Author + information
- Received July 16, 2017
- Revision received August 15, 2017
- Accepted August 18, 2017
- Published online October 30, 2017.
- Ali Vazir, MBBS, PhDa,b,
- Brian Claggett, PhDa,
- Bertram Pitt, MDc,
- Inder Anand, MDd,
- Nancy Sweitzer, MD, PhDe,
- James Fang, MDf,
- Jerome Fleg, MDg,
- Jean Rouleau, MDh,
- Sanjiv Shah, MDi,
- Marc A. Pfeffer, MD, PhDa and
- Scott D. Solomon, MDa,∗ ()
- aDivision of Cardiovascular Medicine, Brigham and Women’s Hospital, Harvard Medical School, Massachusetts
- bRoyal Brompton Hospital, Royal Brompton and Harefield NHS Foundation Trust and Institute of Cardiovascular Medicine and Sciences, National Heart and Lung Institute, Imperial College London, London, United Kingdom
- cUniversity of Michigan Medical School, Ann Arbor, Michigan
- dDivision of Cardiology, University of Minnesota, Minneapolis
- eDivision of Cardiovascular Medicine, University of Arizona, Tucson, Arizona
- fDivision of Cardiovascular Medicine, University of Utah, Salt Lake City, Utah
- gNational Heart, Lung, and Blood Institute, Division of Cardiovascular Sciences, Bethesda, Maryland
- hDepartment of Medicine/Cardiology, Montréal Heart Institute, Université de Montréal, Montréal, Québec, Canada
- iDivision of Cardiology, Northwestern University Feinberg School of Medicine, Chicago, Illinois
- ↵∗Address for correspondence:
Dr. Scott D. Solomon, Brigham and Women’s Hospital, Cardiovascular Division, 75 Francis Street, Boston, Massachusetts 02115.
Objectives The aim of this study was to examine the relationship between baseline heart rate (HR), change in HR from a preceding visit, and time-updated HR with subsequent outcomes in patients with heart failure with preserved ejection fraction (HFpEF) in the TOPCAT (Treatment of Preserved Cardiac Function Heart Failure With an Aldosterone Antagonist) trial.
Background Higher resting HR and increase in HR over time in patients with heart failure are associated with adverse outcomes. Whether these relationships between HR and prognosis are also observed in patients with HFpEF requires further assessment.
Methods In 1,767 patients enrolled in the TOPCAT trial from the Americas, the associations between baseline resting HR and change in HR from the preceding visit and clinical outcomes were examined using Cox proportional hazards models, along with the association between HR at each visit and outcome.
Results Both baseline HR (adjusted hazard ratio: 1.08; 95% confidence interval: 1.04 to 1.12) and change in HR from the preceding visit (adjusted hazard ratio: 1.09; 95% confidence interval: 1.05 to 1.14; p < 0.001 per 5 beats/min higher HR), after adjusting for covariates, were associated with a higher risk for the primary endpoint of cardiovascular death, hospitalization for HF, or aborted cardiac arrest. Time-updated resting HR at each visit was also associated with risk (adjusted hazard ratio: 1.11; 95% confidence interval: 1.07 to 1.15; p < 0.001 per 5 beats/min higher HR). Furthermore, a rise in resting HR of approximately 10 beats/min, beginning approximately 10 days prior to the primary endpoint, was observed.
Conclusions Baseline resting HR and change in HR over time predict outcomes in patients with HFpEF, as does time-updated HR during follow-up. These data suggest that frequent outpatient monitoring of HR, possibly with remote technologies, may identify patients with HFpEF who may be at increased risk for rehospitalization or death.
Elevated resting heart rate (HR) is a known risk factor for adverse outcome in patients with cardiovascular (CV) disease (1–5). The association between temporal change in HR and mortality has been assessed in subjects without known CV disease (6,7), as well as in subjects with hypertension (8) and heart failure (HF) (9). These studies showed that an increase in HR over time was associated with higher risk for adverse events.
Whether baseline HR and change in HR over time could be useful prognostic markers in patients with a contemporary definition of HF with preserved ejection fraction (HFpEF) requires further assessment. The availability of new remote monitoring strategies and devices that can measure and track HR provides a practical platform for assessing HR in real time in patients with HFpEF. Thus the objective of this analysis was to determine whether temporal change in HR from the preceding visit is of prognostic importance independent of baseline resting HR in patients with HFpEF within the TOPCAT (Treatment of Preserved Cardiac Function Heart Failure With an Aldosterone Antagonist) trial (10).
