Author + information
- Received February 10, 2015
- Revision received April 30, 2015
- Accepted May 16, 2015
- Published online September 1, 2015.
- Michael R. Gold, MD, PhD∗∗ (, )
- Amie Padhiar, MSc†,
- Stuart Mealing, MSc†,
- Manpreet K. Sidhu, MBA, PhD‡,
- Stelios I. Tsintzos, MD, MSc§ and
- William T. Abraham, MD‖
- ∗Division of Cardiology, Medical University of South Carolina, Charleston, South Carolina
- †Health Economics and Epidemiology, ICON/Oxford Outcomes, Oxford, United Kingdom
- ‡Health Economics and Epidemiology, ICON/Oxford Outcomes, Morristown, New Jersey
- §Global Economics, Reimbursement and Evidence, Medtronic Global CRHF Headquarters, Mounds View, Minnesota
- ‖Division of Cardiovascular Medicine, Ohio State University, Columbus, Ohio
- ↵∗Reprint requests and correspondence:
Dr. Michael R. Gold, Medical University of South Carolina, Division of Cardiology, 114 Doughty Street—MSC 592, Charleston, South Carolina 29425.
Objectives This study sought to assess the lifelong extrapolated patient outcomes with cardiac resynchronization therapy (CRT) in mild heart failure (HF), beyond the follow-up of randomized clinical trials (RCTs).
Background RCTs have demonstrated short-term survival and HF hospitalization benefits of CRT in mild HF. We used data from the 5-year follow-up of the REVERSE (REsynchronization reVErses Remodeling in Systolic left vEntricular dysfunction) study to extrapolate survival and HF hospitalizations. We compared CRT-ON versus CRT-OFF and CRT defibrillators (CRT-D) versus CRT pacemakers (CRT-P).
Methods Multivariate regression models were used to estimate treatment-specific all-cause mortality, disease progression, and HF-related hospitalization rates. Rank-preserving structural failure time (RPSFT) models were used to adjust for protocol-mandated crossover in the survival analysis.
Results CRT-ON was predicted to increase survival by 22.8% (CRT-ON 52.5% vs. CRT-OFF 29.7%; hazard ratio [HR]: 0.45; p = 0.21), leading to an expected survival of 9.76 years (CRT-ON) versus 7.5 years (CRT-OFF). CRT-D showed a significant improvement in survival compared with CRT-P (HR: 0.47; 95% confidence interval [CI]: 0.25 to 0.88; p = 0.02) and were predicted to offer 2.77 additional life-years. New York Heart Association (NYHA) functional class II patients had a 30.6% higher HF hospitalization risk than class I (I vs. II incident rate ratio [IRR]: 0.69; 95% CI: 0.57 to 0.85; p < 0.001) and 3 times lower rate compared with class III (III vs. II IRR: 2.98; 95% CI: 2.29 to 3.87; p < 0.001).
Conclusions RPSFT estimates yielded results demonstrating clinically important long-term benefit of CRT in mild HF. CRT was predicted to reduce mortality, with CRT-D prolonging life more than CRT-P. NYHA functional class I/II patients were shown to have a significantly reduced risk of HF hospitalization compared with class III, leading to CRT reducing HF hospitalization rates.
- cardiac resynchronization therapy
- health economics
- heart failure
- health policy
- rank-preserving structural failure time
Cardiac resynchronization therapy (CRT) is a well-established treatment option for patients with advanced heart failure (HF) and interventricular conduction delay, on the basis of the results of randomized trials and consensus guidelines (1–6). CRT alone (CRT-P) or in combination with an implantable cardioverter defibrillator (CRT-D) has been shown to be cost effective in advanced HF (7–10). Subsequently, 3 multicenter randomized studies were performed to evaluate the role of CRT in mild HF with a reduced ejection fraction: REVERSE (REsynchronization reVErses Remodeling in Systolic left vEntricular dysfunction) (11), MADIT CRT (Multicenter Automatic Defibrillator Implantation Trial—Cardiac Resynchronization Therapy) (5), and RAFT (Resynchronization-Defibrillation for Ambulatory Heart Failure) (12). The results of these trials were largely complementary, showing the benefit of CRT in this patient population, particularly in the presence of left bundle branch block (LBBB) (12,13). However, there were important differences in the study designs and follow-up of these trials.
