Heart failure (HF) is an increasing worldwide health problem and is responsible for an increasing health care burden in the United States (1). It accounted for more than 1 million hospital discharges in 2006 and was associated with an overall annual cost of $39.2 billion in 2010 (1). Despite the success of pharmacological therapy, the rate of repeat hospital stays for HF has not declined during the past decade (1- 2). During the past 10 years, randomized cardiac resynchronization therapy (CRT) trials have consistently shown that treatment with CRT is associated with a significant reduction in the risk of morbidity and mortality among patients with ischemic or nonischemic cardiomyopathy, advanced HF symptoms, and QRS ≥120 ms (3- 10). Recently, MADIT-CRT (Multicenter Automatic Defibrillator Implantation Trial with Cardiac Resynchronization Therapy) extended these findings and showed a significant 34% reduction in risk of the combined endpoint of a first heart failure event (HFE) or death with cardiac resynchronization therapy-defibrillator (CRT-D) as compared with defibrillator-only therapy in patients with milder HF symptoms (11). Similar to the MADIT-CRT trial, prior clinical trials of CRT have been limited mainly to an assessment of the combined endpoint of a first HFE or death (10- 13). Thus, it remains unknown whether the benefit of CRT is maintained after the occurrence of a first post-implantation HFE. Evaluation of recurring HFEs in clinical trials has important clinical, quality of life, and cost-effectiveness considerations that cannot be derived from an evaluation of the risk associated with a first event endpoint. Accordingly, the assessment of the benefit of CRT-D for the prevention of recurring HFEs was a pre-specified substudy of the MADIT-CRT trial.

The design and primary results of the MADIT-CRT trial have been recently published (11). Briefly, the MADIT-CRT trial was designed to determine whether CRT-D would reduce the risk of death or nonfatal HFEs in patients with mild cardiac symptoms, a reduced ejection fraction, and wide QRS complex, when compared with implantable cardioverter-defibrillator (ICD) therapy. A total of 1,820 patients were enrolled at 110 hospital centers in North America and Europe and randomly assigned in a 3:2 ratio to receive either CRT-D or ICD. Patients of either sex who were at least 21 years of age were enrolled in the study if they had ischemic cardiomyopathy (New York Heart Association [NYHA] functional class I or II) or nonischemic cardiomyopathy (NYHA functional class II only), sinus rhythm, an ejection fraction ≤0.30, and prolonged intraventricular conduction with a QRS duration ≥130 ms. All eligible subjects met the guideline indication for ICD therapy (12). Patients were excluded from enrollment for a variety of reasons as previously reported (11). The present study comprises all 1,820 patients enrolled in the MADIT-CRT trial.

The protocol was approved by the institutional review board at each participating organization, and each patient provided written informed consent before enrollment.

The MADIT-CRT trial was carried out from December 22, 2004, through June 22, 2009. After the study was stopped on the recommendation of the data and safety monitoring board, complete data collection and adjudication of HFE and mortality endpoints was continued throughout 2009. Thus, the present study provides extended follow-up data for all MADIT-CRT participants through December 31, 2009. The incremental data provided during the post-trial period, representing a 15% increase in follow-up times, are summarized in (6).

Data regarding cross-overs between the allocated treatment arms were recorded for all study participants throughout follow-up and are presented in (6). Among the 731 patients randomized to ICD-only therapy, 596 (82%) were treated with an ICD throughout the follow-up period; and among the 1,089 study participants who were allocated to CRT-D therapy, 959 (88%) were treated with the same device throughout follow-up. Notably, among the 186 patients assigned to ICD-only therapy who had a first HFE, 68 (37%) crossed over to the CRT-D arm after the occurrence of a first HFE (in 4.8 months, on average), whereas among patients allocated to CRT-D therapy, a change in treatment arm occurred most frequently before the occurrence of a first HFE (8% of all patients allocated to CRT-D therapy crossed over to ICD-only therapy before the occurrence of a first HFE, as compared with only 1% after a first HFE).

The MADIT-CRT protocol did not call for evaluating a potential differential effect of CRT-D with baseline QRS morphology. However, after publication of the primary report (11), analyses were done comparing left bundle branch block (LBBB) patients—comprising 70% of the study patients—with non-LBBB patients (30%; including 13% with right bundle branch block and 17% with an intraventricular conduction defect), revealing evidence that the CRT-D benefit of reducing risk of the primary endpoint seemed to be largely or even entirely in the LBBB subgroup (13). Consequently, on September 16, 2010, the Food and Drug Administration extended approval of the use of the CRT-D only in the LBBB subgroup of the MADIT-CRT–defined population (14). Each analysis in the present study allowed for possible differential effects by QRS morphology—specifically, LBBB or non-LBBB at baseline—consistent with this finding and recommendation.

