Circadian modulation on T-wave alternans activity in chronic heart failure patients

Average TWA activity has been shown to be an independent predictor of sudden cardiac death (SCD) in chronic heart failure (CHF) patients. However, the influence of circadian rhythms on TWA remains understudied. In this work, we assessed circadian TWA changes in a CHF population and evaluated whether the prognostic value of TWA indices is sensitive to the circadian pattern. Holter ECG recordings from 626 consecutive CHF patients (52 SCD) were analyzed. The index of average alternans (IAA), quantifying the average TWA level, was measured in 4 consecutive 6-hour intervals using a multilead fully-automated method. Survival analysis was performed considering SCD as an independent endpoint. IAA changed along the day, with statistically significant lower values during the night than during daytime. This pattern is similar to the one observed in the mean heart rate (HR). However, a low correlation (r=.18) was found between IAA and HR in windows of 128 beats. After dichotomization of patients based on the third quartile of IAA indices, IAA indices computed between hours 06-12 (IAA06-12) and 18-24 (IAA18-24) successfully predicted SCD (Hazard Ratio, HaR:2.34(1.33-4.13)per μV, andHaR:1.87(1.04-3.36) per μV, respectively). In conclusion, circadian variation should be considered for SCD risk prediction.


Introduction
A great part of deaths in patients with mild-to-moderate chronic heart failure (CHF) are represented by sudden cardiac deaths (SCD), being most of them as a consequence of ventricular tachyarrhythmias. In this context, implantable cardioverter-defibrillators (ICD) have been shown to be highly effective at terminating these life threatening arrhythmias. However, the cost-effectiveness of this therapy is still low and identifying patients at risk who would bene-fit the most from ICD therapy remains a clinical challenge.
T-wave alternans (TWA), a consistent beat-to-beat alternation in the amplitude or morphology of the ST segment and/or the T wave, reflects temporal and spatial heterogeneity of ventricular repolarization. It is presently regarded as a noninvasive risk marker for identifying patients at risk for SCD and ventricular vulnerability [1]. Although the long-term averaging of TWA activity in ambulatory recordings has been shown to be an independent predictor of SCD in CHF [2], the influence of circadian rhythms on TWA remains still understudied.
Our aim in this study was to assess circadian TWA changes in a CHF population as well as to evaluate whether the prognostic value of TWA is sensitive to this circadian pattern. To do that, a novel signal processing methodology, that improves TWA estimation under noisy conditions, has been proposed.

Study population
Consecutive patients with symptomatic CHF corresponding to New York Heart Association (NYHA) classes II and III were enrolled in the multicenter MUSIC (MUerte Súbita en Insuficiencia Cardiaca) study, a prospective study designed to assess risk predictors for cardiovascular mortality in ambulatory CHF patients [3]. The study protocol was approved by institutional investigator committees and all patients gave written informed consent. The 24-hour Holter ECG recordings of 626 patients (52 victims of SCD, 63 of other cardiac causes, 25 non-cardiac deaths and 486 survivors) with sinus rhythm and aged 18-89 years (62.7±11.9 years) were available for the present study. ECG signals were acquired by using SpiderView records (ELA Medical, Sorin Group, Paris, France) and two or three orthogonal leads (X, Y, Z) sampled at 200 Hz were available for each subject. Collection of clinical data for this population was reported in previous studies [3,4].
Patients were followed up every 6 months for a median of 48 months. SCD, defined as (1) a witnessed death occurring within 60 minutes from the onset of new symptoms unless a cause other than cardiac failure was obvious, (2) an unwitnessed death (< 24 hours) in the absence of preexisting progressive circulatory failure or other causes of death, or (3) death during attempted resuscitation, was considered as an independent endpoint in this study. Endpoints were reviewed an classified by the MUSIC Study Endpoint Committee.