The REVERSE study was the only randomized trial in mild HF in which all patients were implanted with CRT devices, with patients programmed for active CRT (“CRT-ON”) or control (“CRT-OFF”). The double-blind design has certain advantages from a trial interpretation perspective (14), but the protocol-mandated crossover to active therapy of those randomized to CRT-OFF following the primary period of blinded assessment presents a major challenge for quantification of the long-term benefits of therapy.
Statistical techniques exist that adjust for crossover effects when treatment effect estimates are generated (14–17). These techniques, particularly rank-preserving structural failure time (RPSFT) models, are increasingly being accepted (18) but remain, to the best of our knowledge, unused in cardiology. Accordingly, an RPSFT-based analysis was performed on the REVERSE patient population to estimate the long-term benefit of CRT on survival. Additional analyses were also undertaken to evaluate long-term HF hospitalization rates as well as assess the benefits of device type (CRT-P vs. CRT-D) on outcomes.
The REVERSE trial was a prospective, randomized, double-blind study of 610 patients. The key inclusion criteria were New York Heart Association (NYHA) functional class I/II HF, QRS duration of ≥120 ms and left ventricular ejection fraction (LVEF) of ≤40%. All patients were implanted with a CRT device and randomly assigned (2:1) to active CRT (“CRT-ON”) or control (“CRT-OFF”). Randomization ended with all patients having CRT programmed ON at pre-specified post-implant times (12 months in North America and 24 months in Europe). Overall, 37 centers in North America randomized 348 patients and 35 centers in Europe randomized 287 patients. All patients were followed for a maximum of 5 years post-implant per study protocol, and the choice of using a CRT-P or CRT-D device was determined by local guidelines at implantation (19,20).
Covariates of interest and subgroup definition
On the basis of expert input and previous analyses (21,22), LBBB morphology (yes/no), HF etiology (ischemic/nonischemic), sex (male/female), and QRS duration were used in subgroup definitions, with QRS duration defined by quartiles in study data (<138, ≥138 to <152, ≥152 to <167, and ≥167 ms). Baseline age was included as a continuous variable, and treatment was included in an appropriate manner for the required comparison. Statistical significance of covariates was investigated, but all covariates were included in all analyses regardless of statistical significance to facilitate the generation of subgroup results. All analyses were undertaken in STATA 12.0 (StataCorp LP, College Station, Texas).
Quantification of all-cause mortality
An RPSFT model was used to adjust for the impact of protocol-mandated treatment switching when lifetime survival estimates were generated for the CRT-ON versus CRT-OFF comparison. RPSFT models belong to a family of statistical techniques that are fundamentally based on the assumptions that: 1) the observed time to event and information on what treatment was offered and when to patients can be used to predict the treatment effect if patients never switched (on the basis of “counterfactual” event times; thus, “structural failure time”); 2) clinical events in both arms will happen in the same sequence as observed for each patient, and only the time of occurrence will differ when adjustment is made for treatment switching (thus “rank preserving”); and 3) the treatment effect is common between the patients who were randomized to treatment initially and patients who switched to receive treatment. RPSFT models compare the time it takes patients in the active arm of a trial to reach an endpoint with the same time in the control arm. It then calculates an “acceleration parameter” and relates the actual times to event with the “counterfactual” times calculated to ultimately predict how a patient switching treatments after a period of randomization would have behaved in the absence of that switching (23). A detailed discussion of this methodology and its suitability for application to the REVERSE study data is presented in the Online Appendix.
Common parametric survival models were explored to extrapolate the REVERSE study mortality data beyond trial follow-up (exponential, Weibull, Gompertz, log-logistic, log-normal, and gamma). The ultimate choice of parametric function was on the basis of “goodness of fit” to the observed data assessed through Akaike and Bayesian information criteria (AIC and BIC, respectively) and inspection of the Cox-Snell residual plots and long-term predictive plausibility. Both univariate and multivariate survival models were explored for the comparison between CRT-ON and CRT-OFF, for which the multivariate models contained all core covariates discussed previously to generate subgroup results alongside the treatment covariate.
