Cause-of-death analysis in patients with cardiac resynchronization therapy with or without a defibrillator: a systematic review and proportional meta-analysis

Abstract

Aims

The additional benefit of a defibrillator in cardiac resynchronization therapy (CRT) patients is a matter of debate. Cause-of-death analysis in a CRT population has been recently proposed as a useful approach to gain insight into this problem. We performed a systematic review and meta-analysis looking at cause of death in studies involving CRT subjects with (CRT-D) or without (CRT-P) a defibrillator.

Methods and results

Literature search performed from inception to 31 March 2016 for relevant studies. Proportional and conventional meta-analyses were performed to obtain and compare causes of death in CRT-D vs. CRT-P patients, including sudden cardiac death (SCD), all-cause mortality, heart failure, cardiovascular, and non-cardiovascular mortalities. The systematic review included a total of 44 studies and 18 874 patients (13 248 receiving CRT-D and 5626 receiving CRT-P), representing 48 504 patient-years of follow-up. CRT-D recipients were younger, more often male, had lower NYHA class, less atrial fibrillation, more ischaemic heart disease and were more often on beta-blockers than those receiving CRT-P. There were an additional 42 deaths per 1000 patient-years in the CRT-P group compared with CRT-D (97 ± 9, 95% CI 79–115 vs. 55 ± 5, 95% CI 44–65, respectively), of which 35.7% were due to SCD (20 ± 2, 95% CI 15–24 vs. 5 ± 1, 95% CI 3–6) and the remaining 64.3% due to non-SCD. Of all deaths reported in CRT-D and CRT-P patients, 9.1% and 20.6% were due to SCD, respectively. The extent of SCD in CRT-P patients significantly increased in studies with higher percentage of males, ischaemic cardiomyopathy and NYHA class 3.

Conclusion

Overall, compared with CRT-D patients, unadjusted mortality rate was almost two-fold higher in CRT-P recipients, with SCD representing one third of the excess mortality. Rate of SCD was significantly higher in certain subgroups (males, ischaemic cardiomyopathy, NYHA class 3), where a CRT-D may be of more pronounced benefit. This deserves further focused investigation.

Introduction

The implantable cardioverter-defibrillator (ICD) is a widely accepted treatment for the prevention of sudden cardiac death (SCD) in heart failure patients.¹ However, the additional benefit of the ICD in patients receiving cardiac resynchronization therapy (CRT) has not been as extensively studied as in patients without CRT.

There has not been any specifically designed randomized study comparing CRT-defibrillator (CRT-D) vs. CRT-pacemaker (CRT-P). The CeRtiTuDe cohort study has recently shown that only a very small proportion of the increased mortality seen in CRT-P patients, compared with CRT-D, was actually related to SCD.³ CRT-D patients have lower all-cause mortality rates, which could be related to differences in patient characteristics and/or the presence of the ICD.⁴ A detailed cause-of-death analysis in a large cohort of patients receiving CRT, with and without an ICD, would allow us to gain a better insight into the relative contribution of SCD in these populations and thus the potential added interest of a defibrillator. This in turn would help select the patients who are more likely to benefit from this therapy.

We therefore performed a systematic review and proportional meta-analysis of the causes of death among CRT recipients with (CRT-D) or without (CRT-P) a defibrillator across multiple studies. Using pooled data on cause-of-death, we evaluated the extent to which the addition of the ICD may contribute to reduced risk of SCD.

Methods

Data sources and study selection

We searched MEDLINE (via PubMED), EMBASE, clinicaltrials.gov and COCHRANE databases (from inception to 31 March 2016) using the following search strings: ‘cardiac resynchronization therapy’ OR ‘CRT-D’ OR ‘CRT-P’ OR ‘biventricular pacemaker’ OR ‘biventricular defibrillator’ OR ‘implantable cardioverter-defibrillator’ AND ‘mode of death’ OR ‘cause of death’ OR ‘sudden cardiac death’ OR ‘sudden death’ OR ‘sudden arrhythmic death’ OR ‘cardiovascular death’. Reference lists of all accessed full-text articles were searched for sources of potentially relevant information and experts in the field were contacted about further potentially eligible studies. Authors of full-text papers were also contacted by email to retrieve additional information when required.

