Multiple randomized clinical trials have demonstrated a significant reduction in the risk of death with an implantable cardioverter-defibrillator (ICD) among patients with a left ventricular ejection fraction (LVEF) ≤35% (1- 3). However, the benefit of the ICD within the low-LVEF population may not be uniform because this group of patients is a heterogeneous group comprising symptomatic and asymptomatic HF patients who may also have different comorbidities and different prognoses. We have recently developed a risk score for the assessment of ICD efficacy in the MADIT-II (Multicenter Automatic Defibrillator Implantation Trial II) population (4). The study showed that 5 simple clinical parameters (comprising age, heart failure functional class, blood urea nitrogen [BUN], QRS duration, and atrial fibrillation) can be used to stratify patients into a low-risk group (with none of the 5 clinical risk factors) who did not derive a significant survival benefit from the ICD in MADIT-II; an intermediate high-risk group (1 to 2 risk factors) who derived a pronounced benefit from primary ICD implantation; and a very high-risk group (with advanced comorbidities), in whom the benefit of the ICD was attenuated. This study, however, was based on the in-trial phase of MADIT-II, comprising a median follow-up of only 1.5 years.

We have recently shown that the beneficial effects of the ICD were sustained during an 8-year follow-up of the MADIT-II population, thus extending the evidence for ICD efficacy at least throughout the life span of the implanted device (5). The present study was designed to evaluate the yield of the previously identified risk score for the assessment of the long-term survival benefit of primary ICD therapy.

MADIT-II enrolled 1,232 patients with a myocardial infarction 1 month or more before entry into the study and an LVEF ≤30%. Patients were randomly assigned in a 3:2 ratio to receive either an implanted defibrillator or non-ICD conventional medical therapy. Details of the design, methods, and results of the MADIT-II trial have been previously reported (1). The original MADIT-II publication was based primarily on the 0- through 4-year trial period, with a median follow-up of 1.5 years (interquartile range 0.8 to 2.5 years). The newly acquired long-term data comprise a median follow-up of 7.6 years (interquartile range 3.5 to 9.0 years) (5). Data regarding crossover between allocated treatment arms were recorded for all study participants during the study and following trial closure and is provided in the Online Appendix. Of the 1,232 study participants, 41 were excluded due to missing data on 1 of the 5 clinical risk factors; thus, the current study includes 1,191 (97%) patients who were enrolled in MADIT-II.

The present study employs a previously developed risk score (4) for the endpoint of all-cause mortality that was developed based on follow-up data from the in-trial phase of MADIT-II. A simple risk score was constructed as a count of risk factors identified in each patient, among the selected following 5 risk factors: New York Heart Association (NYHA) functional class >II; age >70 years; BUN >26 mg/dl; QRS duration >0.12 s; and atrial fibrillation. For the present study, patients were categorized into 3 subgroups: 1) low-risk subgroup (n = 345), including patients with no risk factors; 2) intermediate-risk subgroup (n = 646), including patients with 1 or 2 risk factors; and 3) high-risk subgroup (n = 200), including patients with 3 or more risk factors. Risk score counts were truncated at ≥3 because of small numbers of patients with counts of 4 or 5 risk factors. The primary endpoint was time to all-cause mortality during the 8-year extended follow-up.

We also evaluated the proportion of patients from each risk group who changed category during the MADIT-II in-trial period for 4 of the 5 pre-specified risk factors with available follow-up data (NYHA functional class, age, BUN, atrial fibrillation). The last ascertainment of risk score was obtained at the end of the MADIT-II trial.

