Do GLP-1s Beat Metformin for Prediabetes? No Trial Yet

No randomized trial has compared GLP-1 drugs against metformin for prediabetes. The study behind this week's coverage is an observational analysis of medical records, presented as a conference abstract that the American Heart Association itself describes as "not peer-reviewed" and "preliminary until published as a full manuscript in a peer-reviewed scientific journal."¹

Every number in that coverage is a relative difference with no confidence interval, no p value and no absolute risk attached to it anywhere.

What did the study report?

Seven outcomes, all as relative differences against metformin. The figures come from the American Heart Association's own release on the abstract.¹

None of those seven carries an interval, a p value or an underlying event count in any source read for this report.¹²³

All seven point the same direction, and consistency is not the same as precision. A real effect looks like that, and so does channeling by prescription access when the healthier group receives the newer drug. Separating those two explanations needs the randomization this study did not have.

Scale is the thing to hold in view alongside them. About 115 million US adults have prediabetes, "more than 2 in 5," and "8 in 10 people don't know they have it," so small relative shifts in progression apply across a very large population.⁴

What kind of study was it?

The study was observational, built from electronic health records instead of a trial. Researchers drew on the TriNetX network covering "104 institutions from 2010 to 2023," assembling 256,492 adults with prediabetes and obesity, average age 52, with 76% women.¹

Patients were paired on characteristics instead of randomized. Matching covered "similar age, health conditions" including high blood pressure, sleep apnea and nicotine dependence, plus other medications, with roughly half the participants in each group.¹ Neither of the two trade accounts names the matching technique, and the release does not either.²³

Nobody was assigned a drug by chance. That single fact is what separates this from a trial. In records research the question of who was prescribed what, and why, sits inside every result.

Matching narrows that problem without closing it. Pairing on age, blood pressure, sleep apnea, nicotine dependence and concurrent medications removes the differences somebody thought to measure, and leaves every difference nobody recorded.¹ Income, insurance tier, prior authorization success, baseline weight and motivation to pursue a newer drug are not diagnosis codes.

The data window matters as well. Records run from 2010 to 2023, a period during which these drugs went from unavailable to heavily prescribed, so the GLP-1 group is weighted toward the later years of the window and the metformin group toward the whole of it.¹

What is missing from every figure?

Three pieces of information that would let a reader judge the size of these findings are absent. No confidence interval appears in the release or in any trade coverage read here, so the precision of each estimate is unknown.¹²³

No absolute risk appears either. A 31% relative reduction describes a ratio between two rates. Without the rates themselves a reader cannot tell whether the gap represents a large number of cases or a small one. Reporting elsewhere notes the same absence, with one outlet stating plainly that the percentages are "not labeled as hazard ratios" and that "no adjustment or statistical method is described."³

Event counts are absent as well. How many people in each group developed diabetes, or died, is not in the release.¹

Those three gaps compound instead of sitting side by side. Without an interval a reader cannot tell whether 31% is a precise estimate or a wide guess; without absolute rates the reader cannot tell whether it matters; and without event counts neither of the first two can be reconstructed. A full manuscript would ordinarily carry all three.

Why does the comparator matter so much here?

Metformin is the comparator, and an active treatment is not a placebo. That changes what the headline means. Beating a drug that already works is a different claim from beating nothing. A reader who hears "31% lower risk" without the comparator attached will reach for the wrong conclusion.

Put placebo in that slot and the sentence means something else entirely. A 31% reduction against placebo would describe the drug's effect on diabetes progression. A 31% reduction against metformin describes only the gap between two treatments, and says nothing about how either compares with no treatment.

Active-comparator studies are often the more useful design, and that is a separate point from how they get reported. A patient deciding between two offered drugs wants exactly this comparison. Coverage that drops the comparator and leaves a bare percentage is where this breaks, since "31% lower risk of diabetes" and "31% lower risk than metformin" describe different magnitudes of claim.

