Prediction of heart rate variability on cardiac sudden death in heart failure patients: A systematic review

Sudden cardiac death occurs frequently in patients with heart failure [1,2]. Determining suitable factors for identifying high risk heart failure patients is necessary. Heart rate variability (HRV) [3] was previously demonstrated to predict sudden cardiac death in heart failure patients, though findings have been inconsistent [4].

We conducted a systematic review for clarifying the predictive value of HRV. A comprehensive literature search up to December 2013 using the PubMed (MEDLINE) and CENTRAL with restrictions to humans’ studies and English publications was carried out, with key words containing Sudden Cardiac Death, Heart Failure, and Heart Rate Variability. Among the 138 publications identified, 119 were excluded based on screening of titles and/or abstracts. 19 potential relevant articles and an additional article identified from references were fully reviewed. Six articles were excluded from our study due to different reasons. Jiang et al. included only death or a life-threatening cardiac event as the outcome but not covers sudden cardiac death [5]. Arsenos et al. included surrogate of sudden cardiac death as outcome [6]. Yamada et al. and Nessler et al. did not provide sufficient information [7,8]. La Rovere et al. did not provide specified data for outcome of sudden cardiac death [9]. Szabo et al. did not provide sufficient information for multivariate adjusted result of HRV parameters [10]. After excluding these 6 articles, 14 appropriate articles were retained (Fig. 1) [11–24]. Among them 4 articles have overlapping for participants of two studies [14,19,21,23]. We included all of them as in each article they demonstrated different variables/parameters of HRV. Further quality assessment in standardized manner [25] did not find deviations among these studies.

Fig. 1.

Articles identified through comprehensive literature search.

In total, 14 articles representing 12 prospective studies were included. Detailed information for these articles is demonstrated in Table 1. These studies used quite different variables/parameters of HRV in their report, making an overall quantitative evaluation challenging [26,27]. As the main variable of HRV, standard deviation of all normal-to-normal intervals (SDNN) was mostly studied. While this variable was presented as numerical variable in several studies [12,13,21,22], in other studies it was categorized according to different values, making an overall numerical estimation difficult [11,14,18,20,23]. However, in all these 9 studies, SDNN did not show a predictive role; thus, this variable should have a limited role in predicting sudden cardiac death among heart failure patients. Similarly, some other studied variables, including SD derived from time-domain analysis [12], NN [12,14,23], SDANN [12–14,23,24], pNN50 [12–14,20], rMSSD [12,13,18,23], HRV index [13], sNN50 [18], log HRV [17] and RR [21], were demonstrated to be not predictive in most studies except that one study supported a marginal predictive role of SDANN and rMSSD [20]. Thus, overall these variables are less likely to be predictive of sudden death in heart failure patients.

Table 1.

Studies investigating predictive value of heart rate variability on cardiac sudden death in heart failure patients.