Only longitudinal studies performed in humans and written in English were considered for inclusion. The population, intervention, comparison, and outcome (PICO) approach was used.⁵ The population of interest included patients with guideline indication for CRT, with or without an ICD. The intervention was CRT implantation and the comparison was CRT-D vs. CRT-P.

The primary outcome of interest was SCD, while secondary outcomes included all-cause mortality, heart failure death, cardiovascular mortality and non-cardiovascular mortality, evaluated at the longest follow-up available. SCD was defined as any sudden unexpected death presumed to be of cardiovascular origin and fulfilling at least one of the following criteria³: (i) occurring within 1 h of the initiation of cardiac symptoms in the absence of progressive cardiac deterioration; (ii) occurring in bed during sleep; or (iii) occurring within 24 h after last being seen alive and stable. Progressive heart failure death was defined as any death due to progressive circulatory failure. Cardiovascular death was defined as any death due to a cardiac or vascular cause, including SCD, heart failure death, deaths due to coronary, cerebrovascular, aortic events, and thrombo-embolism.

A meta-analysis of proportions was conducted to obtain average incidence rates per patient-year [and 95% confidence intervals (CI)] of the different endpoints in both groups. In addition, a conventional meta-analysis was also performed to provide a comparison between patients receiving CRT-D vs. CRT-P in studies reporting causes of death in both groups. The majority of meta-analyses aim to establish the effects of interventions by getting a pooled estimate of effect size (for example, relative risk, odds ratio, risk difference). However, meta-analyses can also be useful to get a more precise estimate of disease frequency, such as disease incidence rates and prevalence proportions.

In order to be eligible, studies needed to report information on the direct cause of death of patients receiving CRT. Incidence or number of SCD, the primary endpoint, had to be reported, or provided by the authors, for a study to be included. Registries, observational studies, and randomized trials were considered eligible for analysis. The methods sections of evaluated studies were reviewed to confirm the suitability and composition of the reported endpoint. Studies reporting causes of death of CRT patients but without specification of device type (CRT-P or CRT-D) were excluded unless the authors could provide such information. Studies comprising only patients implanted either with CRT-P or CRT-D but providing information on cause of death were included in the proportional meta-analysis but not the conventional meta-analysis.

Two independent reviewers (SB, RD) screened all abstracts and titles to identify potentially eligible studies. The full text of all potentially eligible studies was then evaluated to determine the eligibility of the study for inclusion in the systematic review.

Data extraction and quality assessment

Data extraction and presentation for the preparation of this manuscript followed the recommendations of the PRISMA group. The following data were extracted for characterizing each patient sample in the selected studies, whenever available: demographics and sample characterization, LV ejection fraction (EF), New York Heart Association (NYHA) class, QRS duration, aetiology (ischaemic or non-ischaemic dilated cardiomyopathy), history of atrial fibrillation, treatment with beta-blockers, angiotensin-converting-enzyme inhibitors or angiotensin type-2 receptor blockers and antiarrhythmic medication, and follow-up duration.

Quality Assessment was performed using the Delphi Consensus criteria for randomized controlled trials and a modified Newcastle–Ottawa Quality Assessment Scale for Cohort Studies by three reviewers (SB, RP, and RD).

Data synthesis and statistical analysis

A random-effects model was used to calculate a pooled estimate of the incidence rate from the combined studies. We summarized the incidence rate of each of the study endpoints at the study level and produced an average incidence rate for each outcome. Comparisons of the two device groups were performed using the raw mean difference of the incidence of SCD and respective 95% CIs. We then estimated the patient-based number needed to treat (NNT) to prevent one SCD based on its cumulative incidence, taking into account a median CRT-D battery life of 5 years⁶. We computed the cumulative incidence of the event from incidence rates using the formula described by Suissa et al.⁷ and then used the cumulative incidence to calculate the patient-based NNT to prevent one SCD per battery-life. For the conventional meta-analysis, the rate ratio (RR) with respective 95% CI was used as measurement of treatment effect.

