensors first involves the calibration of sensors. In the lab, we directly measure energy expenditure or directly observe behavior while participants perform a variety of locomotion, household, sport activities and sedentary behaviors (e.g. sitting performing computer work, filing). Simultaneously, signals from the wearable accelerometers are collected continuously. We then use features from the accelerometer signal to generate machine learning algorithms (e.g. neural networks) to predict energy expenditure, intensity of physical activity or to identify activity type. As a final step we validate these algorithms on independent samples to test the precision and accuracy of these algorithms. To examine the performance of these algorithms in the field, we have subjects carry out their normal activities in a natural setting while wearing the monitors (subjects are observed and the direct observation data are recorded continuously). This presentation will describe data from these laboratory and field studies to demonstrate performance of algorithms for predicting activity levels and sedentary behavior in free-living settings. Evidence supporting the sensitivity of these algorithms in estimating change in activity level will also be presented.
Citation Format: Patty S. Freedson. Technology to assess physical activity and sedentary behavior. [abstract]. In: Proceedings of the AACR Special Conference on Post-GWAS Horizons in Molecular Epidemiology: Digging Deeper into the Environment; 2012 Nov 11-14; Hollywood, FL. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2012;21(11 Suppl):Abstract nr IA12.