A similar approach was used to generate lifetime survival estimates in the CRT-D versus CRT-P comparison, the difference being that an adjustment for crossover was not required. To evaluate the impact of adjusting for crossover, we also created a mortality risk equation using a conventional intention-to-treat (ITT)–based approach (i.e., crossover unadjusted).
The results from the raw parametric survival analysis are presented as hazard ratios (HRs), with the long-term predicted values reported in terms of absolute survival rates and number needed to treat (NNT) to avoid 1 death estimates. A consequence of the lifetime time horizon is everyone being predicted to be dead at the end of the extrapolation process; therefore, a truncated time horizon of 10 years was used in all NNT calculations.
Modeling disease progression
Multinomial logistic regression models were used to predict the changes in NYHA functional class over time for both comparisons. Because the REVERSE study enrolled only NYHA functional class I and II patients, there was a paucity of patients with advanced HF early in the trial. Accordingly, NYHA functional class III and IV status were pooled into 1 group (“III/IV”), and NYHA functional class II was used as the baseline category in all analyses. Univariate and multivariate models were explored, with SEs clustered to account for intrapatient correlation. In all multivariate models, baseline NYHA functional class and time were included into the list of predictor variables, in addition to the core covariates. A range of approaches to include time into the statistical models were explored.
For ease of interpretation, covariate estimates are presented both as raw coefficients and relative risk ratios (RRRs). The latter has an intuitive interpretation: a value greater than 1 corresponds to an increased likelihood of being in a given NYHA functional class and vice versa for a value between 0 and 1.
Monthly likelihood of HF hospitalization
Monthly hospitalization rates were estimated using a similar approach to that for NYHA functional class distribution. Count-based Poisson regression models were assessed, with the predictor covariates being augmented by the inclusion of NYHA functional class as a time-varying covariate and the exclusion of any treatment covariates. SEs were again clustered in all analyses, and covariate estimates are presented on both the natural and incident rate ratio scales. Interpretation of the latter is very similar to that for RRRs, with a value greater than 1 corresponding to an increased likelihood of experiencing an HF-related hospitalization (and vice versa for a value between 0 and 1).
A comparison of patient baseline characteristics in each of the analyzed groups is presented in Table 1. No significant differences existed between patients randomized to CRT-ON or CRT-OFF. Within the nonrandomized CRT-D versus CRT-P subgroups of CRT-ON, significant differences were observed in baseline LVEF and overall mean QRS duration. The baseline characteristics were similar, with no significant differences in quartile-specific mean QRS duration. Nevertheless, the European sample was significantly younger and had longer QRS durations and fewer comorbidities (lower prevalence of ischemia, less peripheral artery disease, and less history of prior myocardial infarction and hypertension). Europeans had a smaller proportion of CRT-D implanted (24). Details of these baseline differences are presented in Online Table 1.
Mortality rates and projected survival
There were 84 deaths observed in the 610 patients enrolled over the entire duration of the study. Of those, 63 deaths (75%) occurred during time periods before treatment-mandated crossover. There were 67 deaths among CRT-D patients (10%) and 17 in CRT-P patients (11.8%). Further details are provided in Online Table 2.
CRT-ON versus CRT-OFF
The Gompertz model was selected as most appropriate for crossover-unadjusted survival estimation (univariate: AIC 608.9, BIC 622.1; multivariate: AIC 580.2, BIC 624.4). The final adjusted and unadjusted multivariate models for the ITT and RPSFT analyses are provided in Table 2. The goodness of fit to the trial data is presented in Figures 1A and 1B, and the 10-year extrapolations are presented in Figures 1D and 1E for the ITT and RPSFT analyses, respectively.
CRT was associated with decreased mortality, although this association did not reach statistical significance. This was noted for patients in the ITT analysis, with the magnitude of the treatment effect being very similar in both the univariate and multivariate models (univariate: HR: 0.75; 95% confidence interval [CI]: 0.47 to 1.17; multivariate: HR: 0.73; 95% CI: 0.47 to 1.15). After adjustment for crossover with the RPSFT model, the magnitude of the treatment effect was increased (HR: 0.45; 95% CI: 0.13 to 1.58). Hence, the crossover-adjusted analysis showed evidence of a greater benefit in mortality for CRT patients compared with the ITT analysis.
