Most obesity trials answer whether a drug works. STEP 4 answers a different question, and it is the one a person about to enter a card number actually has: what happens if the injections stop. The design puts everyone on semaglutide first and only then splits them, so the comparison is not drug against no drug but continuing against stopping in people who were already responding. That turns a monthly figure into a horizon, which is the arithmetic laid out in the annual cost article.
The design, and the filter built into it
STEP 4 was a randomized, double-blind, 68-week phase 3a withdrawal study conducted at 73 sites in 10 countries between June 2018 and March 2020. Eligibility required a body-mass index of at least 30, or at least 27 with a weight-related comorbidity, and excluded diabetes.[1]
A total of 902 participants received once-weekly subcutaneous semaglutide during the run-in: 16 weeks of dose escalation followed by 4 weeks at the maintenance dose. At week 20, the 803 participants (89.0%) who had reached 2.4 mg weekly were randomized 2:1 to 48 further weeks of semaglutide (n = 535) or a switch to placebo (n = 268), with lifestyle intervention continuing in both groups.[1]
That 89.0% is the sentence to carry into any generalization. Roughly one participant in nine never got to the randomization, so every figure reported after week 20 describes a cohort pre-selected for having tolerated the full dose for a month. The people for whom nausea decides the outcome — the subject of the tolerability timeline — are underrepresented in the result by construction.
The randomized cohort had a mean age of 46 years (SD 12), was 634 (79%) women, and weighed a mean 107.2 kg (SD 22.7). Mean weight loss during the run-in was 10.6%. Retention was unusually high: 787 participants (98.0%) completed the trial and 741 (92.3%) completed treatment.[1]
The rare trial whose cohort resembles the buyer
Most of the evidence quoted on GLP-1 pages was generated in populations the reader does not belong to: cardiology cohorts averaging 61 years of age and three-quarters male, or nephrology cohorts defined by protein in the urine. STEP 4 is the exception. Its randomized participants averaged 46 years, were four-fifths women, carried a mean weight of 107.2 kg and had no diabetes — which is close to the profile of someone filling in a telehealth intake form for weight loss.[1]
That is why a withdrawal design matters more here than a dose-response one. The question a person in that cohort faces is not how much can be lost in a year, which is answered elsewhere, but whether the result is theirs to keep. STEP 4 is the only randomized trial of semaglutide built to answer it, and it answers it in the arm that stops.
The 2:1 allocation is a design choice worth noticing: twice as many participants were assigned to keep taking the drug as to stop. That keeps the trial ethical and the safety database large, and it means the withdrawal estimate rests on 268 people rather than on half the cohort. The confidence interval on the treatment difference — 2.5 percentage points wide — is the measure of how much that cost in precision, and it is narrow enough that the direction is not in question.[1]
What each arm did over the next 48 weeks
The primary endpoint was percent change in body weight from week 20 to week 68 — measured from the post-run-in weight, not from baseline. With continued semaglutide the mean change was −7.9%. With the switch to placebo it was +6.9%. The treatment difference was −14.8 percentage points (95% CI, −16.0 to −13.5; P < .001).[1]
Two facts are folded into that pair. The continuing arm was still losing weight at week 68, 15 months into treatment, so the trial ended before any plateau rather than at one — a pattern examined in the plateau article. And the withdrawal arm regained while still receiving the same lifestyle counseling, which is the cleanest available answer to whether diet and exercise alone hold a pharmacological result in place.
The endpoints that reversed alongside the weight
Three confirmatory secondary endpoints were prespecified, and all three favored continuing. Waist circumference differed by −9.7 cm (95% CI, −10.9 to −8.5), systolic blood pressure by −3.9 mm Hg (95% CI, −5.8 to −2.0), and the SF-36 physical functioning score by 2.5 points (95% CI, 1.6 to 3.3), each at P < .001.[1]
What comes back on withdrawal is therefore broader than a number on a scale. Waist circumference, blood pressure and self-reported physical function all track the treatment rather than the prior weight loss, which is the argument against treating a course of this drug as a finite project with an end date.
What continuing cost in side effects
Gastrointestinal events were reported in 49.1% of those who continued semaglutide against 26.1% of those switched to placebo. The proportions discontinuing because of adverse events were nevertheless almost identical, at 2.4% and 2.2%.[1]
Read carefully, that is a statement about a survivor cohort. Among people who had already completed a 20-week titration, gastrointestinal symptoms stayed common but rarely ended treatment. It says nothing about the roughly 11% who never reached randomization, and it is not transferable to someone in week three of an escalation.
A year off the drug, measured
STEP 4 stopped at week 68. Its sibling trial followed participants further. STEP 1 randomized 1,961 adults without diabetes 2:1 to 68 weeks of semaglutide 2.4 mg or placebo plus lifestyle intervention, and recorded a mean change in body weight of −14.9% against −2.4%.[4] At week 68 both the treatment and the lifestyle intervention were discontinued, and an off-treatment extension followed a subset of 327 participants for another year.[2]
From week 0 to week 68, mean weight loss was 17.3% (SD 9.3) with semaglutide and 2.0% (SD 6.1) with placebo. By week 120 — a full year after withdrawal — the two groups had regained 11.6 (SD 7.7) and 1.9 (SD 4.8) percentage points of lost weight, leaving net losses from baseline of 5.6% (SD 8.9) and 0.1% (SD 5.8). Cardiometabolic improvements reverted toward baseline for most variables.[2]
Both halves of that are true and they are usually quoted separately. About two-thirds of the weight came back within a year, and about a third was still gone. The standard deviation of 8.9 around a mean net loss of 5.6% is the more honest summary: that interval spans participants who held nearly everything and participants who ended above where they started, and an average conceals which one a given reader would be.
The class pattern: regain scales with response
A 2025 systematic review and meta-analysis pooled 8 randomized controlled trials and 2,372 participants, all with a body-mass index of 27 or greater, to measure what happens after a GLP-1 receptor agonist is stopped. Regain was proportional to the original weight loss. Participants who had taken liraglutide regained 2.20 kg (95% CI, 1.69 to 2.70; P < 0.00001), while those on semaglutide or tirzepatide regained 9.69 kg (95% CI, 5.78 to 13.60; P < 0.00001).[3]
Proportionality is an uncomfortable finding for anyone choosing a molecule on potency alone. The more effective drug had the larger rebound, in kilograms, which means the choice between agents is partly a choice about how much there is to lose on stopping. The tirzepatide version of the same experiment is set out in the SURMOUNT-4 article.
What STEP 4 does not answer
It tested one thing: continuing the full 2.4 mg dose against stopping outright. It did not test a taper, a reduced maintenance dose, an every-other-week schedule or any of the off-label maintenance protocols sold online, which is why the claims examined in the microdosing article have no randomized support behind them. A trial that compares two endpoints of a spectrum says nothing about the middle of it.
It also ran on branded, FDA-approved semaglutide at a labeled dose, with supervised titration and a drug supply that never lapsed. Compounded semaglutide is not FDA-approved, and compounded drugs are not reviewed by the FDA for safety, efficacy or quality before they are dispensed — the distinction is set out in the compounded-versus-brand article.
The practical consequence sits on the invoice rather than in the journal. If the effect depends on continuing, the relevant comparison between sellers is not the first month but the standing rate at a maintenance dose, which is why a price that holds across doses matters more the longer treatment runs. How each figure here was established is described in the methodology.