The most misused figure in this market is not a fabricated one. It is a real result, from a real trial, applied to the wrong people. STEP 2 ran semaglutide 2.4 mg for 68 weeks in adults who had type 2 diabetes and produced a mean weight change of −9.6%. The famous −14.9% came from a trial that excluded diabetes. Both are quoted from the same molecule at the same dose, and the weight-loss summary reports them side by side. Establishing why they differ is what this page is for.
What STEP 2 actually was
STEP 2 was a double-blind, double-dummy, phase 3 superiority study that recruited from 149 outpatient clinics in 12 countries across Europe, North America, South America, the Middle East, South Africa and Asia. Eligibility required a body-mass index of at least 27 and a glycated hemoglobin of 7% to 10%, with type 2 diabetes diagnosed at least 180 days before screening.[1]
Between June 4 and November 14, 2018, 1,595 patients were screened and 1,210 were randomly allocated 1:1:1, stratified by background glucose-lowering medication and glycated hemoglobin, to semaglutide 2.4 mg (n = 404), semaglutide 1.0 mg (n = 403) or matching placebo (n = 403), once a week for 68 weeks with a lifestyle intervention.[1]
Estimated change in mean body weight to week 68 was −9.6% (SE 0.4) with semaglutide 2.4 mg against −3.4% (0.4) with placebo, an estimated treatment difference of −6.2 percentage points (95% CI, −7.3 to −5.2; p < 0.0001).[1]
The variable is the eligibility criterion
The comparator trial enrolled 1,961 adults with a body-mass index of 30 or greater, or 27 or greater with a weight-related coexisting condition, and without diabetes, for the same 68 weeks at the same 2.4 mg dose.[2] Its design and estimand are set out in the STEP 1 article.
Line the two protocols up and one thing changes. The dose is identical, the duration is identical, the sponsor is the same, the lifestyle intervention runs in every arm of both. What differs is that one trial made type 2 diabetes an exclusion and the other made it a requirement, with a glycated hemoglobin floor of 7% attached. The 5.3-point gap between −14.9% and −9.6% is therefore an estimate of what that eligibility criterion is worth, not an estimate of how well the drug works.
That distinction has a practical edge. A person with diabetes reading the larger figure is being shown a result measured in people selected for not having their condition, and the two indications behind those trials are separate approvals with separate labeling, which is the subject of the two-approvals article.
The third arm nobody quotes
STEP 2 was not only a placebo comparison. Its middle arm received semaglutide 1.0 mg, the dose approved for treating diabetes, which makes the trial a dose comparison inside the same population as well.
The cost of the larger dose shows in the tolerability column. Adverse events were reported in 353 of 403 patients (87.6%) on 2.4 mg, 329 of 402 (81.8%) on 1.0 mg and 309 of 402 (76.9%) on placebo. Gastrointestinal adverse events, mostly mild to moderate, were reported by 63.5%, 57.5% and 34.3% of those arms respectively.[1] Going from the diabetes dose to the obesity dose added about six percentage points of gastrointestinal events.
An endpoint that moved the other way
An exploratory analysis looked at kidney measures. In STEP 2, where 1,205 patients (99.6% of the cohort) had data, the geometric mean baseline urine albumin-to-creatinine ratio was 13.7, 12.5 and 13.2 mg/g in the 1.0 mg, 2.4 mg and placebo groups. At week 68 that ratio had changed by −14.8% and −20.6% on the two semaglutide doses and by +18.3% on placebo, giving between-group differences of −28.0% (95% CI, −37.3 to −17.3; p < 0.0001) and −32.9% (−41.6 to −23.0; p = 0.003).[3]
Pooling STEP 1, 2 and 3, among the 3,379 participants with data there was no difference between semaglutide 2.4 mg and placebo in estimated glomerular filtration rate trajectory at week 68.[3] The authors' own reading is that semaglutide improved albuminuria in people with overweight or obesity and type 2 diabetes, and did not affect filtration rate decline in people with normal kidney function.
Read together, those two results invert the ranking. The population with the smaller weight figure is the population where a kidney endpoint moved at all, because it is the population whose kidneys were under strain. A smaller number on one axis is not a smaller benefit, and the broader kidney evidence sits in the kidney article.
Two tempting explanations that do not hold
The trials differ in racial composition, and that is the first place an argument usually goes. STEP 2 participants reported race as White (59.4%), Asian (27.3%), Black (8.9%) or another group (4.4%), against 75.3% White and 10.6% Asian in the pooled STEP 1 and 3 population. But the same analysis found no significant interaction between treatment effect and race in STEP 2 (p ≥ 0.15) or in STEP 1 and 3 (p ≥ 0.07), and none by ethnicity (p ≥ 0.85 and p ≥ 0.40).[4] Composition differs; responsiveness did not.
The second explanation is background medication, and metformin is the usual candidate. A post hoc analysis pooled 16,996 participants across four trials including STEP 2, of whom 12,928 (76%) were treated with metformin. Concomitant metformin did not increase the percentage of participants developing gastrointestinal adverse events, their severity, or the rate of discontinuation; numerically, metformin nonusers had more of both.[5] How these drugs interact with other tablets is covered in the oral medications article.
The same gap appears in the other molecule
SURMOUNT-2 ran tirzepatide at 10 mg and 15 mg for 72 weeks in 938 adults with a body-mass index of 27 or higher and a glycated hemoglobin of 7% to 10%. Mean age was 54.2 years (SD 10.6), 476 (51%) were female, baseline body weight was 100.7 kg (SD 21.1), body-mass index 36.1 (SD 6.6) and glycated hemoglobin 8.02% (SD 0.89). Least-squares mean change in body weight at week 72 was −12.8% (SE 0.6) and −14.7% (0.5) against −3.2% (0.5) on placebo.[6]
Against the non-diabetes trial described in the SURMOUNT-1 article, whose participants had a mean baseline weight of 104.8 kg, the top arm lands several points lower. Two sponsors, two molecules, the same direction. Whatever is happening is a property of the population rather than of either drug.
Why comparing across trials is the actual error
There is a formal way to compare two trials that never met, and someone has done it. A 2025 indirect treatment comparison applied the Bucher method to the efficacy estimand, setting tirzepatide 10 and 15 mg at week 72 in SURMOUNT-2 against semaglutide 2.4 mg at week 68 in STEP 2. It reported mean differences in percent weight reduction of 2.57% and 4.79% (p < 0.01) favoring tirzepatide, and differences in glycated hemoglobin of 0.47% and 0.56% (p < 0.001).[7]
The threshold outcome is where it wobbles. Odds of reaching a 5% reduction favored tirzepatide 15 mg at an odds ratio of 1.76 (95% CI, 1.04 to 2.97; p = 0.035) but not 10 mg, at 1.24 (0.75 to 2.04; p = 0.407), and the authors note that sensitivity analyses diverged when the populations were matching-adjusted and power fell.[7] This is a comparison built by the manufacturer of one of the two drugs, with a declared method, a declared estimand and matched endpoints, and parts of it still do not survive a sensitivity check.
Which is the standard against which a marketing comparison should be read. Setting −20.9% beside −9.6% on a landing page uses two different populations, two different trial lengths, two different estimands and no adjustment of any kind. The result of doing that is not an approximation of the right answer. It is a number that corresponds to nothing that was measured.
What none of this describes
All of these results were produced with branded, FDA-approved products at labeled doses under supervised escalation. The sellers listed on the semaglutide board mostly dispense compounded preparations, which are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed. A person with type 2 diabetes buying one of them is two steps away from the trial being quoted: the wrong population, and a preparation no trial tested.