Ischemic etiology and female sex were significant predictors of mortality in both models (RPSFT multivariate model: HR: 1.96 and 0.29, respectively; p < 0.03 throughout). After correction for other covariates, LBBB morphology, QRS duration, and age were not significant predictors of mortality.
The predicted lifetime survival estimates generated using the fitted models are reported in Table 3. In the unadjusted analysis, the improvement in overall survival for CRT-ON compared with CRT-OFF was estimated as 4% (0.40 years), whereas CRT was predicted to increase life expectancy by approximately 30% (2.46 years) in the crossover-adjusted analysis. The cumulative proportion of patients predicted to have died at years 1 to 10 is presented in Table 4. At 10 years, the proportion of patients predicted to be alive for CRT-ON and CRT-OFF was 52.5% and 29.7%, respectively. Thus, the NNT to avoid 1 death was 4.5.
CRT-D versus CRT-P
Gompertz models were again the best fit to the trial data (univariate: AIC 400.0, BIC 412.2; multivariate: AIC 371.9, BIC 412.3) in this patient group. Treatment with CRT-D, although beneficial, was not significant in the univariate analysis but became so after accounting for the impact of other covariates (univariate HR: 0.63; 95% CI: 0.34 to 1.18; multivariate HR: 0.47; 95% CI: 0.25 to 0.88). The full risk model is presented in Table 2, with the goodness of fit and long-term extrapolation plots in Figures 1C and 1F. In addition to treatment with CRT-D, female sex and LBBB morphology were significant predictors of mortality.
Lifetime survival estimates are presented in Table 3, with CRT-D predicted to provide, on average, an additional 2.8 years of life (+27%). On the basis of the 10-year survival proportions (Table 4), the NNT for CRT-D to avoid 1 death was 5.2.
For the unadjusted comparison of CRT-ON with CRT-OFF, conditional on being alive, the likelihood of being in NYHA functional class III compared with NYHA functional class II significantly increased over time (RRR: 1.71; 95% CI: 1.45 to 2.01). A similar result was generated for the comparison of CRT-D with CRT-P. The full risk equations used to predict NYHA functional class over time are presented in Online Table 3, with time included in the final model using a natural log transformation.
After adjustment for baseline characteristics and the effect of time, compared with those who did not receive CRT, individuals receiving active therapy had a significantly increased likelihood of regressing to NYHA functional class I (RRR: 1.39; 95% CI: 1.05 to 1.86) and a decreased directional likelihood of being in NYHA functional class functional III (RRR: 0.62; 95% CI: 0.38 to 1.03) relative to NYHA functional class II. No significant differences on the rate of disease progression were predicted for patients receiving CRT-D or CRT-P. As such, CRT therapy was shown to have a positive impact on delaying disease progression (as measured using NYHA functional classes), whereas the presence of a defibrillator backup (i.e., CRT-D) did not show a significant impact on NYHA functional class progression.
The predicted distribution across each NYHA functional class during years 1 to 10 for patients receiving CRT-ON or CRT-OFF are presented in Figures 2A and 2B, with the predicted values for patients receiving CRT-D or CRT-P being presented in Figures 2C and 2D. The values on these plots are contingent on patients being alive; therefore, the predicted rates accounting for death are presented in Table 4, with CRT therapy again showing a positive impact in terms of delaying disease progression.
Univariate and multivariate Poisson models were fitted for both comparisons, with the final risk equations reported in Online Table 4. For the comparison of CRT-ON with CRT-OFF, the monthly event rate varied significantly across each NYHA functional class. The rate for individuals in class I was predicted as 30.6% lower than that in class II (incident rate ratio: 0.69; 95% CI: 0.56 to 0.85); for patients in class III, the monthly rate was nearly 3 times higher than that for a similar individual in class II (IRR: 2.98; 95% CI: 2.29 to 3.87). For all patients in the CRT-ON versus CRT-OFF comparison, the average absolute monthly event rates were 0.04 (NYHA functional class I), 0.05 (NYHA functional class II), and 0.16 (NYHA functional class III). The NYHA functional class–specific absolute monthly values in the comparison of CRT-D versus CRT-P were very similar (0.04, 0.05, and 0.17 for NYHA functional classes I, II, and III, respectively).