The primary endpoint of the present study was the occurrence of recurring HFEs after a first post-enrollment HFE, hereafter referred to as subsequent HFEs. The effects of first and subsequent HFEs on the risk of subsequent all-cause mortality were also assessed. The diagnosis of an HFE, made by physicians aware of the implanted devices, required signs and symptoms consistent with congestive HF that was responsive to intravenous decongestive therapy on an outpatient basis or an augmented decongestive regimen with oral or parenteral medications during an in-hospital stay. Adjudication of the endpoints was carried out by an independent HFE committee blinded to device implantation information. Data for patients who experienced subsequent HFEs within 2 weeks of a prior HFE were further reviewed to distinguish ongoing HFEs (defined as events that occurred within 2 weeks of a prior HFE in which the patient continued to experience HF symptoms at hospital discharge or after outpatient intravenous diuretic therapy) from new HFEs (defined as events occurring at any time after a prior HFE in which the patient was considered to be fully stabilized at hospital discharge or after outpatient intravenous therapy). Forty-seven subsequent HFEs (15% in the ICD-only arm and 14% in the CRT-D arm) were identified with these definitions as ongoing HFEs and therefore were combined with the prior event. In the analyses of all-cause mortality, 14 patients (8 in the ICD-only arm and 6 in the CRT-D arm) who died within 2 weeks of device deactivation due to terminal HF were categorized as being allocated at the time of death to the device that was employed before deactivation.

Because risk of HFEs increased measurably after a first HFE (due to identification of patients with high on-going risk and/or changes due to occurrence of the event), the risk for subsequent HFEs (among patients who experienced a first HFE) was analyzed separately from the first HFE endpoint. No further distinctions in risk for second, third, or additional HFEs were made, because there was lesser evidence of distinctions among them and data become sparse with increasing HFE count. Deaths subsequent to a first HFE were also analyzed.

Because many patients with an ICD-only had CRT capability added to their devices after experiencing a first HFE, intention-to-treat (ITT) analyses of events subsequent to a first HFE have limited interpretability. Therefore, primary analyses in this study were carried out on an efficacy basis, comparing risks of events occurring while specific devices were active. Efficacy analyses were further categorized as full (incorporating complete follow-up data for each patient after cross-over between devices) and modified (with follow-up censored upon the first occurrence of cross-over between device-types or transition to no device). Daily records were kept for each patient as to whether an ICD-only was active, a CRT-D was active, or no device was active (e.g., before any implantation, during temporary device inactivation, or after device explantation). Follow-up for each patient was allocated into these 3 categories, with risk for a first HFE evaluated from the time of enrollment and risk for subsequent HFEs evaluated from the time of the first HFE (for a more detailed explanation of the follow-up methodology that was employed in the efficacy analyses, see Figure 1). Secondary analyses were carried out on an ITT basis, comparing devices by original treatment allocation, regardless of cross-over during follow-up.

For first HFEs, Cox proportional hazards regression modeling was used, tracking active-device types as time-dependent covariates; for subsequent HFEs, Andersen-Gill proportional-intensity regression modeling was similarly used (15) (this separation of risk of a first event and risk of later events is similar to what is done in a modified conditional Prentice, Williams, and Peterson analysis [16] but allows for different contributions from covariates to the respective risk analyses). For subsequent HFEs, time of first HFE was included as a covariate. All models included, besides device type, baseline LBBB status and its interaction with device type. Twenty-seven baseline covariates with <5% missing data were selected as potential risk factors (brain natriuretic peptide, a known risk factor for HFEs, was the only covariate that was omitted due to substantial missing data). Missing baseline values for covariates were imputed by linear or logistic regression (so as not to omit these patients from further analyses), and then all variables were dichotomized at pre-specified cutpoints. The list was reduced with best-subsets regression methods, separately for first HFEs and for subsequent HFEs, and the union of the 2 resulting sets of covariates was used as a common set of risk factors in regression analyses of both first and subsequent HFEs. Interaction terms between device types and each selected covariate were later considered, including interaction by sex and by QRS, because these interactions had been identified and reported earlier (11), before consideration of LBBB status in regression modeling. Because follow-up with no active device was limited (1.4% of the total), and with few events occurring during it (14 HFEs [2.3% of total HFEs] and 20 deaths [12.5%]), no evaluations of relative risk of ICD-only to no active device are reported.

