Modelling change processes in multivariate interrupted time series data using a multivariate dynamic additive model: An application to heart rate and blood pressure self-monitoring in heart failure with drug changes

Abstract

Abstract Heart rate (HR) and blood pressure (BP) measured by a patient can be used to monitor response to pharmacologic therapies. Continuously measured HR and BP are multivariate time series as sequences of values at regularly spaced intervals over time and changes in pharmacologic therapies interrupt the multivariate time series and create phases. In the multivariate interrupted time series, there are change processes in multiple phases, such as level changes, linear or non-linear trend changes, and time-varying serial dependence. This paper presents an application of a multivariate dynamic additive model to account for these change processes. In addition, a simulation study is conducted to evaluate the model’s parameter recovery and to demonstrate the consequences of ignoring the time-varying serial dependence in HR and BP when detecting level changes and trend changes. The results of the simulation study show that the accuracy and precision of parameter estimates are satisfactory. Furthermore, the simulation results present that ignoring time-varying serial dependence in the same conditions as those found in the application results in biased estimates and standard errors for the level changes and trend changes.