At the European Association for the Study of Diabetes meeting in Milan, Eli Lilly presented ACHIEVE-4, the largest and longest trial so far of its oral GLP-1 medication orforglipron. The results were published in The Lancet at the same time. Coverage has leaned on two numbers: a “16% lower risk” of major cardiovascular events and a much larger drop in deaths.
Both numbers deserve a slower read than the headlines gave them.
What the trial did
ACHIEVE-4 randomized 2,749 adults with type 2 diabetes and overweight or obesity. All of them had established cardiovascular disease or chronic kidney disease. The average participant was 63 years old, 38% were women, and average HbA1c was 8.2%. Participants were randomized to orforglipron, a once-daily pill, or to insulin glargine, a long-established basal insulin.
The question was not “does this pill protect the heart?” It was narrower: is the pill’s cardiovascular safety acceptable compared with insulin?
What “noninferior” means
The primary endpoint was MACE-4, a composite of major adverse cardiovascular events. The hazard ratio was 0.84, with a 95% confidence interval of 0.59 to 1.20.
A noninferiority trial sets a margin in advance. Here, the upper end of that interval had to stay below 1.8. It did, so the trial met its goal: orforglipron did not look meaningfully worse than insulin on heart events.
That is a real and useful result. It is also a different claim from “reduced heart risk.” The interval runs from 0.59 to 1.20, which means the data are compatible with a large benefit, no difference, or a modest harm. The same is true of the MACE-3 secondary result (hazard ratio 0.77, interval 0.52 to 1.13). Neither interval excludes 1.0, so neither is a statistically significant reduction. Describing 0.84 as a “16% lower risk” reads a point estimate as if it were settled.
The death-rate figure
The number that travels furthest is all-cause death: a hazard ratio of 0.43 (95% CI 0.25 to 0.75), which sounds like 57% fewer deaths.
Three reasons to hold it loosely:
- It was not protected against chance. The analysis was prespecified, but it was not controlled for multiple comparisons. When a trial examines many endpoints, some will look good by luck, and formal testing exists to account for that. Coverage of the results itself calls this analysis hypothesis-generating.
- The trial was open-label. Participants and clinicians knew who was on a pill and who was on insulin. That can change how people are monitored and treated in ways unrelated to the drug itself.
- The comparator matters. Insulin carries its own risks, including low blood sugar and weight gain. In this trial, weight moved in opposite directions (about 8.8% lost on orforglipron versus 1.7% gained on glargine). A comparison against insulin tells you less about how the drug would fare against other diabetes treatments.
None of this means the death signal is false. It means it is a reason for a dedicated trial, not a conclusion.
What it showed on blood sugar, weight, and tolerability
On HbA1c, orforglipron lowered it by 1.6 points versus 1.0 with glargine, and Lilly reports the effect persisted to 104 weeks. Participants on orforglipron lost weight; those on insulin gained it. Discontinuation because of adverse events was 10.6% in the orforglipron group during the first 52 weeks, with the usual gastrointestinal complaints for the class: nausea, vomiting, diarrhea, decreased appetite, and constipation. The press release does not report the glargine group’s rate or the per-arm GI percentages, so a fair side-by-side tolerability comparison has to wait for the full paper’s tables.
Also worth noting: the trial was sponsored by the company that makes the drug, and several of the figures in public coverage trace back to its press release. That is normal for pharmaceutical trials, and a Lancet publication means peer review, but it is part of how to weigh the framing.
Who this does not describe
This was a population with type 2 diabetes and existing cardiovascular or kidney disease. It does not tell you what orforglipron would do in someone without diabetes, and it says nothing about weight-loss use specifically. Averages across 2,749 people also do not predict any one person’s response.
It also bears repeating that this evidence belongs to a branded, manufactured product. It does not transfer to compounded semaglutide or tirzepatide, which are different drugs prepared by state-licensed pharmacies, are not FDA-approved, and have not been through trials like this. Our GLP-1 education page explains that distinction in more detail.
The takeaway
ACHIEVE-4 is a reassuring safety result for an oral GLP-1 in a high-risk group, and a reasonable reason to expect follow-up trials on heart outcomes. It is not proof that the pill protects hearts or extends life. If you want to know how cardiovascular history figures into whether a GLP-1 could be appropriate for you, that is a question for a licensed provider who can review your records. Our FAQ covers how that review works.
This post is general information, not medical advice. It does not describe a benefit of any NoBsRx program and is not a claim that any medication will protect your heart or improve any health outcome. Compounded medications are not FDA-approved. Treatment is never guaranteed. Eligibility and every treatment decision are made by independent licensed providers based on an individual clinical assessment, and availability varies by state. Do not start, stop, or change any treatment based on an article.