Italian study links stopping GLP-1s to higher event risk

People with type 2 diabetes who stopped a GLP-1 receptor agonist had about twice the rate of a composite of nonfatal heart attack, nonfatal stroke, or death from any cause, compared with people who stayed on treatment, according to a large Italian retrospective cohort study. The analysis is observational: it shows an association, not that stopping the medicines caused those events.

The paper, by Antonio Ceriello of IRCCS MultiMedica and colleagues, with Antonio Nicolucci as last author, appeared in The Lancet Regional Health – Europe on Nov. 9, 2026, the Science Media Centre said. The DOI is 10.1016/j.lanepe.2026.101881. We checked the published abstract on ScienceDirect and two open figures from the article. The journal’s full text and PDF were blocked from our fetch, so figures that appear only in the body of the paper are either attributed to an outlet or omitted.

What the published paper reports

Using records collected between 2004 and 2024 from 301 diabetes clinics in Italy, the authors identified 80,302 adults with type 2 diabetes and no established cardiovascular disease who had started a GLP-1 receptor agonist. At the last visit, 31,687 people (39.5%) were still on therapy and 48,615 (60.5%) had discontinued it. In that full cohort, mean time on the drug was 2.97 ± 2.93 years among people who continued and 1.21 ± 2.13 years among those who stopped. Mean follow-up was 3.65 ± 3.03 years and 4.25 ± 3.41 years, respectively.

The primary endpoint was a composite of myocardial infarction, stroke, or death from any cause. The authors said they accounted for immortal time bias with a treatment-exposed person-time approach and used propensity score matching to obtain balanced groups. They then ran Poisson models (incidence rate ratios) and Cox models (hazard ratios).

After matching, each group had 28,713 people. For the composite, the incidence rate ratio was 2.01 (95% confidence interval 1.77 to 2.28) and the hazard ratio was 2.03 (95% CI 1.92 to 2.14). For death from any cause, the incidence rate ratio was 4.67 (95% CI 4.11 to 5.30) and the hazard ratio was 4.81 (95% CI 4.39 to 5.26). Those four ratios all come from the matched analysis in the published abstract; a figure in the paper shows the same two hazard ratios. Some recaps have called 2.01 and 4.67 hazard ratios. The abstract labels them incidence rate ratios.

The abstract also says risk factors were not meaningfully different after discontinuation in people who later had a primary endpoint and in people who did not. Part of the work was supported by the Italian Ministry of Health (Ricerca Corrente) to IRCCS MultiMedica.

What we could not verify in the full text

MedicalXpress, in a report bylined by Lancet, put the combined incidence near four events for every 100 people each year in the group that stopped, versus two in the group that continued, and put all-cause deaths at about 2.2 versus 0.5 per 100 people per year. The same report said the authors argue that stopping alone is “unlikely to explain a gap this large.” Those per-year rates and that comment are not in the published abstract or in the two figures we opened, so they are attributed to MedicalXpress.

That report also said that, among people who stopped, both risks were higher in those with obesity (a body mass index of 30 or over) than in those with a BMI under 30, and that weight, blood sugar, and blood pressure changed only modestly after stopping, by similar amounts whether or not someone later had an event. ABC News carried a similar account of modest, similar changes. We could not check the subgroup tables or risk-factor plots in the full paper.

MedicalXpress identified the source as the Italian Association of Clinical Diabetologists (AMD) database. The published abstract refers to 301 diabetes clinics in Italy and does not use that name in the text we have.

ABC News, writing about the paper, said the study did not include people who took tirzepatide and that only a small number of participants used semaglutide, with most using liraglutide or dulaglutide. That mix is not stated in the abstract or the two figures, so it is attributed to ABC.

Limits, and a comment on residual confounding

Because the design is observational, people who stopped may have differed from people who continued in ways the matching did not capture, including illness that led them to stop. MedicalXpress said the researchers did not know why people discontinued treatment. We are not using a claim in one secondary outlet that patients in this study named gastrointestinal side effects, cost, or a sense the drugs were not working as their main reasons, because that is not in the published abstract and sits uneasily with the Lancet-bylined statement that the reasons were unknown.

Naveed Sattar, professor of cardiometabolic medicine at the University of Glasgow, commenting through the Science Media Centre, said the reported excess risks “could readily reflect residual confounding.” “People who discontinue treatment, for whatever reason, may differ systematically from those who continue, with characteristics associated with a higher underlying risk of future disease,” he said. In the same statement, he said the findings are observational and should be read as associations, not as evidence that stopping the medication itself leads to an immediate increase in risk, or that risks become higher than they would have been had the same people never started treatment. He also noted that randomized trials support starting approved medicines in selected groups, but that randomized evidence on the consequences of discontinuation remains limited.