A trial can only answer the question it was powered for, and SUSTAIN-6 was powered for a question about harm. Regulators required new diabetes drugs to demonstrate that they did not raise cardiovascular risk, and the statistical target was accordingly an upper bound: show that the top of the 95% confidence interval for the hazard ratio sits below 1.8. It cleared that bound comfortably and then kept going, and the number it kept going to — a 26% relative reduction — is now the sentence the whole class is sold on. The result is real. The test that produced it was run after the fact, and the composite it lives inside does not hold together when its four parts are separated. How the class-wide cardiovascular argument is assembled across several trials belongs to the heart benefits article; this page is about what this one trial was and was not built to show.
What was run, and for how long
SUSTAIN-6 randomly assigned 3,297 patients with type 2 diabetes on standard care to once-weekly semaglutide at 0.5 mg or 1.0 mg or to matching placebo, for 104 weeks, across 229 sites in 20 countries. Four groups were run rather than three: 826 and 822 on the two semaglutide doses, 824 and 825 on their matched placebos, pooled into 1,648 against 1,649 for the primary analysis.[1][2]
At baseline, 2,735 patients (83.0%) had established cardiovascular disease, chronic kidney disease, or both.[1] This was a secondary-prevention population in poor glycemic control, which is the population a regulator cares about and not the population a weight-loss subscription sells to. The distance between the two is the subject of the diabetes-versus-weight-loss article.
The margin the trial was sized against
The primary composite was the first occurrence of cardiovascular death, nonfatal myocardial infarction or nonfatal stroke. It occurred in 108 of 1,648 (6.6%) on semaglutide against 146 of 1,649 (8.9%) on placebo — a hazard ratio of 0.74 (95% CI, 0.58 to 0.95).[1]
The trial’s own statistical plan describes what that comparison was sized for: assuming the true hazard ratio equaled 1, a minimum of 122 events was needed for at least 90% power to show that the upper two-sided 95% limit fell below 1.8. The reported P < .001 is the noninferiority test against that margin, not a test of benefit.[1][2]
The superiority figure exists, and its provenance is stated plainly in the results posting: a two-sided Wald test of no difference, returning P = 0.0167, described there as a post hoc analysis of superiority performed on the prespecified Cox model.[2] The distinction is not pedantry. A prespecified superiority claim and a post hoc one carry different weight precisely because the second is chosen after the numbers are visible, and this is the trial the phrase “proven to cut cardiovascular events” most often traces back to.
Four components, one significant direction
Take the composite apart and the picture changes shape. Cardiovascular death occurred in 1.6% against 1.9% — hazard ratio 0.98 (95% CI, 0.65 to 1.48; P = 0.92). Nonfatal myocardial infarction ran 2.5% against 3.7% — 0.74 (95% CI, 0.51 to 1.08; P = 0.12), which does not reach significance. Nonfatal stroke ran 1.5% against 2.5% — 0.61 (95% CI, 0.38 to 0.99; P = 0.044), with an upper bound a hundredth away from 1. Revascularization, part of the expanded composite rather than the primary one, ran 2.6% against 4.2% — 0.65 (95% CI, 0.50 to 0.86).[2]
Hospitalization for heart failure went the other way: 2.7% on semaglutide against 2.4% on placebo.[2] None of these components was powered individually and none should be read as a standalone finding. Read together, they say that the composite’s movement came from strokes and revascularizations, and that nobody in this trial was shown to be less likely to die of a cardiac cause.
This is not the dose being sold
Semaglutide for weight management is 2.4 mg weekly. SUSTAIN-6 tested 0.5 mg and 1.0 mg, and its weight column reads accordingly. Estimated mean change in body weight was −3.57 kg at 0.5 mg and −4.88 kg at 1.0 mg against −0.62 kg on placebo — treatment differences of −2.95 kg (95% CI, −3.47 to −2.44) and −4.27 kg (95% CI, −4.78 to −3.75). Glycated hemoglobin fell 1.09 and 1.41 points against 0.44 and 0.36.[2]
Roughly three to five kilograms over two years is a glycemic trial’s weight column, not a weight trial’s. The doses share a molecule with the product sold for weight loss, and the difference between the two labels is set out in the Wegovy-versus-Ozempic article. Any sentence that moves a cardiovascular hazard ratio from this trial onto a 2.4 mg prescription is extrapolating across a dose it did not test.
