PARAGON-HF: Heart Failure With Preserved EF Proves a Tough Foe

John M. Mandrola, MD


September 01, 2019

In a study of nearly 5000 patients who had heart failure with preserved ejection fraction (HFpEF), the effect size of sacubitril/valsartan vs valsartan to reduce heart failure hospitalizations or cardiovascular death did not reach statistical significance. We know better than to call a trial "negative" or "positive" based on a P value alone, but PARAGON HF confirmed the challenges of treating patients with HFpEF.

This was a global trial[1,2] of patients who had well-defined heart failure with a left ventricular ejection fraction (LVEF) greater than 45%, elevated brain natriuretic peptide, and documented structural heart disease. The investigators used equal-length run-in periods starting with valsartan, then sacubitril/valsartan, a protocol similar to that of previous HFpEF trials.[3,4,5,6]

Their choice of primary endpoint was unusual. While four previous trials in HFpEF used time to first heart failure hospitalization or cardiovascular (CV) death, the PARAGON-HF authors chose the composite of total heart failure admissions and death from cardiovascular causes.

They offered three reasons for the change: (1) because patients with HFpEF experience recurring hospital admissions, an endpoint that counts all admissions better reflects disease burden; (2) in the CHARM-Preserved[3] trial of candesartan vs placebo, the primary endpoint of time to first heart failure admission and CV death did not reach statistical significance, but a post hoc analysis that counted all hospital admissions did;[7] and (3) counting all admissions for heart failure increases the trial's statistical power.

The Main Results

Over a median follow-up of 35 months, there were 894 primary endpoint events in the sacubitril/valsartan arm and 1009 in the valsartan arm (12. 8 vs 14.6 per 100 patient-years). This corresponded to a relative risk (RR) reduction of 13% (RR, 0.87; 95% confidence interval [CI], 0.75 - 1.01; P = .06).

The numerically lower event rate stemmed mostly from fewer heart failure hospitalizations in the sacubitril/valsartan arm (690 vs 797; RR, 0.85 [95% CI, 0.71 - 1.00]). Death from CV causes was similar in both arms (8.5% vs 8.9% for valsartan alone; hazard ratio, 0.95 [95% CI, 0.79 - 1.16]).

The authors specified that "if the primary outcome reached significance," further analysis of the secondary endpoints looking at functional class was planned. That didn't happen, so one cannot put much weight on the slightly higher number of patients in the sacubitril/valsartan arm who had an improvement in functional class.

Similar numbers of patients in both arms stopped taking their assigned drug, but hypotension, as defined by systolic blood pressure less than 100 mm Hg, occurred in 15.8% of sacubitril/valsartan arm vs 10.8% of the valsartan arm. Angioedema occurred rarely but was threefold more common with sacubitril/valsartan (0.6% vs 0.2%). Worsening creatinine and hyperkalemia were slightly higher in the valsartan arm.

Two subgroups looked interesting. Patients with LVEF in the lower range (45% to 57%) and women had  lower rates of the primary endpoint on sacubitril/valsartan (RR, 24% and 34%, respectively); both these were significant for a treatment-covariate interaction.


I've learned the hard way that—even with large, well-conducted trials—it's difficult to know the exact treatment effect. In PARAGON-HF, the CIs for the primary endpoint range from a clinically meaningful 25% lower rate of events to no effect. The threshold P value is arbitrary. Truth: there is uncertainty.

There are other clues in the data. In the published supplement, the authors show the results of various analyses of the primary endpoint. If they used the time-to-first event method, as the four previous trials had done, the hazard ratio was decidedly less impressive at 0.92 (95% CI, 0.81 - 1.03) and sacubitril/valsartan would be added to the list of ineffective drugs for HFpEF without much debate.

Another clue pointing to a non–clinically meaningful effect of sacubitril/valsartan is the lack of signal for lower CV and overall death. PARAGON-HF enrolled older patients (mean age, 72 years) with multiple comorbidities. If sacubitril/valsartan is that much better than valsartan for reducing hospital admissions, why is there not a clear signal of lower rates of death from CV causes?

Then there is the question of subgroup effects. HFpEF is a heterogenous condition; it's tempting to see signal rather than noise in the observation that women and patients with LVEF in the lower range had more reduction in the primary endpoint. Again, maybe it is real, but the case for a true heterogenous effect would be much stronger had the overall results demonstrated clear benefit in both heart failure admissions and survival.

Clinicians deciding to adopt a new therapy must consider adverse effects and cost. Even after careful run-in periods and well-defined selection criteria, hypotension was more common with sacubitril/valsartan. Indeed, low blood pressure seems to be the biggest impediment to sacubitril/valsartan use in patients with systolic heart failure.

The significant cost difference in the two drugs matters. Health systems cannot print money. Money spent on one therapy means less for others. Spending for expensive drugs is fine if there are clear net benefits. The results of PARAGON-HF do not clear that bar.


Cardiology in 2019 boasts many excellent therapies. Contemporary medical therapy, such as high-dose valsartan, can be tough to beat in trials. What's more, I can't help but think much of HFpEF is due to long-term effects of exposure to life—in other words, aging. With that frame of reference, it's easier to understand the challenge of treating patients with this condition.

PARAGON-HF stands out as an example of clinical science done well. But as a clinician caring for older complex patients with HFpEF, I find its results do not persuade me to adopt a new therapy.

Editor’s Note: An earlier version of this column misstated the rate of hypotension as 0.8% in the valsartan arm; it was 10.8%. It also mistakenly listed valsartan as the drug in the CHARM trial instead of candesartan. Corrections have been made.



Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.