The TOPCAT trial
The details of the design, and overall findings have been previously reported (10). In brief, the TOPCAT trial enrolled a total of 3,445 patients with a contemporary definition of HFpEF (symptomatic HF and a left ventricular ejection fraction ≥45%) with either an admission with decompensated HF in the past 12 months or elevated natriuretic peptide (brain natriuretic peptide level ≥100 pg/ml or N-terminal pro–brain natriuretic peptide level ≥360 pg/ml), assigned to either spironolactone (15 to 45 mg/day) or placebo. The primary outcome was a composite of death of CV causes, aborted cardiac arrest, or hospitalization for the management of HF.
The main exclusion criteria were severe systemic illness with a life expectancy of <3 years, severe renal dysfunction (estimated glomerular filtration rate <30 ml/min/1.73 m2 body surface area or serum creatinine level ≥2.5 mg/dl [221 μmol/l]), and specific coexisting conditions, medications, or acute events (11).
The scheduled follow-up consisted of up to 16 trial visits, including an initial baseline visit followed by visits 2 and 3 occurring at 4 and 8 weeks, visit 4 at 4 months, and subsequent visits occurring every 4 months thereafter, up to visit 16. The median time interval between visits was 135 days (interquartile range [IQR]: 61 to 182 days). The follow-up time from the initial visit was a median of 3.3 years. Resting HR was recorded at each visit as part of the physical examination. At each visit, HR was recorded at rest by palpation, auscultation of the heart, or electrocardiography.
HR at any clinic visit and calculation of temporal change in HR
To assess temporal change in HR, we created a time-updated covariate representing the most recent available HR value for each patient at each visit over the course of the trial. We called this time-updated variable “HR at any visit.” A patient’s baseline HR was carried forward until the day of the first follow-up visit, at which time the new “current” value was used and subsequently carried forward until the next visit. Because there were up to 16 trial visits in the program, the resting HR was updated up to 15 times after baseline for each patient. We calculated temporal change in resting HR from the preceding visit (ΔHR) by subtracting the time-updated visit HR value from the value from the preceding visit. These changes in HR reflect changes occurring between visits.
Because of previously reported significant regional differences between the Americas and Russia and Georgia, and with very few events in Russia and Georgia (12), we performed the data analysis on subjects recruited from the Americas (n = 1,767) to have a more homogenous group of subjects with a contemporary definition of HFpEF. Sensitivity analyses were also performed using subjects from Russia and Georgia (n = 1,678) as well as the full TOPCAT cohort (n = 3,445), with data presented in the Online Appendix.
We related resting baseline HR, HR at any visit (i.e., time-updated HR) and ΔHR to several clinical outcomes. These included the primary endpoint of the TOPCAT trial, the composite of CV death, aborted cardiac arrest, and admission for worsening HF, as well as the outcome of all-cause mortality, CV death, hospitalization for HF, non-CV death, fatal and nonfatal myocardial infarction (MI), and stroke. The basis for using non-CV death as an outcome was to assess whether ΔHR was predictive of general ill health or whether it was specific to CV outcomes alone. Incidence rates were calculated per 100 patient-years.
The associations between outcomes and resting baseline HR, HR at any visit, and ΔHR as continuous covariates were assessed using multivariate Cox proportional hazards models. For baseline HR, HR at any time, and ΔHR, the estimated hazard ratio for each of these covariates correspond to a 5 beats/min difference in HR. We also modeled as 5 categories of HR change (decrease >10 beats/min, decrease of 5 to 10 beats/min, <5 beats/min change, increase of 5 to 10 beats/min, increase of >10 beats/min), on the basis of the rationale that changes in HR >5 beats/min would be considered clinically meaningful. In models using categorical covariates, the no-change-in-HR category (<5 beats/min change) was used as the reference for change in HR from the preceding visit.
