The inert arm of this trial was the dangerous one. Serious adverse events were recorded in 35 of 263 participants on semaglutide and 71 of 266 on placebo — 13.3% against 26.7% — in a population where a serious adverse event is most often the disease itself arriving at a hospital.[1] That inversion is the shape of heart failure with preserved ejection fraction, and it is why a trial measuring a questionnaire rather than a death rate still changed practice. What the two heart-failure syndromes are and why this class helps one of them is set out in the heart failure article; this page is about how STEP-HFpEF was built and what its own numbers refuse to say.
A questionnaire and a scale, promoted to primary
STEP-HFpEF randomized 529 patients with heart failure, a left ventricular ejection fraction of at least 45% and a body-mass index of 30 or higher to once-weekly semaglutide 2.4 mg or placebo for 52 weeks. Entry also required New York Heart Association class II to IV symptoms and a Kansas City Cardiomyopathy Questionnaire clinical summary score below 90, so nobody already feeling well could be enrolled. Diabetes, or a glycated hemoglobin of 6.5% or higher, was an exclusion; the companion trial took that population instead.[1][3]
The dual primary endpoints were the change in that questionnaire score and the percentage change in body weight. Confirmatory secondaries were the six-minute walk distance, a hierarchical composite of death, heart-failure events and differences in score and walk distance, and the change in C-reactive protein.[1] Notice what is absent from that list: no arm of it is a count of hospitalizations or deaths. The trial was designed to measure how people feel and what they weigh, and those are the only two things it was sized to settle.
Both endpoints moved, by very different amounts
The symptom score rose 16.6 points on semaglutide and 8.7 on placebo, an estimated difference of 7.8 points (95% CI, 4.8 to 10.9; P < 0.001), while mean body weight changed −13.3% against −2.6%, a difference of −10.7 percentage points (95% CI, −11.9 to −9.4). Walk distance improved 21.5 m against 1.2 m, the hierarchical composite returned a win ratio of 1.72 (95% CI, 1.37 to 2.15), and C-reactive protein fell 43.5% against 7.3% — an estimated treatment ratio of 0.61 (95% CI, 0.51 to 0.72).[1]
Retention was near-total. The registry records 256 of 263 completing on semaglutide and 254 of 266 on placebo, with three deaths in the drug arm and four in the placebo arm.[7] Nausea was reported by 45 of 263 against 7 of 266, and acute kidney injury appeared as a serious event in 5 against 1 — a small count running against the drug inside a safety table that otherwise runs the other way.[7]
The companion trial breaks the obvious explanation
STEP-HFpEF DM ran the same protocol in 616 participants who also had type 2 diabetes. Weight fell 9.8% against 3.4%, a difference of 6.4 percentage points, and the symptom score rose 13.7 points against 6.4, a difference of 7.3 points (95% CI, 4.1 to 10.4).[2] Put the two trials side by side and the weight advantage shrank by four percentage points while the symptom advantage barely moved.
The prespecified pooled analysis of all 1,145 participants states this formally. Treatment effects were consistent across the two trials for the symptom score, the walk distance, the hierarchical composite and C-reactive protein. For weight they were not: the test for interaction between trials returned P < 0.0001 with an I² of 95.77%.[3] Two endpoints measured in the same people at the same visit, and only one of them behaved as though the trials were interchangeable.
The subgroup results point the same way. Pooled, the symptom effect varied significantly by baseline NT-proBNP concentration, by loop diuretic use, by renin–angiotensin–aldosterone system inhibitor use and by a history of atrial fibrillation — participants with the more severe phenotype gained more on the questionnaire despite similar weight loss. Weight reduction varied by sex and by race instead.[3]
The analysis that argues the other side
A prespecified analysis of STEP-HFpEF alone reports the association that every summary of this trial quotes. Among semaglutide-treated participants, each 10% reduction in body weight came with a 6.4-point improvement in the symptom score (95% CI, 4.1 to 8.8) and a 14.4-meter improvement in walk distance (95% CI, 5.5 to 23.3), with benefits consistent across obesity classes I, II and III and no significant interaction with baseline body-mass index.[4] Its authors read that as support for weight loss being the mechanism.
