Impact of QRS Duration on Clinical Event Reduction With Cardiac Resynchronization Therapy:  Meta-analysis of Randomized Controlled Trials FREE

Heart failure currently affects approximately 6 million people in the United States and 6.5 million people in Europe.^1- 2 Not only does heart failure contribute to death and poor quality of life, but it also contributes to a significant utilization of resources, with costs related to heart failure estimated to be in excess of $39 billion in 2010 in the United States.¹ Over the past decade, in addition to medical therapy, implantable device therapy has become a cornerstone of the treatment for this disease. Quiz Ref IDCardiac resynchronization therapy (CRT), also known as biventricular pacing, has been shown to improve hemodynamics, promote reverse remodeling, and reduce clinical events including death in patients with prolonged QRS duration on the electrocardiogram.^3- 6

Quiz Ref IDTraditionally, treatment guidelines endorsed by the American College of Cardiology, American Heart Association, Heart Rhythm Society, European Society of Cardiology (ESC), and Heart Failure Society of America (HFSA) recommended CRT in patients with systolic heart failure, New York Heart Association (NYHA) class 3 or 4 symptoms, and a QRS duration of 120 milliseconds or greater.^7- 10 About one-third of patients with systolic heart patients have a QRS duration above this cutoff of 120 milliseconds.^11- 13 Soon after CRT became a popular treatment for heart failure, it was recognized that one-third to one-half of patients receiving CRT based on the guidelines did not respond to this treatment.^14- 15 The recommendation for the QRS cutoff of 120 milliseconds or greater for implantation of CRT devices was based on the entry criteria of 2 major trials.^16- 17 Recently, the HFSA¹⁸ and ESC¹⁹ guidelines for management of heart failure were revised in response to the MADIT-CRT trial.²⁰ These new guidelines introduced a new recommendation for CRT in NYHA 1 and/or NYHA 2 systolic heart failure, but this time with a new QRS cutoff of greater than 150 milliseconds for this population. The new cutoff was based on the subgroup analysis of the MADIT-CRT trial,²⁰ in which patients with a QRS interval shorter than 150 milliseconds had no reduction in heart failure events with CRT. These updated guidelines continued to recommend CRT for patients with NYHA 3 and 4 heart failure with the old QRS cutoff of 120 milliseconds or greater. However, studies using surrogate measures of response (ie, hemodynamics or peak oxygen consumption) suggest that patients with a QRS duration between 120 and 150 milliseconds do not benefit from CRT, regardless of their NYHA functional class.^3,21 In this context, to our knowledge, the impact of the degree of QRS prolongation on the effect of CRT for reducing adverse clinical events (ie, death and hospitalization) has never been analyzed systematically. Therefore, our objective was to determine whether the impact of CRT on clinical end points is affected by the degree of baseline QRS prolongation by performing a meta-analysis of randomized trials testing CRT in heart failure.

Systematic searches were made of MEDLINE, SCOPUS (covering EMBASE), Cochrane Central Register of Controlled Trials, and Cochrane Database of Systematic Reviews databases to retrieve all published randomized controlled trials of CRT that reported clinical events according to baseline QRS duration. The search terms and other search strategies are described in detail for each database in the eAppendix, eTable and eFigure). The results of the literature search are depicted in Figure 1.

To evaluate the efficacy of CRT in relation to QRS duration, we included trials that reported clinical outcomes of subgroups stratified by QRS duration. Studies were excluded if they (1) were not randomized; (2) did not have a non-CRT control group; (3) enabled implantable cardioverter defibrillator (ICD) implantation only in one study arm and not in the other(s) (trials enabling ICD implantation in both arms were eligible); (4) had cross-over study design; (5) did not report the clinical outcomes of interest such as death and hospitalization; and/or (6) reported clinical outcomes without any relation to specific limited QRS ranges.