The presence of ischemic etiology also had a significant impact on the monthly hospitalization rate (IRR: 1.63; 95% CI: 1.27 to 2.09). Similar results were predicted for individuals who received CRT-D or CRT-P, with the magnitude of effect for all clinical variables remaining very similar to that for the CRT-ON versus CRT-OFF comparison.
Survival benefits of CRT, and the additional benefits of CRT-D, made the overall number of HF hospitalizations accrued over a lifetime higher (Table 5). Because CRT-D impacted survival, the total number of events differed.
The primary results of this study are that CRT prolonged survival by more than 2 years (+27%) compared with optimal medical therapy and that CRT-D devices prolonged survival by almost 3 years compared with CRT-P devices (+30%). Advanced statistical techniques have been used to reveal and quantify these effects because the protocol-mandated crossover in the REVERSE study limits the use of randomized data alone to assess long-term outcomes. After a relatively short initial time period, all study patients essentially received CRT for a cumulative period longer than the one of true difference in treatment. This rendered the observed ITT results fundamentally inappropriate for long-term projections. In addition to survival benefits, our analyses showed HF hospitalization rates to be reduced by using CRT as a consequence of significant slowing of disease progression regardless of device type.
The REVERSE trial had several unique aspects that provided both opportunities and challenges for assessment of the long-term impact on therapy. Specifically, this was the only multicenter study of mild HF that included both CRT-P and CRT-D devices. In addition, a 5-year follow-up was pre-planned, with a detailed evaluation of events. However, the design (enabled by all patients undergoing CRT device implantation) included an obligatory crossover of CRT-OFF patients to active CRT therapy after 1 to 2 years, which is significantly shorter than the duration of randomization in the MADIT-CRT or RAFT studies. As already discussed, we used advanced statistical methods to overcome this data limitation and predicted what would have happened had crossover not occurred. These techniques have been widely used and accepted in fields such as oncology (25–27), but not in HF. Hence, this work represents the first use of these techniques in HF. Such analyses allow lifetime extrapolation of outcome data and adjustment for high levels of treatment switching—capabilities important for assessing the population and economic impacts of therapy without limiting the analysis to the duration of a clinical trial.
In our analysis, we focused on quantifying how CRT impacted 3 important aspects of HF: all-cause mortality, rate of disease progression, and, indirectly, the risk of experiencing an HF-related hospitalization. The use of a single study to generate risk equations for all of these outcomes in a mild HF population has not been performed previously.
Previous studies have reported increases in mortality after each HF hospitalization (28). Thus, it could already be ascertained that delaying HF hospitalizations would have a positive impact on disease progression. That was shown in the European subgroup of the REVERSE trial (24), in which a large difference in the number of nonresponders in the CRT-ON group (19% nonresponders with CRT vs. 34% without; p = 0.01) was noted with 2 years of randomization.
After correction for the impact of crossover, as well as other clinical covariates, the all-cause mortality HR for CRT-ON compared with CRT-OFF was 0.45 compared with 0.73 in the unadjusted analysis. Crossover adjustment also increased the lifetime survival benefit associated with CRT therapy from 0.4 years to approximately 2.5 years. Predicted mean life expectancy for CRT-ON and CRT-OFF was 9.76 and 7.5 years, respectively. CRT-ON had a positive impact in terms of delaying disease progression (as measured using the NYHA functional classification system); because NYHA functional class had a significant impact on the rate of HF-related hospitalizations (p < 0.0001), it follows that CRT therapy also had a positive impact on the monthly HF hospitalization rate. Of note, however, due to the survival benefit of CRT, the total number of expected events was higher in CRT patients than in non-CRT patients. The results for the comparison of CRT-D with CRT-P showed that the presence of an implantable cardioverter-defibrillator (ICD) offered a statistically significant improvement in overall survival (HR: 0.47; p = 0.019) but no significant impact on disease progression, as expected. The overall improvement in life expectancy for CRT patients who received a defibrillator was predicted as 2.8 years.
