Estimating the Lifetime Benefits of Treatments for Heart Failure

Objectives This study compared ways of describing treatment effects. The objective was to better explain to clinicians and patients what they might expect from a given treatment, not only in terms of relative and absolute risk reduction, but also in projections of long-term survival. Background The restricted mean survival time (RMST) can be used to estimate of long-term survival, providing a complementary approach to more conventional metrics (e.g., absolute and relative risk), which may suggest greater benefits of therapy in high-risk patients compared with low-risk patients. Methods Relative and absolute risk, as well as the RMST, were calculated in heart failure with reduced ejection fraction (HFrEF) trials. Results As examples, in the RALES trial (more severe HFrEF), the treatment effect metrics for spironolactone versus placebo on heart failure hospitalization and/or cardiovascular death were a hazard ratio (HR) of 0.67 (95% confidence interval [CI]: 0.5 to 0.77), number needed to treat = 9 (7 to 14), and age extension of event-free survival +1.1 years (−0.1 to + 2.3 years). The corresponding metrics for EMPHASIS-HF (eplerenone vs. placebo in less severe HFrEF) were 0.64 (0.54 to 0.75), 14 (1 to 22), and +2.9 (1.2 to 4.5). In patients in PARADIGM-HF aged younger than 65 years, the metrics for sacubitril/valsartan versus enalapril were 0.77 (95% CI: 0.68 to 0.88), 23 (15 to 44), and +1.7 (0.6 to 2.8) years; for those aged 65 years or older, the metrics were 0.83 (95% CI: 0.73 to 0.94), 29 (17 to 83), and +0.9 (0.2 to 1.6) years, which provided evidence of a greater potential life extension in younger patients. Similar observations were found for lower risk patients. Conclusions RMST event-free (and overall) survival estimates provided a complementary means of evaluating the effect of therapy in relation to age and risk. They also provided a clinically useful metric that should be routinely reported and used to explain the potential long-term benefits of a given treatment, especially to younger and less symptomatic patients.

I n randomized controlled trials, the effect of treatment is usually estimated using a time-tofirst-event survival model that compares the hazard rate of the experimental treatment group (or groups) and the control group, which produces a hazard ratio (HR) and corresponding 95% confidence interval (CI) (1). By convention, if the upper 95% CI does not cross unity, the effect of treatment is considered statistically significant. Although this is the standard way of reporting the effect of treatment at medical presentations and in publications, it may exaggerate the effect of therapy (e.g., if the absolute risk reduction is small) and may not be readily interpretable by patients in terms of understanding their survival free of adverse clinical events, including death (2). Reporting the absolute treatment effect, as a percent reduction, reduction in event rate, or number needed to treat (NNT), overcomes the first of these criticisms (although NNT should be standardized for duration of follow-up). However, metrics of absolute benefit will generally look better in a highrisk than in a low-risk population, assuming the proportional risk reduction with the treatment is similar, at least in the relatively short follow-up that typifies most trials. Conversely, treatments started earlier in the course of a disease when patients are at lower risk (or even in younger patients) may have the potential to lead to greater prolongation of life. Another assessment of treatment effect that complements HR and absolute risk reduction or NNT, is the restricted mean survival time (RMST) (3,4). The RMST can be interpreted as the mean event-free survival time up to a pre-specified time point and is equivalent to the area under the Kaplan-Meier curve from the start of the study up to that point. Using age at randomization instead of time, the RMST approach allows for estimation of long-term, event-free survival that can be obtained with a specific intervention compared with a control group, across different age groups (5).
The RMST provides an estimate of the effect of treatment in terms of time "free of an event," years of life gained, or both. Such measures may be more readily interpretable and quantifiable for patients and clinicians. To better understand the use of RMST and how it compares with other conventional measures of treatment effect, we analyzed HR, NNT, and RMST in several large cardiovascular outcome trials.
We also analyzed these metrics in low-risk versus high-risk subgroups to illustrate how the RMST could provide relevant information that is less dependent on the risk of patients, providing a clinically relevant long-term outlook.  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).    Values are n, n (%), or median (interquartile range). The event-free survival was computed using the Kaplan-Meier survivor function over the full data and compared using the log-rank test. The number needed to treat (NNT) to benefit was computed from the cause-specific cumulative incidence functions. Median (25th to 75th percentile) follow-up time (days) and age at randomization (years): PARADIGM-HF: 810 days (564 to 1,069 days) and 64 years (57 to 72 years); EMPHASIS-HF: 639 days (292 to 992 days) and 68 yrs (63 to 74 years); RALES: 714 days (381 to 909 days) and 67 years (59 to 73 years); DIG: 1,152 days (843 to 1,423 days) and 65 years (57 to 71 years). *For consistency, the analysis the follow-up time was capped at 3 years. †For consistency, the RMST used age instead of time and used the same age range from 60 to 80 years in all the studied trials.   years (tau ¼ 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).