Counts of events/100 patient-years of risk were computed, both for first and for subsequent HFEs. These are crude composite descriptive measures of risk, ignoring variation in risk across patients and ignoring the competing risk of mortality; no statistical tests were undertaken.

When no time-dependent factors—such as active device type—were employed, Kaplan-Meier survival curves for the first of HFE or death (but not for first HFE alone due to the competing risk of mortality) were constructed but limited to time from implantation or from first HFE, with censoring at any device change or explantation. Kaplan-Meier curves for death were constructed, from time of enrollment, and for the subset of patients with a first HFE from the time of that HFE and for the further subset with a second HFE from the time of that HFE.

The statistical software used for the analyses was SAS (version 9.2, SAS, Cary, North Carolina). A 2-sided p value <0.05 was used for declaring statistical significance.

The baseline clinical, echocardiographic, and medication characteristics of MADIT-CRT patients appear in (Table 1) of the primary publication (11). (Table 1) herein includes those characteristics considered as potential covariates for the regression analyses, with counts of those experiencing 0, 1, and ≥2 HFEs during follow-up. Patients who experienced HF during follow-up displayed a higher frequency of baseline clinical risk factors as compared with those who did not experience HFEs, including older age, diabetes mellitus, renal dysfunction, advanced HF class >3 months before enrollment, shorter distance walked in a 6-min test, increased frequency of history of atrial fibrillation before enrollment, and higher levels of brain natriuretic peptide. In addition, patients who experienced multiple HFEs displayed more advanced baseline echocardiographic parameters as compared with those who experienced single or no events, including significantly higher left ventricular and atrial volumes and a lower ejection fraction (Table 1).

Of the 1,820 study patients, 347 (19%) had a first HFE during an average follow-up of 2.6 years. After the occurrence of a first HFE, the rate of subsequent HFEs was substantially increased. Thus, of the 347 patients who experienced a first HFE, 142 (41%) had between 1 and 9 additional subsequent HFEs during an average of 1.5 years after the first event (6). Consistent with these findings, the cumulative probability of HF and death was substantially higher immediately after the occurrence of a first HFE, regardless of the type of active device-type at the time of the event ((Figure 2)A to Figure 2C).

Full efficacy analysis (i.e., incorporating complete active device follow-up data) showed that in the total study population active treatment with CRT-D was associated with a significant 46% (p < 0.001) reduction in the risk of a first HFE, as compared with active ICD-only therapy (hazard ratio [HR]: 0.54, 95% confidence interval [CI]: 0.44 to 0.67, p < 0.001) (Table 2) and with a similar magnitude of risk reduction after the occurrence of a first event (38% risk reduction [i.e., HR: 0.62, 95% CI: 0.45 to 0.85, p = 0.003]) (Table 2). Notably, despite a similar magnitude of risk reduction associated with CRT-D therapy before and subsequent to a first HFE, the absolute reduction in event rates (per 100 patient-years) with CRT-D therapy was more pronounced after the occurrence of a first HFE (Figure 3).

Interaction-term analysis, with full efficacy data, showed that the benefit of CRT-D therapy for the prevention of both first and subsequent HFEs was pronounced among patients who had LBBB at enrollment (first: HR: 0.38, 95% CI: 0.29 to 0.49, p < 0.001; and subsequent: HR: 0.50, 95% CI: 0.33 to 0.76, p = 0.001) and nonexistent among non-LBBB patients (first: HR: 1.12, 95% CI: 0.77 to 1.64, p = 0.55; and subsequent: HR: 0.99, 95% CI: 0.58 to 1.69, p = 0.96; p values for interaction: p <0.001 and p = 0.06, respectively) (Table 2). All differences between HRs for first and subsequent HFEs were nonsignificant, and no other substantial differences by baseline covariates were found, once the LBBB/non-LBBB differences were allowed for in the regression models. Specifically, with LBBB status in the analyses, the earlier report (11) of differential effects of CRT-D therapy relative to ICD-only by QRS width was no longer evident, and that by sex was greatly diminished (p value for a differential treatment effect by sex for the endpoints of first and subsequent HFEs; p = 0.07 and p = 0.73, respectively).

Consistent with these findings, active treatment with CRT-D was associated with approximately one-half the rate for both first and subsequent HFEs, as compared with ICD-only therapy among patients with LBBB, but had a less-pronounced effect on the rate of subsequent HFEs among non-LBBB patients (Figure 3).