Sensitivity analyses were performed to assess potential differences in mortality rates between CRT-D and CRT-P depending on study design (randomized vs. non-randomized; single vs. multicentre), date of publication (studies published before vs. from 2008) and in specific scenarios (primary prevention only; prevalence of ischaemic cardiomyopathy >50% vs. ≤50%; percentage of patients in NYHA class >2 ≥75% vs. <75%; percentage of patients on beta-blockers >75% vs. ≤75%; percentage of patients on ACEI/ARA-II >90% vs. ≤90%; these cut-offs took into consideration the means in the study group).

For the conventional meta-analysis, statistical heterogeneity on each outcome of interest was quantified using the I² statistic, which describes the percentage of total variation across studies due to true heterogeneity rather than chance. Values of <25%, 25–50%, and >50% are by convention classified as low, moderate, and high degrees of heterogeneity, respectively.⁸

Funnel plots and meta-regression analyses were obtained using Comprehensive Meta-Analysis software (Version 2). Funnel plots were used for evaluating the presence of publication bias and traced for comparisons including more than 10 studies. A meta-regression (using the Unrestricted ML method) was performed for assessing the possible association of moderator variables with the effect estimate or incidence rates.

Results

Search results and patients’ characteristics

A total of 656 entries were obtained from the initial literature search. Two-hundred and thirty-seven were retrieved for analysis of titles and abstracts and 49 of these were selected for further analysis of the full-length article. Eighteen were considered eligible for inclusion.^3,9–25 A further 26 studies were retrieved after reviewing their reference lists and following manual searches.^2,26–50 Eleven additional studies containing data on cause of death were found, yet they did not fulfil inclusion criteria (details in Supplementary material online, Supplementary file). The systematic review finally included a total of 44 studies. Figure 1 illustrates the study selection process.

The design of selected trials and baseline data are summarized in Table 1 and Supplementary material online, Table S1. Sixteen studies were randomized controlled trials,^{2,10,26,29,31,33,36,37,39,40,45,47,49,51,52} although randomization for CRT-D vs. CRT-P was only performed in one.² The remaining studies were observational. Twenty studies were multi-centre.^{2,3,9,10,26,29–33,36,37,39,40,45,47,49,52,53} Quality assessment of the included studies is shown in Supplementary material online, Table S2. All but one² of the randomized controlled studies had <6 Delphi criteria and 13 cohort studies had a Newcastle-Ottawa score of ≥7.

The final population for the proportional meta-analysis included 18 874 patients (13 248 receiving CRT-D and 5626 receiving CRT-P), representing 48 504 patient-years of follow-up: 33 928 in patients receiving CRT-D and 14 576 in those receiving CRT-P. The conventional meta-analysis included 8143 patients (4947 receiving CRT-D and 3196 receiving CRT-P) with 20 775 patient-years of follow-up: 12 556 in CRT-D patients and 8219 in CRT-P patients.

The CRT-D and CRT-P groups had significant differences in characteristics (Tables 1and2). Patients receiving CRT-D had a mean age in their 60s in all studies, while the mean age of CRT-P patients was in their 70s in eight studies.^{3,14,27,30,33,37,38,50} Overall, those receiving CRT-D were younger (65 years vs. 68.2), more often males (80.3% vs. 72%), had lower NYHA class (60% in NYHA class 2 vs. 88.6%), lower prevalence of atrial fibrillation (21% vs. 24.6%), higher prevalence of ischaemic heart disease (55.3% vs. 45.5%) and were on beta-blockers (78.2% vs. 63.3%) and class III anti-arrhythmic drugs (23.5% vs. 15.2%) more often than those receiving CRT-P (Table 2).

