Two-Thirds of the Weight Comes Back: The Trial Evidence Behind "What Happens If I Stop?"

"What happens if I stop taking Ozempic?" is one of the questions people most reliably ask an AI assistant about a GLP-1, and the answer that comes back is usually a paragraph of hedged prose: appetite returns, weight regain is common, talk to your doctor. All true, and none of it is a number. Two trials actually measured it, and our own predictor's "After Stopping" tab is built directly on them — so this is the evidence behind that tool, not a new claim on top of it.

Two trials, two designs, the same direction

Both are semaglutide (Wegovy) studies. Neither is ours — we didn't run a trial, we read the published results and built a model on top of them.

	STEP-4 (Rubino et al., JAMA 2021)	STEP-1 extension (Wilding et al., Diabetes Obes Metab 2022)
Design	Randomized withdrawal: everyone got 20 weeks of semaglutide first, titrated to 2.4 mg, then was randomized	Observational: everyone who reached this phase had already stopped the drug at trial's end
Who	803 people who completed the run-in, randomized 2:1	327 people, a subset of the original 1,961 STEP-1 participants who completed the full 68-week trial
What changed at the split	535 continued semaglutide, 268 switched to placebo, both for 48 more weeks	All 327 stopped whatever they'd been randomized to (semaglutide or placebo) and were followed drug-free for 52 more weeks
Result	The placebo-switch group's weight rose a mean 6.9% over the 48 weeks, measured from their week-20 weight (the paper's primary endpoint). Measured from the original baseline, the registry posts them at −5.4% at week 68 against the 10.6% lost in the run-in — so roughly half of the run-in loss came back in 48 weeks, while the group that kept the drug went on to −17.7%	The former-semaglutide group had lost 17.3% by week 68 and regained 11.6 of those points by week 120 — 67%, "two-thirds" — versus the former-placebo group, who had lost almost nothing (2.0%) and regained almost nothing back to net 0.1%
Registration	NCT03548987	NCT03548935 (STEP-1 parent trial)

These are different designs answering slightly different questions — STEP-4 is a randomized comparison of continuing versus stopping, over 48 weeks starting from week 20; the STEP-1 extension is a single-arm follow-up of everyone who'd already stopped, over 52 weeks starting from week 68 — and they do not land on the same number. About half of the run-in loss came back in STEP-4 (5.2 of 10.6 points from baseline, in 48 weeks after 20 weeks on the drug); two-thirds came back in the STEP-1 extension (11.6 of 17.3 points, in 52 weeks after 68 weeks on the drug). Our tool's baseline is the STEP-1 extension's own two-thirds — the longer off-drug window after the longer treatment course, and the investigators' own headline conclusion — with STEP-4 as the randomized evidence that the direction and rough size hold, not as a second measurement of the same figure. Be careful with the STEP-4 arithmetic yourself: its +6.9% is a change from week-20 weight, and dividing it by a loss measured from baseline overstates the share regained.

What our tool does with that number — and what it adds that neither trial measured

The predictor's discontinuation tab starts every projection from a 67% terminal regain of your lost weight, approached along a curve that rises faster in the first three months and levels off after. That shape is the model's assumption, not a fit to either trial's visit data: the STEP-1 extension weighed participants at weeks 75, 80, 104 and 120 and reports that regain "continued until the end of follow-up", and STEP-4's published comparison is the single week-20-to-week-68 change. The curve reaches about 98% of its endpoint by week 52, so an unadjusted profile reads roughly 65% at one year, not 67%. From that trial-measured starting point, the tool then adjusts up or down for exercise level (−12 to 0 points), resistance training (−5), how many weeks you were on the drug before stopping (−5 to +5), age over 55 (−3), and diabetes status (−4), clamped to a 30–85% range.

None of those five adjustments comes from a trial. Neither STEP-4 nor the STEP-1 extension reports a subgroup breakdown by exercise habits, resistance training, treatment duration, age, or diabetes status that this tool draws on — we coded plausible directions and magnitudes ourselves, the same way our side-effect risk multipliers are expert judgment layered on an evidenced base rate, not a second trial finding. The tool discloses this in its own confidence note, and we're restating it here rather than only in a tooltip: the 67% is trial-measured, the personalization on top of it is not.

