Information decomposition of short-term cardiovascular and cardiorespiratory variability

We present an entropy decomposition strategy aimed at quantifying how the predictive information (PI) about heart rate (HR) variability is dynamically stored in HR and is transferred to HR from arterial pressure (AP) and respiration (RS) variability according to synergistic or redundant cooperation. The PI is expressed as the sum of the self entropy (SE) of HR plus the transfer entropy (TE) from {RS,AP} to HR, quantifying respectively the information stored in the cardiac system and transferred to the cardiac system to the vascular and respiratory systems. The information transfer is further decomposed as the sum of the (unconditioned) TE from RS to HR plus the TE from SP to HR conditioned to RS. Moreover a redundancy/synergy measure is defined as the difference between unconditioned and conditioned TE from RS to HR. We show that, under the linear Gaussian assumption for the underlying multiple processes, all the proposed information dynamical measures can be calculated analytically, and present a method for their computation from the parameters of a vector autoregressive model. The method is then evaluated on a simulated process reproducing realistic HR, AP and RS rhythms, showing how known cardiovascular and cardiorespiratory mechanisms can be characterized in terms of the proposed information decomposition measures.