Almost everything written about stopping these medicines rests on the same two trials, and almost nothing quotes them directly. Both have posted their results to the public trial registry, where the numbers can be read without going through a press release or a summary of a summary. That is what this page does.
Both figures sets below were read from ClinicalTrials.gov posted results on 2026-09-10. This page reports what those trials measured. It does not advise anyone on whether or when to stop a medicine; that belongs with a prescriber, as our editorial standards set out.
What was gained before anyone stopped is its own question, and the labels answer it as a spread rather than a figure: the posted outcome distributions show how many patients in each arm reached 5%, 10%, 15% and 20%, and how many did not.
Why these two trials answer the question and other studies do not
The obvious way to study what happens after people stop is to find people who stopped and weigh them. It is also close to useless, and the reason is worth a paragraph because it determines which numbers on this subject are worth reading.
People who discontinue a medicine are not a random sample of people who started it. They stopped because of side effects, or cost, or a lapsed prescription, or a supply gap, or because they had reached a goal, or because it was not working. Every one of those reasons is also related to how their weight will behave afterwards. Measure regain in that group and the result contains the reasons for stopping as much as the effect of having stopped. Whether a much smaller weekly dose could hold the weight after a full course is the question behind microdosing a GLP-1, and the one trial designed to test it is not due to finish until 2028.
Randomised withdrawal removes that. Everybody takes the drug openly for a set period. Then a computer, not a person or a circumstance, decides who continues and who receives a placebo — and neither the participant nor the investigator is told which. Anyone who stops in that design stops for no reason at all, which is precisely what makes the comparison mean something.
Two trials of this class exist at scale on these drugs. Both have posted results.
SURMOUNT-4: tirzepatide
NCT04660643, 783 participants. Everyone received open-label tirzepatide for 36 weeks and reached a maximum tolerated dose of 10 mg or 15 mg. At week 36 they were randomised, double-blind, to continue at that dose or switch to placebo for a further 52 weeks. Mean body weight at the start was about 107.8 kg. Serious adverse events in its randomised period were 3.0% in both arms, one row in our record of the longest GLP-1 trials.
Measured from the randomisation point:
| Weeks after the switch | Switched to placebo | Continued tirzepatide |
|---|---|---|
| 28 weeks (week 64) | +9.9% | −6.0% |
| 52 weeks (week 88) | +14.8% | −6.7% |
At one year the two groups were 21.5 percentage points apart, and they were moving in opposite directions the whole time.
Two details in that table are more informative than the headline.
The regain had not finished. Between week 64 and week 88 the placebo group went from +9.9% to +14.8%. Roughly two thirds of the first year's regain happened in the first 28 weeks and a third in the next 24, so the pace slowed — but it did not stop, and the trial ended before the curve did. Any statement about where the weight settles is going beyond what this trial measured, and this page will not make one.
The group that continued was still losing. −6.0% at 28 weeks, −6.7% at 52. Participants who had already been on the drug for 36 weeks and reached maximum tolerated dose had not plateaued; they were still going down, slowly, through a second year. That is a fact about continuing rather than stopping, and it belongs here because the 21.5-point gap is made of both halves. When the curve does flatten in the longest trials, and what has been tested after it, is on our GLP-1 plateau page.
The number that answers the question people actually ask
Percentages of body weight are hard to feel. The trial reported a second endpoint that is not, and it is the most useful figure on this whole subject:
Percentage of participants who maintain at least 80% of the body weight lost during the open-label lead-in period, at week 88: placebo 13.50%, tirzepatide 93.37%.
One year after stopping, roughly one participant in seven had kept 80% of what they lost. Six in seven had not.
That cuts both ways and both should be said. It is a blunt answer to anyone told that the loss is durable without the drug — for six out of seven people in a controlled trial, it was not. It is also an answer to anyone told that everybody regains everything, because one in seven did hold on, and that group existed in a randomised comparison rather than in testimonials.
Measured from the beginning rather than the switch
The percentages above use the randomisation point as zero, which is the right frame for the question "what happens if I stop" and the wrong one for "was it worth doing at all". The trial reports the other frame too, from the original starting weight to week 88:
| Change from starting weight at week 88 | |
|---|---|
| Continued tirzepatide | −26.0% |
| Switched to placebo | −9.5% |
So the group that stopped, a full year later, was still 9.5% lighter than when it began — on a mean starting weight of about 107.8 kg, in the region of ten kilograms. Stopping reversed most of the loss. It did not reverse all of it, and a page that reported only the +14.8% would have left a reader with the impression that it did.
