A hierarchical network model for the estimate of the energy expenditure in individuals with type 1 diabetes.

Total daily energy expenditure (TDEE) is impacted by many medical conditions, such as diabetes. In the case of type 1 diabetes (T1D), individuals need to have an accurate assessment of the energy expenditure in real-time to avoid dietary imbalance, and improve glycemic control. This work proposes a hierarchical Long Short-Term Memory (LSTM)-based modeling approach to predict real-time continuous energy expenditure, expressed as metabolic equivalents (METs), for individuals with T1D on a 24-hour basis by leveraging the step count and heart rate data from a wrist-band smartwatch. To deal with the inter- and intra-individual variability, the proposed model uses three different LSTMs to capture population, activity-type and subject scale information. To evaluate the impact of the components of the hierarchy, the performance of the proposed hierarchical model was assessed at each level. The results show that the combination of population data, such as heart rate and step counts, with individual data in a hierarchical architecture helps to achieve superior prediction performance, than using only individual heart rate and step counts data.Additionally, compared to non-hierarchical modeling, the hierarchical modeling can provide precise and individualized prediction of the METs categories, as it allows the integration of the variation at different levels of the hierarchy. This model can be used to augment current automated insulin delivery (AID) systems to adapt insulin infusion according to the predicted activity intensity and compensate for glycemic perturbations due to exercise.