This prospective, observational, single-center study was approved by the local ethical committee of the Medical Faculty of Niigata University.
Participants
Between January 1, 2015, and December 31, 2017, adult patients with out-of-hospital sudden cardiac arrest who were transferred to the emergency department at Niigata University Hospital and successfully resuscitated were consecutively enrolled. All resuscitated patients were admitted to the Niigata University Hospital ICU.
Patients were eligible for participation if they met the following criteria: more than 16 years old, GCS motor scale of 1 at first evaluation after ROSC, and sinus rhythm. Exclusion criteria were as follows: traumatic cardiac arrest, cardiac arrest due to cerebral origins, normothermic TTM (36 °C), new onset of atrial fibrillation or atrial flutter rhythm, and hemodynamically unstable patients (severe hypotension).
Study protocol
All patients were mechanically ventilated to maintain normocapnia (PaCO2 35–45 mmHg) under sedation with continuous intravenous infusion of midazolam (0.1–0.2 mg kg−1 h−1) with fentanyl (0.1–0.2 μg kg−1 h−1). Rocuronium bromide was continuously infused to mitigate uncontrollable shivering (500–750 μg kg−1 h−1).
TTM targeting 34 °C of bladder temperature for 24 h was introduced and maintained with an intravascular cooling device (Thermogard XP®, Asahikasei Zoll medical, Japan) or a body surface cooling device (Arctic Sun 2000®, IMI, Japan). During hypothermia, a mean arterial pressure (MAP) > 60 mmHg was maintained with fluid resuscitation and/or continuous infusion of noradrenaline. All patients were rewarmed at a rate of 0.25 °C h−1. Patients who could not maintain MAP > 60 mmHg with these methods were excluded from the study. The treating ICU physicians were blinded to the following HRV-related metrics during study period.
Assessment of decelerating capacity; time-, frequency-, and geometrical-domain; and complexity variables of HRV
Electrocardiograms (ECG) were continuously monitored with a bedside monitor (IntelliVue MP70®, Philips, Japan), and the ECG wave data were captured at a sampling frequency of 250 Hz with 14-bit resolution and automatically stored in the dedicated server. RR intervals between 0:00 am and 8:00 am within the first 24 h post-ROSC were identified by wqrs algorithm [
17] and stored as a comma-separated value file (CSV) after 5 points moving averaging. The 8-h recording was started at midnight because this timeframe had fewer external stimuli such as physiological examinations or family visits.
Sinus rhythm was considered only when RR intervals were between 300 and 2000 ms and differed ≤ 20% from the average of five preceding sinus rhythm RR intervals, and consecutive RR interval differences were ≤ 200 ms [
18]. Any RR intervals not based on the above sinus rhythm criteria were replaced with the average value of the five preceding sinus rhythm RR intervals. When the replacement number divided by the entire RR intervals (replacement ratio) was > 20%, the patient was excluded from this study.
Decelerating capacity (DC) was computed using the software program calc-prsa (version 1.3.0) that was developed based on the phase-rectified signal averaging technique [
19].
The method used for time-, frequency-, and geometric-domain HRV analysis has been described elsewhere and adhered to the standards developed by the Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology [
20].
For the time-domain variables, the average of all RR intervals (AVNN), the SDNN, square root of the mean of the squares of differences between adjacent RR intervals (rMSSD), and the percentage of differences between adjacent RR intervals > 50 ms (pNN50) were computed.
For the frequency-domain variables, a Lomb-Scargle periodogram was plotted to measure the spectral power of the ultra-low-frequency range (ULF, 0–0.003 Hz), the very-low-frequency range (VLF, 0.003–0.04 Hz), the low-frequency range (LF, 0.04–0.15 Hz), the high-frequency range (HF, 0.15–0.4 Hz), the total power (TP, 0–0.4 Hz), the ratio of low- to high-frequency power (LF/HF), and the slope of the linear interpolation between 10−4 and 10−2 Hz of the spectrum in a log-log scale (power-law slope, exponent β). All measured powers were expressed as natural logarithm (ln).
For the geometric-domain variables, the total number of all RR intervals was divided by the height of the histogram of all RR intervals measured on a discrete scale with bins of 7.185 ms (triangular index), and RR interval was plotted as a function of the previous one (Poincaré plot). SD1 and SD2 are the two dispersions (standard deviations [SD]) of projections of the Poincaré plot on the line of identity (
y =
x) and on the line perpendicular to the line of identity (
y = −
x), respectively [
21].
For the complexity variables, approximate entropy (ApEn) and sample entropy (SampEn) were computed with a parameter of
m = 2 and similarity criterion = 20% of SD [
22]. Multiscale entropy (MSE) index was defined as the sum of the sample entropy at a scale factor of 1–20 [
23]. Detrended fluctuation analysis (DFA) was measured to quantify fractal scaling properties of the RR interval [
24]. The scaling properties were defined separately for short-term (4 ≤
n ≤ 16 beats,
α1) and long-term (
n > 16 beats,
α2) RR intervals.
DC, rMSSD, pNN50, ln LF power, ln HF power, and LF/HF were computed from the segment of 512 RR intervals, and the averaged values of the entire RR intervals were calculated. AVNN, SDNN, triangular index, ln total power, ln ULF power, ln VLF power, ApEn, SampEn, and MSE index were computed for the entire RR intervals. SD1, SD2, DFA (α1), and DFA (α2) were computed from the segment of 1000 RR intervals, and the averaged values of the entire RR intervals were calculated.
All variables except DC were computed with programs downloaded from PhysioNet (
https://physionet.org /physiotools/matlab/wfdb/wfdb-app-matlab/).
Statistical analyses
Categorical variables were presented as numbers or percentage and compared using the chi-squared test or Fisher’s exact test. Continuous variables were presented as median (interquartile range) and compared using the Mann-Whitney U test. Univariate logistic regression analysis for poor outcome was performed on each HRV-related variable. Variables were included in the multivariate logistic regression analysis if p = 0.001 (both Mann-Whitney U test and univariate logistic regression analysis). Results were presented as odds ratios (OR) and 95% confidence intervals (CI).
The receiver operating characteristic (ROC) curve was plotted, and the area under the curve (AUC) was calculated to evaluate the predictive performance of HRV-related variables for poor outcome. Youden index was used to calculate the optimal cut off value. Sensitivity and specificity were determined for the selected cutoff value. In addition, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and false-positive ratio (FPR) for poor outcome were calculated by dichotomy of the minimum values of the patients with good outcome.
Two-sided p values < 0.05 were considered statistically significant. All analyses were performed using STATA/SE package version 15.0 (StataCorp, College Station, TX, USA).