Author + information
- Received September 18, 2014
- Revision received November 12, 2014
- Accepted November 14, 2014
- Published online April 1, 2015.
- S.M. Afzal Sohaib, MBBS∗,
- Judith A. Finegold, MA∗,
- Sukhjinder S. Nijjer, MBChB∗,
- Ruhella Hossain, MBBS∗,
- Cecilia Linde, MD†,
- Wayne C. Levy, MD‡,
- Richard Sutton, MD∗,
- Prapa Kanagaratnam, PhD∗,
- Darrel P. Francis, MA∗∗ ( and )
- Zachary I. Whinnett, PhD∗
- ∗National Heart & Lung Institute, Imperial College, London, United Kingdom
- †Department of Cardiology, Karolinska University Hospital, Stockholm, Sweden
- ‡Division of Cardiology, University of Washington, Seattle, Washington
- ↵∗Reprint requests and correspondence:
Dr. Darrel P. Francis, International Centre for Circulatory Health, Imperial College London, 59-61 North Wharf Road, London W2 1LA, United Kingdom.
Objectives This study examined the time course of clinical events in cardiac resynchronization therapy (CRT) trials.
Background Recent randomized controlled trial results suggest that in heart failure with narrow QRS, biventricular pacing (CRT) may increase mortality. The authors proposed implant complications as the cause, rather than a progressive adverse physiological effect.
Methods The study identified all trials comparing CRT with no CRT, which reported Kaplan-Meier curves in groups defined by QRS: narrow, non–left bundle branch block (LBBB) broad, and LBBB broad. For each trial, the change in life span every 3 months up to 3.5 years (the longest time for which data are available) was calculated and a power law was fitted, that is, ∝ timen.
Results Four trials (MADIT-CRT [Multicenter Automatic Defibrillator Implantation Trial-Cardiac Resynchronization Therapy], RAFT [Resynchronization-Defibrillation for Ambulatory Heart Failure Trial], REVERSE [REsynchronization reVErses Remodeling in Systolic left vEntricular dysfunction], and EchoCRT [Echocardiography Guided Cardiac Resynchronization Therapy]), totaling 4,717 patients, reported curves for mortality or heart failure–related hospitalization, or for mortality. In patients with LBBB broad QRS (within MADIT-CRT), life span gain increased in proportion to time1.94. In contrast, in patients with non-LBBB broad QRS (within MADIT-CRT) and patients with narrow QRS (EchoCRT), life span was lost in proportion to time1.92 and time,1.96 respectively. Hospitalization-free survival showed similar patterns.
Conclusions The nonlinear growth of life span gained when a CRT device is implanted in patients with LBBB broad QRS is unfortunately mirrored by a similarly progressive loss in life span in narrow QRS heart failure. This suggests the culprit is a progressive physiological effect of pacing rather than implant complications. If these data are not sufficient, a randomized controlled trial of deactivating CRT in patients with narrow QRS may now be needed, with a primary endpoint of increasing survival.
Cardiac resynchronization therapy (CRT) has strong evidence of benefit in patients with symptomatic heart failure with wide QRS and reduced left ventricular ejection fraction (1–6). More recently it has been shown that the benefits may be limited to patients with left bundle branch block (LBBB) (7,8) or very wide QRS (9). The benefits of CRT tend to be progressive with life span gain occurring in a nonlinear manner, suggesting an ongoing beneficial therapeutic effect of biventricular pacing over a number of years (10). Over many years, however, patients with a variety of electrocardiographic (ECG) characteristics received CRT implants. Data from the European CRT Survey suggest that as many as 32% of CRT recipients did not have underlying LBBB (11). Alternative indications might have included mechanical dyssynchrony (11,12). In patients with narrow QRS, even when selected for having mechanical dyssynchrony on echocardiography, CRT increased mortality by 81% (p = 0.02) in an international randomized controlled trial (13). If this is true, a relatively large population with CRT may be at potential risk of adverse effects of CRT instead of benefit.
How these findings should affect clinical practice depends on the cause of increased mortality, specifically whether it is the result of implant complications or an undesirable effect of ventricular pacing. Device implant complications will tend to cluster around the time of implant. For example, for dual-chamber pacing, one study shows that 75% of all complications occurring over a 3-year period occur in the first 3 months post-implant (14), and for CRT, one center has reported that 59% of complications in a mean follow-up of 2.7 years occur in the first 90 days post-implant (15). These event rates in the first 3 months are 33 and 14 times, respectively, the rates in the remaining periods. If early implant complications are responsible, then there remains no issue for the remainder of patients who did not have implant complications to continue with CRT pacing. In contrast, if the excess mortality is driven by a detrimental effect from the action of pacing from CRT, then we may have an opportunity to improve outcomes in surviving recipients by deactivating CRT.
