Background
Elevated heart rate (HR) has been associated with higher mortality in otherwise healthy individuals [
1‐
3] and in those with coronary artery disease [
4,
5]. In the Framingham cohort, higher HR was associated with an increased risk of cardiovascular disease, heart failure, and death [
6]. Our group and others have previously shown that an elevated HR is associated with increased all-cause and cardiovascular mortality as well as with 30-day re-hospitalization in chronic heart failure [
7‐
9]. The negative effect of elevated HR on the myocardial supply-demand balance is implicated in this process and clinically, elevated HR has been associated with progression of atherosclerosis in patients with an MI at a young age [
10] and plaque disruption [
11].
Ivabradine, an I
f current inhibitor that selectively lowers HR was initially shown to improve time to ST-segment depression in patients with stable angina [
12]. More recently, ivabradine was studied in the heart failure (HF) population where it reduced HF hospitalizations (SHIFT) [
13] resulting in its inclusion as part of guideline recommended therapy for HF patients with HR ≥ 70 bpm despite optimal medical therapy [
14,
15]. However, despite the growing interest in heart rate modulation as a target for therapeutic intervention, there is a paucity of literature on the optimal method of HR assessment.
While resting HR is commonly used in the clinical setting due to its simplicity, it has been shown that single time-point HR measurement reproducibility is poor with only modest correlation between resting HR and mean HR obtained on extended monitoring [
16‐
18]. Moreover a recent study examining ambulatory monitoring suggested that continuous HR was an independent predictor of all-cause mortality after adjusting for resting HR while adjusted resting HR was not [
19]. This has implications not only from a therapeutic perspective but also merits investigation to inform clinical trial design. Indeed, the ivabradine trials have shown mixed results both in the setting of ischemic heart disease [
12,
20], and chronic HF [
13,
21]. Therefore we sought to examine the relationship between resting and mean HR on prolonged monitoring in patients with reduced LVEF. However, because correlation does not necessarily imply agreement, we further sought to investigate the level of agreement between these measures in order to ascertain the safety and validity of these methods of HR determination.
Methods
Study population and study design
We prospectively recruited a convenience sample of patients with systolic dysfunction between April 2015 and March 2016 who were followed in the device clinic at Toronto General Hospital, a large tertiary care centre in the University Health Network, Toronto, Canada. Patients were eligible for inclusion if they were older than 18 years old, had an LVEF less than 40%, had an ICD in situ, and were paced in the atrium or ventricle less than 20% of the time. Patients with an LVEF greater than or equal to 40%, who were paced more than 20% of the time, who had a HR less than 40 bpm, or a paced rhythm on resting rhythm strip during their clinic visit were excluded. Patients with a left ventricular assist device or complex congenital heart disease were also excluded.
Baseline characteristics and device indication were obtained from the electronic patient record (EPR). Medication lists were obtained directly from the patient, or if not available, from the chart or EPR. Analyses were limited to those patients whose medication lists were documented within 13 weeks of HR assessment. Beta-adrenoreceptor blockers were stratified into low and intermediate-high dose categories. Carvedilol ≥12.5 mg twice daily, bisoprolol ≥5 mg daily, and metoprolol ≥ 50 mg twice daily were considered to be intermediate-high doses. The intermediate-high dose category which we analyzed together, is consistent with high dose group in the recent OBTAIN study [
22].
Ascertainment of heart rate
Resting HR was defined as that obtained from the device interrogation at the end of the clinic visit in the seated or recumbent position after the patient had been resting both prior to and during the clinic visit. Mean HR was obtained from the HR histogram available upon device interrogation. The HR histogram is divided into 10 bpm heart rate intervals, (for example 60–69 bpm, 70–79 bpm, etc.) and provides the relative amount of time spent in each heart rate interval as a percentage of the total time since last device interrogation. Mean HR was calculated by taking the median of each 10-bpm heart rate interval and multiplying it by the percentage of time spent in that interval. The mean HR was the sum of the products calculated in each interval. The highest % HR interval was defined as the 10-bpm interval that the patient spent the most time in. If multiple 10-bpm intervals had equal frequencies, the one closest to the resting HR was used. An illustrative example of our calculation of mean heart rate can be seen in Additional file
1: Figure S1.
Outcomes
The primary goal was to determine the agreement between resting HR on the ECG at a single time point and mean HR from the ICD interrogation. Secondary analyses included the correlation between resting HR and the highest % HR interval. We also conducted a sensitivity analysis to examine the relationship in important subgroups including HR ≤ 70 bpm, diabetics, patients with arrhythmias on interrogation (atrial fibrillation and ventricular tachycardia), those taking intermediate-high doses of beta-blockers, the subgroups with and without pacing on interrogation, and those with cardiac resychronization therapy (CRT).
Statistical analysis
Continuous variables are reported as mean ± standard deviation and categorical variables as numbers (percentages). The correlation coefficient (r) was determined for the association between resting HR and mean HR, as well as the association between resting HR and the highest % HR interval. Percent agreement between the two HR measurements was assessed using Bland-Altman plots with 95% limits of agreement. Statistical analysis was performed using STATA version 12, StataCorp, College Station, Texas.
