Enhanced detection of abnormalities in heart rate variability and dynamics by 7‐day continuous ECG monitoring

Abstract Background The analysis of heart rate variability (HRV) and heart rate (HR) dynamics by Holter ECG has been standardized to 24 hs, but longer‐term continuous ECG monitoring has become available in clinical practice. We investigated the effects of long‐term ECG on the assessment of HRV and HR dynamics. Methods Intraweek variations in HRV and HR dynamics were analyzed in 107 outpatients with sinus rhythm. ECG was recorded continuously for 7 days with a flexible, codeless, waterproof sensor attached on the upper chest wall. Data were divided into seven 24‐h segments, and standard time‐ and frequency‐domain HRV and nonlinear HR dynamics indices were computed for each segment. Results The intraweek coefficients of variance of HRV and HR dynamics indices ranged from 2.9% to 26.0% and were smaller for frequency‐domain than for time‐domain indices, and for indices reflecting slower HR fluctuations than faster fluctuations. The indices with large variance often showed transient abnormalities from day to day over 7 days, reducing the positive predictive accuracy of the 24‐h ECG for detecting persistent abnormalities over 7 days. Conversely, 7‐day ECG provided 2.3‐ to 6.5‐fold increase in sensitivity to detect persistent plus transient abnormalities compared with 24‐h ECG. It detected an average of 1.74 to 2.91 times as many abnormal indices as 24‐h ECG. Conclusions Long‐term ECG monitoring increases the accuracy and sensitivity of detecting persistent and transient abnormalities in HRV and HR dynamics and allows discrimination between the two types of abnormalities. Whether this discrimination improves risk stratification deserves further studies.

improved by extending the monitoring time (Brachmann et al., 2016;Hariri et al., 2016;Liao et al., 2007). In a study comparing arrhythmia detection with 14-day continuous ECG and 24-h Holter monitoring in 32 patients, Chua et al. (2020) reported that the former detected 202 episodes of paroxysmal atrial fibrillation/ flutterin6patients,whereasthelatterdetectedonly1episodeof paroxysmal atrial fibrillation in 1 patient. Extended ECG monitoring for multiple days may uncover new ECG features that were not observed in conventional 24-h Holter ECG.
The duration of heart rate variability (HRV) analysis has also been standardized to 24 h. Analysis of HRV and nonlinear heart rate (HR) dynamics of Holter ECG has been mainly used to predict increased risk of poor prognosis in cardiac diseases, and the major predictive indices of HRV and HR dynamics including standard deviation of normal-to-normalR-Rinterval(SDNN) (Camm etal.,1996;Kleigeretal.,1987), decelerationcapacity(DC) (Baueretal.,2006;Kantelhardtetal.,2007), scaling exponent α 1 by detrended fluctuation analysis (DFA) (Huikuri et al., 2000;Iyengar et al., 1996;Peng et al., 1995), and very-lowfrequency (VLF) power (Bigger et al., 1992)  showing good reproducibility between two measurements made on differentdays (Kleigeretal.,1991;VanHoogenhuyzeetal.,1991),butthe level of day-to-day variations during long-term monitoring has not been studied. In the present study, we investigated the intraweek variation in the indices of HRV and HR dynamics, particularly of the appearance of their abnormalities predicting poor prognosis, during a week.
The sensor was attached on the upper chest wall with adhesive tape andcontinuouslyrecordedECGat256Hzandtriaxialacceleration at 32 Hz for 7 days and stored in it. The sensors were loaned to clinics, attached to patients, collected after measurements, and returned to JSR Corporation, where stored data were extracted, all QRS complexes in ECG were detected, noise and arrhythmia types were annotated, and annotated R-R interval time series data were generated. These processes were performed on a long-term Holter ECGanalysisviewer(NEY-HEA3000,NexisCo.,Ltd.)usingaHolter analyzerprogram(JMDN36827012,NexisCo.,Ltd.),whichhasbeen approved by the Japanese Ministry of Health, Labour, and Welfare (Medical device approval number 228AGBZX00099000). To annotate the cardiac rhythms, the analyzer program classified QRS complexes by the standard cycle length criteria for supraventricular ectopic heartbeats, grouped them by QRS morphology, and labeled the groups according to the type of arrhythmia. The results of the automated analysis were reviewed and edited by skilled technicians, and the morphological classification table was provided to medical doctors for confirmation. The long-term R-R interval time series thus generated were provided for this study.

