The question behind "long-term side effects" is what happens after years on these drugs, not months. The randomised record now runs to seven years for one GLP-1 and nearly five for semaglutide at its weight-management dose, and every one of the long trials has posted its adverse event counts to ClinicalTrials.gov. We read seven of them on 2026-09-17.
The headline is less alarming than much of the coverage, and it comes with a correction that most coverage leaves out: serious adverse events were equal or lower on drug in six of the seven long trials — and the one exception is the trial whose population had the fewest heart attacks and strokes for the drug to prevent.
The counted adverse reactions from the approval trials, and which are common, are on our side-effects hub. This page's axis is duration. Nothing here is advice about any treatment; that belongs with a prescriber, as our editorial standards set out.
The seven longest trials, and what they recorded
Each figure below is a posted count from the trial's ClinicalTrials.gov record, divided by the number at risk in that arm. The "adverse event window" is the period over which the trial collected adverse events, as its record states it.
| Trial (registry ID) | Drug | Population | Adverse event window | Serious AEs, drug | Serious AEs, placebo |
|---|---|---|---|---|---|
| REWIND (NCT01394952) | Dulaglutide | Type 2 diabetes, 9,901 | Up to 7 years | 40.3% | 41.3% |
| LEADER (NCT01179048) | Liraglutide | Type 2 diabetes, high cardiovascular risk, 9,341 | Up to 5 years + 30 days | 49.7% | 50.4% |
| SELECT (NCT03574597) | Semaglutide 2.4 mg | Overweight or obesity with cardiovascular disease, no diabetes, 17,604 | Up to 240 weeks | 33.4% | 36.4% |
| SCALE, prediabetes arm (NCT01272219) | Liraglutide 3.0 mg | Overweight or obesity with prediabetes | Up to 172 weeks | 15.1% | 12.7% |
| STEP 5 (NCT03693430) | Semaglutide 2.4 mg | Obesity, 304 | Up to 111 weeks | 7.9% | 11.8% |
| SUSTAIN 6 (NCT01720446) | Semaglutide 0.5 / 1 mg | Type 2 diabetes, 3,297 | Up to 109 weeks | 34.3% | 38.0% |
| SURMOUNT-4 (NCT04660643) | Tirzepatide | Obesity, randomised after a 36-week lead-in | Weeks 36 to 92 | 3.0% | 3.0% |
SUSTAIN 6 arms are pooled across its two doses and their matched placebos. STEP 5 is small (152 per arm), so its difference is a handful of people.
Deaths, where the record posts them
| Trial | Deaths, drug | Deaths, placebo |
|---|---|---|
| SELECT | 371 of 8,803 (4.2%) | 460 of 8,801 (5.2%) |
| REWIND | 536 of 4,943 (10.8%) | 592 of 4,949 (12.0%) |
These are raw counts across the whole adverse event window, not the adjusted analyses the trials published as outcomes.
Why the lower totals need a second look
A serious adverse event is any serious medical event during the trial, whatever caused it — including the heart attacks, strokes, hospitalisations and deaths that SELECT, LEADER, REWIND and SUSTAIN 6 were designed to reduce. If the drug prevents some of those events, the drug arm's serious adverse event total falls even if the drug causes some serious events of its own. A lower total is therefore partly a measure of benefit, not purely of safety.
That is what makes the SCALE row the most informative line in the table. It is a weight-management trial in adults with obesity and prediabetes, a population with comparatively few cardiovascular events to prevent over three years. It is also the one long trial where serious adverse events were higher on drug: 230 of 1,524 on liraglutide against 96 of 755 on placebo. The same record shows the same direction in its shorter, non-prediabetes arm (59 of 957 against 19 of 487, 6.2% against 3.9%, to week 70).
Neither pattern settles the question alone. Read together, they say the long-term serious-event picture depends on who is taking the drug and what it is preventing in them.
What rose on drug: the non-serious events, and people stopping
The long trials also post non-serious adverse events above a frequency threshold. On drug, those ran consistently higher:
| Trial | Other (non-serious) AEs, drug | Placebo |
|---|---|---|
| SELECT | 44.8% | 27.7% |
| SCALE, prediabetes arm | 88.1% | 77.6% |
| REWIND | 72.3% | 68.0% |
| LEADER | 14.7% | 4.5% |
Two of these records say their non-serious collection was incomplete. SELECT's record states that "not all the AEs were collected systematically"; LEADER's that non-serious events included non-systematically collected ones. So the gaps are real in direction and uncertain in size.
The discontinuation figure is firmer, because it was a collected category. The Wegovy label, reporting SELECT, states:
"Sixteen percent (16%) of WEGOVY-treated patients and 8% of placebo-treated patients, respectively, discontinued study drug due to an adverse event."
That is over a median exposure of 37.3 months on drug. The same label notes that safety data collection in SELECT "was limited to serious adverse events (including death), adverse events leading to discontinuation, and adverse events of special interest". A long trial built to measure heart outcomes is not the same instrument as a trial built to catalogue every side effect.
When the side effects happen
The labels place most of the burden early, not late. Zepbound's label states that "the majority of patients who discontinued ZEPBOUND due to adverse reactions did so during the first few months of treatment due to gastrointestinal adverse reactions", and that tirzepatide's delay in gastric emptying "is largest after the first dose and this effect diminishes over time". Why the stomach effect fades is on how GLP-1 works.
What the long trials add to that picture is that the difference in stopping rates persists across years, not only in the first months.
What surfaced only after approval
Some reactions on today's labels were not identified in the trials at all. Both the Ozempic tablets label (effective 2026-01-30) and the Zepbound label (effective 2026-08-28) list postmarketing reports of:
- ileus, intestinal obstruction, and severe constipation including faecal impaction
- pulmonary aspiration in patients undergoing elective procedures under general anaesthesia or deep sedation
- acute kidney injury or acute renal failure, sometimes requiring haemodialysis
- acute pancreatitis, including necrotising pancreatitis, sometimes resulting in death
The semaglutide label also lists cholecystitis and gallstones requiring gallbladder removal, and alopecia. Each label adds the same caveat: these reactions "are reported voluntarily from a population of uncertain size", so "it is not always possible to reliably estimate their frequency or establish a causal relationship."
Several have their own pages: pancreatitis, gallbladder, hair loss, thyroid cancer, and aspiration on GLP-1 and surgery.
What stopping does over the long run
Long-term use is also a question about what happens at the end of it. SURMOUNT-4 randomised people who had already taken tirzepatide for 36 weeks either to continue or to switch to placebo, and that design — and what happened to weight after withdrawal — is covered on stopping GLP-1. Its serious adverse event proportions in the randomised period were identical, at 3.0% in both arms.
What the record does not establish
- Randomised data beyond about seven years does not exist for any GLP-1 in this set. REWIND's window is the longest.
- The long trials were mostly in people with diabetes or cardiovascular disease. SELECT is the long trial closest to the weight-management population and it required established cardiovascular disease.
- Tirzepatide's longest randomised adverse event record here runs to 92 weeks, far shorter than the older molecules'.
- Rare harms are best seen in postmarketing data, which cannot give rates. A trial of 10,000 people is too small to count an event that occurs in one in 50,000.
- Posted counts are not adjusted analyses. They are the raw numbers each sponsor submitted to the registry.
How this page was built
Adverse event modules for the seven trials were read from the ClinicalTrials.gov v2 API on 2026-09-17: serious, death and other-event counts with the number at risk per group, plus each record's stated time frame and collection notes. Percentages are our arithmetic on those counts. Label text is quoted from openFDA (Wegovy NDA215256, Zepbound NDA217806, Ozempic tablets NDA213051).