The multivariate analysis adjusted for factors that had significant differences (p < 0.05) across the tertiles of baseline HR. Thus we adjusted for age, sex, previous hospitalization for HF in past 12 months prior to randomization, history of MI, previous percutaneous coronary intervention, use at baseline of beta-blocker and rate-limiting calcium antagonist, and time-updated diastolic blood pressure and weight. We also created a second multivariate model that adjusted for additional covariates from our initial model. These included baseline New York Heart Association functional class III or IV versus I or II, race, smoking status, chest radiography (signs of congestion), presence of atrial fibrillation at baseline, QRS duration, left ventricular hypertrophy at baseline, paroxysmal nocturnal dyspnea, treatment with spironolactone, diabetes mellitus, peripheral vascular disease at baseline, baseline creatinine, and time-updated use and dose change of beta-blocker and rate-limiting calcium antagonist to account for starting or stopping and dose changes. For the second model, we also adjusted for time-updated systolic and diastolic blood pressure and time-updated New York Heart Association functional class at each visit. For the second model, outcomes without MI incident nonfatal MI was also included in the model. The results of the second model are reported in the Online Appendix. We also controlled for baseline HR when modeling for HR at any time for the initial model and also the second model; however, we replaced baseline HR with HR from the preceding visit when modeling ΔHR.
An adjusted model using a restricted cubic spline with 5 knots was constructed to flexibly display the relationship between the hazard of developing an outcome and the continuous covariate of HR at any time, using a reference value of 60 beats/min. For ΔHR, zero was used as the reference. Interaction testing was used to determine whether the relationship between ΔHR and outcomes varied in different subgroups (patients recruited from the Americas vs. Russia and Georgia, with or without atrial fibrillation, diabetes mellitus at baseline, ejection fraction ≥55%, with or without beta-blocker use at any time).
We also performed an analysis to relate the value of resting HR up to 30 days before the occurrence of the primary endpoint of the study; this analysis excluded patients who did not reach the primary endpoint. Relative to the time of randomization, all patients followed a similar study visit schedule. However, because clinical events could occur at any time between scheduled study visits, the amount of time elapsed between a clinical event and that patient’s prior study visit was continuous. Furthermore, for each patient who experienced a clinical event, multiple prior HR values were available, each collected a different number of days prior to the event. We estimated the average values of HR as a function of the number of days prior to a primary endpoint. This was done using a mixed-effects linear regression model, accounting for multiple observations per patient using patient ID as a random intercept term. The exposure variable was the date of the confirmed primary endpoint subtracted from the date of the HR measurement (i.e., days prior to the event). The outcome variable was the reported HR. This methodology has been used previously to estimate changes in other biomarkers prior to an event (13). No measurements obtained after the event were used for this analysis. We conducted sensitivity analyses using only data collected within 30 days of the primary event. To allow for potentially nonlinear changes over time, the exposure variable (days prior to event) was modeled using restricted cubic spline terms.
Differences between baseline characteristics of patients were compared using trend tests. Continuous variables are expressed as median (IQR) and categorical variables as counts and percentages. Assessment of the proportional hazards assumption was performed using Schoenfeld residuals (14). All p values were 2-sided, and a p value <0.05 was considered to indicate statistical significance. Analysis was performed using Stata version 13.1 (StataCorp., College Station, Texas).
Baseline characteristics of patients
The key baseline characteristics of the study group composed of subjects from the Americas are summarized in Table 1. The median age of the cohort was 72 years (IQR: 64 to 79 years), and half of the population was female. At baseline, 72% had entered the study on the basis of an HF admission in the past 12 months. Ninety percent of the patients had histories of hypertension, and one-fifth had previous MIs. The mean left ventricular ejection fraction was 58 ± 8%. Atrial fibrillation was present in just over 40% at baseline, and a third of the patients were anticoagulated.
The baseline characteristics of the patients by tertiles of baseline HR are also described in Table 1. Patients within the highest tertile of baseline HR, compared with lower tertiles of HR, had higher diastolic pressure, had higher values of body mass index, more frequently had histories of hospitalization for HF, had a lower prevalence of ischemic heart disease, and were more frequently on diuretic agents. There was a higher prevalence of atrial fibrillation in patients within the highest tertile of HR.