Annual incidence rates of specific causes of death

The pooled data of studies revealed that CRT-D patients had significantly lower incidence rates of all-cause mortality (55 ± 5 per 1000 patient-years, 95%CI 44-65, vs. 97 ± 9, 95%CI 79-115, P < 0.001), SCD (5 ± 1, 95% CI 3–6 vs. 20 ± 2, 95% CI 15–24, P < 0.001), progressive heart failure death (27 ± 3, 95% CI 21–33 vs. 41 ± 5, 95% CI 30–51, P < 0.001), and non-cardiovascular death (13 ± 1, 95% CI 10–15 vs. 20 ± 3, 95% CI 15–25, P < 0.001) compared with CRT-P recipients. Of all deaths reported in CRT-D patients, 9.1% were SCD, 49.1% due to progressive heart failure, 23.6% represented non-cardiovascular mortality and the remaining consisted of deaths due to cardiovascular causes other than SCD and progressive heart failure death. The distribution of causes of death was different amongst CRT-P patients: 20.6% were SCD, 42.3% due to progressive heart failure, 20.6% represented non-cardiovascular mortality, while the remaining consisted of deaths due to other cardiovascular causes. In the proportional meta-analysis, there were an additional 42 deaths per 1000 patient-years in the CRT-P group compared with CRT-D, of which 35.7% were due to SCD and the remaining 64.3% due to non-SCD. The forest-plots in Figures 2–5 illustrate the incidence rates of all-cause mortality and SCD (per patient-years) in CRT-D and CRT-P patients across studies included in the proportional meta-analysis.

The pooled data of studies included in the conventional meta-analysis revealed that CRT-D patients had a significantly lower risk of all-cause mortality (RR = 0.65, 95% CI 0.52–0.80; P < 0.001; I² = 65%) (Supplementary material online, Figure S1), SCD (RR = 0.34, 95% CI 0.24–0.47; P < 0.001; I² = 12%) (Supplementary material online, Figure S2) and cardiovascular mortality (RR = 0.60, 95% CI 0.44–0.81; P = 0.001; I² = 66%) compared with those receiving CRT-P. There was also a trend towards lower risk of non-cardiovascular mortality (RR = 0.74, 95% CI 0.54–1.02; P = 0.063; I² = 25%). No difference was observed regarding the risk of progressive heart failure mortality (RR = 0.83, 95% CI 0.56–1.23; P = 0.35; I² = 75%). The observed I² values revealed mild heterogeneity for the analysis on SCD, but marked heterogeneity for the analysis on the risk of all-cause mortality, cardiovascular, and progressive heart failure mortality. Funnel plots for the different endpoints revealed the presence of a publication bias.

Sensitivity analyses and meta-regression

Sensitivity analyses and meta-regression were performed in our proportional meta-analysis for the endpoint of SCD, while meta-regression was also performed in the conventional meta-analysis. Table 3 shows the results of our sensitivity analyses. As shown, the incidence rate of SCD in studies where less than 75% of patients were in NYHA class >2 was 7 per 1000 patient-years in CRT-P patients vs. 4 in CRT-D patients.

The incidence rate of SCD was higher in randomized trials compared with observational studies for both CRT-D and CRT-P patients. For CRT-P patients, but not CRT-D, SCD was more frequent in multicentre vs. single-centre studies, and in studies where percentage of patients with ischaemic cardiomyopathy was higher than 50% vs. <50%. In both device groups, SCD was much more frequent in studies published before 2008 compared with studies published after 2008. The incidence rate of SCD was similar in CRT-D studies including PP patients only vs. studies including both PP and SP patients. Table 3 describes these results.

The meta-regression (Supplementary material online, Table S3) revealed that the risk of SCD in CRT-P patients increased in studies with higher prevalence of males, ischaemic cardiomyopathy and NYHA class 3. The incidence rate of SCD decreased in studies with older patients, higher prevalence of atrial fibrillation and higher LV ejection fraction. Likewise, SCD decreased in more recent studies compared with older studies, presumably a result of improved heart failure management. In our conventional meta-analysis, the benefit of the ICD in decreasing the risk of SCD was more pronounced with increasing QRS duration (Supplementary material online, Table S4). No other associations were seen in both meta-analyses.