| Study design
Westudieddatainconsecutive158outpatients(67males,71females, and 20 of unknown gender; age ±SD,64±16years)whounderwent 7-day continuous ECG monitoring for the screening or evaluation of arrhythmiasbetweenAugust2020andJanuary2021inJapan.Theinclusion criteria were (1) aged 20 years or older, (2) provision of written informed consent for the use of the anonymized data in this study, (3) sinus rhythm in 12-lead ECG at the entry, and (4) willingness to comply with up to 7 days of continuous ECG monitoring.
The written informed consent was obtained from each subject by JSR Corporation (Japan). The protocol was approved by the Ethics Review Committee of the Nagoya City University Graduate School ofMedicalSciences,Nagoya,Japan(No.60-18-0211).

| Data selection
The continuous R-R interval time series for a week was divided into 24-h intervals from the beginning of the monitoring. Only those 24-h datasegmentsinwhichthetotaldurationofsinusrhythmwas≥19.2 (24 × 0.8) h were used as valid data segments. Then, only those patients who had six or more valid 24-h data segments during a week were included for the final analysis.

| Computations of HRV and HR dynamics indices
For each valid 24-h data segment, we computed the time-domain and frequency-domain indices of HRV and nonlinear indices of HR dynamics that are used for predicting increased cardiovascular mortality risk and for assessing autonomic functions. They were computed by the methods according to the recommended standard (Camm et al., 1996)andtotheearlierstudies(Iyengaretal.,1996Kantelhardtetal., 2007;Pengetal.,1995).Briefly,fromtheECGdata,thetimeseries was calculated as the standard deviation of the averages of N-N inter-valofnonoverlapping5-minsegmentsover24h,andtherootmean square of successive N-N interval difference (rMSSD) was obtained as the square root of average squared difference between all successive N-N intervals during 24 h. DC was computed by the phase recti-fiedsignalaveragingofthe24-hN-Nintervaltimeseries (Kantelhardt et al., 2007). For a frequency-domain index, we computed the power of ultra-low, very-low, low, and high frequency (ULF, VLF, LF, and HF, respectively) components. For this purpose, 24-h {NNi} time series were interpolated by a horizontal step function, resampled at 2 Hz, filtered with a Hanning window, and converted into the frequency domain by a fast Fourier transform (FFT). The power spectral density was integrated for the power within the ULF ( <0.0033Hz),VLF(0.0033-0.04Hz),LF(0.04-0.15Hz),andHF(0.15-0.4Hz)bands.Thepower of these components was transformed into natural logarithmic values to normalize the distribution. LF/HF was calculated as the ratio of the absolute values of LF power and HF power, and the spectral exponent β was calculated as the slope of the log-log plot of the 24-h power spectrum . For the nonlinear indices, we calcu-latedthefractalcorrelationpropertiesofHRdynamicsusingtheDFA method and measured the short-term (4 to 11 beat) and long-term (>11 beats) scaling exponents (α 1 and α 2 , respectively) (Iyengar et al., 1996;Pengetal.,1995).

| Data analysis
To assess the day-to-day variation in the indices of HRV and HR dynamics, we calculated the intraweek coefficient of variance. First, the indices of HRV and HR dynamics were calculated for each 24-h segment during the monitoring week (7 days). Next, for each index, the average and standard deviation (SD) of the 24-h values during the 7 days were calculated in each patient. Then, the intraweek coefficient of variance was calculated in each patient using the following formula.
We examined the day-to-day variation in the occurrence of abnormal values of HRV and HR dynamics during the monitoring week.
To compare the ability of 7-day ECG to detect abnormalities in HRV and HR dynamics with 24-h ECG, the following two types of abnormalities were defined for each index of HRV and HR dynamics.
1. Index is abnormal in 7-day average (type 1 abnormality) 2. Index is abnormal at least on one day during 7-day monitoring (type 2 abnormality) For type 1 abnormality, the sensitivity, specificity, and positive and negative predictive accuracies were calculated for each index by analyzing the confusion matrix between the presence of type 1 abnormality and the detection of abnormal value by 24-h segmented ECG. For type 2 abnormality, the improvement of the detection of abnormality by 7-day continuous ECG monitoring was estimated by the expected sensitivity ratio, S 7-day /S 24-h , where S 7-day is the sensitivity of 7-day continuous ECG monitoring to detect type 2 abnormality and S 24-h is the sensitivity of 24-h ECG to detect the abnormal value of the index. Since type 2 abnormality was defined by 7-day data, S 7-day is always 100% by that definition in this study. S 24-h was estimated as the probability that the abnormal values would be detected by 24-h ECG monitoring on one day of the 7 days (E 24-h ). Then, the sensitivity ratio was estimated using the following equation.
Since the specificity to detect type 2 abnormality by 24-h ECG is always 100% by the definition of type 2 abnormality, only expected sensitivity was evaluated for type 2 abnormality.
The improvement of the detection rate of the number of indices with abnormal values (abnormal indices) by 7-day continuous ECG monitoring was estimated by the expected detection ratio of abnormal indices. To calculate this ratio, patients were grouped by the number of abnormal indices detected during 7-day ECG monitoring (C 7-day ), using only seven abnormal indices, including SDNN < 70 ms, DC <4.5ms,DFAα1 <0.75,ULF< 8.1 ln (ms 2 ), VLF <5.75ln(ms 2 ), LF <5.5ln(ms 2 ), and LF/HF < 0.43. In each patient, the average and maximum number of abnormal indices detected by 24-h ECG segments (C 24-h ) during the 7 days were calculated. Then, the average and minimum expected detection ratio of abnormal indices of 7-day ECG monitoring (R 7-day ) were computed as (C 7-day /average C 24-h ) and (C 7-day /maximum C 24-h ), respectively, in each patient.

