A lower number on a scale is not the same result as fewer heart attacks. Cardiovascular outcomes trials are designed to answer the latter kind of question, usually by following thousands of people for years. They provide a different level of evidence from changes in weight, blood pressure or laboratory markers. Understanding that difference helps readers recognise both meaningful findings and claims that reach beyond them.
This review considers five randomised human studies of pharmaceutical GLP-1 receptor agonists. Each asks a different question about cardiovascular events, the population studied and the outcome measured.
First identify the event being counted
In these trials, the primary composite commonly combines cardiovascular death, non-fatal heart attack and non-fatal stroke. Combining events makes a study more statistically informative, but creates an interpretation problem: a positive composite does not prove that each component improved separately. A reader should check both the overall result and what contributed to it.
SELECT provides a useful example. It enrolled 17,604 adults with established cardiovascular disease and overweight or obesity, without diabetes. Over mean follow-up of 39.8 months, a primary event occurred in 6.5% with semaglutide versus 8.0% with placebo. The hazard ratio was 0.80. Adverse events caused permanent treatment discontinuation in 16.6% versus 8.2%. That safety and persistence information belongs alongside the benefit. SELECT publication.
Relative and absolute effects are complementary
For SELECT, the observed event proportions differ by 1.5 percentage points. A hazard ratio describes the relative event rate over follow-up; it is not the same calculation as subtracting those proportions. Saying only “20% lower” omits the starting risk and timescale. Saying only “1.5 points” omits the relative comparison. Both can be informative when their meanings remain clear.
LEADER randomised 9,340 people with type 2 diabetes and high cardiovascular risk. During median follow-up of 3.8 years, its primary composite occurred in 13.0% with liraglutide and 14.9% with placebo, with a hazard ratio of 0.87. The displayed event proportions therefore differ by 1.9 percentage points. This was evidence in a diabetes population with substantial existing risk; it did not answer the same population question later addressed by SELECT. LEADER publication.
The statistical question matters
SUSTAIN-6 assigned 3,297 people with type 2 diabetes to semaglutide or placebo for 104 weeks. Primary events occurred in 6.6% versus 8.9%, with a hazard ratio of 0.74. Its prespecified primary testing framework was non-inferiority. The trial also found more retinopathy complications with semaglutide: hazard ratio 1.76, with a 95% confidence interval of 1.11–2.78. A favourable cardiovascular result does not cancel a different safety signal. SUSTAIN-6 publication.
Non-inferiority asks whether an intervention stays within a predefined limit of being worse than its comparator. Superiority asks whether it performs better on the tested endpoint. Confidence intervals, margins and the statistical testing order determine what conclusions are supported. Treating every favourable point estimate as the same kind of proof loses that structure.
Broader eligibility changes applicability
REWIND included 9,901 people with type 2 diabetes and either a prior cardiovascular event or cardiovascular risk factors. Median follow-up was 5.4 years. Its primary composite occurred in 12.0% with dulaglutide and 13.4% with placebo, a 1.4-percentage-point observed difference; the hazard ratio was 0.88. Gastrointestinal adverse events were reported by 47.4% versus 34.1%. The wider entry criteria improve the range of people studied, without making the results universal. REWIND publication.
Route and formulation require direct evidence too
SOUL tested oral semaglutide in 9,650 people with type 2 diabetes and cardiovascular disease, chronic kidney disease or both. Over median follow-up of 49.5 months, the primary event proportions were 12.0% and 13.8%, with a hazard ratio of 0.86. Serious adverse events occurred in 47.9% and 50.3%, respectively. The confirmatory secondary kidney outcomes did not differ significantly. This supports a specific cardiovascular conclusion, not a promise that every organ outcome improves. SOUL publication.
Why a league table would mislead
The absolute differences above come from different populations, background care and observation periods. Ranking medicines by those differences would compare unlike experiments. None of these placebo comparisons directly randomised all the agents against one another. Likewise, a result in people with established disease should not be silently repackaged as prevention evidence for young, low-risk people.
The useful habit is to keep five details beside every headline: who entered the trial, what event counted, how long follow-up lasted, how large the effect was and what harms occurred. That makes cardiovascular research easier to interpret alongside the choices that shaped each trial.
Continue with How to read a clinical trial for a closer look at interpreting research evidence.
Sources and further reading
- SELECT: Semaglutide and cardiovascular outcomes without diabetes — Peer-reviewed randomised human outcomes trial; 2023-11-11.
- LEADER: Liraglutide and cardiovascular outcomes — Peer-reviewed randomised human outcomes trial; 2016-06-13.
- SUSTAIN-6: Semaglutide and cardiovascular outcomes — Peer-reviewed randomised human outcomes trial; 2016-09-15.
- REWIND: Dulaglutide and cardiovascular outcomes — Peer-reviewed randomised human outcomes trial; 2019-06-09.
- SOUL: Oral semaglutide and cardiovascular outcomes — Peer-reviewed randomised human outcomes trial; 2025-03-29.
