Uncovering personalised glucose responses and circadian rhythms from multiple wearable biosensors with Bayesian dynamical modelling

Motivation Wearable biosensors measure physiological variables with high temporal resolution over multiple days and are increasingly employed in clinical settings, such as continuous glucose monitoring in diabetes care. Such datasets bring new opportunities and challenges, and patients, clinicians and researchers are today faced with a common challenge: how to best capture and summarise relevant information from multimodal wearable time series? Here, we aim to provide insights into individual blood glucose dynamics and their relationships with food and drink ingestion, time of day, and coupling with other physiological states such as physical and heart activity. To this end, we generate and analyse multiple wearable device data through the lens of a parsimonious mathematical model with interpretable components and parameters. A key innovation of our method is that the models are learnt on a personalised level for each participant within a Bayesian framework, which enables the characterisation of inter-individual heterogeneity in features such as the glucose response time following meals or underlying circadian rhythms. This framework may prove useful in other populations at risk of cardiometabolic diseases. Summary Wearable biosensors and smartphone applications can measure physiological variables over multiple days in free-living conditions. We measure food and drink ingestion, glucose dynamics, physical activity, heart rate (HR) and heart rate variability (HRV) in 25 healthy participants over 14 days. We develop a Bayesian framework to learn personal parameters that quantify circadian rhythms and physiological responses to external stressors. Modelling the effects of ingestion events on glucose levels reveals that slower glucose decay kinetics elicit larger postprandial glucose spikes, and we uncover a circadian baseline rhythm of glucose with high amplitudes in some individuals. Physical activity and circadian rhythms explain as much as 40-65% of the HR variance, whereas the variance explained for HRV is more heterogeneous across individuals (20-80%). A more complex model incorporating activity, HR and HRV explains up to 15% additional glucose variability, highlighting the relevance of integrating multiple biosensors to better predict glucose dynamics.