Retrospective cohort study with patients admitted for decompensated HF or syncope at two private hospitals with suspected or confirmed diagnosis of HF. The selection of patients with syncope was necessary since these patients are routinely submitted to 24h Holter monitoring. Despite having a different pathophysiological mechanism, syncope can be a marker of arrhythmic events, and be involved in an increased risk of death. ⁶ The study period was from January 2014 to December 2016. Patients who underwent Holter during hospitalization or after hospital discharge (up to 30 days) in a database with 4179 exams were identified. Patients with terminal HF, evidence of another disease whose prognosis does not provide the prospect of one-year survival, acute coronary syndrome for less than two months, Holter record with recording less than 18 hours, non-sinus HR, pacemaker carriers, and total premature beats (atrial + ventricular) ≥ 5,000 were excluded.

Clinical, laboratory, and echocardiographic data of eligible patients were assessed through electronic medical records, admission form, and discharge summary. After analysis, the patient was classified as a patient without HF (without symptoms compatible with HF or echocardiogram without systolic and diastolic dysfunction or serum B-type natriuretic peptide (BNP) levels below the laboratory reference or a patient with HF.

The signals were acquired through the three-channel Holter device, model DR200/HE (NorthEast Monitoring, MA, USA). The recording bandwidth ranges from 0.05 to 70 Hz with 12-bit resolution, with a sampling rate of 180 samples per second. The signal processing was performed using a program based on the Matlab® application (The MathWorks, MA, USA). HRV indicators were recorded by two observers, blinded to the outcomes in two situations: traditional long-term parameters and short-term parameters. In the long-term parameters, the following variables were evaluated: SDNN 24h, SDANN 24h, rMSSD 24h, average HR, total premature ventricular beats (PVB), and total premature supraventricular beats (PSB). 24h Holter recording was segmented into nonoverlapping 10-minutes windows, and the window presenting with the lowest HRV was identified to assess the short duration parameters (moment of highest sympathetic tonus). Each 10-minute window was individually examined and included in the analysis if extrasystoles, artifacts, and inadequate normal beats combined corresponded to less than 5% of the window duration. The short duration variables that were generated were: the lowest SDNN 10min (lowest SDNN recorded in a 10 min window during the 24h signal); lowest rMSSD 10min (lowest rMSSD recorded in a 10 min window during the 24h signal), and lowest pNN50 10min (smallest pNN50 recorded in a 10 min window during the 24h signal).

Age, gender, cause of HF, comorbidities, and use of previous medications were evaluated. Laboratory data and echocardiographic reports were also evaluated; the first echocardiogram performed at admission was considered in the analysis.

The primary outcome of this study was all-cause mortality. The minimum follow-up period in the study was 12 months.

The study protocol was registered on Plataforma Brasil with CAAE 63827617.5.0000.5249 after approval by the local ethics committee of Hospital Copa D’Or on 04/27/2017. As this was a retrospective observational study, the informed consent term was waived by the ethics committee.

The results were expressed as mean ± standard deviation (normal distribution) or median and interquartile range (non-normal distribution) for continuous variables and number of occurrences (with percentage) for categorical variables. Clinical, laboratory, echocardiographic characteristics, and HRV variables in survivors and deaths were evaluated. The chi-square test for categorical variables and the unpaired Student’s t-test for continuous variables were used. When the sample distribution was not normal according to Shapiro-Wilk test, the Mann-Whitney test was used. All variables with p<0.05 were assessed using a univariate Cox regression model. Variables with p<0.01 in univariate model were included in a multivariate Cox regression model to determine the occurrence of the all-cause death. HRV variables with p<0.05 in multivariate Cox’s model for the mortality outcome were also evaluated through the ROC curve to determine the optimal cut-off point, aiming to create a predictive model score. For the addition of the cut-off point, each variable was assigned the integer value corresponding to its beta coefficient obtained in the Cox regression. Survival curves were presented in the subgroups according to the predictive model.

The inclusion flow is summarized in Figure 1 . There were 2049 hospitalizations in both hospitals; of these, 206 performed Holter (10.1%). Ninety patients were excluded from the sample, totaling 116 patients for clinical evaluation. After this evaluation, 48 (41.4%) patients with compensated cardiac function during the hospitalization period and 68 (58.6%) with HF were identified. Sixty-two patients (53.4%) were admitted due to syncope.

Figure 1

Population characteristics are summarized in Table 1 . There was a predominance of elderly and female gender. The most prevalent comorbidity was arterial hypertension, followed by diabetes and ischemic heart disease.

