Association Between Use of Renin-Angiotensin System Antagonists and Mortality in Patients With Heart Failure and Preserved Ejection Fraction FREE

Up to half of patients with heart failure have normal or near-normal ejection fraction,¹ termed heart failure with preserved ejection fraction (HFPEF) or diastolic heart failure. The mortality in HFPEF may be as high as in heart failure with reduced ejection fraction (HFREF) or systolic heart failure,¹ but there is no proven therapy.

Whether HFPEF and HFREF are distinct or similar disorders remains controversial,^2- 3 but there is considerable overlap, with elevated filling pressures and classic heart failure signs and symptoms.⁴ For example, diastolic dysfunction and impaired systolic contractility occur in both HFREF and HFPEF.⁵ Renin-angiotensin-aldosterone system and other neurohormonal activation occurs in both HFREF and HFPEF.^4,6 Renin-angiotensin-aldosterone system is also involved in precursors to and processes associated with HFPEF, such as hypertension, vascular dysfunction, and myocardial hypertrophy and fibrosis,^6- 8 which lead to increased stiffness, impaired relaxation, and diastolic dysfunction.^6- 7 Renin-angiotensin system (RAS) antagonists (angiotensin-converting enzyme [ACE] inhibitors and angiotensin receptor blockers) inhibit the maladaptive remodeling associated with both HFREF and HFPEF and their precursors,^6,8 and improve outcomes in HFREF.^9- 10

These observations provide a rationale for RAS antagonists in HFPEF. However, in 3 randomized controlled trials (RCTs),^11- 13 RAS antagonists did not improve primary outcomes in HFPEF. However, there were signals toward benefits in primary outcomes or significant benefits in secondary outcomes. Selection bias, underpowering, or high crossover rates may have concealed a real benefit.¹⁴ Therefore, we hypothesized that RAS antagonists are associated with reduced mortality in a broad unselected population with HFPEF.

The Swedish Heart Failure Registry has been described.^15- 16 The inclusion criterion was clinician-judged heart failure. Eighty variables were recorded at discharge from hospital or outpatient visit and entered into a web-based database managed by the Uppsala Clinical Research Center in Sweden. The database is run against the Swedish death registry monthly. The protocol, registration form, and annual report are available at http://www.rikssvikt.se (in Swedish). Establishment of the registry and analysis of data were approved by a multisite ethics committee. Individual patient consent was not required, but patients were informed of entry into national registries and allowed to opt out. The registry and this study conform to the Declaration of Helsinki.

Propensity scores for treatment with RAS antagonists were estimated for each patient with logistic regression using 43 clinically relevant baseline variables. The propensity score is the propensity from 0 to 1 to receive a treatment given a set of known variables and is used to attempt to adjust for potential selection bias, confounding, and differences between treatment groups in observational studies.^17- 18 Continuous variables, including N-terminal pro-brain natriuretic peptide, were modeled using restricted cubic splines in the logistic regression. Missing data were handled by estimating separate logistic regressions for each missing variable pattern on all available observations. Each individual then received the propensity score that incorporated all nonmissing variables for that individual.

We constructed an age-matched and propensity score–matched cohort.¹⁹ The matching may be considered a form of adjustment for the propensity score (and the 43 variables used to derive the propensity score). Matching was 1:1 and without replacement, with each untreated patient matched to the closest treated patient in which age differed by 5 years or less and the propensity score differed by 0.1 or less. This yielded 3329 patients in each group.

Crude survival by RAS antagonist use (yes vs no) in the overall cohort was charted using the Kaplan-Meier method and compared in a univariable Cox regression. Thereafter, we performed several adjusted Cox regression analyses. The primary analysis was the matched cohort. Survival for the matched cohort was charted in the same figure as the crude survival in the overall cohort. Because propensity score matching excludes patients and impairs generalizability, we performed a secondary analysis in the overall cohort with adjustment for propensity score as a continuous covariate. Because higher age and lower creatinine clearance are associated with increased mortality and are also common reasons to avoid RAS antagonists, we also performed separate analyses with adjustment for these 2 variables alone in the overall cohort.