Both readings are drawn from the same trial and they are not the same kind of evidence. The dose-response runs within the treated arm, comparing people who lost more weight against people who lost less — a comparison the randomization does not protect, because how much weight a person lost is itself an outcome. The between-trial and between-subgroup contrasts are the randomized ones, and they show the symptom benefit holding steady while the weight effect moves. A careful statement is that weight loss explains part of this result and is not established as explaining all of it.
Women lost more weight and felt the same amount better
A prespecified secondary analysis split the pooled cohort by sex. Of 1,145 participants, 570 (49.7%) were women, who entered with higher body-mass index, higher C-reactive protein and worse symptoms than the men. Semaglutide reduced body weight more in women — −9.6% (95% CI, −10.9 to −8.4) against −7.2% (95% CI, −8.4 to −6.0), P for interaction = 0.006. The symptom score moved identically: +7.6 points (95% CI, 4.5 to 10.7) in women and +7.5 points (95% CI, 4.3 to 10.6) in men, P for interaction = 0.94.[5]
That is a third instance of the same dissociation, this time inside one pooled population rather than between two trials. A 2.4-point difference in weight loss produced a 0.1-point difference in how people felt.
Who stopped, in a trial where quitting was rare
Pooled across both trials, gastrointestinal events led to discontinuation of study treatment 60 times in the semaglutide group — 10.7 per 100 person-years — against 19 in the placebo group, 3.3 per 100 person-years.[3] Serious adverse events numbered 161 on semaglutide (28.7 per 100 person-years) against 301 on placebo (52.7 per 100 person-years), with fewer serious cardiac disorders and fewer infections on the drug.[3]
Both facts are true at once and they answer different questions. Roughly three times as many people abandoned the drug for a gut symptom as abandoned placebo, and roughly half as many had something serious happen to them. Anyone reading a tolerability figure from a weight-loss trial and applying it to this population is comparing against a placebo arm that behaves nothing like the one in the STEP 1 article.
What it was never powered to show
Heart-failure events were counted, not tested. Pooled, adjudicated hospitalization or an urgent visit for heart failure occurred in 8 of 573 participants on semaglutide (1%) against 30 of 572 on placebo (5%), a hazard ratio of 0.27 (95% CI, 0.12 to 0.56); the composite of cardiovascular death or a heart-failure event returned 0.31 (95% CI, 0.15 to 0.62), and the total-event analysis 12 against 41, a mean ratio of 0.30 (95% CI, 0.14 to 0.64).[3] Thirty-eight events across 1,145 people is a signal, not a settled question, and these were exploratory endpoints in trials whose primary outcomes were a questionnaire and a scale.
A larger post hoc analysis assembled every participant with a history of this syndrome from four trials — 3,743 of 22,282, drawn from SELECT, FLOW and the two STEP-HFpEF trials. The combined endpoint of cardiovascular death or a worsening heart-failure event occurred in 103 of 1,914 (5.4%) against 138 of 1,829 (7.5%), a hazard ratio of 0.69 (95% CI, 0.53 to 0.89; P = 0.0045), and worsening heart-failure events alone returned 0.59 (95% CI, 0.41 to 0.82). Cardiovascular death on its own did not move: 59 (3.1%) against 67 (3.7%), a hazard ratio of 0.82 (95% CI, 0.57 to 1.16; P = 0.25).[6] A composite that reaches significance while its mortality component does not is the standard reading trap in this literature, and the trials that do carry mortality arms are described in the cardiovascular article and the FLOW article.
Who the result describes
Pooled median age was 69 years, the cohort split evenly by sex, and 1,026 of 1,145 (90%) participants were White. Sixty-five percent had a body-mass index of 35 or higher and 69% were in New York Heart Association class II. At least three quarters of each group were already taking beta blockers, diuretics and renin–angiotensin system inhibitors, and median follow-up was 401 days.[3]
Every one of those people arrived with an ejection fraction measured on an echocardiogram, a symptom score administered in a clinic and a cardiologist’s diagnosis. None of those exists on a telehealth intake form, and the eligibility floor that a questionnaire score had to sit below 90 means the trial deliberately excluded anyone whose symptoms were mild.
The product was branded, FDA-approved semaglutide 2.4 mg, supplied and supervised. Sellers on the compounded semaglutide board dispense preparations that are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed. A symptom result earned in a 529-patient cardiology trial does not follow a molecule name into a vial, and the gap that separates the two is set out in the compounded-versus-brand article.