Partial information decomposition of physiological time series provides a novel framework for identifying markers of autonomic control in cardiovascular and cardiorespiratory networks.
Abstract The analysis of complex physiological systems increasingly relies on multivariate approaches that capture the dynamic interactions within and between interdependent subsystems. One promising measure in this context is the predictive information (PI), quantifying the amounts of Shannon information that is dynamically shared among past and future states of a multivariate system. While the PI is often decomposed using standard information theory, in this work we demonstrate that the emerging tool of partial information decomposition (PID) allows a better separation of unique and shared (redundant or synergistic) contributions of the past dynamics of the two components of a bivariate system to the present state of a designated target system. The proposed decomposition is applied to the study of the short-term regulation of heart period (HP) variability assessed in conjunction with systolic arterial pressure (SAP) variability to describe cardiovascular (CV) dynamics, or with respiration (RESP) variability to describe cardiorespiratory (CR) dynamics. The framework is implemented using both model-free and model-based estimators and applied to HP, SAP and RESP time series measured in young healthy subjects at rest state and during postural or mental stress. We observe that HP predictive dynamics are strongly determined by unique self-influences, which play a major role in driving the changes in process regularity observed moving from rest to postural and mental stress, and that the variations across conditions of the measures of CV and CR coupling are determined predominantly by the atoms of shared information (redundant and/or synergistic). Overall, our findings highlight the relevance of decomposing the PI by means of the PID tools for assessing information dynamics in physiological networks and their potential for identifying novel markers of autonomic control.
Saputo et al. (Wed,) studied this question.