Characteristics of study patients were compared using the Wilcoxon rank sum, chi-square, or Fisher exact tests, as appropriate. The probability of all-cause mortality by treatment group for each risk group, with follow-up censored by change in treatment arm was graphically displayed according to the Kaplan-Meier method, with comparison of cumulative events by the log-rank test. The number of patients that needed to have been treated with an ICD to save 1 life in each risk subgroup was calculated as the inverse of the survival difference between the 2 treatment arms using the Kaplan-Meier estimates at 8 years. Cox proportional hazards regression modeling was used to evaluate the independent contribution of each of the pre-specified covariates to the occurrence of all-cause mortality during 8 years of follow-up, with follow-up censored by change in treatment arm. Pre-specified covariates in the multivariate models included treatment (ICD vs. non-ICD), and the high risk score covariates: age >70 years; NYHA functional class >II; QRS duration >0.12s; BUN levels >26 mg/dl; and atrial fibrillation. In addition, Cox proportional hazards regression modeling was used to evaluate the independent contribution of the ICD treatment within each risk group category (low, intermediate, and high) to predict the risk of all-cause mortality during 8 years of follow-up. An ICD treatment–by–risk group interaction term was used to both estimate the ICD to conventional treatment hazard ratio for each risk group as well as test for an overall interaction between risk group and ICD treatment. This 1 degree-of-freedom interaction term used risk group as a 3-level ordinal variable. A Bonferroni correction was used when examining the statistical significance of ICD treatment within each risk group, which resulted in a corrected significance level of 0.050/3 = 0.017. A 2-sided p value <0.05 was used for declaring statistical significance for all other analyses. The statistical software used for the analyses was SAS software (version 9.2, SAS Institute, Cary, North Carolina).

The baseline characteristics of study patients by risk score category are shown in (Table 1). As expected, patients in the high-risk subgroup were sicker and had higher prevalence of the 5 pre-specified risk factors as compared with the lower-risk subgroups. In addition, patients in the high-risk subgroup as compared with the low-risk subgroups exhibited a higher prevalence of diabetes mellitus, hypertension, and left bunch branch block; they had lower LVEF and were treated more frequently with diuretics and digitalis and less frequently with beta-blockers.

At 8 years of follow-up, the cumulative probability of mortality among all study patients was 54%, with patients treated with ICD having a significantly lower cumulative probability of mortality as compared with non-ICD patients (50% and 64%, respectively, p < 0.001). When survival was assessed by the risk score (Figure 1), there was a direct relationship between risk categories and long-term mortality rates.

Multivariate analysis showed that each of the 5 pre-specified risk factors included in the risk score was a significant independent predictor of all-cause mortality in the MADIT-II population during 8 years of follow-up (Table 2). Furthermore, when outcome was assessed by the risk score, the intermediate- and high-risk subgroups were shown to have respective 2.4-fold (p < 0.001) and 5-fold (p < 0.001) increases in the risk of long-term mortality as compared with the low-risk subgroup.

Kaplan-Meier survival analyses demonstrated that ICD was associated with a significant survival benefit in the low-risk subgroup (Figure 2). Thus, at 8 years, the cumulative survival rate was 75% for patients treated with ICD as compared with 58% among non-ICD patients (p = 0.004). Among intermediate-risk patients, ICD therapy was also associated with a significant survival benefit (Figure 3). Thus, in this subgroup, the 8-year cumulative survival rate was 47% for patients treated with ICD and 31% among non-ICD patients (p < 0.001). The survival difference corresponded to ICD being associated with 6 patients that needed to have been treated to save 1 life within 8 years in both the low- and the intermediate-risk subgroups. By contrast, among high-risk patients (Figure 4), there was no significant difference in 8-year survival between ICD and non-ICD subgroups (19% vs. 17%, p = 0.50).

Multivariate analysis (Table 3) showed that ICD therapy was associated with a significant long-term survival benefit among low- and intermediate-risk patients (hazard ratio [HR]: 0.52, p < 0.001, and HR: 0.66, p < 0.001, respectively), whereas treatment with an ICD was not associated with a significant benefit among high-risk patients (HR: 0.84, p = 0.25). This analysis shows a p value for the overall ICD-by–risk group interaction = 0.075.

To further evaluate a possible bias that might have contributed to the lack of a significant long-term ICD effect in high-risk patients, we carried out sensitivity analyses and found consistent results with the primary analysis (Online Appendix).