RESULTS
TRIALS ANALYZED. The overall results for each of the 4 trials analyzed are shown in Table 1 and Figure 1.   Table 2). The analysis was also performed using the trial median NT-pro BNP of 1, 615 pg/ml (Supplemental Table 1).
Among patients with a NT-pro BNP <1,000 pg/ml, the levels were 680 pg/ml (quartile 1 to quartile 3 [Q1 to Q3]: 527 to 827 pg/ml), and the mean AE SD age was 61.8 AE 11.2 years. The primary outcome was experi-  Among patients with NT-proBNP below the median, the levels were 888 pg/ml (Q1 to Q3: 642 to Values are n, n (%), or median (interquartile range). The event-free survival was computed using the Kaplan-Meier survivor function over the full data and compared using the log-rank test. The number needed to treat (NNT) to benefit was computed from the cause-specific cumulative incidence functions. *For consistency, the analysis using follow-up time was capped at 3 years. †For consistency, the restricted mean survival time (RMST) using age instead of time used the same age range from 60 to 80 years in the studied subgroups of the PARADIGM-HF (Angiotensin-Neprilysin Inhibition versus Enalapril in Heart Failure) trial, except for the age subgroups in which in patients younger than 65 years, the age range was 50 to 64 years and in patients age 65 or older, the age range was 65 to 80 years (i.e., tau ¼15). This metric is expressed in years. years.
The summary of the main characteristics of the RMST compared with absolute and relative risk metrics is provided in the Central Illustration.

DISCUSSION
In this study, we showed how analysis of RMST can complement conventional ways of describing the benefit of treatment in clinical trials, which expanded on our previous descriptions of using this metric (5,9).
We Higher-risk/more-symptomatic and older patients

Measures of Absolute Benefit RMST Difference Using Age at Risk
Higher event rate but potentially larger short-term absolute risk reduction This is not to diminish the importance of delaying or preventing nonfatal events and prolonging eventfree survival, which is possible even with treatments that do not alter all-cause mortality. This was illustrated by our analysis of the DIG trial, in which the gain in event-free survival was due to a reduction in hospitalizations for HF but not mortality.
STUDY LIMITATIONS. These were post hoc analyses. Our findings were derived from trial data, and their generalizability to a real-world population might be limited. Subgroup analyses might not always provide robust estimates of the true effect of a treatment.
Although we used NT-proBNP and NYHA functional class as proxies for risk, risk was a multivariable construct. The use of the RMST with age instead of follow-up time required a wide range of age in the population analyzed and a sufficiently large number of events across the age spectrum to provide relatively stable age-specific risk estimates. The proposed method made some major statistical assumptions and was therefore only suitable for the exploratory analyses in this study. The key assumption was that although a patient's risk of an event was related to their age and treatment group, it was not related to the length of time they spent in the study. Therefore, this methodology would not be suitable in studies in which the event rate was substantially elevated in the period shortly after randomization (e.g., in surgical trials or trials with a large variation in underlying patient risk). The proposed method would also be unsuitable in the presence of a competing risk (e.g., noncardiovascular death) that was either frequent or imbalanced between treatment groups.