A modified efficacy analysis (Table 2), in which follow-up during active device therapy was censored at the time of the first cross-over, yielded results virtually identical to those obtained from the full efficacy analyses. This analysis was not affected by any cross-over activity after a first HFE, further suggesting that the benefit of CRT-D therapy for the reduction of subsequent HFEs is maintained after the occurrence of a first HFE.

Intention-to-treat-analysis (i.e., on the basis of original treatment allocation) showed results similar to those obtained from the efficacy analyses when the endpoint of a first HFE was assessed (Table 2). However, randomization to CRT-D therapy was not associated with a statistically significant reduction in the risk of subsequent HFEs, as compared with randomization to ICD-only therapy (Table 2), suggesting that assignment of the relatively large proportion of ICD-only patients who crossed over to CRT-D therapy after the occurrence of a first HFE (37%) to the original ICD-only randomization group attenuated the statistical significance of the association between active CRT-D therapy and the risk of subsequent HFEs.

During follow-up, 160 patients died—of whom 50 were treated with an ICD at the time of death, 90 with CRT-D, and 20 had the device explanted/deactivated before death for a reason that was not known to be related to terminal HF (6). Multivariate Cox proportional hazards regression modeling showed that active treatment with CRT-D was not associated with a significant reduction in the risk of death, as compared with ICD-only therapy in the total study population (HR: 0.95, 95% CI: 0.67 to 1.34). However, risk reduction associated with CRT-D therapy was significantly greater among patients with LBBB (HR: 0.75, 95% CI: 0.49 to 1.14) than among non-LBBB patients (HR: 1.65, 95% CI: 0.87 to 3.13; p value for treatment × LBBB interaction: p = 0.03).

The occurrences of first and second HFEs were the most powerful predictors of mortality in the MADIT-CRT population. Among all study patients, the occurrence of a first HFE was associated with a 7-fold (p < 0.001) increase in the risk of subsequent all-cause mortality; and the development of a second HFE was associated with a 2.6-fold (p < 0.001) further risk increase, resulting in a nearly 19-fold increase in the risk of death as compared with mortality risk before the occurrence of a first HFE (Table 3). The pronounced effects of both first and second HFEs on the risk of subsequent mortality were evident among patients treated with either active device-type (Table 3). Results regarding the effect of a first and second HFE on the risk of death were consistent when device-type was assessed on an ITT-basis (6). Additional factors in the multivariate analysis had a less pronounced effect on the risk of all-cause mortality, including distance walked in a 6-min test of <400 m (HR: 1.89, 95% CI: 1.26 to 2.82), NYHA functional class ≥III more than 3 months before enrollment (HR: 1.80, 95% CI: 1.99 to 2.72), ischemic etiology of cardiomyopathy (HR: 1.64, 95% CI: 1.13 to 2.37), and serum creatinine ≥1.4 mg/dl (HR: 1.58, 95% CI: 1.14 to 2.20).

Consistent with the findings from the multivariate models regarding mortality risk associated with the development of HF, Kaplan-Meier survival analysis showed that the cumulative probability of all-cause mortality was 5% 2 years after enrollment among all study patients, 29% 2 years after the occurrence of a first HFE among those with a first HFE, and 46% 2 years after the occurrence of a second HFE among that subgroup (Figure 4).

The MADIT-CRT trial showed that CRT-D therapy is associated with a significant reduction in the risk of a first HFE or all-cause mortality in asymptomatic or mildly symptomatic patients with left ventricular dysfunction (11). The present study is the first to assess the benefit of CRT-D for the prevention of recurring HFEs in a clinical trial and, therefore, extends the observations of the original report. We have shown that treatment with CRT-D was associated with a significant reduction in the risk of recurring episodes of HF among study patients even after the occurrence of a first post-enrollment HFE. Recurring HF was shown to be the most powerful predictor of subsequent mortality in the MADIT-CRT population, further stressing the important long-term implications of identifying measures for the prevention of repeated HF episodes.

Heart failure is a common chronic disease marked by frequent exacerbations often resulting in hospital stay and death (1). Readmission for HF occurs in 20% of patients within 30 days of discharge in those over age 65 years, and this might exceed 50% at 6 months (17- 18). Despite current HF therapies, recent data (2) suggest an increase in the rate of repeat hospital stays after HF. The high rate of repeat hospital stays is a major contributor to the estimated $39.2 billion in cost of HF care in the United States for 2010 (1). Therefore, the identification of therapies that reduce the risk of recurring HF has incremental health, economic, and quality-of-life implications that extend beyond the traditional first event endpoint. The present study shows a very high rate of repeated HF episodes (41%) after the occurrence of a first post-enrollment HFE even in the MADIT-CRT population, which comprised asymptomatic or mildly symptomatic patients. Accordingly, the absolute reduction in the rate of recurring events with CRT-D therapy in the trial was substantially higher than the corresponding reduction in the rate of first events (Figure 3), even though the magnitude of relative risk reduction associated with the device before and after the occurrence of a first HFE was similar.