What does randomized evidence on metformin show?

Metformin has its own trial. Those figures are published. The Diabetes Prevention Program randomized participants to a lifestyle intervention, metformin or placebo, and the National Institute of Diabetes and Digestive and Kidney Diseases reports the results across three follow-up points.⁵

Those are randomized comparisons against placebo. At 15 years the trial also found "no overall difference between treatment groups" on microvascular outcomes, while participants who avoided diabetes altogether "had fewer microvascular complications than those who did" develop it.⁵

Age changed the lifestyle result substantially. Among participants aged 60 and older, losing a small amount of weight through "a low fat, low calorie meal plan and increasing physical activity" lowered diabetes risk by 71%, against 58% in the trial overall.⁵

The two arms also aged differently. Lifestyle's advantage narrowed from 58% to 34% and then 27% across the three follow-up points, while metformin's held flat at 18% at both 10 and 15 years.⁵ A drug effect that does not decay and a behavioral effect that does are different propositions for somebody deciding what to start.

Why do two different 31% figures keep appearing?

Two unrelated comparisons produced the same figure, and conflating them inverts the finding. Metformin reduced diabetes incidence by 31% against placebo in a randomized trial.⁵ GLP-1 drugs were associated with a 31% lower risk than metformin in this observational analysis.¹

Same number, two unrelated comparisons. One is a drug measured against nothing under randomization; the other is one drug group measured against another drug in medical records. Stacking them would imply a combined effect that neither study supports.

Nothing but chance connects the two numbers. No arithmetic relationship connects a randomized placebo comparison from the 1990s to a matched records comparison from a 2010 to 2023 data window, and the identical figure invites a reader to add them or to treat one as confirming the other.¹⁵

What about the 41% lower death figure?

It is the figure most likely to be misread, and the American Heart Association's own caution applies to it directly: the study "is observational and, therefore, cannot prove a cause-and-effect relationship."¹

Scale is the reason for caution here. A 41% difference in death from any cause between two matched drug groups over a few years is a very large gap for any medication to produce. All-cause mortality also captures every reason a person might die, including reasons unrelated to blood sugar.

Prescribing patterns offer a competing explanation that matched records cannot rule out. GLP-1 drugs during the study period were expensive and often required prior authorization, reaching patients who could work through that process. None of that is measured by a diagnosis code. Whether such differences explain part of the mortality gap is unresolved. No source read for this report tests it.

Which GLP-1 drug performed best?

No drug can be named as performing best from this study, and the researchers say so. Semaglutide, tirzepatide, liraglutide and dulaglutide were analyzed as one group, and the release states that because "GLP-1-based medications were grouped together, individual differences between agents were not considered."¹

Four drugs with different potencies and different weight-loss profiles sit inside a single estimate. A reader choosing between Ozempic and Mounjaro finds nothing here that distinguishes them.

Pooling also makes the result hard to interpret even as a group finding. If one agent drove most of the benefit, the pooled estimate understates that agent and overstates the other three. If the benefit were even across all four, the pooled estimate is fair. Nothing reported tells those two apart.¹

Prescribing shares inside the pool matter as well. Two of the four drugs arrived far later in the 2010 to 2023 window than the others, so the pooled group is unlikely to be evenly split, and an estimate weighted toward whichever agent was prescribed most often is not a statement about the class.¹

How long were people followed?

How long people were followed is where the record contradicts itself. The American Heart Association's release describes patients followed for an "average of roughly 3 years," while the summary bullets on the same page cite a five-year period for mortality.¹ Trade coverage then reports "5-year follow-up" and "41% lower all-cause mortality over 5 years."³

Three years and five years are not interchangeable for a mortality claim. A release that contradicts itself is the weaker artifact of the two, since the discrepancy sits inside a single document from the study's own announcer instead of between independent accounts. Until the abstract or a manuscript resolves it, the follow-up duration behind the death figure is unsettled.

