Computational Approaches to the Estimation of the Components of Energy Balance in Humans

Background: Continuous, long-term measurement of energy balance behaviours is a significant challenge and of great scientific interest to the field of energy balance and a multitude of related fields. Methodologies such as doubly labelled water (DLW) are infeasible in large scale studies because of their expense. Recent developments in wearable technologies may offer an opportunity to overcome this issue, but uncertainty exists regarding their accuracy. Should accurate estimates of energy expenditure (EE) be obtainable from such devices, it will be possible to incorporate estimates into validated mathematical models to estimate the change in energy intake (EI), in free-living subjects.

Objectives: This thesis aimed to examine methods to estimate EE from wearable sensors in free-living subjects participating in the NoHoW trial, a weight loss maintenance intervention.

Methods: A series of studies were conducted to investigate the validity of EE estimates from the manufacturer estimates of the Fitbit charge 2, and machine learning models trained on the sensor outputs. Both manufacturer estimates and model predictions were compared in free-living and used to estimate PAEE and ∆EI in the NoHoW trial.
Results: Laboratory validation studies indicated that the manufacturer estimates of the Fitbit charge 2™ were inaccurate and subsequently, that machine learning models could provide more accurate estimates of EE. Comparisons were made to an established research-grade armband, the SenseWear armband mini™ which showed that the manufacturer estimates were in slightly better agreement than the developed algorithms. In the application of several EE estimation methods to the NoHoW dataset, ∆EI could be estimated and this demonstrated that caloric restriction was greatest in the earlier phases of the intervention and this diminished as time progressed.

Conclusions: Digital tracking technologies are providing novel opportunities for physiological research. This thesis took positive steps towards developing a methodological framework for the estimation of free-living EE, which will have implications for energy balance and related fields. Future work will examine the models developed in this thesis against DLW measurements, and evaluate energy balance modelling in a wider range of subjects and circumstances.