SURMOUNT-4 is the tirzepatide version of a deliberately awkward experiment: give everyone the drug until it has clearly worked, then take it away from half of them and watch. It is the only randomized evidence of its kind for this molecule, it produced the largest regain figure in the class, and that figure is routinely misread as proof that tirzepatide rebounds harder than semaglutide. The trial that supports part of that reading also undermines the rest of it, which matters before choosing between the tirzepatide sellers and their semaglutide counterparts on rebound risk.
A long lead-in, then a coin flip
The trial was a phase 3 randomized withdrawal study run at 70 sites in four countries. It opened with a 36-week open-label tirzepatide lead-in at the maximum tolerated dose of 10 or 15 mg weekly, followed by a 52-week double-blind, placebo-controlled period. Eligibility required a body-mass index of at least 30, or at least 27 with a weight-related complication, and excluded diabetes.[1]
A total of 783 participants entered the lead-in. At week 36, 670 were randomized 1:1 to continue tirzepatide (n = 335) or switch to placebo (n = 335) for 52 weeks. Those 670 had a mean age of 48 years, were 473 (71%) women, and weighed a mean 107.3 kg. During the lead-in they had lost a mean of 20.9% of body weight.[1]
The gap between 783 and 670 is the same structural caveat that applies to any withdrawal design: the randomized cohort had already spent nine months tolerating a maximum dose, so the figures below describe responders rather than everyone who starts. The 36-week lead-in is nearly twice the length of the semaglutide equivalent described in the STEP 4 article, which is the first reason the two trials are not interchangeable.
What the two arms did over 52 weeks
The primary endpoint was mean percent change in weight from week 36 to week 88. Continuing tirzepatide produced −5.5%. Switching to placebo produced +14.0%. The difference was −19.4 percentage points (95% CI, −21.2 to −17.7; P < .001).[1]
The key secondary endpoint is the one that translates most directly into a decision. At week 88, 300 participants (89.5%) on tirzepatide had maintained at least 80% of the weight lost during the lead-in, against 16.6% of those switched to placebo (P < .001).[1] Roughly five in six people who stopped had given back more than a fifth of their result within a year.
Measured across the whole 88 weeks, the continuing group ended at −25.3% and the withdrawal group at −9.9% from their original baseline.[1] Those two numbers are worth reading together with the regain figure: +14.0% sounds like a reversal, and the arm that experienced it still finished the trial roughly a tenth lighter than it started. Both descriptions are accurate.
The same result in kilograms
Percentages compound in ways that mislead, so it is worth walking the trial in weight. The randomized cohort started at a mean 107.3 kg. A 20.9% lead-in reduction puts it near 85 kg at week 36. Adding back 14.0% of that figure returns roughly 12 kg, landing near 97 kg — which is about 9.9% below the original baseline, exactly the figure the trial reports for the placebo arm across all 88 weeks.[1]
So the withdrawal arm lost about 22 kg and gave back about 12 of them in a year, while the continuing arm lost about 22 and took off another 5. Neither group ended where it started, and the difference between the two after 88 weeks is on the order of 16 kg — the weight of the decision to keep paying for the injections.
The comparison to STEP 4, done carefully
Placed side by side, the two withdrawal arms read +14.0% for tirzepatide and +6.9% for semaglutide, and the obvious inference is that one molecule rebounds twice as hard. Four things block that inference. The lead-ins differed in length (36 weeks against 20). They differed in what they removed (a mean 20.9% against 10.6%). The randomized windows differed (52 weeks against 48). And no trial has ever randomized the same people to withdraw from both drugs, so this is a comparison between separate cohorts recruited years apart.[1][2]
Expressed as a share of what each lead-in had removed, the two look alike: about two-thirds of the lead-in loss returned in each. That arithmetic is approximate, because a lead-in loss is a percentage of baseline weight while a regain is a percentage change from the randomization weight, but the mismatch runs the same way in both trials, so the comparison of shapes survives it.