What did the researchers say about the limits?

Four limitations appear in the release itself, in the study team's own framing.¹ The analysis "cannot prove a cause-and-effect relationship." Outcomes came from diagnosis codes, and those "can be incomplete." Prescription records do not show "how consistently patients took medication or at what dose." And the drugs were grouped, so differences between agents went unexamined.

Limitations stated by the authors carry more weight than limitations raised by a critic, because they are concessions instead of objections. All four of these came from the announcement, not from outside commentary.

Each one bites in a specific way. Incomplete diagnosis codes mean the outcome counts themselves are approximate, since a case of diabetes recorded at one institution in the network and not another is a missing event. Unknown adherence means a prescription in the record is not a drug in the patient, and the two groups may differ in how closely they match.

The adherence gap runs in a particular direction here, and the release does not address it. Metformin is a daily tablet and the GLP-1 drugs were weekly or daily injections during the study window, so the practical burden of staying on each one differed. Whether that affected the comparison is untested by anything read for this report.

What did the study's own author say about metformin?

Metformin is still a first-line treatment, by the study author's own account. He says this is not about replacement. Vinicius de Sousa Barzon Serra, the Mayo Clinic research fellow credited as study author, is quoted in the release saying "Metformin is still a first-line treatment choice. It is safe, inexpensive and has a long track record."¹

His framing of the implication is narrower than the headline the findings produced. He describes the effect as landing "on the shared decision-making discussion between doctors and patients," says the work is "less about replacing one medication with another" and "more about discussing additional options with specific patients," and suggests it "may add to the case for considering GLP-1 medications earlier for certain patients."¹

He also puts lifestyle ahead of the drug question. Serra is quoted saying "It's important to note that an essential component of prediabetes treatment is lifestyle changes to manage weight," with medications "including metformin" considered "based on each individual's overall health."¹

An author damping his own headline is the most useful thing in this announcement. One drug beat another by 31%. That single sentence is what reaches a reader. The person who produced it qualified it before any outside critic touched it.

What does an independent expert say is driving the effect?

Weight loss is driving the effect, says the one independent expert quoted. Robert Eckel, a professor of medicine emeritus at the University of Colorado Anschutz Medical Campus and a past American Heart Association volunteer president who was not involved in the study, says the benefit "likely relates to the patients losing weight," which "is known to help prevent Type 2 diabetes."¹

Attributing it to weight reframes the comparison. GLP-1 drugs produce substantially more weight loss than metformin does, so a weight-mediated benefit would make this partly a weight-loss comparison between two drugs of very different potency on that axis. Neither the release nor any coverage read here reports how much weight either group lost.¹²³

Eckel places the drugs as an addition instead of a substitution, saying "GLP-1s provide another option for prevention, perhaps as an add-on therapy."¹ He attaches one practical limit to that: "However, costs and access to these medications are important factors."¹

What is not in the comparison at all?

Lifestyle intervention is absent from the comparison, despite being the arm with the strongest randomized evidence behind it. The Diabetes Prevention Program found a 58% lower diabetes incidence for lifestyle against placebo at three years, nearly double metformin's 31%, and it held a lead at 10 and 15 years as well.⁵

No lifestyle arm appears in this analysis. A reader comparing options for prediabetes is therefore looking at a study that ranks two medications against each other while omitting the intervention that outperformed one of them in a randomized trial.

Serra raises the same point unprompted, placing lifestyle change ahead of the drug question in his own comments on the findings.¹ A design that compares drug against drug is a legitimate design; a reader who takes it as the full set of options has been given an incomplete picture by the coverage, not by the researchers.

What would settle this?

One thing would settle this: a randomized trial comparing a GLP-1 drug against metformin in people with prediabetes, with diabetes progression as a prespecified endpoint. Nothing of that kind has been published.

