Generalized Linear Mixed Models for Changepoint Analysis of Biomedical Time Series Data

Changepoint analysis for time series data identifies the time points for which properties about the data shift. In this paper, we are motivated by two distinct types of biomedical time series data: digital health monitoring and neuroimaging. The neuroimaging study consists of multivariate time series obtained from resting-state functional magnetic resonance imaging (fMRI). The digital health monitoring data consists of binary univariate time series observed within an individual throughout the day across multiple days of activity. The goal of the analysis is to model how physical activity changes within a day over a period of 56 days. The goal in the analysis of that data is to determine the association between dynamic functional connectivity (dFC), which is the interaction between different brain regions and how this interaction changes over time, and severity of non-suicidal self-injury (NSSI). In this paper, we leverage the framework of generalized linear mixed models to incorporate structure that is present in biomedical time series data for changepoint analysis. This structure could be, for example, the repeated measures of the time series data within an individual across multiple days or the interconnected community structure of brain networks. We embed the mixed model in a dynamic programming algorithm for detecting multiple changepoints. We evaluate the performance of our proposed method across several scenarios using simulations. Finally, we show the utility of our proposed method on our two distinct motivating applications.