Estimating the Lifetime Benefits of Treatments for Heart Failure

João Pedro Ferreira, MD, PHD; Kieran F. Docherty, MBCHB; Susan Stienen, MD, PHD; Pardeep S. Jhund, MBBCH, PHD; Brian L. Claggett, PHD; Scott D. Solomon, MD; Mark C. Petrie, MBCHB; John Gregson, PHD; Stuart J. Pocock, PHD; Faiez Zannad, MD, PHD; John J.V. McMurray, MD


JACC Heart Fail. 2020;8(12):984-995. 

In This Article


Trials and Endpoints

The incidence rate and the effect of treatment illustrated as a relative risk reduction, absolute event difference, NNT, and RMST were calculated (see Statistical Analysis section) in 4 heart failure (HF) trials: PARADIGM-HF (Angiotensin–Neprilysin Inhibition versus Enalapril in Heart Failure), RALES (The Effect of Spironolactone on Morbidity and Mortality in Patients with Severe Heart Failure), EMPHASIS-HF Eplerenone in Patients with Systolic Heart Failure and Mild Symptoms-Heart Failure), and DIG (The Effect of Digoxin on Mortality and Morbidity in Patients with Heart Failure). A brief description of each of these trials and their outcomes is provided in the Supplemental Appendix. In this study, we examined 2 outcomes: 1) the composite of time to first occurrence of either HF hospitalization or cardiovascular death (used to estimate event-free survival); and 2) all-cause mortality (used to estimate overall survival). To homogenize the estimates, all follow-up times were capped at 3 years. The RMST, using age instead of time, was performed from 60 to 80 years in all trials, except for the age subgroups in PARADIGM-HF, in which in patients younger than 65 years, the age range was 50 to 64 years and in patients aged 65 or older, the age range was 65 to 80 years.

These trials were selected because they allowed comparison of the different means of quantifying the effect of treatment, as listed previously, and how patient characteristics could influence these. Because of its large sample size, we believed that the PARADIGM-HF trial would give reasonably robust estimates of the effect of treatment in subgroups (at least for the composite hospitalization and/or death outcome). This allowed us to test the hypothesis that patient risk and patient age would influence potential gains in event-free survival and overall survival, and specifically, that these gains would be smaller in higher risk and older patients compared with that of lower risk and younger patients. We tested this hypothesis by examining patients in PARADIGM-HF with a baseline N-terminal pro–B-type natriuretic peptide (NT-proBNP) level of <1,000 pg/ml (and also below the median of 1,615 pg/ml) versus ≥1,000 pg/ml (and equal or above the median), who were in New York Heart Association (NYHA) functional class I and/or II versus class III and/or IV and age younger 65 years versus age 65 years or older. In the subgroups, we focused on event-free survival, because there was more statistical power for this outcome than for all-cause mortality.

The RALES and EMPHASIS-HF trials provided an indirect comparison of how the effects of a mineralocorticoid receptor antagonist might be modified by patient risk. RALES was a trial that enrolled a high-risk advanced HF population (all patients were in NYHA functional class III and/or IV) who received suboptimal treatment by today's standards (e.g., 10% beta-blocker use). The EMPHASIS-HF trial randomized a lower risk, relatively well-treated patient cohort (all NYHA functional class II; 87% treated with a beta-blocker).

Because digoxin did not reduce all-cause mortality, the DIG trial was used to illustrate the effects of therapy on event-free survival versus overall survival.

All of these trials had similar patient age ranges and follow-up times.

Ethics approval was obtained for each individual trial.

Statistical Analysis

Cox proportional hazards models were used to estimate HRs and relative risk reductions. Event rates and the differences, with the respective 95% CIs were calculated using the quadratic approximation to the Poisson log-likelihood for the log-rate parameter. Event-free survival was computed using the Kaplan-Meier survivor function over the full data and compared using the log-rank test. The NNT to benefit was computed from the cause-specific cumulative incidence functions. The RSMT is a measure of average event-free survival from time 0 to a pre-specified time point.[6] We computed the RMST using the within-trial follow-up time and also used age instead of time (the age-specific event rates were then estimated). These estimations were obtained by multiplying the annual conditional survival probabilities of patients included in the studies, starting from a specific age, from projections of the expected duration of event-free survival (up to a pre-specified time horizon [tau]), and were non-parametrically estimated by calculation of the area under the Kaplan-Meier survival curve.[5,7] Subsequently, differences in the estimated duration of event-free survival for patients treated with active drug and treatment versus those in a placebo and/or control group were interpreted as mean years of life gained (or lost). In other words, for any given age, a survival curve was estimated that represented the survival probabilities over time for patients alive at that age and who receiving an active drug and/or treatment. A corresponding survival curve was then estimated using data from patients who were in the placebo and/or control group. From each survival curve, the average time spent before the event of interest was estimated by the area under the survival curve. Because patient age and treatment assignment were independent of one another due to randomization, the difference in the areas under the survival curve could be interpreted as the effect of treatment on time spent event free. The key statistical assumption required to obtain valid long-term survival projections based on short-term follow-up data was that a patient's risk for a given event depended on age and treatment but not on the duration of exposure to the treatment.[5,8] This meant that a 79-year-old patient who took sacubitril-valsartan since the age of 60 years (i.e., 19 years of treatment exposure) had the same assumed risk of dying by age 80 years as a 79-year-old patient who started taking sacubitril-valsartan at age 78 years (i.e., 1 year of treatment exposure). For consistency, we assessed the survival data over a maximum follow-up of 3 years and an age range from 60 to 80 years (tau = 20) in all trials. In the PARADIGM-HF trial, age projection estimates were also performed for the previously described subgroups described. The p values <0.05 were considered statistically significant. All analyses were conducted using Stata version 16 (StataCorp, College Station, Texas).