Resting HR measured at any time and temporal change in HR
The median values of resting HR measured at any visit for the total Americas cohort were almost identical to the resting baseline HR of 68 beats/min (IQR: 61 to 76 beats/min) (Table 1). The distribution of ΔHR occurring over a median of 135 days (IQR: 61 to 182 days) is summarized in Figure 1. The majority of patients within the cohort did not change their HRs from the preceding visit (median 0 beats/min; IQR: −6 to 6 beats/min).
Temporal changes in HR from the preceding visit (ΔHR) and outcomes
Over a median follow-up period of 3.3 years, 522 patients experienced the primary endpoint of the study, a composite of CV death, aborted cardiac arrest, or hospitalization for HF. The numbers of events for other outcomes are summarized in Table 2. As a continuous covariate, change in HR from the preceding visit was associated with all outcomes measures, except for fatal and nonfatal MI (Table 2). For example, each 5 beats/min increase in HR from the preceding visit was associated with 9% higher risk for the primary outcome and 17% higher risk for all-cause mortality. Furthermore, each 5 beats/min increase in HR from the preceding visit was also associated with a 14%, 11%, and 20% higher risk for CV death, hospitalization for HF, and non-CV death, respectively (Table 2).
The restricted cubic spline model showed that the relationship between ΔHR and the primary endpoint was nonlinear (Figure 2A). Any rise in HR was associated with elevated risk; however, a decline in HR was not significantly associated with lower risk for the primary endpoint. In patients experiencing the primary endpoint, a rise in HR of >10 beats/min was seen 5 to 10 days prior to the primary endpoint (Figure 2B).
When we analyzed the data categorically, ΔHR was also associated with outcomes (Figure 3). A >10 beats/min rise in HR from the preceding visit compared with the group with no change in HR was associated with a 70% higher risk for the primary endpoint and a 120%, 116%, and 118% higher risk for all-cause mortality, CV death, or non-CV death, respectively (Figure 3). A >10 beats/min rise in HR from the preceding visit compared with the group with no change in HR was associated with 60% higher risk for hospitalization for HF. A drop in HR of >10 beats/min was significantly associated with a 32% and 51% reduced risk for all-cause mortality and non-CV death. There was no association with drop in HR and reduction in risk for CV death.
The use of a beta-blocker at baseline in the study did not affect the relationship between ΔHR and outcomes (p for interaction = 0.90 for the primary endpoint), nor did the presence of atrial fibrillation at the time of randomization (p for interaction = 0.41) for the primary endpoint. Ejection fraction >55% versus ≤55% did not modify the relationship between change in HR from the preceding visit and the primary endpoint (p for interaction = 0.90). No interactions were detected between the other specified subgroups and ΔHR, such as patients with or without diabetes mellitus.
Sensitivity analysis assessing the relationship between ΔHR and the primary endpoint within the total cohort of the TOPCAT study detected no significant interaction between geographic region and the relationship between ΔHR and primary endpoint (p for interaction = 0.86) (Online Table 1). Furthermore, sensitivity analysis using the second model, which adjusted for time-updated changes in doses of beta-blocker and rate-limiting calcium antagonist and many more covariates, had very minor differences in the quality of the results (Online Table 2).
HR at any time and outcome
As a continuous covariate, both baseline HR and time-updated HR, which represents HR at any visit, were associated with most adverse outcomes (Table 2). However, baseline HR was not associated with fatal and nonfatal MI or with fatal and nonfatal stroke, and HR at any time was not associated with fatal and nonfatal MI. For each 5 beats/min increase in HR at any time, the primary endpoint and all-cause mortality were 11% and 17% higher, respectively.
A resting HR of 61 to 76 beats/min at any time during follow-up was not associated with a higher risk for the primary endpoint compared with a resting HR of 60 beats/min (Figure 2C). However, a resting HR >76 beats/min at any time during follow-up was associated with a higher risk for the primary endpoint (Figure 2C). In patients with atrial fibrillation at baseline, a higher HR at any time was associated with the primary endpoint, such that for each 5 beats/min increase in HR, the adjusted hazard ratio for the primary endpoint was 1.09 (95% confidence interval: 1.03 to 1.16; p < 0.01). The presence of atrial fibrillation at baseline did not modify the relationship between HR at any time and the primary outcome (p for interaction = 0.50). The presence of diabetes mellitus or preserved versus reduced ejection fraction or the presence of a beta-blocker produced no statistically significant interaction between HR at any time and the primary endpoint.
