Session S82.1

Quantifying the Complexity of Short-Term Heart Period Variability through K-Nearest Neighbor Local Linear Prediction

L Faes*, S Erla, G Nollo

Università di Trento
Trento, Italy

The complexity of heart period variability over a short time scale (~5 min) was quantified exploiting the paradigm that associates the degree of unpredictability of a time series to its dynamical complexity. Stationary time series of 300 consecutive heart periods were measured from ECG recordings in young healthy subjects during two experimental protocols: supine position (SU) vs. upright position after head-up tilt (UP) (15 subjects); spontaneous breathing (SB) vs. paced breathing at 0.2 Hz (PB1) and 0.3 Hz (PB2) breathing frequency (14 subjects). Prediction of each series was performed by k-nearest neighbor local linear approximation, i.e. fitting a local linear model to the k nearest neighbor states of the current state vector to predict its future value. Linear and nonlinear predictions were performed choosing a small (local prediction, k=30) and a large (global prediction, k~270) neighborhood size, respectively. Predictability and unpredictability indices were obtained taking the normalized mean squared prediction error (E) and the squared correlation between actual and estimated data (C), respectively. Considering local prediction, the predictability increased markedly moving from SU (E=0.43±0.13, C=0.58±0.12) to UP (E=0.15±0.08, C=0.85±0.08) positions, documenting the reduction of heart period complexity evoked by head-up tilt. The predictability also increased significantly from SB (E=0.48±0.18, C=0.53±0.17) to PB1 (E=0.32±0.14, C=0.68±0.13) conditions, while it decreased during PB2 (E=0.45±0.13, C=0.56±0.13), suggesting that the periodic forcing input given by paced breathing is effective in reducing heart period complexity only at slower breathing rates. Using global prediction we found comparable predictability values and variations among conditions. Nevertheless, the comparison between global and local prediction suggested the involvement of nonlinear dynamics in the generation of heart period variability in all conditions, as the predictability was larger for local than for global prediction in a significant number of subjects (5 and 6 out of 15 during SU and UP; 8, 7, and 9 out of 14 during SB, PB1 and PB2).

(Abstract Control Number: 169)