First author, publication year (reference), country	Cases/subject (age), duration of follow-up	Method for obtaining data	Exposure categories	RR/HR (95% CI)	Adjusted factors
Bilchicket al., 200211, US	21/127 (mean 64.7 y), mean 34 months	24 h Del-Mar Holter recording	SDNN: SDNN >65.3 ms (Ref) SDNN < 65.3 ms	2.40 (p = 0.088)	LVEF, systolic BP, HR, age, treatment group, presence of ischemic cardiomyopathy
Anastasiou-Nana et al., 200512,Greece	4/52 (56 ± 12 y), 2years	24 h ambulatory ECG monitoring	High frequency power SD derived from time-domain analysis	0.31 (0.101-0.954) 0.913 (0.831-1.004)	None
			Mean NN, SDNN, SDANN, pNN50, rMSSD, low power, total power, normalized low or high power, low/high power ration	All NS
Tamaki et al, 200913, Japan	18/106 (~64y), 65 ± 31 months	24 h ambulatory ECG monitoring	Mean RR, SDNN, SDANN, SDNN index, rMSSD, pNN50, HRV index, TP, ULFP, n-ULFP, VLFP, n-VLFP, LFP, n-LFP, HFP, n-HFP, LFP/HFP	All NS	Age; sex; underlying causes (ischemic or nonischemic); New York Heart Association (NYHA) functional class; heart rate; systolic and diastolic blood pressure; LVEF; presence of nonsustained ventricular tachycardia on Holter monitoring; echocardiography data; plasma noradrenaline concentration; serum uric acid, sodium, and creatinine levels; and the results of cardiac MIBG imaging, SAECG, HRV, and QT dispersion
Brouwer et al, 199614, Netherlands	11/95 (~60 y), 2-4y	24 h ambulatory Holter recording	Poincare plot: normal (REF) abnormal plot	5.3 (1.0-27.5)	LVEF (<0.30), plasma norepinephrine ( > 450 pg/ml), ventricular premature beats (>20/h), ventricular tachycardia (present)
			Mean NN interval (<750 ms), SDNN (<110ms), SDNN (<50 ms), SDANN (<100 ms), pNN50 (<2.0%), total power (<2500 ms2), very low frequency power ( <1500 ms2), low frequency power (<300 ms2), high frequency power ( <100 ms2)	All NS	None
Soejima et al, 200015,Japan	7/52 (mean 61.5 y), mean 3.8 y	24 h Holter monitoring	LF, HF: normal (Ref) Abnormal LF, HF	NS	Age
Kuch et al, 200916,Polanda	33/158 (49-80 y), 2 y	24 h ECG monitoring	HRV: normal (Ref) Low	P < 0.001	Diabetes status, AMI location
Tereshchenko et al, 201217, Spain	52/533 (mean 62.8 ± 12 y), median 44 months	High-resolution orthogonal ECG recordings	Log HRV: 3 higher quartile (Ref) Lowest quartile of log HRV	1.53 (0.85-2.77)	None
Nolan et al, 199818, UK	18/433 (62 ± 9.6 y), 484 ± 161days	24 h ECG recording	SDNN (> 100ms),sNN50, rMSSD	All NS	Cardiothoracic ratio, LVEDD, NSVT, potassium
La Rovere et al, 200319, Italy	19/202 (52 ± 9y), 3 y (derivation sample)	ECG recording	Controlled-breathing LF power: >13 ms2 (Ref) ≤13 ms	3.7 (1.5-9.3)	Left ventricular end-diastolic diameter
	20/242 (~54 y), 3 y (validation sample)		Controlled-breathing LF power: >11 ms2 (Ref) ≤ 11 ms2	3.0 (1.2-7.6)	Ventricular premature contractions
Galinieret al, 200020, France	21/190 (mean 61 ± 12 y), mean 22 ± 18 months	24 h HolterECG recordings	Day-time low frequency power: ≥3.3 ln (ms2) (Ref) <3.3 ln (ms2)	2.8 (1.2-6.8)	Aetiology (CAD)
			SDANN (<55 ms)	2.5 (1-5)	None
			RMSSD (<14 ms)	2.4(1-5.5)	None
			Day-time total power (<4.8ln(ms2))	2.4 (1-5.6)	None
			Mean heart rate (>89 beats/min), SDNN (<67 ms), SD (<30 ms), pNN50 ( < 2%), night-time total power (<5.3 ln (ms2)), night-time low frequency power (<3.6 ln (ms2)), day-time high frequency power (<2.7 ln (ms2)), night-time high frequency power (< 3.1 ln(ms2))	All NS	None
Guzzetti et al, 200521, Italyb	29/352 (47-59 y), median 34 months	Holter ECG recordings	LF night(≤20 ms2) HF night((≤60 ms2)	2.7 (1.3-5.6) 2.2 (1.0-4.6)	None None
			RR 24-h, RR night, SDNN 24-h, SDNN night, VLF 24-h, VLF night, LF 24-h, HF 24-h, LF 24-h nu, LF night nu, HF 24-h nu, HF night nu, LF/HF 24-h, 1/f slope	All NS	None
Shehab et al, 200422, UK	5/34 (mean 68 y), 1 y	24 h ECG monitoring	SDNN (mean)	NS	None
Smilde et al, 200923, Netherlandsc	28/90 (mean 60 ± 8 y), mean 11.7 y	Holter ECG recordings	Mean NN (<750 ms), SDNN (<110 ms), SDNN (<50 ms), SDANN (<100 ms), RMSSD (<25 ms), total power (<2500 ms2), very LF power (<1500 ms2), LF power (<300 ms2), HF power (<100 ms2)	All NS	None
Wooetal,199724, US	19/113 (53 ± 10 y), 1y	24 h Holter monitoring	Poincare plot (nonlinear method) SDANN	P < 0.0001 NS	Mean right atrial pressure at hemodynamic optimization, serum level ofsodium, LVEF, results of the 6-minute walk

RR: relative risk; HR: hazard ration; CI: confidence interval; NA: not available; NS: not statistically significant.

^aStudied patients are post myocardial infarction patients with heart failure.

^bPart of La Rovere et al., 2003 as shown above.

^cPart of Brouwer et al., 1996 as shown above.

Several other variables of HRV were demonstrated to be potentially predictive in more studies. High frequency power (HF/HFP) was shown to be predictive in the 2005 Greece study [12], and night HFP seemed to be predictive in the 2005 Italy study [21], while a null effect was shown in the 2000 and 2009 Japan study [13,15], the 1996 and 2009 Netherlands study [14,23], and the 2000 France study [20]. It thus makes HFP being predictive of sudden cardiac death in heart failure patients less possible, since the 2009 Japan study with a long following-up (65 months) showed null result [13]. On the other hand, low frequency power (LF/LFP) was predictive in the 2003 Italy study (2 independent samples) [19], besides day-time LFP in the 2000 France study [20] and night-time LFP in the 2005 Italy study [21]. However, the prediction was not demonstrated in the 2005 Greece study [12], the 2000 and 2009 Japan study [13,15], and the 1996 and 2009 Netherlands study [14,23]; also the 2005 Italy study showed that all LFP did not have a predictive effect [21]. Thus overall evidences supporting the predictive effect of LFP are also weak. Another HRV variable, Poincare plot, demonstrated strong prediction in available studies (the 1996 Netherlands and the 1997 US studies) [14,24]. Though limited by a small number of studies, the relatively small sample size and short following-up make this effect is likely real.

To our knowledge, this is the first systematic review for the prediction value of heart rate variability on sudden cardiac death in heart failure patients. Although a quantitative meta-analysis is impossible due to that heart rate variability variables/parameters used in each study are quite different, all findings from these studies were carefully evaluated in our systematic review. We found that variables of SDNN, SD derived from time-domain analysis, NN, SDANN, pNN50, rMSSD, HRV index, sNN50, log HRV and RR should have limited predictive roles. Other two HRV variables, high frequency power and low frequency power, although showing a predictive value in several studies, are less likely to effectively predict sudden cardiac death in heart failure patients. On the other hand, Poincare plot can potentially predict well according to current evidences. Further studies are warranted to determine whether Poincare plot can really be a representative HRV variable in effectively predicting sudden cardiac death in heart failure patients.