Discontinuation Outcomes in GLP-1 Trials: What Extension Studies Show About Weight Regain

The question patients ask most often is the one the original registration trials were not designed to answer: what happens after you stop? Over the past four years a specific class of study — the randomized withdrawal design and the off-treatment extension — has produced a reasonably consistent evidence base. This is a review of what those studies actually measured.

Why withdrawal designs exist

A standard placebo-controlled obesity trial tells you how much weight a drug produces relative to placebo over a fixed window. It tells you nothing about what happens at the end of that window. To answer the discontinuation question you need a different structure, and two have been used.

The first is the off-treatment extension: run the trial to completion, stop all study product, and keep measuring. The second is the randomized withdrawal: run an open-label lead-in where everyone receives active drug, then randomize at the end of the lead-in to either continue or switch to placebo. The second design is more informative, because both arms have already achieved weight loss on drug and the only variable that changes at randomization is whether the drug continues.

The STEP 1 off-treatment extension

STEP 1 randomized adults with overweight or obesity without diabetes to once-weekly semaglutide 2.4 mg or placebo for 68 weeks, alongside lifestyle intervention. The primary results, published in the New England Journal of Medicine in 2021, reported a mean body-weight change of about −14.9% on semaglutide versus about −2.4% on placebo.

A subset of participants was then followed for a further year with no study drug and no continued lifestyle intervention. That extension, published in Diabetes, Obesity and Metabolism in 2022, is the single most-cited piece of discontinuation evidence in the field. One year off treatment, participants who had been on semaglutide had regained approximately two-thirds of their prior weight loss. Improvements in cardiometabolic variables followed the same direction of travel.

Two caveats matter when reading it. First, the extension was a subset, not the full trial population, which limits precision. Second, the lifestyle intervention stopped at the same time the drug did — so the extension measures the combined loss of both, not the loss of pharmacotherapy alone.

STEP 4: randomized withdrawal on semaglutide

STEP 4, published in JAMA in 2021, addressed the confounding directly. All participants received semaglutide during a 20-week run-in and escalated to 2.4 mg. At week 20 they were randomized either to continue semaglutide or to switch to placebo, with lifestyle support maintained in both arms through week 68.

From randomization to week 68, the continuation arm lost additional weight while the placebo-switch arm regained. The divergence was substantial and began early. Because lifestyle support was held constant in both arms, the gap is attributable to the drug rather than to the withdrawal of behavioral support.

SURMOUNT-4: the same design on tirzepatide

SURMOUNT-4, published in JAMA in 2024, applied an equivalent structure to tirzepatide. Participants completed a 36-week open-label lead-in on maximum tolerated tirzepatide, then were randomized to continue or switch to placebo for a further 52 weeks.

The continuation arm achieved further weight reduction over the randomized period. The placebo-switch arm regained a large share of what had been lost during the lead-in. The magnitude of the divergence in SURMOUNT-4 is larger than in STEP 4, but the lead-in was also longer and the starting weight loss deeper, so the two are not directly comparable numbers.

What the pattern means, and what it does not

Across designs, drugs, and populations, the direction is uniform: stopping produces regain, regain begins quickly, and the metabolic improvements that accompanied the loss track back with it. The most defensible reading is that these agents produce a treatment effect rather than a cure — closer in kind to antihypertensives than to a course of antibiotics.

What the evidence does not show is equally worth stating. No published trial demonstrates that everyone regains everything. Mean regain is not universal regain, and the distributions in these studies include participants who maintained a meaningful share of their loss. Nor has any trial isolated the effect of a structured taper against abrupt cessation — that comparison has not been run at scale, which means claims about tapering preventing regain currently rest on mechanism and clinical experience, not randomized evidence.

Reading the literature critically

Three methodological points recur and are worth carrying into any article you read on this subject.

Percentage of loss regained is not percentage of body weight regained. "Regained two-thirds of the loss" and "gained back 10% of body weight" describe different quantities. Sources frequently blur them.

Trial populations are not the general population. Registration-trial participants meet inclusion criteria, attend scheduled visits, and receive structured support. Real-world discontinuation happens for reasons — cost, side effects, supply — that also predict worse outcomes independently.

Follow-up windows end. A one-year extension tells you about one year. Whether the regain curve plateaus, continues, or overshoots baseline beyond that window is not something the current published data answers.