Non-Linear Dynamics. Other analysis techniques being applied to RR interval data come from the theory of non-linear dynamics, or "chaos theory" as it is popularly known (34,35). These techniques make a fundamentally different assumption about the data: that much of its behavior is aperiodic, but still not random. It remains to be seen whether this approach will provide additional useful physiologic or prognostic information. It is clear that such information is less intuitive in its interpretation, the measures less understandable in direct terms, and the statistical theory less advanced. Because these techniques are still exploratory, it is important that the complete details of computational technique be provided in publications reporting results obtained with these methods.
The CHRONOS algorithms
Considering the methodologies available, choices were make in developing CHRONOS for cardiovascular research: (1) prediction of death and arrhythmic events in coronary heart disease, and (2) physiologic and pharmacologic research. Time series measures were already standardized on RR intervals when we began to adapt our power spectral methods. We decided to work with time series of RR intervals rather than instantaneous heart rate in order to work in the same time and frequency domain dimensions. We have used Marquette algorithms for ÀËÎ conversion of the ECG signal and for labeling the QRS complexes. Editing is done with Marquette software and subsequently with algorithms developed at Columbia University (xx). The sampling of the RR time series is done with algorithms developed at MIT by Berger et al. (xx). CHRONOS uses methodology described by Albrecht and Cohen at MIT for splining across noisy segments of ECG recordings (xx). For prediction of risk of death or sustained ventricular arrhythmias from 24-hour continuous ECG recordings, we have used an in tote) FFT method developed at ÌÈÃ by Albrecht et al. (xx). For physiologic and pharmacologic research, we use methods developed by Rottman et al. (xx) for averaging adjacent, non-overlapping 5-minute segments of continuous ECG recordings. In recordings with modest amounts of noise or ectopic QRS complexes, the average LF or HF power calculated from the 288 5-minute ECG segments by CHRONOS are essentially identical to the values of LF or HF power obtained from an in toto FFT done on the entire 24-hour continuous ECG recording (xxRottman). Also, we did experiments to determine that excluding segments with more than 20% ectopic complexes or noise was the best compromise between consistent results and minimal exclusion of 5-minute segments with large amounts of ectopic complexes and/or noise.
The choices made for CHRONOS were finalized 1988 and been used with change since that time. The CHRONOS algorithms have proved to be effective in processing ECG recordings for physiologic and pharmacologic research and for predicting death and arrhythmic events in coronary heart disease populations.
Physiological Basis of High and Low Frequency Power
Power spectral measures of the RR time series can delineate cyclic fluctuations in the RR intervals in terms of their frequency and amplitude. Physiologic perturbations and pharmacologic interventions help to define physiologic systems responsible for cyclic fluctuations in the RR intervals. Studies of this kind have shown that high frequency power (0.15 - 0.40 H-1) represents a pure vagal efferent signal that is modulated by ventilation (respiratory sinus arrhythmia). Low frequency power (0.04 - 0.15 Hz) has contributions from vagal and sympathetic modulation of RR intervals. The LF/HF ratio gives an index of autonomic balance: high values indicate sympathetic nervous system predominance and low values indicate parasympathetic nervous system predominance. Although somewhat crude, high and low frequency power in a power spectrum of the RR time series together with mean RR interval provide important information about the autonomic nervous system in man. The information for spectral analysis can be obtained noninvasively and inexpensively which makes the test feasible to use in large-scale epidemiologic studies and clinical trials. The CHRONOS algorithms have been validated for this purpose.