The scaled Schoenfeld residuals and difference of the freedom β from the models were investigated to detect violations to the proportional hazards assumption and possible influential outliers, respectively; none were found. Continuous variables, including the propensity score, were modeled using restricted cubic splines. Interactions between RAS antagonist use and ejection fraction and other selected baseline variables in the matched cohort were estimated by Cox regression and displayed in a forest plot.

To test the null hypothesis that there is no difference in mortality between use and nonuse of RAS antagonists, 3170 patients or more would be needed in the smallest group. This assumed a hazard ratio (HR) for mortality associated with use of RAS antagonists of 0.90, 1-year mortality of 20%, an accrual period of 7 years (from our population), a power level of 80%, and 2-sided significance level of .05. The HR of 0.90 was conservatively estimated from the 95% confidence interval for the primary end points in the Perindopril in Elderly People with Chronic Heart Failure (PEP-CHF) study at 1 year (95% CI, 0.474-1.010),¹² in the Candesartan in Heart Failure: Assessment of Reduction in Mortality and Morbidity (CHARM) preserved study throughout follow-up (95% CI, 0.77-1.03),¹¹ and in the CHARM overall study¹⁰ throughout follow-up (95% CI, 0.83-1.00) without heterogeneity with respect to ejection fraction. The 20% 1-year mortality was based on 1-year mortality of greater than 20% in previous epidemiological studies of HFPEF^20- 21 and our overall registry.¹⁵

Propensity scores were calculated from known and measured potential confounders (n = 43 in our study). However, unmeasured confounders may affect the results if they (1) are unrelated to or not fully accounted for by measured confounders, (2) affect the decision to prescribe a RAS antagonist, and (3) are associated with mortality independent of RAS antagonist use.^22- 24 Therefore, we quantified the effects of hypothetical unmeasured confounders in the matched cohort^16,25 (eTables 1-2. In patients with HFREF, use of a higher dose of a RAS antagonist is associated with greater benefit.^26- 27 Therefore, we performed an analysis in our patients with HFPEF who were matched by age and propensity score for a target dose of 50% or greater and a target dose of less than 50% vs no treatment (eTables 3 and eFigures 1–2).

In patients with HFREF, use of RAS antagonists has been shown to reduce mortality.^9- 10 Therefore, we performed a positive-control analysis of the association between use of RAS antagonists and mortality in patients with HFREF in the same registry (eFigure 1 and eFigure 3).

The level of significance was set at .05 and all reported P values and 95% confidence intervals are 2-sided. Statistical analyses were performed using R version 2.14.2 (R Foundation for Statistical Computing).

Between May 11, 2000, and October 10, 2011, there were 64 140 registrations from 64 of 77 hospitals and 84 of 1011 primary care outpatient clinics in Sweden. We included 16 216 first registrations with an ejection fraction of 40% or greater for the main analyses and the dose consistency analysis. We included 20 111 first registrations with ejection fraction of less than 40% for the HFREF consistency analysis.

Table 1 and Table 2 show baseline characteristics and standardized differences between use of RAS antagonists (yes vs no). Of 16 216 patients, 12 543 (77%) did receive RAS antagonists and 3673 (23%) did not receive RAS antagonists. Of the 12 543 patients, 73% received an ACE inhibitor, 25% received an angiotensin receptor blocker, and 2% received both. Specific agents and doses are listed in eTables 3.

In the overall population, there were several differences between the groups, generally consistent with better health and more specialized care and planned follow-up in patients receiving RAS antagonists. The distribution of propensity scores was therefore different. In the matched cohort, the standardized differences between the groups were considerably smaller and few were statistically significant. The distribution of propensity scores was therefore similar, with lower propensity scores for treated patients and higher propensity scores for untreated patients compared with the overall cohort (Figure 1).

The patients who were excluded due to missing ejection fraction or RAS antagonist treatment (eFigure 1) had a mean (SD) age of 82 (10) years, 53% were women, and had overall worse health and less treatment with neurohormonal antagonists (eTables 4–5).