These 2 possibilities should generate different time courses of effect on mortality. Implant complications predominantly occur early, whereas progressive consequences of the detrimental activation sequence of CRT compared with intrinsic conduction may occur gradually throughout follow-up. Mathematically, the pattern can be quantified by fitting the change in life span gain to a power law of time. In this study, we did this with the data published by the randomized controlled trials assessing CRT in heart failure.
Eligibility and search strategy
We searched MEDLINE and Google Scholar from inception to April 2014 using the following search criteria: cardiac resynchronization therapy, survival, mortality, left bundle branch block, right bundle branch block, and QRS morphology. Reference lists of the retrieved articles were hand-searched for additional publications.
We identified all randomized controlled trials comparing CRT with no CRT (either CRT-pacemaker or CRT-defibrillator) and reported Kaplan-Meier survival curves for mortality stratified by QRS morphology (LBBB, non-LBBB broad QRS, and narrow QRS). We similarly identified studies that provided Kaplan-Meier curves for a combined endpoint such as death or heart failure hospitalization. Where studies stratified results by ECGs, we checked whether they described blinded analysis of ECG morphology.
Calculation of life span gain or lost
The segmental area between the 2 curves was calculated for 3-month intervals. This represented life span gain or loss per patient randomized during that period. The cumulative area between the curves up to each time point (life span gain up to that point per patient) was also calculated. The process has been described (10,16) and is illustrated in Figure 1. As an example, to calculate the life years gained between 3 months and 6 months, the following calculation is used, with each survival rate expressed as a proportion between 0 and 1:
The cumulative gain in life span at any time point was defined as the sum of the gains in each 3-month period from zero to that time point. This cumulative life span gain at each 3-month interval was then re-expressed as a proportion of the total life span lost or gained at the end of the period analyzed.
Where studies randomized patients to CRT versus control and separately reported results for LBBB and non-LBBB broad QRS, we analyzed the Kaplan-Meier curves for the 2 ECG morphologies separately. Each trial had different follow-up durations. The latest time point at which overall mortality data were available consistently in all the trials was 3.5 years. The latest time point at which there were data describing death or heart failure hospitalization available consistently in all the trials were 2 years.
The effect on life span loss or gain per year was calculated for each ECG morphology group and fitted to a power law. All statistical analyses were performed using the R software for statistical computing version 3.0.2 (R Foundation for Statistical Computing, Vienna, Austria).
Data were available from 5 groups in 4 trials totaling 4,717 patients. Three trials each showed data for patients with LBBB and non-LBBB broad QRS (5,8,17), and 1 trial showed data for patients with narrow QRS (13) (Table 1). In all cases, the data were randomized comparisons between CRT and no CRT. All trials stratifying subjects by ECG morphology had sufficient blinding of ECG morphology to treatment and outcome: ECGs for MADIT-CRT (Multicenter Automatic Defibrillator Implantation Trial-Cardiac Resynchronization Therapy) were analyzed in a core laboratory (7); ECGs for REVERSE (REsynchronization reVErses Remodeling in Systolic left vEntricular dysfunction) were analyzed by investigators blinded to treatment allocation and outcome, with a further 50 randomly chosen ECGs assessed for intraobserver and interobserver variability (8); and QRS verification in RAFT (Resynchronization-Defibrillation for Ambulatory Heart Failure Trial) was performed by 3 different investigators blinded to treatment allocation and outcomes (17).
Life years gained from CRT in LBBB
In patients with LBBB, the increase in life span achieved with CRT grew with time, and much more than linearly. This is evident on assessment of the Kaplan-Meier data of the individual trials (Figure 2). The best-fit power-law relationship was life span gain proportional to follow-up time1.94 (R2 = 0.998, p < 0.0001) (Figure 3). This finding is consistent with a favorable physiological effect of CRT, with progressively more life span gain increment as the window of observation lengthens.
Life year impact of CRT in non-LBBB broad QRS
In the non-LBBB broad QRS group, life span gain was numerically shorter in the patients randomized to CRT than control. This is evident in the curves of the individual trials (Figure 2). The best-fit power-law relationship was life span gain proportional to follow-up time1.92 (R2 = 0.996, p < 0.0001) (Figure 3). This is consistent with there being an adverse effect of pacing from CRT, reducing life span further as the window of observation lengthens.
