Comparing model-free and model-based transfer entropy estimators in cardiovascular variability

Information flow between heart period (T), systolic pressure (S) and respiration (R) variability in a head-up tilt (HUT) protocol is assessed by transfer entropy (TE). Two estimates of TE are compared: the model-based (MB) approach using linear regression under the Gaussian assumption, and the model-free (MF) approach combining binning estimates of entropy and non-uniform delay embedding. The approaches were applied to 300-beats series of T, S, R measured in the supine (su) and upright (up) positions during HUT. Both MB and MF approaches detected a unidirectional information transfer from R to T and from R to S, and a significant decrease of the TE from R to T, as well as a significant increase of the TE from S to T, moving from su to up. For the MF approach, these trends were supported by the statistical test for TE significance. These results suggest that TE estimated from T, S and R variability can successfully describe the physiological mechanisms involved in the short term cardiovascular and cardiorespiratory regulation during HUT.