Data from studies meeting the selection criteria were extracted and verified independently by 2 investigators (I.S. and T.P.C.). In cases where the point estimates and the confidence intervals (CIs) for subgroups were not specifically stated, forest plots were used, if available, to extract this information using electronic calipers. Among the included trials, there were slight differences in the cutoffs used for QRS subgroup reporting. In an attempt to standardize the cutoff, we contacted the corresponding authors of the trials that did not report QRS subgroups with the exact 150-millisecond cutoff (ie, COMPANION¹⁶ reporting with 147 milliseconds, REVERSE²³ with 152 milliseconds, and CARE-HF¹⁷ with 159 milliseconds) and asked for the effect sizes with the 150-millisecond cutoff. However, this information was not provided for any of these trials. Thus, we defined the subgroup with a QRS duration of less than 150 milliseconds in most cases as moderately prolonged and those with a QRS duration of 150 milliseconds or greater in most cases as severely prolonged. One trial¹⁶reported 3 subgroups; the “middle” subgroup of this trial (148-168 milliseconds) was included among our severely prolonged QRS subgroups, since the QRS duration was greater than 150 milliseconds for most patients in the subgroup.

We generated funnel plots according to different QRS subgroups to examine the possibility of publication bias. We supplemented testing for publication bias with the Begg Rank Correlation test.²⁴

Four of the included trials reported hazard ratios (HRs), and 1 trial reported odds ratios (ORs). The values for ORs are similar to those for HRs (ie, instantaneous risk ratio [RR]) when the outcome is uncommon. Given that the outcome in the trial that reported ORs was less than 20%, we were able to combine the ORs from this trial with the HRs from the other trials to obtain a meta-analytic RR.²⁵

Statistical heterogeneity was tested by the Cochran Q statistic and reported as I². Fixed-effect models were used, unless there was evidence of heterogeneity (ie, I² >> 40%) where random-effects models were used. The difference in the meta-analytic effect size in patients with severely vs moderately prolonged QRS intervals was assessed with heterogeneity analysis. A meta-regression analysis was performed to examine the relationship between the QRS duration (ranked according to the degree prolongation among all subgroups) and log-transformed RR. Statistical tests were considered significant if the 2-sided P value was less than .05. Data were analyzed using Comprehensive Meta Analysis software, version 2.2.048 (Biostat Inc, Englewood, New Jersey).

The results of the literature search are shown in Figure 1. Of the 412 results, 70 reports without the exclusion criteria were then subjected to a detailed investigation looking for the parameters of interest (ie, reporting of clinical events according to specific baseline QRS ranges). Accordingly, a total of 5 randomized controlled trials enrolling a total of 5813 patients were included in the meta-analysis.

The characteristics of the included trials are summarized in Table 1. The COMPANION¹⁶ trial had 3 arms (medical therapy vs CRT only vs CRT-ICD). Data from the medical therapy vs CRT only arms are included in this analysis. In REVERSE,²³ all patients received a CRT device, but the left ventricular lead was turned off in the control arms. The control group patients did not receive a CRT device in COMPANION,¹⁶ CARE-HF,¹⁷ MADIT-CRT,²⁰ and RAFT.²² The REVERSE²³ and RAFT²² trials were double-blind trials, whereas the COMPANION,¹⁶ CARE-HF,¹⁷ and MADIT-CRT²⁰ trials were not blinded. However, all of the unblinded trials had blinded end points committees. The composite clinical end point reported in subgroup analyses according to different baseline QRS duration ranges always included all-cause mortality and heart failure hospitalization but varied across trials with respect to other included events, as indicated in Table 1. A total of 3624 patients were in the severely prolonged QRS group (62.3%), and 2189 were in the moderately prolonged QRS group (37.7%).