Life year impact from CRT in narrow QRS
In patients undergoing CRT with narrow QRS, life span gain was numerically shorter in the patients randomized to CRT than control. This is visible in the curves of the individual trials (Figure 2). The best-fit power-law relationship was life span gain proportional to follow-up time1.96 (R2 = 0.994, p < 0.0001) (Figure 3). This is consistent with there being an adverse effect of pacing from CRT, which produces progressively more life span decrement as the window of observation lengthens.
Impact of CRT on survival time free of first hospitalization
In the LBBB broad QRS group, time free from hospitalization and mortality was numerically longer in the patients randomized to CRT than those randomized to control (Figure 4B). The narrow QRS group showed the opposite direction of effect (Figure 4A, left). The non-LBBB broad QRS group was mixed (Figure 4A, right).
Life span gain from CRT develops progressively with time after implantation. The same shape of time course is seen in all 3 groups of patients, although the direction of the effect is different. This is consistent with the effect being mediated by the physiological consequences of pacing.
The nonlinear growth of life span gain from CRT was originally documented in trials that recruited patients with predominantly LBBB (10). The present study separately analyzes patients with narrow QRS and patients with non-LBBB broad QRS. In these patients, CRT affects life span in a similarly shaped time course, but, crucially, the effect is inverted, meaning there is a loss of life span that expands with approximately the square of time. This shape implies that this is due to the pacing effect of CRT rather than the initial risk associated with device implantation.
Mechanisms for adverse impact on mortality
EchoCRT (Echocardiography Guided Cardiac Resynchronization Therapy) was the landmark study demonstrating a clear increase in mortality in patients undergoing CRT with narrow QRS, despite being designed to identify and recruit those patients with the best prospect of showing a benefit, namely, those with mechanical dyssynchrony.
The discussion of the EchoCRT publication suggested that implantation or subsequent lead manipulation might have caused the increased mortality (13). However, this is not a plausible cause. First, the controls also underwent implantation. Second, mortality from implantation would be expected to manifest early and not many years later as evidenced in the data reported. Moreover, lead manipulation is a relatively unlikely cause of excess mortality, with infection being the most likely adverse outcome (18). Our interpretation of the EchoCRT mortality data differs, because we observe that the reported mortality was driven by more than 20 excess cardiovascular deaths (p < 0.01), of which the great majority were classified as “heart failure” or “arrhythmic events” rather than infection. Instead, the pattern suggests that CRT in these patients unintentionally contributes to heart failure progression.
The main study publication does not appear to specify that the increase in mortality was a progressive process more suggestive of a pathophysiological consequence of pacing rather than a procedural consequence of the implant. By using a systematic approach, we identified the articles citing this publication and examined them to see if they drew this inference for themselves. We used Google Scholar to identify citing documents. The 37 citing documents that were accessible from Imperial College, London, were read independently by 2 authors (S.M.A.S. and R.H.), with disagreements resolved by a third author (D.P.F.). A total of 25 documents (68%) mentioned that harm could be caused by CRT, whereas the remainder stated that there was no benefit or did not comment. Of the 37 documents, only 2 (5%) demonstrated awareness that the harm was due to the ongoing effects of pacing rather than implantation (Online Appendix).
Difference between device and medical therapy
If a drug is found to increase mortality, administration can be stopped. Moreover, information on side effects in established drug therapy is readily available for patients. Understanding benefits and adverse effects of implanted devices is more complex. For CRT in heart failure, for example, this information is not fully established. If an adverse effect is suspected, interrupting therapy from a device that has already been implanted may appear unfamiliar and uncomfortable to both patients and physicians, unless the evidence to do so is very compelling. However, as information accumulates on who will benefit or be harmed by CRT, it is becoming increasingly clear that randomization to stopping CRT in an already implanted device may indeed be ethical, and perhaps may even be an ethical imperative. Thus, if the ongoing action of an implanted device is found to be a progressive increase in mortality, there should be no reluctance in turning off the device in the survivors.