A recent multivariate subanalysis of the REVERSE study comparing CRT-D with CRT-P devices using the 5-year follow-up data only in the CRT-ON arm showed that ICD backup was associated with increased survival (23). Whereas the present analyses were on the basis of some of the same data used previously, we used a different modeling approach that also incorporated CRT-OFF patients. The high degree of similarity in the results was supportive and helps to cross-validate both analyses.
Assessing the covariate estimates, LBBB morphology was associated with improved outcomes in other REVERSE analyses and in the RAFT and MADIT-CRT trials. Sex played a significant role in improved survival with CRT-D, and QRS duration was a strong predictor of better outcomes in a recent meta-analysis (29). Approximately one-half of the patients receiving CRT therapy in this meta-analysis had NYHA functional class III/IV HF; therefore, direct comparison with the results from our work should be made with caution. Both LBBB morphology and prolonged QRS duration were associated with improved prognosis after CRT-D implantation in Medicare claims data (30). Long-term outcomes of LBBB patients who received a CRT-D device are also available from the MADIT-CRT study (31), with reported 5- and 7-year cumulative death probabilities of approximately 11% and 18%, respectively. The corresponding values from our model were in general agreement with these MADIT-CRT observations at 8.7% and 15.2%, respectively. The MADIT-CRT study results raise our confidence in the robustness of our calculations because results appeared generally aligned and our statistical models used no MADIT-CRT information. However, comparisons may be difficult. In the MADIT-CRT trial, the patients were older than those in the CRT-D subgroup of the REVERSE study (50% of MADIT-CRT CRT-D patients were ≥65 years of age and the mean age of REVERSE CRT-D recipients was 62.7 years of age). Thirty-one percent of MADIT-CRT CRT-D patients had QRS duration of <150 ms, whereas 52.2% of REVERSE CRT-D patients had QRS duration of ≤151 ms. LVEF was more comparable, with 61% being equal to or less than 25% in the MADIT-CRT study, and CRT-D patients in the REVERSE study had a mean LVEF of 26%.
The unique analysis in this study was the use of an RPSFT model to evaluate the REVERSE trial. The analysis arose from the large amount of treatment switching in the trial, the relatively short follow-up before the protocol-mandated switch, little evidence of a significant difference in arms before switching, and the overall low event rates. We believe our approach may be appropriate for 4 reasons: preservation of randomization, preservation of event ordering, its robustness to a large degree of crossover, and preservation of treatment effects.
Differences in patient baseline variables between North America and Europe may have played a role in our analyses. As noted previously, patients in Europe were younger and had fewer comorbidities, longer QRS durations, and better performance in the 6-min walk test. The fewer comorbidities (especially the lower percentage of patients with ischemia in Europe) may explain the fewer CRT-Ds implanted in these countries, on the basis of implantation guidelines at the time of enrollment (28). Some of these differences occurred in covariates studied. We believe our analyses are not jeopardized because adjustments were made for them in all of the statistical models employed.
Another important limitation of this analysis is the long-term clinical pathway uncertainty. Specifically, it can be expected that a large portion of the class II patients enrolled in the REVERSE study would have progressed to a “traditional” class III indication for biventricular pacing in subsequent years. Our analyses have not looked at the incremental benefit of a class II implant compared with implanting when the patient reaches class III.
We applied a widely reported analytic method to control for protocol-mandated crossover that had not been previously published for cardiology trials, with the end goal of using all available REVERSE study data to quantify the long-term impact of CRT in patients with mild HF. Our analyses demonstrated that implantation of a CRT device in a patient with NYHA functional class II status, a reduced LVEF, and QRS prolongation is expected to prolong survival by, on average, 2.26 years (+27%). A patient with CRT-D is expected to survive 2.77 additional years (+30%) compared with a patient with CRT-P. CRT was also projected to delay progression from one NYHA functional class to the next and to reduce HF hospitalization rates. As expected, the models showed CRT-D to have no significant impact on disease progression compared with CRT-P. In addition to providing estimates of the long-term clinical benefit of CRT, this method may be useful for extrapolation of economic benefit of therapies.
