Estimation of Energy Expenditure and Physical Activity Classification with Wearables

Physical activity (PA) is defined as any bodily movement produced by skeletal
muscle that requires energy expenditure. The scientific evidence for the
beneficial effects are irrefutable. Regular PA is proven to help prevent and
treat several non-communicable diseases such as heart disease, stroke, diabetes
and different forms of cancer.
PA is a complex behaviour that is characterized by frequency, intensity, time
and type (FITT). In order to understand the effect of PA on health and our
general well-being, it is essential to monitor all four characteristics of PA.
A PA classification algorithm can assess the amount of time spent in different
body postures and activity. Making it possible to assess frequency, time and
type. In order to completely characterize PA, intensity needs to be estimated.
This can be done by the estimation of energy expenditure (EE).
Wearables play a crucial role in the monitoring of PA. They are practical way
to collect objective PA data in daily life, in an unobtrusive way, at a
relatively low cost. Furthermore they can be applied as a motivational tool to
increase PA. Accelerometry has been routinely used to quantify PA and to
predict EE using linear and non-linear models. However, the relationship
between EE and acceleration differs from one activity to another. For example,
cycling can generate the same acceleration amplitude as running, but the EE may
differ greatly. It is clear that acceleration alone has a limited accuracy to
estimate EE from different activities.
Improving the estimation of EE could be achieved by first classifying the
activity type. For each type of activity, different estimations can be used.
There are numerous methods to classify PA and estimate EE. Literature describes
the use of regression based equations combined with cut-points, linear models,
non-linear models, decision trees, artificial neural networks, etc. It is still
unclear what would be the best method to estimate EE, not to mention which
features would contribute to the model.
Another possibility is to add a relevant bio-signal to the estimation model.
Heart rate, breathing rate, temperature are all signals that have a response
related to an increase in PA. Heart rate has been used previously to improve
the EE estimation in combination with accelerometry. The breathing rate and
temperature could contribute to the estimation of EE is still unclear.
Therefore, the goal of the current study is twofold. Firstly, to explore the
contribution of different variables (physiological signals) to the estimation
of EE and the classification of PA. Secondly, develop and validate a model to
estimate EE and classify PA in simulated free-living conditions based on the
relevant variables.

Primary Objective:
The primary objective of this study is to develop and validate an energy
expenditure estimation and physical activity classification algorithm based on
wearable sensors. To do so the relevant signals contributing to the
classification of physical activity and the estimation of energy expenditure
will be identified.

Secondary Objectives:
Based on the collected data secondary objectives are:
- Design and validate a heart rate (variability) algorithm
- Assess the contribution of different bio signals to the estimation of energy
expenditure
- Investigate the feasibility of modelling the instantaneous energy expenditure

Study Design:
Cross-sectional exploratory and validation study. Using stratified, repeated
k-fold cross validation the models for the classification of PA and the
estimation of EE will be developed.

Setting:
Data will be collected at the facilities of the Metabolic Research Unit
Maastricht (MRUM).

Duration:
12 months from the start of inclusion, planned March 2022, after approval of
the METC.

The possible harms, burden and risks for the subject are minimal. The wearables
and electrodes are attached with medical grade patches. However, subjects might
experience some burden form the patch. The activities included in the simulated
free-living protocol do not induce additional risks, since they are all
activities of daily living. Subjects will need to walk and run at a moderate
speed on a treadmill. Therefore, we will use the physical activity readiness
questionnaire (PAR-Q+) as an inclusion criterium.
No major risks are involved and the burden for the subject is low to moderate.