STEP 4: semaglutide
NCT03548987, 902 participants. The same design on the other molecule: a 20-week run-in on semaglutide 2.4 mg, then randomisation to continue or switch to placebo, measured to week 68.
From randomisation to week 68 — 48 weeks after the switch:
| Change from randomisation | |
|---|---|
| Switched to placebo | +6.5% |
| Continued semaglutide | −8.3% |
A gap of 14.8 percentage points, and the same shape: one group up, the other still going down.
The trial also reported waist circumference over the same window, which the tirzepatide trial's posted results do not present in the same form: placebo +3.2 cm, semaglutide −6.9 cm. Weight regain in that group was accompanied by waist expansion rather than being some other kind of tissue. Waist circumference is a shape measurement rather than a tissue one, and the separate question of what the lost weight was made of is one the approved labels answer in eleven unquantified words.
The comparison to resist
Placing +14.8% next to +6.5% invites the conclusion that stopping tirzepatide is more than twice as bad as stopping semaglutide. That conclusion is not supported, and the reason is in the design of the two trials rather than in the drugs.
The lead-in periods differed: 36 weeks of open-label tirzepatide against 20 weeks of semaglutide. Participants in SURMOUNT-4 had therefore lost substantially more weight before the switch than participants in STEP 4, which means they had substantially more available to regain. A percentage of body weight regained is not a fixed quantity across two groups who arrived at the switch point in different places.
The windows differ too — 52 weeks against 48 — and, as the SURMOUNT-4 timepoints show, four extra weeks at that stage of the curve are not free.
The same caution applies wherever figures from two trials are set beside each other, which is most of this subject: the placebo arms alone range across nearly two percentage points, as the ranking of every labelled weight-loss result sets out. These are separate trials, run by different sponsors, with different populations and different protocols. What they share is a design, and what the shared design supports is the common finding: in both, randomised withdrawal was followed by substantial regain, and continuation was followed by further loss. That much is a replication. The size of the difference between the two drugs is not — for that, only the one trial that randomised people to one molecule or the other is admissible evidence, and it measured people who stayed on treatment.
What the trials do not say
They do not say where the regain stops, because both ended while it was still happening. They do not say whether people who restarted returned to where they had been. They do not identify who the one in seven were, or what distinguished them. They do not compare stopping abruptly with tapering — both simply switched to placebo. And they say nothing about what happens on a treatment break of weeks rather than a discontinuation of months, which is a different experiment nobody has run.
The gap between what is measured and what people want to know is wide here, and it is worth being clear about which side of it any particular claim sits on.
Both trials also counted kilograms rather than tissue, so what is regained is as unmeasured as what was lost. That matters because the intervention most often proposed to change the composition of either — eating more protein — has no randomised answer in this population until 2029, and the one published study with both a lifestyle arm and a drug arm gives opposite answers depending on whether lean mass is read in kilograms or as a share of the weight lost.
And they say nothing about the commonest scheduled reason of all to consider an interruption, which is an operation. On that the drug labels decline to take a position at all, and the professional guidance disagrees with itself across borders.
They also say nothing about the one group with a scheduled reason to stop. The Wegovy label instructs discontinuation at least two months before a planned pregnancy, which lands hardest on people taking these drugs off-label for polycystic ovary syndrome while hoping to conceive — a discontinuation with a deadline attached, and one no trial has followed.
One thing the trials do settle is that continuation and discontinuation are different states with different results, which makes the recurring cost of continuing a live question rather than a footnote; every current price, dated is tracked separately on this site.
Regain is also the outcome behind the one genuinely positive randomised result in the supplement market — about 2 kg less put back over six months, in people who had already dieted the weight off — which is a much narrower claim than the marketing around it suggests. We read that trial, and the large berberine trial that found nothing, on GLP-1 supplements.
Sources
- SURMOUNT-4, NCT04660643, ClinicalTrials.gov posted results, read 2026-09-10 — percent change from randomisation at weeks 64 and 88, percent and absolute change from baseline at week 88, the 80% maintenance endpoint, baseline body weight, and design and masking details.
- STEP 4, NCT03548987, ClinicalTrials.gov posted results, read 2026-09-10 — change from randomisation to week 68 in body weight and waist circumference, enrolment and run-in design.