In the overall HFPEF cohort, crude 1-year survival was 86% (95% CI, 86%-87%) for patients receiving RAS antagonists vs 69% (95% CI, 68%-71%) for patients not receiving RAS antagonists; and 5-year survival was 55% (95% CI, 54%-56%) vs 32% (95% CI, 30%-34%), respectively (unadjusted HR, 0.48 [95% CI, 0.45-0.51]; P < .001) (Figure 2 and Table 3).

In the matched HFPEF cohort, 1-year survival was 77% (95% CI, 75%-78%) for treated patients vs 72% (95% CI, 70%-73%) for untreated patients. Five-year survival was 36% (95% CI, 34%-38%) vs 34% (95% CI, 32%-37%) (Figure 2). The HR was 0.91 (95% CI, 0.85-0.98; P = .008). In the overall cohort adjusted for propensity score, the HR was 0.90 (95% CI, 0.85-0.96, P = .001).

Figure 3 and Figure 4 show the HRs associated with RAS antagonists in the matched cohort for clinically relevant subgroups. Use of RAS antagonists interacted significantly only with mean blood pressure. There was no interaction with ejection fraction (P = .12), but the HRs were nominally different in patients with ejection fractions of 40% to 49% (HR, 0.85 [95% CI, 0.76-0.95]; P = .004) vs patients with ejection fractions of 50% or greater (HR, 0.95 [95% CI, 0.87-1.04]; P = .26).

eTables 1-2 show quantified effects of hypothetical confounders in the matched cohort. To invalidate our findings (ie, for RAS antagonists not to be significantly associated with reduced mortality), a binary confounder with a HR for mortality of 1.1 would have to be present in at least 40% (absolute) more in untreated patients vs treated patients. A normally distributed continuous confounder would have to be at least 2 units higher in untreated patients and be associated with a HR of at least 1.03 per unit increase.

In the analysis of propensity for dose-matched HFPEF, 1-year survival was 82% (95% CI, 80%-83%) for high dose, 80% (95% CI, 78%-81%) for low dose, and 76% (95% CI, 75%-78%) for no treatment; and 5-year survival was 43% (95% CI, 40%-46%), 39% (95% CI, 36%-42%), and 40% (95% CI, 37%-42%), respectively (eFigure 2, matched cohort). The HR was 0.85 (95% CI, 0.78-0.83) for high dose vs no treatment (P < .001) and 0.94 (95% CI, 0.87-1.02) for low dose vs no treatment (P = .14; Table 3).

In the analysis of propensity for treatment-matched HFREF, 1-year survival was 67% (95% CI, 64%-69%) for patients receiving RAS antagonists vs 58% (95% CI, 56%-60%) for patients not receiving RAS antagonists; and 5-year survival was 29% (95% CI, 26%-32%) vs 23% (95% CI, 21%-26%), respectively (eFigure 3, matched cohort) (HR, 0.80 [95% CI, 0.74-0.86]; P < .001; Table 3).

In this large, prospective registry of unselected patients with HFPEF, use of RAS antagonists was associated with reduced all-cause mortality. Our findings were consistent, with narrow confidence intervals, in a cohort matched based on age and propensity score and in the overall cohort after adjustment for propensity score. Our findings were strengthened by a positive dose-response relationship and by a separate analysis in patients with HFREF, in which the associated mortality reduction was consistent with published data.

The 77% rate of patients receiving RAS antagonists was as high as the overall population (including patients with HFREF) in our registry,¹⁵ and higher than the 44% to 62% rate in patients with HFPEF and similar to the 75% to 82% rate in patients with HFREF in large surveys.^2- 3 In the unadjusted analysis, there was a large separation between the survival curves and use of RAS antagonists associated with an HR for death of only 0.48. Because treated patients were younger and had better renal function, this benefit was attenuated after adjustment for age or creatinine clearance alone, and because treated patients were healthier in other regards, the benefit was further attenuated but remained significant and consistent in the matched cohort (HR 0.91; 95% CI, 0.85-0.98) and in the overall cohort after adjustment for propensity scores (HR 0.90; 95% CI, 0.85-0.96). Furthermore, in the matched cohort, the mean age was 79 years and the mean creatinine clearance was 52 to 53 mL/min, with a similar benefit to the overall cohort. There was no interaction with age or creatinine clearance, suggesting that age or renal insufficiency alone should not be contraindications to use of RAS antagonists.

