Develop and apply novel analytic methods for device-based data on sitting and physically-active time.
Improve self-report and device-based measurement methodologies.
State of the science
Accurate measurement is necessary for documenting levels of sitting time, understanding dose-response relationships with specific health parameters, and evaluating the efficacy and effectiveness of programs to reduce this behaviour. Both self-report and device-based measures of sitting time provide important and complementary insights, with self-report providing contextual information, and device-based measures providing quantitative information on the duration and patterns of sitting time.
However, the ability of both self-report and device-based measures to capture intervention change remains poorly understood, and additional research is needed to identify the optimal method for combining self-report and device-based data. Further, the research potential of device-based measures has been severely under-utilised, with analyses of free-living data typically restricted to the application of count thresholds or regression-based prediction models.
An emerging approach to data reduction that promises to revolutionise device-based measurement of physical activity and sedentary time is pattern recognition. With mobile phones and accelerometer-based motion sensors providing easy access to large volumes of high frequency raw tri-axial accelerometer signal, pattern recognition has enormous upside potential for device-based measurement of sitting and other activity-related behaviours.
Recent work by CI-C Trost has demonstrated standard supervised learning algorithms such as artificial neural networks_ENREF_8, and support vector machines_ENREF_19 to be highly accurate in recognising activity type (e.g. sitting s. standing) and predicting energy expenditure. However, before these models can be deployed to the field, additional research is needed to identify features and/or algorithms that have sufficient time resolution to reliably detect free-living activities and transitions from one activity to the next.
Moreover, user-friendly systems for storing and processing these vast volumes of data need to be established.
Proposed research in this theme
To further advance the development of valid, yet practical measures of sitting and physically-active time, and significantly enhance research capacity in chronic disease prevention in Australia and worldwide, we will address the following research priorities:
- Evaluate self-report sitting time measures within intervention studies and across varying population sub-groups (e.g. adults in aged care) and delivery modes (e.g. online).
- Advance methods for integrating self-report and device-based measures.
- Devise and test novel pattern recognition or machine learning data processing techniques to classify activity type (e.g. sitting, standing, walking) and predict energy expenditure in age-diverse free-living samples.
- Evaluate the predictive validity of previously trained algorithms in independent samples. The extent to which previously trained algorithms require additional training and tuning prior to use in a new sample or different study population (e.g. children versus older adults) has important implications for software design and uptake by the broader research community.
- Develop a computing infrastructure that simplifies the application of pattern recognition approaches for public health researchers and accommodates the massive data upload, storage and processing requirements associated with this method.
Our multidisciplinary research team is uniquely qualified to address the aforementioned research priorities. Theme chairs CI-C Trost and CI-F Healy are internationally recognised leaders in sedentary behaviour and activity measurement, and their broad network of collaborators and industry partners will facilitate uptake and application of findings arising from the CRE.
AI Matthews will provide expertise in both population surveillance of sitting time, as well as novel applications (e.g. online) for the collection of contextual data48,49. CI-C Trost and AI Wong have successfully developed and validated several supervised learning algorithms to detect activity type and predict energy expenditure from accelerometer data18,20.
The involvement of AI Wong in the CRE will allow our team to extend this work and develop novel data analytic approaches to reliably detect sedentary time in age-diverse free-living samples. Our collaboration with AI Wong’s laboratory will also provide a unique opportunity for our Research Fellows to work with researchers and postgraduate students from computer science and engineering, and thus enhance their multidisciplinary research skills.
AI Chastin provides expertise into measuring the patterns of sitting time change following intervention (Themes 2 and 3); he also provides a link with activity monitor industry partners for potential device refinement and development based on our CRE findings. Deakin University have pledged a 3-year ‘value-add’ domestic PhD scholarship to assess sitting time patterns in children using these devices which will provide the opportunity for CIs Salmon and Timperio to team with CI-C Trost, CI-F Healy and AI Chastin.
AI Friedenreich will provide expertise and guidance to our Research Fellows on the development and implementation of innovative sedentary behaviour outcome measures for large-scale intervention trials.
Utilising our strengths in both self-report and device-based methodologies, our CRE will develop valid and practical measures of sitting and activity time, and capacity in the use and application of such measures, within population-based investigations, intervention trials, and mechanistic studies.