Sensitivity analysis using the total cohort of the TOPCAT study demonstrated that geographic region was associated with a differential relationship between baseline HR and the primary endpoint (p for interaction <0.01) and also the relationship between HR at any time and the primary endpoint (p for interaction <0.01) (Online Table 1). The magnitude of the hazard ratio associated with baseline HR was larger in subjects from Russia and Georgia compared with subjects from the Americas.
In a large cohort of patients with HFpEF, we found that baseline resting HR and change in resting HR from the preceding clinic visit occurring over a median of 135 days (IQR: 61 to 182 days) were independent predictors of the composite endpoint of CV death, aborted cardiac arrest or hospitalization for HF, and other endpoints such as all-cause mortality and non-CV death. Although increases in HR were associated with higher risk, reductions in HR were not significantly associated with lower risk. Higher HR at any time was also predictive of adverse outcomes. These findings suggest that resting HR may be a useful and easily measured prognostic biomarker in the management of patients with HFpEF.
Our study further supports that changes in HR over time from preceding clinic visits are of prognostic importance in patients with HFpEF, irrespective of cardiac rhythm. Patients with HFpEF and atrial fibrillation were at a similar risk compared with those in sinus rhythm; the presence of atrial fibrillation did not modify the relationship between baseline HR, HR at any time, or change in HR from the preceding visit and adverse outcome. Increases in HR in patients with HF may reflect higher sympathetic tone due to decompensated HF or further progression of HF (15). Another possible cause of an increase in HR is the onset of an atrial arrhythmia such as atrial fibrillation. In contrast, a decline in HR may reflect improving cardiac function and lower sympathetic tone. However, in this cohort, a decrease in HR was associated with a lower risk only for non-CV death, not for any of the adverse CV outcomes.
We controlled for both the use and dose changes of beta-blockers and rate-lowering calcium antagonists in a time-updated analysis, as shown in Online Table 2. Thus, the risk associated with change in HR was independent of both the use of and dose changes of these drugs. Furthermore, interaction testing showed that the use of a beta-blocker at any time in the study did not modify the relationship between ΔHR and outcome.
As previously demonstrated in the analysis of HR and its change over time in the CHARM (Candesartan in Heart Failure: Assessment of Mortality and Morbidity) program (9), an increase in HR from the preceding visit was also associated with increased risk for non-CV death, which further supports that such an increase in HR is likely to be a nonspecific signal of deteriorating health or episodes of acute infection or other systemic stress.
Our study also showed that a temporal drop in HR over a median of 135 days was associated with reduced risk only for non-CV death but was not associated with lower risk for adverse CV events. The explanation for these findings is unclear, and the results differ from those found in the CHARM population, in which a decline in HR was associated with reduced risk for the composite of CV death or hospitalization for HF as well as for non-CV death. Perhaps the decrease in HR over time has prognostic significance only in patients with HF and reduced ejection fraction, but this will require further examination. Another explanation that could be considered is the presence of chronotropic incompetence in patients with HFpEF, which is poorly tolerated. In this study, beta-blocker therapy was associated with poorer outcome (unadjusted hazard ratio: 1.11; 95% confidence interval: 1.04 to 1.18; p = 0.001).
Our analysis provides additional evidence that the value of HR recorded at any time during the study was also of prognostic importance, in keeping with findings in the CHARM program (9). In the TOPCAT trial, patients with HR at any time <76 beats/min had the lowest risk for adverse outcomes when 60 beats/min was taken as the reference point. This finding is identical to that in the analysis of the CHARM program. Thus, a resting HR >76 beats/min appears to be associated with a higher risk for adverse events. In our analysis, HR appeared to be stable over time, except that 5 to 10 days before an adverse event occurred, resting HR appeared to rise, as demonstrated in Figure 2B. This method of analysis provides further support that a rise in HR is a marker of adverse events and could provide a warning sign to both patients and providers of such impending events. Our observation suggests that monitoring HR over time, as a biomarker of severity of HF, in the clinic setting or perhaps remotely, may be useful in identifying patients with HFpEF at greatest risk for readmission and death. Further research is required to assess the utility of such an approach.