Three additions would also make the current analysis readable on its own terms. Each estimate would need a confidence interval, both groups would need absolute event rates, and one agent would need analyzing separately instead of four pooled. The full abstract may carry some of that. It could not be retrieved for this report.⁶

A useful trial would also have to specify things this analysis leaves open: the agent and the dose, how much weight each arm lost, and whether a lifestyle arm runs alongside both drugs, given that the randomized evidence already puts lifestyle ahead of metformin.⁵

Whether such a trial is already underway could not be checked. The public trial registry was not reachable from this environment, so no statement is made here about registered studies either way.

What has not been established?

Four gaps sit under this story.

Whether GLP-1 drugs reduce diabetes progression more than metformin has no randomized answer. Whether the 41% mortality association reflects the drugs or reflects who received them is untested by anything read here.¹ No part of the study's own account examines which of the four pooled agents drives the pooled result.¹ The abstract itself also went unread for this report, so every figure above comes from the announcement and not from the research record.¹⁶

A fifth item belongs to the abstract's identity. Its release names the abstract as MDP1524 while its own abstract link uses the identifier P10382.¹ Nothing in that inconsistency bears on the findings, and it does matter for anybody trying to locate the source document.

Common questions

Are GLP-1 drugs better than metformin for prediabetes?

Can Ozempic or Mounjaro reverse prediabetes?

Was this a clinical trial?

Has the study been peer reviewed?

Does the study prove GLP-1s cut deaths by 41%?

How many people were in the study?

What confidence intervals were reported?

Which GLP-1 medications were studied?

How well does metformin work for prediabetes?

Is lifestyle change better than metformin?

Why do I keep seeing 31% for two different things?

How long were participants followed?

What did the authors list as limitations?

Who conducted the study?

Where was it presented?

Should somebody with prediabetes switch medications based on this?

Sources

1. American Heart Association, "GLP-1 meds linked to lower diabetes, heart risks than standard diabetes prevention med." News release, Dallas, November 2, 2026, embargoed until 5 a.m. ET that day. Read at source on newsroom.heart.org. Source for every figure, the study design, the participant characteristics, the author attribution and all four stated limitations. The release carries two internal inconsistencies, naming the abstract as both MDP1524 and P10382, and describing follow-up as an average of roughly three years while its highlights cite five years for mortality. ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩
2. The American Journal of Managed Care, "GLP-1 Use Tied to Lower Diabetes, CV Risk vs Metformin in Prediabetes," November 2, 2026. Read at source. Cited here for confirming the absence of hazard ratios, confidence intervals, p values, absolute event counts and any named matching method. Its own sourcing is the American Heart Association release, so it is not independent corroboration of the findings. ↩ ↩ ↩ ↩ ↩
3. Drug Topics, "GLP-1s Outperform Metformin in Prediabetes Risk, MACE." Read at source. Cited for confirming that no confidence interval appears anywhere, for the observation that the percentages are "not labeled as hazard ratios" with "no adjustment or statistical method described," and for its five-year follow-up figure. Its only reference is the American Heart Association release. ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩
4. Centers for Disease Control and Prevention, prediabetes and type 2 diabetes prevention page. Read at source on cdc.gov. Used only for the population scale figures. This page carries no figure for the rate at which prediabetes progresses to type 2 diabetes, so the absolute baseline the abstract omits could not be supplied from it either. ↩
5. National Institute of Diabetes and Digestive and Kidney Diseases, "Evidence Supporting Prevention." Read at source on niddk.nih.gov. A federal primary, used for the Diabetes Prevention Program and DPP Outcomes Study figures at three, 10 and 15 years. It carries no information about the 2026 abstract. ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩ ↩
6. Abstract MDP1524, American Heart Association Scientific Sessions 2026. Not retrieved. Publisher access is restricted and the identifier in the release is inconsistent with the one in its own link, so the abstract could not be located or read. Every figure in this report therefore comes from the announcement and not from the research document. ↩ ↩