Our findings differ from those of 3 previous RCTs: CHARM preserved,¹¹ PEP-CHF,¹² and Irbesartan in Heart Failure with Preserved Ejection Fraction Study (I-PRESERVE).¹³ However, 3 observations suggest that RAS antagonists may nevertheless be beneficial in patients with HFPEF: (1) signals toward benefit in these RCTs, (2) potential selection bias and underpowering in these RCTs; and (3) potentially more severe heart failure and a more unselected and representative population in our study.

In the CHARM preserved study, candesartan did not reduce the primary end point (cardiovascular death or hospitalization for heart failure) but covariate-adjusted hospitalization for heart failure was reduced. In the CHARM overall study, the covariate-adjusted primary end point (all-cause mortality) was reduced without heterogeneity across the component trials, including the CHARM preserved study.¹⁰ In the PEP-CHF study, there was a nominally large and borderline significant benefit in the primary end point at 1 year, but 26% of patients randomized to the placebo group crossed over to open-label treatment with an ACE inhibitor and the benefit was not sustained. In the I-PRESERVE study, irbesartan was not beneficial but the authors suggested underpowering, citing a 34% drug discontinuation and a 40% concomitant use of ACE inhibitors. In a meta-analysis of these trials, the reduction in combined cardiovascular mortality and hospitalization for heart failure reached significance,²⁸ confirming the hypothesis made by us and others¹⁴ that the individual trials were underpowered. Another meta-analysis suggested reduced all-cause mortality.²⁹ In a previous study from our registry, candesartan was associated with lower mortality than losartan, potentially due to greater angiotensin II receptor antagonism and supporting the role of RAS antagonists in patients with HFPEF.¹⁶

In these RCTs, patients were considerably healthier than in epidemiological surveys and in our study, which is consistent with documented systematic selection bias and poor generalizability in RCTs.³⁰ Mean or median age ranged from 67 to 75 years, with a majority having a New York Heart Association classification of I or II, only 17% to 30% had atrial fibrillation, with mean or median renal function near normal, and significant renal insufficiency as exclusion criterion. In contrast, in our matched cohort, mean age was 79 years, most had a New York Heart Association classification of II or III, more than 50% had atrial fibrillation and mean creatinine clearance was 52 to 53 mL/min.

In the RCTs, patients also had milder heart failure and possibly no heart failure at all. Median N-terminal pro-brain natriuretic peptide level was only 335 to 453 pg/mL in the PEP-CHF study and 320 to 360 pg/mL in the I-PRESERVE study compared with 2464 to 2522 pg/mL in our study. The European Society of Cardiology guidelines suggest that N-terminal pro-brain natriuretic peptide level of less than 300 pg/mL makes heart failure unlikely.³¹ N-terminal pro-brain natriuretic peptide is proportional to the severity of heart failure.³² In the I-PRESERVE study, 80% had a New York Heart Association classification of III to IV, surprisingly severe for such a low N-terminal pro-brain natriuretic peptide level and suggesting other reasons than heart failure for poor functional status and prognosis.

In the RCTs, healthier patients may also explain the low outcome rates, leading to underpowering. In the CHARM preserved study, only 11% in both groups experienced cardiovascular death over 37 months. In the PEP-CHF study at 1 year, only 4.0% vs 4.5% died. In the I-PRESERVE study, 11% died in both groups over 49.5 months. This is in stark contrast to epidemiological surveys with rates of approximately 25% for 1-year mortality and 60% for 5-year mortality. In our study, 1-year mortality ranged from 11% to 31% and 5-year mortality ranged from 39% to 68%, depending on cohort and treatment. In addition, the close monitoring and specialist follow-up in RCTs may dilute the benefits of RAS antagonists, contributing to underpowering, whereas in our study, a majority of patients were cared for by an internal medicine physician or a geriatrician and follow-up referral was to primary care in half of the patients.

