The pivotal obesity trials were built to last 68 to 72 weeks, which is long enough to win an approval and far short of the question people ask. Two things extend the record. First, the cardiovascular trials ran years longer, in tens of thousands of patients, collecting adverse events the whole time: LEADER at a median 3.8 years, REWIND at 5.4, SELECT at a mean 3.3 on the obesity dose itself. Second, three large database studies look for serious gastrointestinal complications, suicidality, and the thyroid cancer that a rodent finding put in the boxed warning, events too rare for any trial to detect. Two structural limits apply to all of it. Long safety data mostly comes from people with type 2 diabetes, not from healthy adults taking these drugs for weight. And 5.4 years is the ceiling. For a drug someone may take for thirty, nobody yet knows what happens past that point.
Ranked papers
#1STEP-5
Garvey WT, Batterham RL, Bhatta M, et al. · Nat Med · 2022
What it measured: Weight and side effects across two full years
The two-year randomized read on semaglutide 2.4 mg, extending the STEP-1 design to 104 weeks in 304 adults. Nothing new appeared in year two: gastrointestinal effects stayed dominant (82.2% against 53.9% on placebo), mild to moderate for the most part and clustered around dose escalation, with serious adverse events at 7.9% against 11.8% on placebo. The useful finding is the absence of one: side effects did not accumulate with continued dosing, which is the premise chronic use rests on.
PMID 36216945 ↗NCT03693430 ↗
#2SELECT
Lincoff AM, Brown-Frandsen K, Colhoun HM, et al. · N Engl J Med · 2023
What it measured: Cardiac events and serious side effects over an average of 3.3 years
The largest and longest safety dataset at the obesity dose: 17,604 adults followed a mean 39.8 months. Semaglutide produced fewer serious adverse events (33.4% against 36.4%), while twice as many people stopped over side effects (16.6% against 8.2%), almost entirely gastrointestinal. Gallstones were slightly more frequent (2.8% against 2.3%). No excess in pancreatitis, in thyroid cancer, or in psychiatric events. Both halves matter. There were fewer serious problems, and more people quit.
PMID 37952131 ↗NCT03574597 ↗DOI 10.1056/NEJMoa2307563 ↗
#3LEADER
Marso SP, Daniels GH, Brown-Frandsen K, et al. · N Engl J Med · 2016
What it measured: Cardiac events and reviewed side effects over a median 3.8 years
The reference safety dataset for the class, and the one FDA labeling still leans on: 9,340 patients on liraglutide over a median 3.8 years. Serious adverse events came out even (49.7% against 50.4%), gallstone disease was more common on the drug (3.1% against 1.9%), pancreatitis was numerically lower, and medullary thyroid carcinoma occurred in no one on liraglutide and one person on placebo. Pancreatic cancer was numerically more common on the drug (13 cases against 5), though the gap fell short of statistical significance. It is a diabetes population at a diabetes dose, so it is the closest thing to a long safety record in this class but not a direct read on Wegovy.
PMID 27295427 ↗NCT01179048 ↗DOI 10.1056/NEJMoa1603827 ↗
#4SUSTAIN-6
Marso SP, Bain SC, Consoli A, et al. · N Engl J Med · 2016
What it measured: Cardiac events and reviewed side effects over two years
The trial that produced the one long-term signal in this class that changed clinical practice. Across 3,297 high-risk patients over 104 weeks, retinopathy complications ran 3.0% against 1.8% (a 76% relative increase), concentrated in patients who already had retinopathy and whose blood sugar dropped fast. The mechanism is thought to be the speed of glycemic improvement rather than the drug itself, a phenomenon known from insulin. It is why eye screening before starting is now standard in patients with existing retinopathy.
PMID 27633186 ↗NCT01720446 ↗DOI 10.1056/NEJMoa1607141 ↗
#5REWIND
Gerstein HC, Colhoun HM, Dagenais GR, et al. · Lancet · 2019
What it measured: Cardiac events and reviewed side effects over a median 5.4 years
The longest randomized follow-up anywhere in the class: 9,901 patients on dulaglutide over a median 5.4 years, and the only long dataset drawn mostly from people who had not already had a cardiac event. Serious adverse events matched placebo (about 40% against 41%), the three conditions that carry class warnings (pancreatitis, pancreatic cancer, medullary thyroid carcinoma) came out balanced, and more people stopped over side effects on the drug (9.1% against 6.3%, mostly gastrointestinal, per the European Medicines Agency's review). Five and a half years is the current outer edge of randomized evidence for this drug class. Everything beyond it is observational.