COMPETENCY IN MEDICAL KNOWLEDGE: CRT has been shown to delay HF disease progression, reduce HF hospitalization rates, and increase survival among patients with mild or advanced HF. For estimation of outcomes beyond the duration of clinical trials, advanced statistical techniques can be applied to extrapolate results and adjust for treatment crossover.
TRANSLATIONAL OUTLOOK: Current data suggest that these long-term projections are sufficiently accurate; therefore, they can be used to inform decision making and cost-effectiveness analyses.
For an expanded Methods section, and supplemental tables and references, please see the online version of this paper.
The REVERSE study was funded by Medtronic. ICON/Oxford Outcomes conducted the analysis after receiving funding from Medtronic. Dr. Gold has received research funding and consulting fees from Medtronic, Boston Scientific, and St. Jude Medical. Ms. Padhiar, Mr. Mealing, and Dr. Sidhu are employees of ICON/Oxford Outcomes. Dr. Tsintzos is an employee of and owns stock in Medtronic. Dr. Abraham has received consulting fees from St. Jude Medical.
- Abbreviations and Acronyms
- cardiac resynchronization therapy
- cardiac resynchronization therapy defibrillator
- cardiac resynchronization therapy pacemaker
- heart failure
- left bundle branch block
- left ventricular ejection fraction
- New York Heart Association
- rank-preserving structural failure time
- Received February 10, 2015.
- Revision received April 30, 2015.
- Accepted May 16, 2015.
- American College of Cardiology Foundation
- Brignole M.,
- Auricchio A.,
- Baron-Esquivias G.,
- et al.
- Calvert M.J.,
- Freemantle N.,
- Yao G.,
- et al.
- Feldman A.M.,
- de Lissovoy G.,
- Bristow M.R.,
- et al.
- Linde C.,
- Gold M.,
- Abraham W.T.,
- Daubert J.C.
- Linde C.,
- Abraham W.T.,
- Gold M.R.,
- St John Sutton M.,
- Ghio S.,
- Daubert C.
- ↵Latimer NR, Abrams KR. NICE DSU Technical Support Document 16: Adjusting Survival Time Estimates in the Presence of Treatment Switching. NICE Decision Support Unit. 2014. Available at: http://www.nicedsu.org.uk/TSD16_Treatment_Switching.pdf. Accessed January 31, 2015.
- Hunt S.A.
- Vardas P.E.,
- Auricchio A.,
- Blanc J.J.,
- et al.
- Gold M.R.,
- Daubert J.C.,
- Abraham W.T.,
- et al.
- ↵Robins JM. The analysis of randomized and non-randomized AIDS treatment trials using a new approach to causal inference in longitudinal studies. Health Service Research Methodology: A Focus on AIDS. Eds: Sechrest L., Freeman H., Mulley A. Washington, D.C.: U.S. Public Health Service, National Center for Health Services Research., pp. 113-159 (1989). Available at: http://cdn1.sph.harvard.edu/wp-content/uploads/sites/343/2013/03/nchsr.pdf. Accessed April 27, 2015.
- Daubert C.,
- Gold M.R.,
- Abraham W.T.,
- et al.
- ↵National Institute for Health and Care Excellence. Gemcitabine for the Treatment of Metastatic Breast Cancer: TA116. Available at: http://www.nice.org.uk/guidance/ta116/resources/guidance-gemcitabine-for-the-treatment-of-metastatic-breast-cancer-pdf. Accessed January 31, 2015.
- National Institute for Health and Care Excellence. Sunitinib for the Treatment of Gastrointestinal Stromal Tumours: TA179. Available at: http://www.nice.org.uk/guidance/ta179/resources/guidance-sunitinib-for-the-treatment-of-gastrointestinal-stromal-tumours-pdf. Accessed January 31, 2015.
- National Institute for Health and Care Excellence. Pazopanib for the First-line Treatment of Advanced Renal Cell Carcinoma: TA215. Available at: http://www.nice.org.uk/guidance/ta215/resources/guidance-pazopanib-for-the-firstline-treatment-of-advanced-renal-cell-carcinoma-pdf. Accessed January 31, 2015.
- Solomon S.D.,
- Dobson J.,
- Pocock S.,
- et al.
- Cleland J.G.,
- Abraham W.T.,
- Linde C.,
- et al.