Discussion

This study has shown that, compared with CRT-D, CRT-P patients have an almost two-fold higher unadjusted risk of all-cause mortality; SCD accounts for roughly a third of the excess mortality, while non-SCD and non-cardiovascular mortality account for two-thirds. Given the significant differences in population characteristics between the two groups and the significant competing 8.4% annual non-SCD risk in CRT-P patients, the extent of mortality reduction contributed by the presence of the ICD is difficult to infer. However, our results suggest that the primary prevention ICD may be more cost-effective in young male CRT patients with ischemic cardiomyopathy who are on stable NYHA class III despite optimal medical treatment and have few comorbidities, while the cost-effectiveness ratio of routine CRT-D implantation (compared with CRT-P) in elderly patients or those with mild heart failure may be less attractive. Considering that CRT-D associates with a higher risk of complications^44,54,55 and a significantly higher cost compared with CRT-P, our results reinforce the importance of selecting the right patient for the procedure and suggest that providing every CRT candidate with an ICD is unlikely to be clinically beneficial or cost-effective.

It is also useful to look at the presumptive magnitude of mortality benefit which could be conferred by the ICD. Assuming a median CRT-D battery life of 5 years⁶ and an incidence rate of 5 SCD per 1000 patient-years in CRT-D and 20 in CRT-P, after computing the cumulative incidence of the event we calculated a patient-based NNT of 13.5 per battery life, or an annual NNT of 67.5. These represent the number of CRT patients who would need an ICD for one SCD to be prevented during the battery-life or a 12-month period, respectively. However, it needs to be borne in mind that this represents a ‘best case scenario’ where every additional SCD noted in the CRT-P population would be prevented by the presence of an ICD and, furthermore, this gain would not be offset by the competing risk of non-SCD. However, the ICD does not prevent all cases of SCD. In SCD-HeFT, the ICD was able to prevent approximately 60% of all SCDs compared with placebo,¹ a similar reduction to what we have seen. Furthermore, almost one tenth of all deaths in CRT-D patients are still due to SCD.^1,19,61 Therefore, it seems logical that, to be cost-effective, candidates for CRT-D have to be carefully chosen where there is a significant additional risk from SCD and a lower competing risk of non-SCD. The results from our cumulative data suggest that this could be the case in selected subgroups such as younger patients, males, those with ischaemic cardiomyopathy and in stable NYHA class 3. This can be informative in planning targeted RCTs to evaluate this question further.

The differences in patient characteristics explain the higher risk of non-SCD among CRT-P subjects. Patients receiving CRT-P are in general older, more often of female sex, have more advanced heart failure and comorbidity and are less often on beta-blockers and class III antiarrhythmic drugs. In the CeRtiTuDe cohort study, the higher all-cause mortality rate in CRT-P patients (mean age 75 years) was almost entirely due to much higher number of progressive heart failure related- or non-cardiac deaths, while SCD was only slightly more frequent.³ The CeRtiTuDe findings illustrate that the benefit of the ICD may dramatically decrease with increasing number of comorbidities to a point where patients may cease to benefit from it.^62,63 It is noteworthy that the mean pooled age of our CRT-P group was relatively low when compared with more recent studies such as CeRtiTuDe, possibly leading to a greater difference in observed SCD incidence rates.³ The extent of the benefit of adding the ICD will be lower as the population receiving CRT gets older. Moreover, male patients and those with ischaemic cardiomyopathy have been shown to obtain a more pronounced benefit from the ICD compared with females^4,65 and those with non-ischaemic cardiomyopathy, respectively.⁴ The lower use of beta-blockers and class III antiarrhythmic drugs amongst CRT-P patients may also help explain their higher risk of SCD. Differences in the use of these drugs may be a reflection of the higher prevalence of sustained or non-sustained arrhythmias amongst CRT-D patients (in fact, as our CRT-P patients did not have a secondary prevention indication for the ICD, we can assume that poorly tolerated sustained VT had not occurred in this group) and the lower age and degree of comorbidity of the CRT-D group and therefore better tolerance to these drugs.

Response to CRT leads not only to improved LV systolic function and heart failure symptoms but also to a decrease in the risk of potentially life-threatening arrhythmias.^56,59 CRT alone decreases the risk of SCD even in the absence of an ICD.¹⁰ However, approximately one third of all deaths among CRT-P patients in CARE-HF were sudden (equivalent to 7% of all CRT patients),⁶⁰ a rate similar to that observed among CRT-P patients in COMPANION.² As such, the addition of the ICD may in theory provide incremental protection. Nevertheless, this comes at the expense of an increased risk of device-related complications^44,54,55 and significantly higher cost; therefore this question merits careful scrutiny.

