Read the top lines of the semaglutide obesity program side by side and STEP 7 looks like the weakest entry: a treatment difference of 8.47 percentage points where the pivotal trial returned 12.4 and the Japan and South Korea trial returned 11.06. Two facts account for most of that gap and neither is about the drug. STEP 7 stopped its clock at 44 weeks rather than 68, and its control group lost 3.6% of body weight, more than either of the control arms in those two 68-week trials. Both are design choices. Neither is visible in the headline percentage.
What was run
STEP 7 was a double-blind, multicenter, randomized, placebo-controlled phase 3a trial recruiting from 23 hospitals and trial centers in China, Hong Kong, Brazil and South Korea. Adults with overweight or obesity, with or without type 2 diabetes, were assigned 2:1 to once-weekly subcutaneous semaglutide 2.4 mg or placebo for 44 weeks, plus a diet and physical-activity intervention. Allocation used blocks of six through an interactive web response system, stratified by diagnosis of type 2 diabetes, and participants, investigators and the sponsor stayed masked until database lock.[1]
Between 8 December 2020 and 23 August 2022, 448 were screened and 375 randomized — 249 to semaglutide, 126 to placebo.[1] The preregistered primary endpoints were the percentage change in mean body weight from baseline to week 44 and the proportion of participants reaching a reduction of at least 5% of body weight at week 44. The registry entry for NCT04251156 lists those same two, in that order, with the same week-44 window.[2]
The clock stopped 24 weeks early
Every other trial cited above ran 68 weeks. This one ran 44, and the difference is not a rounding detail: on the weight trajectories these trials report, the curve is still descending at week 44 and has begun to flatten by week 68. A 44-week endpoint is measured on the way down.
That single fact disqualifies the obvious comparison. Anyone who lines STEP 7’s 12.5% up against a 68-week figure from a different trial is comparing two different points on two different curves, and the arithmetic that results does not mean anything. How the percentages across this program should and should not be read against each other is the subject of the weight-loss expectations article.
The control group that lost 3.6%
Participants assigned to placebo lost 3.6% of body weight over 44 weeks — 3.4 kg — with a standard deviation of 5.9.[2] Thirty-six of 116 evaluable placebo participants, 31%, reached a 5% reduction. Twelve reached 10%, seven reached 15% and two reached 20%, all on diet and activity counseling and a dummy injection.[2]
Set that against the control arms it is usually compared with. The pivotal trial’s placebo group lost 2.4% over 68 weeks,[4] and the Japan and South Korea trial’s pooled placebo group lost 1.9% over the same 68.[5] STEP 7’s control arm therefore lost half again as much weight in two thirds of the time. Since a treatment difference is the gap between two arms, a livelier control arm shrinks it without the treated arm doing anything differently.
Why it was livelier is not answerable from inside the trial. A younger cohort, a more intensive lifestyle program, a different relationship with the trial site, different food environments — the report contains no analysis that separates these, and nothing was randomized that could. The honest statement is that the control condition in STEP 7 worked better than the control condition elsewhere in the program, and that no one can say from this trial whether that is about the participants or about the protocol.
Who was enrolled
The registry records a cohort with a mean age of 41, with 205 of 375 men (55%), 340 participants recorded as Asian, 31 as White and four as Black or African American, and 37 as Hispanic or Latino by ethnicity.[2] The trial title describes the population as predominantly east Asian, and the numbers bear that out precisely: 91% Asian, not 100%. Brazil is in the trial.
Baseline weight, body-mass index and waist circumference for all 375 are not published in any source open to a reader. The abstract does not carry them and the registry’s baseline module posts only age, sex, race and ethnicity. What exists is the China subset described below, and its figures should not be printed as the trial’s.
Two accounts of the primary endpoint
The abstract reports an estimated mean percentage change in body weight of −12.1% (SE 0.5) against −3.6% (SE 0.7), an estimated treatment difference of −8.5 percentage points (95% CI, −10.2 to −6.8), at P < 0.0001.[1] The registry posts observed means of −12.5% (SD 7.4) against −3.6% (SD 5.9), with a treatment-policy estimand difference of −8.47 (95% CI, −10.17 to −6.76).[2] Modeled means and observed means are not interchangeable; the treatment differences are the same quantity.
On the second primary endpoint, 203 of 238 evaluable participants on semaglutide (85%) and 36 of 116 on placebo (31%) reached a 5% reduction, an odds ratio of 13.07 (95% CI, 7.40 to 23.10).[2] Deeper thresholds: 151 against 12 at 10%, 82 against seven at 15%, and 34 against two at 20%.[2] Waist circumference fell 11.1 cm against 3.8, glycated hemoglobin 0.8 percentage points against 0.1, and systolic blood pressure 7 mmHg against 2.[2]
The China analysis, and what it is allowed to say
A separate report covers the 300 Chinese participants — 195 on semaglutide, 105 on placebo, 90% from mainland China and 10% from Hong Kong — as a prespecified analysis outside the statistical testing hierarchy. Its own authors state the consequence plainly: because the analysis sits outside the hierarchy, superiority cannot be ascertained from it.[3] It is prespecified rather than post hoc, which is better; it is still not a tested result.