Although our study may be more representative and generalizable than the RCTs, patients excluded due to unknown ejection fraction or treatment with RAS antagonist were (as expected) older, more frequently women, and overall less healthy. This suggests that even in our less selective registry, we cannot completely eliminate the selection bias and limited generalizability that may occur in RCTs.³⁰

Observational studies are subject to confounding and may exaggerate benefits seen in RCTs.^22- 24 Therefore, we performed a dose-response analysis in the HFPEF cohort and a positive-control consistency analysis in patients with HFREF. Specific dosing trials suggest greater benefit with higher dose in patients with HFREF.^26- 27 In our HFPEF cohort, we observed a proportional associated reduction in mortality with doses of 50% or greater of target vs less than 50% of target vs no treatment, suggesting patients with HFPEF may respond similarly to patients with HFREF.

In our HFREF analysis, the HR for mortality of 0.80 (ie, 20% reduction in mortality) was similar to previous RCTs and precisely matches that seen in a large meta-analysis in HFREF,⁹ confirming that our registry and findings are representative of other rigorous settings.

The HR range of 0.90 to 0.91 in patients with HFPEF represents a lower reduction in all-cause mortality than in patients with HFREF. This may reflect a lower degree of RAS and aldosterone activation in patients with HFPEF,⁶ lower all-cause^2- 3 or cardiovascular³³ mortality in patients with HFPEF, or higher age, comorbidity, and noncardiovascular mortality in patients with HFPEF,^20- 21,33 which is less likely to be prevented by RAS antagonists.

While the prognosis in patients with HFPEF may be similar^20- 21 or better^2- 3 than in patients with HFREF, once ejection fraction is above 40%, it does not affect prognosis further.^2,34 In the HFPEF cohort, there was no heterogeneity with regard to ejection fraction, but the nominal benefit of RAS antagonists was greater in patients with ejection fractions of 40% to 49% than in those with ejection fractions of 50% or greater. Together with the yet greater benefit in our HFREF cohort and other HFREF studies,⁹ this suggests that heart failure may be on a continuous ejection fraction spectrum, with greater benefit of RAS antagonists the lower the ejection fraction. Currently, RAS antagonists are recommended only in patients with ejection fractions of less than 40%.³¹ Even though ejection fractions of 40% to 49% may not be considered normal, the benefit in this group has not previously been demonstrated.

In the hierarchy of strengths of evidence, RCTs are superior to observational studies.^22- 24 An observational registry study is subject to confounding.²⁴ Our study contains matching and propensity score adjustment for most variables that we are aware may affect the choice to prescribe a RAS antagonist (bias) and to affect outcome independent of RAS antagonist (confounding). However, we cannot rule out residual confounding from unknown or unmeasured variables.²⁴ To invalidate our findings, such confounding would have to be of reasonable magnitude, associated with mortality independently of all variables measured in our study and independently of treatment, and likely to affect the choice to prescribe a RAS antagonist. Although RCTs eliminate bias and confounding, they may have limited generalizability and may be complemented by rigorous registry studies with greater power.^22,24

As in the RCTs discussed,^11- 13 heart failure was a clinical diagnosis, and although median N-terminal pro-brain natriuretic peptide level was high, we cannot be completely sure that all patients had heart failure. Dosing of RAS antagonists was suboptimal, but we observed a dose response. Treatment may have changed over time, but this should, if anything, underestimate the true benefit. We assessed all-cause mortality but not cause of death, hospitalization, or cause of hospitalization. In the 3 RCTs, mortality in both treated and untreated patients with HFPEF was mainly cardiovascular.^12,35- 36 The relative effect of RAS antagonists on all-cause mortality vs cardiovascular mortality remains to be determined. One potential explanation for the benefits of RAS antagonists may be blood pressure lowering, as seen in studies not on heart failure.^37- 38 In our study, there was a significant interaction yielding a greater benefit with use of RAS antagonists and lower mean blood pressure. Although this may be due to chance, it may suggest that RAS antagonists are more effective in sicker patients (eg, those with lower blood pressure and ejection fraction).

There is currently no consensus on the use of RAS antagonists in patients with HFPEF. In our study, use of RAS antagonists was associated with reduced all-cause mortality in a broad unselected population of patients with HFPEF. Our results together with the signal toward benefit in RCTs suggest that RAS antagonists may be beneficial in patients with HFPEF, but this should be confirmed in an appropriately powered randomized trial.