That shape is also what the pooled evidence reports. A 2025 meta-analysis of 8 randomized trials and 2,372 participants found regain after discontinuation to be proportional to the original weight loss, at 2.20 kg for liraglutide and 9.69 kg for semaglutide or tirzepatide.[4] A more effective drug leaves more to give back, which is a different claim from the drug having a worse withdrawal profile.
One analysis does report a direct contrast. A 2026 systematic review pooled six studies covering 8,993 patients — 5,553 who discontinued and 3,440 who continued — and found a weight difference between the groups of 17.90% (95% CI, 14.11 to 21.69; P < 0.0001), with significant heterogeneity (P = 0.0082). Its subgroup analysis reported the rebound after tirzepatide to be significantly larger than after semaglutide.[5] That subgroup is built from the same separate trials, so it inherits their differences rather than resolving them; it is a signal worth knowing about and not a head-to-head result. The molecule-level evidence is set out in the comparison article.
What the lead-in figure means on its own
A mean 20.9% over 36 weeks is consistent with the pivotal trial. In SURMOUNT-1, 2,539 adults with obesity and without diabetes were randomized to 5, 10 or 15 mg of tirzepatide or placebo for 72 weeks, with mean weight changes of −15.0%, −19.5% and −20.9% against −3.1% on placebo. A reduction of 20% or more was reached by 50% of the 10 mg group and 57% of the 15 mg group, against 3% on placebo.[3] The dose-by-dose picture is in the tirzepatide weight article.
Weight is not the only thing that returns
The three-year SURMOUNT-1 analysis followed 1,032 participants who had both obesity and prediabetes for 176 weeks on treatment, then for a 17-week off-treatment period. On treatment, mean weight change was −12.3%, −18.7% and −19.7% at 5, 10 and 15 mg against −1.3% on placebo, and type 2 diabetes was diagnosed in 1.3% of the tirzepatide groups against 13.3% on placebo (hazard ratio 0.07; 95% CI, 0.0 to 0.1; P < 0.001).[6]
After only 17 weeks off drug, the tirzepatide figure had risen to 2.4% against 13.7% on placebo, and the hazard ratio had moved to 0.12 (95% CI, 0.1 to 0.2).[6] Seventeen weeks is a short interval for a diagnosis rate to nearly double, and it says that what unwinds on withdrawal includes metabolic status rather than a number on a scale alone. The same direction of travel appears in the pooled withdrawal data, where overall and gastrointestinal adverse reactions fell after discontinuation while cardiovascular event incidence was unaffected.[5]
Who the result describes
The randomized cohort averaged 48 years of age and was 71% women, with diabetes excluded and a weight-related complication required of anyone entering below a body-mass index of 30.[1] That is much closer to a telehealth weight-loss patient than the cardiology and nephrology cohorts behind most GLP-1 headlines, which is the main reason the trial is worth reading closely rather than quoting once.
Two differences still matter. Everyone reached and held a maximum tolerated dose of 10 or 15 mg for nine months before the randomized phase began, and the trial ran at 70 sites in four countries with a supervised supply, free product and scheduled visits. Adherence under those conditions is not adherence under a subscription that renews monthly, and a person who stops because a shipment lapses or a price rises is running the placebo arm of this trial without having volunteered for it.
What SURMOUNT-4 does not settle
It tested continuation of a maximum tolerated dose against full withdrawal, and nothing between. No arm stepped down to a lower maintenance dose, lengthened the interval between injections or tapered, so the maintenance protocols advertised by telehealth services have no randomized support from this trial in either direction. The trial also excluded diabetes and ran for 52 weeks after randomization, which is a year rather than a lifetime.
It studied branded, FDA-approved tirzepatide, dosed under supervision. Compounded tirzepatide is not FDA-approved, and compounded drugs are not reviewed by the FDA for safety, efficacy or quality before they are dispensed — a gap described in the compounded-versus-brand article. A withdrawal curve measured on one product is not a property of another that shares its active ingredient.
The consequence for a buyer is a horizon rather than a monthly figure. If holding a 20% result requires continuing to pay for it, the comparison between sellers should be run over years, which is what the cost calculator is for. How each figure on this page was established is described in the methodology.