We used a previously developed risk score among MADIT-II patients to explore ICD efficacy in reducing all-cause mortality during 8 years of follow-up among patients who had low-, intermediate-, and high-risk clinical characteristics at enrollment in the trial. Our findings suggest that: 1) important time-dependent changes in the clinical characteristics of study patients occurred during the trial, resulting in corresponding changes in risk categories during follow-up; 2) ICD therapy conferred a long-term survival benefit in patients who had low-risk characteristics at enrollment, in whom the benefit of the ICD was not apparent during the early phase of follow-up; 3) the early benefit of ICD therapy among intermediate-risk patients, shown in the earlier report (4) was maintained during 8 years of follow-up; and 4) we observed no significant long-term ICD benefit among the high-risk subgroup comprising 17% of study patients, who experienced very high mortality rates during long-term follow-up, regardless of treatment group.

ICDs are costly technologies, with implantation costs of more than $25,000 per patient (6); thus, although ICD implantation can reduce the risk of sudden cardiac death, it may do so at a relatively high cost. Improved cost-effectiveness estimates for primary ICD implantation can be obtained if appropriate risk subgroups are identified. Accordingly, we have previously identified 5 high-risk clinical factors—NYHA functional class >II, age > 70 years, BUN >26 mg/dl, QRS duration >0.12 s, and atrial fibrillation—that were independently associated with increased mortality in the non-ICD arm of MADIT-II, and we developed a simple risk score as a count of risk factors identified in each patient (4). In that study, we observed a U-shaped pattern for ICD efficacy, in which ICD therapy was shown to be associated with a significant reduction in the risk of death only among patients with 1 or 2 risk factors.

In the present study, follow-up of the MADIT-II population was extended to 8 years. Similar to the earlier report, we have also shown a significant survival benefit with an ICD among intermediate-risk patients (1 to 2 risk factors) with 34% mortality risk reduction (p < 0.001) and no significant ICD benefit among high-risk patients (3 or more risk factors; HR: 0.84, p = 0.25). By contrast, among low-risk patients (no risk factors), primary ICD was associated with a pronounced 48% reduction in 8-year mortality (p < 0.001), possibly due to the development of higher-risk clinical characteristics in this subgroup during long-term follow-up (Online Appendix). Thus, ICD was most effective in the low-risk group, was associated with moderate benefit in the intermediate-risk group, and not associated with any benefit in the high-risk group (p for ICD-by–risk group interaction = 0.075).

Our analysis did not incorporate data on the mode of death following trial closure, which could have facilitated further clarification of the relationship between ICD efficacy and risk score categories. We had no data on risk factors following trial closure, thus we were limited in our ability to calculate the percentage of patients who moved to other risk categories during the post-trial period. It should also be noted that, similar to the original MADIT-II risk score (4), the current study population was not sufficiently powered to show significant ICD treatment–by–risk subgroup interactions. However, we did identify a trend across risk groups providing, for the first time, data regarding the relative effectiveness of the ICD across risk groups during long-term follow-up.

Risk stratification among the low-LVEF population has important clinical implications for predicting outcome, selection of treatment, and reduction in the costs of treatment. In the present study, we have shown that employing a simple risk score for the assessment of ICD efficacy results in the identification of low- and intermediate-risk patients with ischemic cardiomyopathy, who derive a pronounced long-term survival from the device, with a number-needed-to-treat of 6 to save 1 life during 8 years of follow-up. By contrast, patients with high-risk clinical characteristics prior to ICD implantation do not appear to derive a significant survival benefit from the nonresynchronization device during long-term follow-up, possibly due to the competing risks of heart failure death and additional causes of noncardiac mortality that are more prevalent in this population. In this patient subset, additional therapeutic interventions, including combined cardiac resynchronization–defibrillator therapy, may be necessary to provide improved long-term survival.

For data regarding crossover between the allocated treatment arms of MADIT-II and additional secondary analyses, please see the online version of this article.