Discussion
In the present, hypothesis generating study, we examined a novel method of mean HR determination from prolonged ICD monitoring. Although our data demonstrated good correlation between resting and mean HR with minimal systematic bias between the 2 methods, the limits of agreement were unacceptably wide. Since there is no gold standard for HR determination we further assessed whether resting HR obtained at a single time point could predict the highest frequency HR interval over the duration of ICD monitoring. Our findings demonstrated that resting HR only reflected the highest frequency HR interval half of the time as determined from the HR histogram. Importantly, resting HR more frequently overestimated the HR interval than underestimated it. From a clinical perspective, this suggests that single time point resting HR is an imperfect surrogate for mean HR when titrating pharmacotherapies.
Our findings of correlation between methods are consistent with other small studies comparing resting and HRs obtained by Holter monitoring. Carlson et al., found partial correlation between resting ECG HR and both mean daytime HR as well as resting HR determined by Holter monitoring in a cohort of otherwise healthy participants [
17]. The strength of correlation was weaker than in the present study, which may be explained by the lower activity levels of the patients in our study as well as lower heart rate variability. Moreover, the routine use of beta-blockers in this population would be expected to further contribute as evidenced by the strong correlation between methods in the subgroup on an intermediate-to-high dose beta-blocker. Furthermore, in a telemonitoring study of ICD patients with NYHA class III heart failure, resting HR correlated with mean 24-h HR [
18]. More recently Jolly et al., showed a positive correlation between ambulatory Holter mean HR and ECG resting HR in patients with atrial fibrillation [
23].
Our study extends the literature by demonstrating that correlation between methods for determining HR does not necessarily imply that the two different approaches yield the same measurements. After confirming correlation, we determined that the limits of agreement between resting single time point HR and mean HR on prolonged monitoring were wide. While there is no consensus on how wide a limit of agreement is acceptable, from a clinical perspective, a 10-bpm difference would be likely to alter clinical decision-making. When the limits of agreement were set to a clinically meaningful interval of ±5 bpm, 37% of patients fell outside of this range. Furthermore, we demonstrated that resting HR was frequently above the mean HR, which is consistent with the findings of Pastor-Perez et al., who identified a subgroup of patients with HF in whom resting HR was above the mean with 7-day ambulatory HR monitoring [
24].
The present findings, both using Bland-Altman analysis and comparing resting HR to the highest frequency heart rate interval, raise concern for iatrogenic bradycardia when resting ECG heart rate is used to titrate therapies. This issue was clinically highlighted in both the SHIFT [
13] (Systolic Heart failure treatment with the I
f inhibitor ivabradine Trial) and SIGNIFY [
20] (Study Assessing the Morbidity-Mortality Benefits of the I
f Inhibitor Ivabradine in Patients with Coronary Artery Disease) trials where resting ECG was used to measure heart rate. In these trials, the rate of bradycardia in the treatment arm was 10.3% and 18.9% respectively. Thus, uptitrating HR lowering therapies, such as ivabradine, using a resting ECG has the potential for detrimental outcomes particularly given that it has never demonstrated an independent mortality benefit in this population. Clinically this iatrogenic bradycardia may result in failure to augment cardiac output sufficiently to maintain adequate perfusion with ensuing presyncope and falls. Moreover, while there is no consensus on the lower HR limit for a mortality benefit, our group found a non-significant trend towards increased 30-day mortality in a cohort of community-based HF patients with HR < 60 bpm [
7]. Furthermore, in the MESA cohort, there was a U-shaped mortality distribution amongst individuals taking HR-modifying therapies with increased mortality at HR < 50 bpm on resting ECG [
25]. Finally, in a heart failure population such as ours, who frequently are recipients of an ICD or pacemaker, over titration of medications may result in increased RV pacing which also carries negative prognostic implications [
26]. In the context of these concerns, our findings suggest that an alternative method of HR assessment to supplement single time point measurements, should be considered by clinicians. Since many of these patients will have an ICD, either the mean HR or highest frequency HR interval could be used in order to reduce the risk of iatrogenic bradycardia. Long-term studies are then required to assess whether this method will also have prognostic implications.
Our study has several limitations. Firstly, mean HR could not be directly assessed from the ICD, as most commercially available devices do not offer this function. However, we developed a novel method of mean HR determination using the HR histogram which demonstrated strong correlation with resting HR, confirming its validity as a tool to measure HR. Secondly, the presence of tachyarrhythmias such atrial fibrillation or ventricular tachycardia could have influenced the results. With respect to atrial fibrillation, the documented burden was low and only three of these patients had single lead device in which case it would have been difficult to assess the atrial fibrillation burden. The episodes of NSVT/VT were also brief and unlikely to have affected the overall results. Our sensitivity analyses of these subgroups demonstrated results similar to the overall cohort, with moderate to good correlation but wide limits of agreement.
Thirdly, resting HR was taken at a single time point near the end of the clinic visit after the patient had been seated or recumbent and thus, in some instances the patient may have moved or become anxious, thus transiently raising their resting HR. However, this would have been expected to weaken the correlation between the two methods of measurement, which nonetheless remained robust. Fourthly, it is possible that the beta-blocker dose was adjusted during the time between device interrogations. However, it is equally likely that the doses would be increased or decreased. Given that this would result in non-differential misclassification, it would bias the results toward the null and thus is unlikely to affect the finding of the present study.