All 5 trials were analyzed using intention-to-treat principle. In COMPANION,¹⁶ prior to reaching the primary end point, 13% of patients in the medical therapy group withdrew, and 2% in the CRT group (without ICD) withdrew. In REVERSE,²³ 7.3% of patients crossed over from CRT off to CRT on, and 1.4% patients crossed over from CRT on to CRT off at 12 months. In MADIT-CRT,²⁰ 1% did not receive a device in the CRT-ICD arm, and 2.6% did not receive a device in the ICD only arm. In this trial, 12.4% of patients assigned to ICD only were switched to a CRT-ICD device before study end, whereas in the CRT arm, 7.5% of patients crossed over to ICD only. In RAFT,²² 4% of patients crossed over and received CRT in addition to an ICD, and in the CRT-ICD group, 6.0% did not receive CRT. In this trial 0.6% and 1.1% either withdrew or were lost to follow-up in the ICD and CRT-ICD arms, respectively. Dropout or cross-over rates were not presented in the CARE-HF trial.

The baseline characteristics of patients enrolled in the trials are listed in Table 2. Within each arm of the included trials, there were no statistically significant differences with regard to age, sex, ejection fraction, functional class, QRS duration, or medication use.

No evidence of publication bias was detected with the Begg Rank Correlation method (P >> .50). Funnel plots examining publication bias according to the QRS groups are presented in the eFigure.

The impact of CRT on clinical events in patients with severely prolonged QRS is shown in Figure 2. For these patients, there was a statistically significant reduction in risk for composite clinical events in each individual trial with the exception of the middle QRS subgroup of COMPANION¹⁶ (ie, the subgroup with the least severely prolonged QRS among the subgroups with severely prolonged QRS), where there was a statistically insignificant benefit (P = .09). On meta-analysis, patients with severely prolonged QRS randomized to CRT had a 40% risk reduction in clinical events (I² = 32.1%; RR, 0.60[95% CI, 0.53-0.67]) (P < .001 by fixed-effect model). On the contrary, there was no statistically significant benefit for patients with moderately prolonged QRS in any of the individual trials (Figure 3). There was a statistically insignificant benefit in the CARE-HF¹⁷ trial, which had the most prolonged QRS interval within the moderately prolonged QRS subgroups (P = .06). On meta-analysis, there was no significant benefit of CRT for reduction in clinical events in this group of patients (I² = 0%; RR, 0.95 [95% CI, 0.82-1.10]) (P = .49 by fixed-effects model). When directly compared with heterogeneity analysis, the overall effect of CRT on clinical events was significantly different in patients with moderately vs severely prolonged QRS intervals (P < .001).

The relationship between the magnitude of QRS prolongation and the impact of CRT on the risk of composite clinical events assessed by meta-regression analysis is presented in Figure 4. There was a statistically significant relationship between the QRS duration and log RR (slope, −0.07 [95% CI, −0.10 to −0.04; z = −4.60) (P < .001). Accordingly, groups with QRS durations less than 150 milliseconds did not benefit from CRT. Beneficial effects of CRT on reduction of clinical events became evident when cases with QRS intervals of 150 milliseconds or greater were included, and the magnitude of benefit became more prominent with further increases in QRS duration.

The findings of the meta-analysis remained robust to sensitivity analysis (Table 3). When the analysis was limited to NYHA 3 and 4 cases (COMPANION¹⁶ and CARE-HF¹⁷), there was still a highly significant benefit of CRT in patients with severely prolonged QRS and no statistically significant benefit in patients with moderately prolonged QRS. The same was observed in NYHA 1 and 2 cases. Similarly, when the analysis was limited to trials with nearly universal use of ICDs in both arms of trials, statistically significant benefit with CRT was seen only in patients with severely prolonged QRS and not in those with moderately prolonged QRS. When the analysis was limited to trials without background ICD therapy, again the benefit of CRT was observed only in the patients with severely prolonged QRS. In each sensitivity analysis performed using the 1-study-out method, there remained a significant difference in the impact of CRT on clinical events according to degree of QRS prolongation. When analysis was limited to trials uniformly reporting HRs (ie, when REVERSE²³ was left out), there was again benefit in patients with severely prolonged QRS and not in patients with moderately prolonged QRS.