One in five stopped treatment, in every arm
The published abstract reports only the direction — more discontinuations for adverse events on semaglutide, mainly gastrointestinal — without rates.[1] The trial’s posted results carry the counts. Premature treatment discontinuation for any reason occurred in 164 of 826 (19.9%) at 0.5 mg and 186 of 822 (22.6%) at 1.0 mg, against 151 of 824 (18.3%) and 159 of 825 (19.3%) in the matched placebo arms. Trial completion itself was near-universal in all four groups, above 97%, because people who stopped the injections were permitted to stay enrolled.[2]
Two readings follow. The drug’s excess over placebo is a few percentage points, not a chasm. And roughly one participant in five stopped an injection they were being given free, at a fraction of the weight-loss dose, with a study team managing escalation. What happens to the numbers once the injections stop is the subject of the stopping article.
The finding that points the other way
Retinopathy complications — vitreous hemorrhage, blindness, or conditions requiring an intravitreal agent or photocoagulation — were significantly more frequent on semaglutide, at a hazard ratio of 1.76 (95% CI, 1.11 to 2.78; P = 0.02).[1] A trial quoted for protecting hearts also reported a signal against eyes, and which patients that applies to is the entire question. It is worked through in the eye health article and not repeated here.
The class-level picture disagrees with SUSTAIN-6 on this point. A meta-analysis of eight cardiovascular outcome trials covering 60,080 patients with type 2 diabetes found GLP-1 receptor agonists reduced major adverse cardiovascular events by 14% (HR 0.86; 95% CI, 0.80 to 0.93) and all-cause mortality by 12% (HR 0.88; 95% CI, 0.82 to 0.94), with no increase in the risk of retinopathy across the pooled trials.[3] One trial’s safety signal and a pooled estimate across eight are not the same evidence, and a page reporting either one alone is reporting half of what is known.
What SUSTAIN-6 never established
It did not show a reduction in cardiovascular death, and it was not designed to detect one. It did not establish superiority under a prespecified plan. It ran 104 weeks, short for an outcome trial, and counted 254 primary events in total across both arms — a small number of events carrying a large claim.[1]
It also did not settle the question for injectable semaglutide in diabetes, because the two purpose-built superiority trials that followed changed something fundamental each time. SELECT enrolled 17,604 patients with cardiovascular disease and overweight or obesity without diabetes, at 2.4 mg, and reported 6.5% against 8.0% (HR 0.80; 95% CI, 0.72 to 0.90).[4] SOUL enrolled 9,650 patients with type 2 diabetes but tested the oral formulation at up to 14 mg, reporting 12.0% against 13.8% (HR 0.86; 95% CI, 0.77 to 0.96; P = 0.006).[5] Both were declared superiority trials in advance. Neither repeated SUSTAIN-6’s combination of injectable semaglutide and type 2 diabetes, so the prespecified confirmation of that specific pairing does not exist. Those two trials have their own pages: SELECT and SOUL.
What was in the syringe
Every figure above belongs to branded, FDA-approved semaglutide at 0.5 mg or 1.0 mg weekly, supplied free, escalated on a protocol, and added to standard diabetes care that included oral agents and insulin in most participants.[1] Sellers listed on the semaglutide board largely dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed.
The practical consequence is narrow and worth stating exactly. A cardiovascular claim made for a compounded vial rests on a post hoc superiority test, in a different formulation, at a fraction of the dose, in a population selected for existing heart or kidney disease, over two years, on top of standard diabetes care nobody buying weight-loss telehealth is necessarily receiving. Each of those five gaps is small on its own and none of them is closed by the trial.