Call for a trial: deactivating CRT in patients with non-LBBB
It has been reported that one third of recipients who already have CRT in Europe have neither underlying LBBB nor broad QRS (11). This amounts to approximately 20,000 individuals in Europe alone (19). In North America and elsewhere, there may be a similarly nontrivial number. This poses 2 opportunities. First, we may still be able to provide extra life span to a larger number of patients enrolled in these trials. Second, there is a large pool of patients suitable for enrollment in a randomized controlled trial to answer definitively the hypothesis generated in this study. Such a randomized controlled trial would randomize these patients into leaving on versus turning off the CRT mode. This trial could be done at low cost because there is no need to implant a device, rather only reprogramming of the currently implanted device. The study would only require informed consent and online randomization with follow-up limited to all-cause mortality in the interests of simplicity and to avoid the substantial costs of segregating causes of death or determining hospitalization. If the hazard ratio of 1.8 seen in EchoCRT is representative, then the hazard ratio for switching off CRT would be approximately 0.6, and therefore the number of patients and duration of follow-up needed would be modest. Moreover, the potential enrollees are already under regular routine device follow-up, and therefore the device community could conveniently approach all approximately 40,000 patients promptly with minimal additional visits and cost.
A perceived difficulty may be explaining to the patient why, having received a device, there is a proposal to deactivate the CRT element. One option might be to explain that the device can be programmed in a variety of different ways, and it is not known which is best. On the one hand, the atrioventricular (AV) delay could be set to always capture the ventricle. On the other hand, it could be set to only capture the ventricle when natural ventricular activation fails. Thus, the trial would have one arm using the currently programmed AV delay and the other arm set to an AV delay longer than the intrinsic AV delay, or set to a low backup rate solely to protect against bradycardia. In the case of narrow QRS, the informed consent process would also require that patients are informed that new clinical trial evidence, which might not have been available at the time of their implant, suggests that such a device would not usually be implanted with their current ECG morphology, and that therefore switching off the pacing may be beneficial at present. However, should the QRS later broaden or the AV delay prolong unacceptably, the device would already be in place and the CRT function could then be usefully switched on. In the meantime, the patient would be protected against asystole and (if it is a defibrillator) tachyarrhythmias.
Such a trial might enroll patients who were not reliant on pacing, whose left ventricular lead was functioning, whose native QRS was below a threshold duration and not in an LBBB pattern, and who are free of serious noncardiac pathology that would limit life span. It might exclude patients with long PR intervals because in this subgroup of non-LBBB there seems to be a beneficial effect of CRT (20). To minimize cost, the baseline data collected could be a simple set of widely available clinical variables, such as the elements of the modified Seattle Heart Failure score validated in SCD-HeFT (Sudden Cardiac Death in Heart Failure Trial) (21,22).
It may be tempting to plan to restrict such a trial to those who had not experienced a favorable symptomatic response. However, this may be unwise. Randomized trial data show that when compared with a placebo control arm, the incremental rate of patients who experience a symptomatic response with CRT pacing is only approximately 15% (23). This means that the remainder of patients experiencing a symptomatic response with CRT, who are twice as numerous in trials and may be more numerous outside carefully monitored trial environments, have not necessarily received any symptomatic benefit from the pacing itself because they would have felt as well without it. Excluding symptomatic responders would have the undesirable effect of causing patients who express an optimistic view of their condition and their care to miss the opportunity to participate in a trial that might have given them an opportunity for additional life span. Blinded randomized controlled data on these patients would be the most valuable and practice-changing information from such a trial because they may show that patients who have CRT switched off gain a better symptomatic state than those who continue CRT. If the consent process is designed carefully with as much opportunity for positive placebo, there might be net symptomatic improvements for both arms. Not every specialist might consider this trial necessary. Some may consider the information in Figures 2 and 3 to be a sufficient indication of harm to merit routine deactivation of biventricular pacing in this cohort.
There is some evidence to suggest that there is a benefit for CRT in those with a non-LBBB ECG, but a very broad QRS (>140 to 150 ms) (9). This trial would also present an opportunity to address this question.
This study uses the published mortality time course data from randomized controlled trials of CRT versus no CRT, in studies that present Kaplan-Meier curves for both arms of strata defined by QRS characteristics. There are other trials that did not present data stratified by ECG morphology (3,24,25) and therefore were not included. Such information is likely to add further information, particularly in those with non-LBBB QRS widening.
Moreover, the evidence of increased mortality in subgroups of patients receiving CRT is only in the short term. It is not known whether such effects might halt or reverse over longer periods of time. However, when the effects of CRT are beneficial, they tend to grow with time (10,26). It might not be prudent to hope that physiological harm would behave differently.