Quiz Ref IDThis meta-analysis shows that while CRT was very effective in reducing adverse clinical events in patients with systolic heart failure and a baseline QRS duration of 150 milliseconds or greater, it did not reduce such events in patients with a QRS interval less than 150 milliseconds. The difference in benefit between the 2 QRS groups was statistically significant (P < .001). These results were consistent among all the randomized trials included in this meta-analysis, regardless of enrollment criteria for functional class.

The lack of benefit of CRT in patients with QRS durations less then 150 milliseconds has been observed in several hemodynamic and echocardiographic studies, as well as in studies using cardiometabolic stress tests and quality of life measures. Auricchio et al³ observed that when the QRS duration was less than 150 milliseconds, biventricular pacing did not improve either the maximum left ventricular pressure derivative or aortic pulse pressure, whereas those with longer QRS intervals had increases in both.³ In a more recent randomized study, Auricchio et al²¹ also showed that peak oxygen consumption as assessed by cardiometabolic stress test did not improve with left ventricular pacing in patients with a QRS duration between 120 and 150 milliseconds. In contrast, both parameters improved significantly in patients with QRS intervals greater than 150 milliseconds. Similarly, distance walked in 6 minutes and quality-of-life score improved only in patients with QRS intervals greater than 150 milliseconds. REVERSE study investigators²⁶ showed that there was no significant reverse remodeling with CRT in patients with moderately prolonged QRS, contrasting with the remarkable reverse remodeling in those with longer QRS durations. Our meta-analysis extends these previous observations of lack of benefit on surrogate measures in patients with QRS interval less than 150 milliseconds to the lack of reduction in clinical events, including death and hospitalizations in such patients, in the setting of randomized controlled clinical trials. On the other hand, it was observed that there was a trend for benefit in the moderately prolonged QRS subgroup (ie, 120-159 milliseconds) in the CARE-HF trial.¹⁷ In this context, it should be pointed out that CARE-HF mandated the presence of at least 2 predefined echocardiographic criteria for mechanical dyssynchrony if baseline QRS was between 120 and 149 milliseconds, unlike the other included trials. Of the 290 patients with QRS intervals between 120 and 159 milliseconds in this trial, only 92 of them had a QRS between 120 and 149 (32%), and the remaining 198 had a QRS of 150 milliseconds or greater (68%).²⁷ Therefore, it is not completely clear whether this trend for benefit was driven by the patients with QRS durations between 150 and 159 milliseconds or by the use of echocardiographic criteria in patients with QRS between 120 and 149 milliseconds or both. One recent nonrandomized study²⁸ suggests that echocardiographic parameters of dyssynchrony (Yu index, radial strain) may help identify patients with moderately prolonged QRS who might respond to CRT.

The initial guidelines advising on the indication for CRT in heart failure were primarily directed by the two trials that reported significant reductions in clinical events in NYHA III and IV patients.^7- 10 These two trials used a QRS duration of >120 milliseconds as the enrollment criterion. The writing committees subsequently endorsed the same cutoff of ≥120 milliseconds in their guidelines with the strongest level of recommendation (ie, Class I: procedure should be performed).^16- 17 However this cutoff set forth in these trials appears to be arbitrary in that other clinical trials have used different QRS cutoffs such as 130 or 150 milliseconds.^29- 31Quiz Ref IDIn contemporary practice, approximately 40% of CRT devices are implanted in patients with a QRS duration <150 milliseconds.³²

Soon after CRT was approved as a treatment for heart failure, it was recognized that one-third to one-half of patients do not respond to these devices implanted according to the current indications.^14- 15 This has led to intense research examining the reasons for non-response and led to the creation of special “non-responder clinics” in some institutions.³³ Predicting response to CRT is complex and is related to both substrate and procedural factors. Sweeney, et al. recently demonstrated that the probability of LV reverse remodeling is linearly related to baseline left ventricular activation time and is <50% with a left ventricular activation time <90 milliseconds (corresponding to a QRS duration of approximately <150 milliseconds according to their regression formula).³⁴ The current meta-analysis of randomized controlled clinical trials that assessed clinical endpoints, along with the previous studies using surrogate outcomes, suggest that a predominant reason for CRT non-response is a suboptimal patient selection criterion for QRS duration.