In our sensitivity analysis, presented in the Online Appendix, we found that geographic region did not produce a differential relationship between the primary endpoint and change in HR, while region did modify the relationship between baseline HR and the primary endpoint and also between HR at any time and the primary endpoint (Online Table 1). The magnitudes of the hazard ratios for both baseline HR and HR at any time for the primary endpoint were larger in subjects from Russia and Georgia compared with subjects from the Americas, suggesting that baseline HR and HR at any time are more important predictors of the primary endpoint in subjects from Russia and Georgia compared with the Americas. A plausible explanation for this relationship could be greater heterogeneity of the TOPCAT population from Russia and Georgia, with patients with true HFpEF having higher baseline HRs and going on to develop the primary event (Online Figure 1).
First, we were dependent on investigator-reported HR at each visit, which may have been measured in different ways, at different times of the day, and under different circumstances. Another limitation is that time-updated analysis limits the detailed characterization of the patients who have increases or decreases in HR from baseline or the preceding visit. However, an advantage of time-updated analysis is that it uses HR data from all visits, allowing the calculation of change in HR over short periods of time and the total time spent in each category with the associated number of events.
Strengths of this study included the large sample of patients with HFpEF, including a large number who experienced adverse events. In addition, our analysis controlled for rate-lowering medications such as beta-blockers and calcium antagonists using a time-updated analysis, taking into account both the use and changes in dose of these drugs. Thus, the importance of HR change appears to be independent of the use of beta-blockers and calcium antagonists.
In patients with HFpEF, including those with atrial fibrillation, both baseline resting HR and change in resting HR over time from the preceding clinic visit were independently associated with clinical outcomes, such that a higher baseline HR and an increase in HR were associated with elevated risk for CV events. However, a decline in HR over time was not associated with a lower risk for CV events. Our findings support the importance of measuring resting HR in everyday clinical practice, potentially with remote monitoring, as a way to identify patients with HFpEF at greatest risk for readmission and death.
COMPETENCY IN MEDICAL KNOWLEDGE: Resting HR and change in HR over time were associated with adverse outcomes in patients with a contemporary definition of HFpEF. Furthermore, we observed a rise in resting HR of approximately 10 beats/min beginning approximately 10 days prior to the endpoint of HF hospitalization, aborted sudden cardiac death, and CV death. These data suggest that frequent outpatient monitoring of HR, possibly with remote technologies, may identify patients with HFpEF who may be at increased risk for rehospitalization or death.
TRANSLATIONAL OUTLOOK: Prospective studies are required to assess whether remote tracking of resting HR in patients with HFpEF can help identify those at increased risk for hospitalization or death.
TOPCAT was funded by the National Institutes of Health, National Heart, Lung, and Blood Institute (contract number HHSN268200425207C). The contents of this article are solely the responsibility of the authors and do not necessarily reflect the official views of the National Heart, Lung, and Blood Institute, the National Institutes of Health, or the U.S. government. Dr. Pfeffer has received consulting fees from Aastrom, Abbott Vascular, Amgen, Bristol-Myers Squibb, Cerenis, Concert, Fibrogen, Genzyme, GlaxoSmithKline, Hamilton Health Sciences, Medtronic, Merck, Novo Nordisk, Roche, Salix, Sanderling, Serono, Servier, Teva, and the University of Oxford; and has received grant support from Amgen, Celladon, Novartis, and Sanofi. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- change in heart rate from the preceding visit
- heart failure
- heart failure with preserved ejection fraction
- heart rate
- interquartile range
- myocardial infarction
- Received July 16, 2017.
- Revision received August 15, 2017.
- Accepted August 18, 2017.
- 2017 American College of Cardiology Foundation
- Castagno D.,
- Skali H.,
- Takeuchi M.,
- et al.
- Fox K.,
- Borer J.S.,
- Camm A.J.,
- et al.
- Poole-Wilson P.A.,
- Uretsky B.F.,
- Thygesen K.,
- et al.
- Shah S.J.,
- Heitner J.F.,
- Sweitzer N.K.,
- et al.
- Pfeffer M.A.,
- Claggett B.,
- Assmann S.F.,
- et al.
- Lewis E.F.,
- Claggett B.,
- Parfrey P.S.,
- et al.
- Kaye D.M.,
- Lambert G.W.,
- Lefkovits J.,
- Morris M.,
- Jennings G.,
- Esler M.D.