Table 1

	All (n=116)	No HF (n=48)	HF (n=68)	p-value*
Age (Years)	71.9 ± 16.3	68.3 ± 21.3	74.5 ± 11.0	0.042
Male gender	53 (45.7%)	15 (31.3%)	38 (55.9%)	0.007
Arterial Hypertension	90 (77.6%)	31 (64.6%)	59 (89.4)	0.001
Diabetes	39 (33.6%)	9 (18.8%)	30 (44.1%)	0.004
Ischemic Heart Disease	21 (18.1%)	0 (0.0%)	21 (31.8%)	<0.001
Chronic Renal Failure	13 (11.2%)	1 (7.7%)	12 (17.6%)	0.007
Stroke	10 (8.6%)	2 (4.2%)	8 (11.8%)	0.135
LVEF (%)	56.3±17.2	68.5±6.8	47.7±17.2	<0.001

HF: heart failure, LVEF: left ventricular ejection fraction. * comparison between HF and No HF.

Regarding ventricular function, most patients had preserved left ventricular systolic function. The use of beta-blockers was only identified in one third of the population. Few patients manifested nonsustained ventricular tachycardia (NSVT) on Holter monitoring.

Mean Holter duration was 23.23 ± 1.48 hours, and the mean follow-up was 2.83 ± 1.27 years. Most Holter monitoring studies were performed in hospital and half were performed in the intensive care unit. There were 39 deaths.

The characteristics were also assessed according to the final classification of the patients. There was no age difference between the groups. A higher prevalence of HF, arterial hypertension, diabetes, ischemic heart disease, and chronic renal failure was observed among male than female patients.

Heart failure patients had reduced HRV parameters when compared to patients without HF. There was no difference in the mean HR between these two groups. The number of PVB was higher in patients with HF ( Table 2 ).

Table 2

	No HF (n=48)	HF (n=68)	Valor p
SDNN 24h (ms)	88 (71-128)*	83.5 (59-109)*	0.061
SDANN 24h (ms)	79.5 (64-112)*	71.5 (49.5-91)*	0.037
rMSSD 24h (ms)	27 (15.25-43.25)*	24.5 (18-35.5)*	0.723
Lowest SDNN (ms)	16.5 (11.5-24.5)*	13.0 (8.5-16)*	0.002
Lowest rMSSD (ms)	10 (8-19.5)*	9 (7-13.5)*	0.070
Lowest pNN50 (%)	0.01 (0.001-1.17)*	0.01 (0.001-0.17)*	0.040
Mean 24h HR (bpm)	72 (64-76.5)*	68 (62-76)*	0.159
PVB	8 (0-74.8)*	152 (7-553)*	<0.001
PSB	48 (3.75-201)*	70.5 (12-318)*	0.347

HF: heart failure; HR: heart rate; PVB: premature ventricular beats; PSB: premature supraventricular beats * median and interquartile range; SDNN: Standard deviation of the NN (R-R) intervals; SDANN: Standard Deviation of the 5 minute Average NN intervals; rMSSD: Root mean square of the successive differences; pNN50: The proportion of NN50 divided by the total number of NN (R-R) intervals; NN50: The number of pairs of successive NN (R-R) intervals that differ by more than 50 ms.

Results related to the all-cause mortality outcome are described in Table 3 . Mean age was higher in the group of patients who died, as well as the prevalence of chronic renal failure and reduced left ventricular ejection fraction (LVEF), and these factors were statistically significant in the univariate regression of Cox. Regarding HRV parameters, those with statistical significance in the univariate Cox regression were lowest SDNN and lowest rMSSD (p<0.05 for both).