In summary, given the marked differences in characteristics between patients receiving CRT-D and CRT-P, and the fact that both devices have slightly different objectives, with the former focusing on both quality and duration of life while the latter focuses especially (but not exclusively) on quality of life, some arguments can be put forth. Patients for CRT-D have to be carefully selected, taking into account the competing risk of non-SCD and using well described risk stratification scores to determine the probability of obtaining a benefit from the ICD.^65,66 The decision-making has to be individualized, with role for patient empowerment and informed-decision making, discussing the specific objectives of the two types of devices. A thorough discussion with the patient and his/her family on the benefits and risks of each device would be useful, explaining that the addition of the ICD will allow a small number of patients to live longer at the expense of potentially increased comorbidity and a higher likelihood of death due to heart failure, infection or malignancy. In their Editorial on the CeRtiTuDe cohort study, Upadhyay and Singh⁶⁷ emphasized the need for an individualized, patient-centric decision-making model, and a prospectively constructed risk scoring system to identify patients more likely to benefit from the addition of the defibrillator. An experienced physician should consider not only the results of their mortality and SCD risk stratification but also the expectations of the patient, the risk of device-related complications and the cost-effectiveness of the proposed treatment.

Limitations

Several limitations should be taken into account when interpreting the results of our proportional and conventional meta-analyses. Firstly, the overall study quality is limited by the fact that, with the exception of the COMPANION trial, no study randomized patients for CRT-D vs. CRT-P. This limitation cannot be overcome and should be accepted. The very recently published Defibrillator Implantation in Patients with Nonischemic Systolic Heart Failure (DANISH) randomized controlled trial, which has not been included in the present meta-analysis, suggested that the lack of benefit of the ICD in patients with non-ischaemic cardiomyopathy was independent of CRT status.⁶⁸

Secondly, the cause-of-death definition and classification across studies were based on criteria specified by each group of authors. Although the definition of SCD was relatively uniform between studies, cause-of-death analysis is always a challenging task even when data is prospectively adjudicated, such as in the CeRtiTuDe cohort study. Most studies included in this meta-analysis did not use an adjudication process with predefined definitions of the different causes of death. However, the extent to which those differences in the adjudication process may significantly alter the main message of the paper is unclear. The SCD rate in CRT-P patients and therefore the potential benefit of the ICD was higher in randomized and multicentre studies, partly because of the more rigorous cause-of-death collection and adjudication which lead to fewer SCD being misspecified. Furthermore, it should be kept in mind that, although the defibrillator aims to prevent sudden arrhythmic death, in some cases it may prevent death due to heart failure of infection by treating a life-threatening arrhythmia and allowing time for the underlying condition to be overcome.

Thirdly, when interpreting the outcomes of CRT patients and the results of our cause-of-death analyses, the reader should always take into account the very significant differences in baseline characteristics between CRT-D and CRT-P patients. It is likely that many unmeasured factors may differ as well.

Finally, the inclusion of both primary and secondary prevention patients adds some heterogeneity to the analysis. However, (i) none of the CRT-P patients included in the meta-analysis had a secondary prevention indication for an ICD; (ii) only three studies on CRT-D recipients included a majority (>50%) of SP patients; (iii) there was no difference in SCD risk between studies including PP patients only and those including both PP and SP.

Conclusion

SCD represents approximately twenty per cent of all deaths in patients receiving CRT-P and one third of all excess mortality in this group compared with CRT-D patients. The cost-effectiveness of CRT-D implantation in patients otherwise eligible for CRT-P is highly dependent on the baseline risk of SCD, competing risks of non-sudden mortality and device battery longevity. Amongst patients with CRT indication, those of male gender, with ischaemic cardiomyopathy and in stable NYHA class 3 may be more likely to benefit from the addition of the ICD, while in some subgroups of patients, such as those of advanced age and with mild heart failure, the cost-effectiveness ratio may be much less favourable, emphasizing the need to evolve a tailored strategy for device selection in individual patients.

Supplementary material

Supplementary material is available at Europace online.

Conflict of interest: none declared.

References