PMID 31189511 ↗NCT01394952 ↗DOI 10.1016/S0140-6736(19)31149-3 ↗
#6SURMOUNT-4
Aronne LJ, Sattar N, Horn DB, et al. · JAMA · 2024
What it measured: Weight and side effects over the year after dose escalation had finished
88 weeks of tirzepatide exposure in 670 adults: a 36-week open-label run-in followed by 52 weeks randomized to continue or switch to placebo. Continued dosing was well tolerated, with serious adverse events at 3.0% in both arms during the randomized year, and gastrointestinal effects were milder than in the original escalation phase because everyone had already titrated up. The other half of the result is why it is quoted so often. Stopping produced 14% weight regain over the year.
PMID 38078870 ↗NCT04660643 ↗DOI 10.1001/jama.2023.24945 ↗
#7STEP-1 extension
Wilding JPH, Batterham RL, Davies M, et al. · Diabetes Obes Metab · 2022
What it measured: Weight, blood pressure, lipids and blood sugar a year after stopping
The paper that made 'indefinite' the default framing. 327 STEP-1 completers were followed for a year after everything stopped. Having lost 17.3%, they regained 11.6 percentage points of it (two-thirds), ending at 5.6% below where they started, with blood pressure, lipids, A1C and inflammatory markers drifting back toward baseline alongside. Adverse events were not systematically collected after the drug stopped. Its significance is about cost rather than safety: it is the evidence insurers are shown when the question is whether to cover treatment that never ends.
PMID 35441470 ↗NCT03548935 ↗
#8
Sodhi M, Rezaeianzadeh R, Kezouh A, et al. · JAMA · 2023
What it measured: Rates of gallbladder disease, pancreatitis, bowel obstruction and stomach paralysis
The most-cited post-marketing safety signal, drawn from a claims database of roughly 16 million US patients and comparing semaglutide and liraglutide against a non-GLP-1 weight drug rather than against no treatment. Pancreatitis, bowel obstruction and gastroparesis were all substantially more frequent; biliary disease was not. The relative increases are large, and the absolute rates stay low.
PMID 37796527 ↗DOI 10.1001/jama.2023.19574 ↗
#9WHO VigiBase replication
McIntyre RS, Mansur RB, Rosenblat JD, et al. · J Affect Disord · 2025
What it measured: Whether suicidality reports stood out against other drugs in a global database
The check on the 2023 suicidality scare. After the same authors found a signal in US adverse-event reports, they went looking for it in the World Health Organization's global database and found a mixed pattern. Reports of suicidal ideation and suicidal behavior were disproportionately frequent for semaglutide and liraglutide (and ideation for tirzepatide), while reports of suicide attempts and completed suicide were disproportionately rare. Spontaneous reports cannot establish a rate in either direction, and the authors say causation cannot be judged from them either way.
PMID 39433133 ↗
#10
Baxter SM, Lund LC, Andersen JH, et al. · Thyroid · 2025
What it measured: Thyroid cancer rates against a comparable drug class, across six countries
The largest multinational check on the boxed warning. Population databases in six places (Denmark, Norway, Ontario, South Korea, Sweden and Taiwan) compared 98,147 GLP-1 users with type 2 diabetes against 2,488,303 DPP-4 inhibitor users (a comparator chosen because it is prescribed to similar patients), with median follow-up of 1.8 to 3.0 years. Thyroid cancer risk came out at 0.81, with a confidence interval running 0.59 to 1.12, and did not rise with cumulative dose. The warning originates in rodent toxicology. This is reassurance for the short term only: the authors say follow-up was too short to rule out excess risk with long-term use, and the data could not separate thyroid cancer types.
PMID 39772758 ↗
About this list
Obesity medicine throws up the same handful of questions over and over, and we answer them with ranked lists anchored to the papers themselves. On 2026-11-10 every citation below was put back through PubMed. Where a trial has a registry entry, the card carries that link too, alongside the PubMed one.
Browse our full index of research lists or our long-form research articles.