Table 3

	Survival (n = 77)	Deaths (n = 39)	p	Cox regression Exp(b); (p)
Mean age (years)	68.6 ± 18.0	78.4 ± 9.6	0.002	1.04 (p = 0.003)
Male gender	32 (41.6%)	21 (53.8%)	0.145
Hypertension	58 (75.3%)	32 (86.5%)	0.130
Diabetes	24 (31.2%)	15 (38.5%)	0.280
Previous MI	14 (18.2%)	7 (18.9%)	0.557
CKD	3 (3.9%)	10 (25.6%)	0.001	4.5 (p<0.001)
Previous stroke	4 (5.2%)	6 (15.4%)	0.070
Betablocker use	24 (31.2%)	14 (35.9%)	0.378
LVEF (%)	59.9 ± 15.9	49.3 ± 17.7	0.002	0.98 (p = 0.003)
Lowest SDNN (ms)	18.7 ± 11.4	12.3 ± 5.2	0.001	0.90 (p = 0.001)
Lowest rMSSD (ms)	14.1 ± 10.7	9.8 ± 6.1	0.020	0.92 (p = 0.012)
Lowest pNN50 (%)	1.44 ± 4.35	0.43 ± 1.60	0.165
Mean HR (bpm)	69.7 [67.6-71.9] ¥	69.9 [66.6-73.3] ¥	0.940
SDNN 24h (ms)	92.3 [84.6-100.7] ¥	78.2 [67.0-91.2] ¥	0.046	0.99 (0.123)
SDANN 24h (ms)	80.2 [73.2-87.9] ¥	69.8 [57.3-85.0] ¥	0.200
rMSSD 24h (ms)	26.6 [23.0-30.7] ¥	24.6 [18.7-32.5] ¥	0.620
PVB	15 [1-181] ¥	156.5 [22-678] ¥	0.002	1.0003 (0.11)
PSB	60 [22-90.5] ¥	94 [33-161] ¥	0.206

MI: myocardial infarction; CKD: chronic kidney disease; HR: heart rate; PVB: premature ventricular beats; PSB: premature supraventricular beats. SDNN: Standard deviation of the NN (R-R) intervals; SDANN: Standard Deviation of the 5 minute Average NN intervals; rMSSD: Root mean square of the successive differences; pNN50: The proportion of NN50 divided by the total number of NN (R-R) intervals; NN50: The number of pairs of successive NN (R-R) intervals that differ by more than 50 ms; LVEF: left ventricular ejection fraction; ¥ Mann-Whitney test.

Continuous variables that were associated with all-cause death were dichotomized using the ROC curve. The area under the curve (AUC), the cut-off points, and the statistical significance of this analysis are summarized in Table 4 . The cut-off points were defined automatically by identifying the Youden index.

Table 4

	AUC (95% CI)	Cutoff point	p	Sensibility	Specificity
Lowest SDNN (ms)	0.313 (0.213-0.413)	≤12ms	0.001	58.97%	68.83%
Lowest rMSSD (ms)	0.332 (0.230-0.434)	≤12ms	0.003	87.18%	44.16%
Age (years)	0.658 (0.558-0.759)	>69 anos	0.006	82.05%	44.16%
LVEF (%)	0.319 (0.217-0.432)	≤57%	0.002	64.10%	71.43%

LVEF: left ventricular ejection fraction; AUC: area under the curve; SDNN: Standard deviation of the NN (R-R) intervals; rMSSD: Root mean square of the successive differences.

A Cox regression model was built with the variable chronic renal failure and the dichotomized variables according to the value recommended by the ROC curve. The model is shown in the Table 5 .

Table 5

	b	HR (95%CI)	p
Lowest SDNN ≤ 12 ms	-0.03929	0.91 (0.47-1.94)	0.9135
Lowest rMSSD ≤ 12 ms	1.7126	5.54 (2.04-15.08)	0.0008
CKD	0.1643	1.18 (0.45-3.06)	0.7372
Age > 69 years	1.3747	3.95 (1.64-9.52)	0.0023
LVEF ≤ 57%	1.5472	4.70 (2.38-9.28)	<0.0001

CKD: chronic kidney disease; LVEF: left ventricular ejection fraction; SDNN: Standard deviation of the NN (R-R) intervals; rMSSD: Root mean square of the successive differences;

The parameters evaluated using Cox regression and that showed statistical significance were considered in the predictive model. For the addition of the cut-off point, each variable received the integer value corresponding to its beta coefficient obtained in the Cox regression ( Table 5 ). Thus, the predictive model was built ( Table 6 ).

Table 6

Lowest rMSSD ≤ 12ms?	1
LVEF ≤ 57%?	1
Age > 69?	1
SCORE	SUM

rMSSD: Root mean square of the successive differences; LVEF: left ventricular ejection fraction

This score was evaluated using the ROC curve to estimate its accuracy and determine the best cut-off point. A score <3 showed an AUC = 0.802 (95% CI 0.72-0.87) for all-cause death with a sensitivity of 46.2% and specificity of 97.4%. This score was also assessed using Cox regression after dichotomization proposed by the ROC curve. Patients with a score of three, as compared to those with a score <3 ( Figure 2 ), were 9.3 times more likely to die (HR 9.31; 95% CI 4.89-17.75).

Figure 2