The model also does not vary by drug, even though the molecule picker on every tab of the predictor lets you choose semaglutide, tirzepatide or liraglutide. Both source trials used semaglutide only. If you select tirzepatide and open the After Stopping tab, you get the same semaglutide-derived 67% baseline — not because we have evidence the two drugs behave identically after stopping, but because the model draws on no tirzepatide trial. One exists: SURMOUNT-4 (Aronne et al., JAMA 2024) ran the same randomized-withdrawal design with tirzepatide — 670 people randomized after a 36-week lead-in that took off 20.9%, and the placebo-switch group ended 52 weeks later at −9.9% from baseline, so again roughly half of the lead-in loss came back. Folding it into the model is on our list; until then we'd rather show one honestly-labelled number than a second number with no evidence wired behind the difference.

What this is not

This is a different model from the one behind our published side-effect rates. Our 15-effect corpus — 2,600-plus points, pooled clinical rates with published n and distinct-source counts — answers "how often does this happen while you're on the drug." Discontinuation regain is a separate question, built from two named trials rather than pooled across the corpus, and it has no "n distinct sources" figure to publish because it draws on exactly two (the tool's own source line also names Look AHEAD maintenance data; neither the 67% nor the curve uses it). If you're looking for a citable rate with a denominator behind it, that's the side-effect corpus, not this tool; the citation for this number is the two semaglutide trials the model uses, not our data API.

Honest limitations

Semaglutide only. Both trials the model uses studied semaglutide 2.4 mg. A tirzepatide withdrawal trial has been published (SURMOUNT-4, above) and is not yet in the model; we have not found one for liraglutide, which is why the model doesn't vary by drug.

The STEP-1 extension is a subset, not the whole trial. 327 of the original 1,961 STEP-1 participants completed the full 68 weeks, consented to the extension, and were followed to week 120. Described as "a representative subset" by the investigators, not a randomly re-drawn sample.

No continued-treatment arm at week 120. By the extension phase everyone had stopped their originally assigned treatment, so it measures regain after stopping, not regain-versus-continuing at that later timepoint — STEP-4 is the trial that measures the continuing-versus-stopping comparison, over its own shorter 48-week window.

Five personalisation adjustments and a 30–85% clamp are ours, not trial-derived, as stated above. Treat the personalized number the tool gives you as an informed estimate shaped by our judgment, not a subgroup finding from either trial.

Standard deviations are wide in both trials: 7.7 points on the STEP-1 extension's 11.6-point regain (4.8 on the placebo arm's 1.9), and 7.3 points on the STEP-4 placebo arm's −5.4% baseline-to-week-68 change as posted on the registry — meaning individual outcomes varied substantially around these means, and a mean is not a guarantee for any one person.

Sources: Rubino D, Abrahamsson N, Davies M, et al. "Effect of Continued Weekly Subcutaneous Semaglutide vs Placebo on Weight Loss Maintenance in Adults With Overweight or Obesity: The STEP 4 Randomized Clinical Trial." JAMA. 2021;325(14):1414-1425. PubMed 33755728; baseline-relative week-68 figures from the posted results at NCT03548987. Wilding JPH, Batterham RL, Davies M, et al. "Weight regain and cardiometabolic effects after withdrawal of semaglutide: The STEP 1 trial extension." Diabetes Obes Metab. 2022;24(8):1553-1564. PubMed 35441470. Aronne LJ, Sattar N, Horn DB, et al. "Continued Treatment With Tirzepatide for Maintenance of Weight Reduction in Adults With Obesity: The SURMOUNT-4 Randomized Clinical Trial." JAMA. 2024;331(1):38-48. PubMed 38078870. Try the model yourself at magistra.health/en/predictor ("After Stopping" tab); the side-effect corpus these trials do not feed into is at magistra.health/en/data-api, with full eligibility methodology at magistra.health/en/methodology.

This article is educational and not a substitute for medical advice. Do not stop or change a GLP-1 medication without talking to your doctor — tapering plans and monitoring needs are individual.