It should also be noted that in the MADIT-II (Multicenter Automatic Defibrillator Implantation Trial) the significant survival benefit of the ICD was shown to be associated with a significant increase in the risk of recurrent HFEs (19). Furthermore, we have recently shown in an extended 8-year follow-up of the MADIT-II population that the long-term survival benefit of the ICD is attenuated among patients who develop HF after device implantation (20). These data further stress the importance of timely intervention with CRT for the reduction in the risk of recurrent HFEs and death in appropriately selected candidates for primary prevention with an ICD.

Response to CRT is usually assessed with a variety of clinical and echocardiographic measures, which usually includes the occurrence of a first HFE (2- 4,21). Prior studies suggested that approximately one-third of the patients who receive CRT in clinical trials can be considered nonresponders to treatment with the device, due to the occurrence of symptomatic HF after implantation. Currently, however, there are limited data regarding the subsequent outcome of this high-risk patient subset. In the present study we employed the Andersen-Gill proportional-intensity regression model to assess the clinical course of patients enrolled in the MADIT-CRT trial who experienced a first HFE. Findings from this analysis show that treatment with CRT-D was associated with a significant 38% reduction (50% in LBBB patients) in the risk of repeated HF episodes subsequent to a first event. Thus, our data suggest that patients who were considered in some prior studies as nonresponders (due to the occurrence of a HFE after CRT implantation) might continue to derive a pronounced clinical benefit from CRT even after the occurrence of a post-implantation HFE. Notably, the time to a first HFE was not found to be influential in assessing the risk of subsequent HFEs, suggesting that reduction of subsequent of HFEs with CRT-D is maintained regardless of when the first post-implantation HFE occurs.

Treatment with CRT-D was associated with a pronounced reduction in the risk of both first and subsequent HFEs among patients with LBBB but had no detectable effect on either of the 2 endpoints among non-LBBB patients. No other statistically significant interactions, including gender and QRS width, were identified once the LBBB/non-LBBB differences were allowed for in the regression models. The pronounced effect of CRT-D therapy among women noted in the earlier report (11) might be attributed to the fact that 87% of the women in the trial had LBBB at enrollment. These findings are consistent with the recent Food and Drug Administration approval of the use of the CRT-D only in the LBBB subgroup of the MADIT-CRT–defined population (14).

The present study shows that first and second post-enrollment HF episodes were associated with 7- and nearly 19-fold respective increases in the risk of subsequent all-cause mortality in the MADIT-CRT population. Furthermore, Kaplan-Meier survival analysis showed the cumulative rate of all-cause mortality after the occurrence of a second HFE was nearly 50% within 2 years. Thus, although CRT-D therapy was not associated with a significant reduction in the risk of all-cause mortality during the trial (11), it is likely that the reduction in the risk for repeated HF episodes with CRT will translate into a survival benefit during longer-term follow-up.

We designed the present study primarily as an efficacy analysis in which the outcome of study patients was assessed during active device therapy rather than by the original treatment allocation, due to the relatively large proportion of ICD-allocated patients who experienced HFE and subsequently crossed over to the CRT-D arm (37%). Potential imbalances were addressed by using covariate adjustments rather than depending on randomization and stratification.

Although risk seems to further increase after a second HFE, most analyses treated all subsequent HFEs uniformly, due to smaller numbers and for simplification in presentation. A sensitivity analysis substantiated this.

Big differences were uncovered between responses in LBBB and non-LBBB patients. The primary results were confirmed by repeating the main analyses within the LBBB subgroup, thereby avoiding proportional hazards assumptions between QRS morphology subgroups.

Randomized clinical trials of CRT are generally designed to evaluate an endpoint that comprises initial clinical events, even though prevention of recurring episodes with device therapy has incremental clinical and prognostic implications. The results of the present MADIT-CRT sub-study show for the first time that CRT is associated with a pronounced reduction in the risk of repeated HF episodes. Furthermore, the present results confirm the prognostic implications of recurring HFEs in patients with left ventricular dysfunction. These findings should be incorporated in the risk assessment and management of patients who receive device therapy for the prevention of HF and sudden cardiac death and stress the importance of evaluating risk for recurrent events in clinical trials.

For supplementary tables, please see the online version of this article.