Except for the EchoCRT study, the studies eligible for our analysis covered the milder parts of the spectrum of heart failure. An even more pronounced effect on mortality in non-LBBB might be seen when the trials restricting inclusion to individuals in New York Heart Association functional class III and IV are included.
Our study relied on the evaluation of the ECG as performed in the original clinical trial. If that classification was incorrect, then our analysis would suffer accordingly.
The impact of CRT on survival time is nonlinearly dependent on the time window over which it is calculated, growing approximately with the square of time. In patients with underlying LBBB, this impact is a benefit, but in those without underlying LBBB, this nonlinearly expanding impact on survival duration is adverse. The time course fits a progressive adverse physiological effect of pacing rather than implant complications. This suggests an opportunity for benefit by deactivating pacing in such patients. We should consider a randomized controlled trial of deactivating CRT in recipients with narrow QRS or without underlying LBBB, with a primary endpoint of survival.
The authors thank the National Institute for Health Research Biomedical Research Centre, based at Imperial College Healthcare NHS Trust and Imperial College London, for infrastructural support.
For a list of supplemental references, please see the online version of this article.
Drs. Sohaib, Finegold, Francis, and Whinnett are supported by the British Heart Foundation (FS/10/038, FS/13/44/30291, FS/11/92/29122, FS/14/25/30676). Dr. Nijjer is supported by the Medical Research Council (UK) (G1100443). Dr. Linde is supported by the Swedish Heart Lung Foundation (Grants 20080498 and 20110406) and the Stockholm County Council (Grants 20090376 and 20110610); was the principal investigator of REVERSE, a CRT study sponsored by Medtronic, Inc.; has received research grants, speaker honoraria, and consulting fees from Medtronic, Inc.; has received speaker honoraria and consulting fees from St. Jude Medical, Inc.; and is on the advisory board of Cardio 3. Dr. Levy has received research grants from the National Institutes of Health, Amgen, Thoratec, ResMed, Impulse Dynamics, Medtronic, Inc., and HeartWare, Inc.; is a consultant to Novartis, HeartWare, Inc., GE Healthcare, Magellan Health, and PharmIn; and holds equity in PharmIn. The University of Washington has received licensing for the Seattle Heart Failure Model from Impulse Dynamics, Thoratec, and Epocrates. Dr. Sutton holds a research grant from and is a consultant to Medtronic, Inc.; is on the Speakers Bureaus of Medtronic, Inc. and St Jude Medical, Inc.; and is a shareholder of Boston Scientific, Inc. and the American Society for Clinical Investigation. Dr. Francis is a consultant to Medtronic, Inc. and Sorin. Dr. Whinnett acts as a consultant to St. Jude Medical, Inc. and Medtronic, Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- cardiac resynchronization therapy
- left bundle branch block
- Received September 18, 2014.
- Revision received November 12, 2014.
- Accepted November 14, 2014.
- American College of Cardiology Foundation
- Linde C.,
- Abraham W.T.,
- Gold M.R.,
- St John Sutton M.,
- Ghio S.,
- Daubert C.
- Daubert C.,
- Gold M.R.,
- Abraham W.T.,
- et al.
- Zareba W.,
- Klein H.,
- Cygankiewicz I.,
- et al.
- Gold M.R.,
- Thébault C.,
- Linde C.,
- et al.
- Cleland J.G.,
- Abraham W.T.,
- Linde C.,
- et al.
- Finegold J.A.,
- Raphael C.E.,
- Levy W.C.,
- Whinnett Z.,
- Francis D.P.
- Ahsan S.Y.,
- Saberwal B.,
- Lambiase P.D.,
- et al.
- Salukhe T.V.,
- Dimopoulos K.,
- Sutton R.,
- Coats A.J.,
- Piepoli M.,
- Francis D.P.
- Birnie D.H.,
- Ha A.,
- Higginson L.,
- et al.
- Baddour L.M.,
- Epstein A.E.,
- Erickson C.C.,
- et al.
- Arribas F.,
- Auricchio A.,
- Boriani G.,
- et al.
- Kutyifa V.,
- Stockburger M.,
- Daubert J.P.,
- et al.
- Levy W.C.,
- Lee K.L.,
- Hellkamp A.S.,
- et al.
- Levy W.C.,
- Mozaffarian D.,
- Linker D.T.,
- et al.
- Linde C.,
- Gold M.R.,
- Abraham W.T.,
- et al.