Preprocessing
Preprocessing of ECG recordings included heart beat detection and labeled using the Aristotle ECG analysis software [5] and linear filtering of baseline wander. Finally, the ECG was low-pass filtered (with cut-off frequency of 15 Hz) and down-sampled to remove noise out of TWA frequency range.
For each day interval, ECG signals were processed in segments of 128 consecutive beats (50% overlapped) with additional stability criterion imposed, as defined in [2], in order to consider only suitable segments for automatic analysis.
For estimating TWA amplitude of each segment (V k ), we replicated the analysis performed in [2]: the three orthogonal leads were linearly combined in order to maximize the TWA content over noise by using πCA [6] and then, the LLRM [7] was applied in the new combined lead to estimate the TWA waveform of each segment, expressed as the median difference between ST-T complexes of even and odd beats.
The non-visible microvolt range of TWA, sometimes comparable to the noise level, makes the TWA detection a challenging task. At this point, a novel methodological step for the computation of the TWA amplitude was included in the analysis, consisting in the phase alignment of all TWA estimated waveforms before averaging.
In vector notation, the TWA estimated waveform of the k th segment was defined as First, a detrended version of each y k , denoted as y ′ k (n) was computed: The coefficients a k and b k were chosen as those that defined the linear interpolation between y k (1) and y k (N ), whith the aim of eliminating any possible residual baseline component at the alternans frequency.
Then, the spatial correlation matrix of y ′ k , R y ′ , was estimated as being K the total number of suitable segments for the analysis.
To obtain the dominant waveform, that is, the first principal component, the following eigenvector equation for R y ′ was solved: where w 1 corresponds to the first eigenvector and λ 1 the first eigenvalue of R y ′ . At this point, the phase-aligned waveform, y a k , was estimated as: Finally, the IAA was defined as the mean value of the average waveform of all y a k :

Statistical analysis
Data is presented as median (25 th ; 75 th percentiles) for continuous variables, unless otherwise specified. Nonparametric Friedman test and Wilcoxon signed rank paired test with Bonferroni correction were applied to evaluate differences among time intervals. Survival analysis was performed by using Kaplan-Meier estimator and comparison of cumulative events by log-rank test. Prognostic value of TWA indices in predicting SCD was determined with univariate Cox proportional hazards analysis. For all tests, the null hypothesis was rejected for p≤ 0.05.

Results
Distribution of HR and IAA for the four defined intervals are presented in Fig. 1. Significant differences were found between the night and all daytime periods. IAA was found minimal during the night period, reaching its maximum value for the interval defined from 12.00 to 18.00   Fig. 1, left panel). The same trend was observed when considering average HR (Fig. 1, left panel). However, a weak correlation was found between IAA and average HR (r=0.18, p<0.001) and when considering IAA HR-restricted indices (IAA X , X = {70, 80, 90}, HR ranging from X-10 to X beats/min) a similar oscillation pattern remains still visible. Additionally, an increasing tendency from IAA 70 to IAA 90 in all day periods, was evidenced.

Discussion and conclusion
Average TWA activity computed during both the morning (from 06.00 to 12.00 h) and the evening periods (18.00 to 24.00h) is associated with the risk of SCD in CHF patients. Interestingly, it coincides with the two circadian periods of elevated risk for SCD, whose distribution has been reported to present a primary peak of incidence from 7 to 11 a.m. and a secondary peak from 5 to 7 p.m. [8]. It should be remarked that the predictive value of both IAA 06-12 and IAA 18-24 indices, not only considerably reduces the required time for ECG monitoring from 24 to 6 hours, which implies some advantages in the clinical practice, especially for the patient, but also the computational cost of TWA analysis, also an important factor that should be considered.
The HR-restriction of IAA indices presented the same pattern for IAA 70 , being similar when considering IAA 80 and IAA 90 . The increasing IAA indices from IAA 70 to IAA 90 are also consistent with the fact that TWA amplitudes for the same subject rises with instantaneous HR. However, only a weak correlation was found between average HR and IAA indices along the day, suggesting the influence of other HR-independent factors in IAA, which are known to be also determinant for TWA and related to higher cardiac electrical instability. Changes in autonomic neurotransmitters, with and increased activity of sympathetic nervous system during the day, especially more prominent at waking time, could be other determinant factors that significantly influence TWA [9]. Also, fluctuations in symptomatic and silent myocardial ischemia episodes, well-known to be linked to TWA phenomenon, have been reported to present this similar behavior [10]. All this supporting the hypothesis that IAA modulation along the day could be actually due to a circadian effect.
Automatic long-term averaging have been already shown to provide a reliable characterization of TWA in 24h-ECG recordings [2], avoiding the posterior visual verification usually needed when other methods, such as the modified moving average, are used [11]. The novel approach for computing IAA indices proposed in this work, including the phase alignment of TWA waveforms before averaging, presents a more insensitive to noise solution, even in such an unfavorable condition when the TWA amplitude is at the noise level (i,e, a few microvolts).
In summary, results from a fully-automatic robuts methodology for TWA analysis suggest that circadian pattern modulates the IAA index, and time of the day should be considered for SCD risk prediction.