Very recently treatment guidelines advising on CRT were updated primarily to incorporate the findings of the MADIT-CRT trial and extended the indication for CRT to NYHA I and/or II patients.^18- 19 For these patients a new QRS cutoff of ≥150 milliseconds was advised given the subgroup analysis of the MADIT-CRT trial showing lack of benefit with a QRS<150 milliseconds. However, these guidelines continued to recommend a QRS cutoff of 120 milliseconds for NYHA III and IV patients. Our meta-analysis shows that the lack of benefit in patients with QRS<150 milliseconds is a more pervasive phenomenon and is not limited to only NYHA I and II patients but is also observed in NYHA III and IV patients. It appears that the degree of QRS prolongation is more important than the level of functional impairment for selection of patients for CRT. Modification of the current guidelines that reflect these findings can have important consequences for resource utilization. We think that an individual patient level analysis of existing clinical trials to examine whether a subset of patients with moderately prolonged QRS might benefit from CRT (perhaps offset by another subset with increased risk resulting in a net neutral effect in the moderately prolonged QRS group) will be helpful to further specify the new recommendations.

Quiz Ref IDNot all randomized CRT trials reported clinical events according to different QRS subgroups, and these trials could not be included in this meta-analysis. However, all the long-term and large-scale trials could be included. For example, the meta-analysis could incorporate QRS-specific data for more than 85% of the total number of deaths recorded in all the randomized CRT trials reporting on this outcome.³⁵ Therefore, publication bias with regard to reporting according to QRS ranges is highly unlikely to account for the observed differences.

The composite outcome varied across the included trials. However, despite the differences in the inclusion of other events besides all-cause mortality and heart failure hospitalization to the composite outcome, the RR was always lower in the severely prolonged QRS group of all the trials, and there was no statistically significant reduction in any of the composite outcomes in the moderately prolonged QRS subgroup of any trial.

The exactitude of the 150-millisecond cutoff observed in the current analysis for predicting clinical benefit with CRT is likely to be imperfect for the individual patient. Because we did not have access to individual-level patient data, we used the ranges of QRS durations reported in the publications of clinical trials to determine a cutoff. With this approach, an approximate value of 150 milliseconds emerged as a cutoff, below which clinical events were not reduced by CRT. In this context, Varma³⁶ has recently shown that despite similar QRS durations, patients with left bundle branch block have left ventricular activation times that are on average 36 milliseconds longer than patients with right bundle branch block. Consequently, the QRS duration above which CRT will be beneficial is probably significantly different with different types of conduction abnormalities. Therefore, we believe that a meta-analysis of individual patient-level data of all relevant clinical trials can further refine the QRS cutoffs for different types of conduction abnormalities.

When performing this meta-analysis, we were faced with the problem of dealing with 2 different types of association measures (ie, HR and OR) reported in different trials. We believed that including only the 4 trials that reported HRs and excluding REVERSE²³ reporting ORs would introduce bias and would be less robust. Given the similarities of the 2 measures in many situations, we combined these measures and reported the meta-analytic effect size as RR. We addressed this limitation using sensitivity analysis (where we excluded REVERSE²³), which revealed very similar results. It is noteworthy that the REVERSE trial also had a broad clinical end point, not only including mortality and heart failure hospitalization but also worsened heart failure symptoms or NYHA functional class.

While CRT was very effective in reducing clinical events in patients with systolic heart failure and a baseline QRS duration of 150 milliseconds or greater, it did not reduce such events in patients with QRS intervals less than 150 milliseconds. This finding was observed not only in trials that enrolled patients with NYHA 1 and 2 disease but also in those that enrolled patients with NYHA 3 and 4 disease. These results have implications regarding patient selection for this important treatment technique.