Within that subset, body weight changed −11.8% against −3.5%, a difference of 8.3 percentage points (95% CI, −10.2 to −6.4), and 85.4% against 26.8% reached a 5% reduction (odds ratio 16.1; 95% CI, 8.4 to 30.9). Mean body-mass index fell 4.0 against 1.2 kg/m².[3] The subset entered at a mean weight of 96.4 kg, a mean body-mass index of 33.8 and a mean waist of 107.5 cm, at a mean age of 40, 59% men, with 26% carrying a diagnosis of type 2 diabetes at screening. Eighty-one percent met the World Health Organization definition of obesity; under the Working Group on Obesity in China thresholds, which set overweight at 24 and obesity at 28, essentially all of them would.[3]
One operational number from that report is worth keeping: 95.3% of semaglutide participants were on the full 2.4 mg dose at week 20 and 91.2% at week 44.[3] Whatever the tolerability table says, nine in ten were still at target dose when the trial ended.
Who left, and which arm left more
Completion was 244 of 249 on semaglutide and 121 of 126 on placebo, with five non-completions in each arm: four withdrawals and one loss to follow-up on each side.[2] Separately, the registry posts permanent discontinuation of trial product at 18 of 249 on semaglutide (7.2%) and 16 of 126 on placebo (12.7%).[2]
That asymmetry runs opposite to what a reader expects from a drug with a gastrointestinal profile, and it comes with no test, no interval and no stated reason breakdown, so it should be read as a fact about this trial rather than as evidence about tolerability. What is certain is that leaving the drug and leaving the trial were different events here: most of the 34 who stopped injecting stayed in for the week-44 measurement, which is exactly what the treatment-policy estimand is built to accommodate. What happens to weight after the injections end is covered in the discontinuation article.
Side effects
Adverse events were reported by 231 of 249 (93%) on semaglutide and 108 of 126 (86%) on placebo, with gastrointestinal disorders the most common category at 168 of 249 (67%) against 45 of 126 (36%).[1] Serious adverse events ran 14 of 249 (5.6%) against eight of 126 (6.3%), and no participant died.[2]
The frequent individual terms at 2.4 mg were diarrhea 65 of 249 (26%) against 13 of 126, nausea 60 (24%) against nine, decreased appetite 43 against five, constipation 29 against seven, vomiting 21 against none, and dyspepsia and eructation 14 each against none.[2] Three terms with a zero in the placebo column is what a drug-specific effect looks like in a table this size. In the China subset, adverse events led seven of 195 (3.6%) to stop trial product permanently against two of 105 (1.9%), with gastrointestinal events accounting for three of the seven.[3] What these symptoms feel like week to week is in the side effects article.
Where STEP 7 and STEP 6 diverge
The two trials share a description — semaglutide 2.4 mg against placebo in adults with overweight or obesity, with or without type 2 diabetes, in east Asian populations — and they share one recruiting country, South Korea. Anyone reading them as one body of evidence should hold four differences in view.
The schedules differ: 44 weeks against 68. The allocations differ: two groups at one dose against four groups at two doses with pooled placebo. The cohorts differ by a decade, 41 against 51, and the enrolled populations differ in composition — all 401 in the earlier trial were recorded as Asian, against 340 of 375 here.[2][5] And the control arms differ by nearly a factor of two. Those four differences move the headline number in the same direction, which is why the smaller treatment difference in STEP 7 is not evidence that the drug works less well in the populations it enrolled.
What STEP 7 did not establish
It has no active comparator. No arm received another weight-loss drug, so the trial ranks semaglutide against counseling and a dummy injection and against nothing else.
It has no 68-week readout, so it cannot say what these participants would have weighed on the schedule the rest of the program used, and it has no off-treatment period, so it says nothing about what happens after week 44. It ran no cardiovascular, renal or mortality endpoint; 375 participants over 44 weeks could not power one. Its Brazilian, Hong Kong and South Korean components are far too small to carry estimates of their own, and none is reported separately. The one geographic analysis that exists, for China, sits outside the testing hierarchy by design.
It also cannot separate its own two explanations. The short clock and the active control arm both compress the treatment difference, and no analysis in the trial, and no arm of it, tells you how much of the 8.47 points belongs to each.
What was in the injection
These figures describe branded semaglutide supplied by the sponsor, titrated on a fixed escalation schedule, injected weekly for 44 weeks alongside a structured diet and activity program, with more than nine in ten participants still at the full dose when the trial closed. Sellers listed on the compounded semaglutide board dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they reach a patient.
The figure to carry away is 8.47 points, with an interval from 10.17 down to 6.76, over 44 weeks. The other 3.6 points of the headline 12.5% belong to a group that got counseling and a dummy injection — and in this trial that group did better than the equivalent group in either 68-week comparison above.