| Statistical analysis
SASprogrampackageversion9.4(SASInstitute)wasusedforstatistical analyses. The sex differences in mean values and frequency were evaluated by t-test and chi-squared test, respectively. Repeated measuresanalysisofvariance(ANOVA)byGLMprocedurewasused to compare the coefficient of variance among different HRV and HR dynamics indices. Post hoc multiple comparisons between two coefficients were performed with generating contrasts between adjacent levels of the coefficient. Considering the non-Gaussian distribution of variables, the correlations between two variables were evaluated by Spearman's coefficient of rank correlation. The confusion matrices were generated by FREQ procedure. Statistical significance was determined with an α <.05withBonferroniadjustment.      Figure 2 shows the estimated sensitivity ratio of each abnormal index in total sample, sorted by the magnitude of the ratios. The ratio differed for each index, and was negatively correlated with the positive rate of type 2 abnormality (Table 3; Spearman's coefficient of rank correlation = −0.78, p = 0.01), but it was not associated significantly with the intraweek coefficient of variance ( Figure 1; Spearman's coefficient of rank correlation =0.52, p = .1). EarlierstudiesreportedstabilityovertimeofHRVindices (Kleiger et al., 1991;Van Hoogenhuyze et al., 1991). In a study comparing These values are also comparable to the intraweek variations in the present study.

24-h HRV indices measured at baseline and on placebo medication
The difference between these earlier studies (Kleiger et al., 1991;Van Hoogenhuyze et al., 1991) and the present study lies not only in the duration of the continuous monitoring, but also in the possible effects of monitoring on subjects' behaviors. In separated 24-h monitoring, the subject has to visit the laboratory twice, before and at the end of the monitoring, for the installation and removal of electrodes, which restricts subjects' daily schedule and activities accordingly, whereas in 7-day continuous monitoring, the subjects Nevertheless, the levels of day-to-day variation in the HRV indices were similar between them. This suggests that the changes in the subject's activity due to monitoring itself do not have a large effect on the day-to-day variation in the HRV indices.
Therefore, day-to-day differences in physical activities and emotional states may have stronger impact on the indices reflecting fast HR fluctuations than on the indices reflecting slow HR fluctuations. The actual maximum C 24-h were < C 7-day , indicating partial overlap of multiple abnormal indices on the same days during the week.
Consequently, it was estimated that 7-day ECG monitoring could detect about twice as many abnormal as 24-h ECG, as indicated by R 7-day .InarecentstudyofpredictivevalueofHRVin138post-AMI patients with preserved (≥35%) left ventricular ejection fraction (LVEF), Liu et al. (Liu et al., 2020) reported that decreased SDNN, VLF, and DC were independently associated with increased risk of sudden arrhythmic death and that combination of SDNN, VLF, and DC may help identify a high-risk patient group. In an earlier study in 687post-AMIpatientsofENRICHDcohort (Berkmanetal.,2003), we also examined the predictive value of combinations of abnormal HRV and HR dynamics indices and found that the combinations of best predictive power of mortality risk differs between patients with reduced (≤35%) and preserved (>35%) LVEF (Hayano et al., 2021). and HR dynamics indices were defined using reported cutoffs to predict mortality risk in patients after acute myocardial infarction.
Therefore, the prognostic value of these indices in the present population itself was unknown.

ACK N OWLED G M ENT
We thank JSR Corporation, Japan, for providing the anonymized long-term ECG data for this study.

CO N FLI C T O F I NTE R E S T S
Heart Beat Science Lab, Co. Ltd. has an advisory agreement with JSR Corporation.

E TH I C S
The protocol of study was approved by the Ethics Review Committee of the Nagoya City University Graduate School of Medical Sciences, Nagoya,Japan(No.60-18-0211).Basedontheprotocol,informed consent for the research use of the anonymized ECG data was obtained from each subject by JSR Corporation (Japan).

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from JSR Corporation, Japan. Restrictions apply to the availability of these data, which were used under license for this study. Data are available from the authors with the permission of JSR Corporation, Japan.