The figure this trial is quoted for is not the one it was built to produce. STEP 8 was powered to answer whether weekly semaglutide 2.4 mg takes off more weight than daily liraglutide 3.0 mg, and it answered that with a tested endpoint and a confidence interval. What circulates instead is its adherence column — the claim that twice as many people abandon the older drug — and that column was never a tested endpoint at all. It was tabulated, not analyzed, and the protocol had already handed the two arms different ways out. The dosing-interval reading of this trial, along with the counterexample that contradicts it, belongs to the weekly-versus-daily article. What follows is the trial itself.
A small trial, an open label, and two placebos
STEP 8 was a phase 3b, randomized, open-label, 68-week trial at 19 sites in the United States, enrolled between 11 September and 26 November 2019 with follow-up ending 11 May 2021. Eligible adults had a body-mass index of 30 or greater, or 27 or greater with at least one weight-related comorbidity, and did not have diabetes. Of 387 people screened, 338 were randomized in a 3:1:3:1 ratio to semaglutide 2.4 mg (n = 126), liraglutide 3.0 mg (n = 127), or one of two matching placebo groups that were then pooled into a single arm of 85.[1]
The cohort had a mean age of 49 years (SD 13), was 265 women (78.4%), and carried a mean body weight of 104.5 kg (SD 23.8) at a mean body-mass index of 37.5 (SD 6.8). Of the 338, 319 (94.4%) finished the trial and 271 (80.2%) finished on treatment — a fourteen-point gap between staying enrolled and staying on the drug.[1]
Only the active-versus-placebo comparisons were blinded. Nobody was masked as to which of the two drugs they were taking, because a weekly injection and a daily one cannot be disguised as each other. That is an unavoidable design constraint rather than a flaw, and it matters most for the endpoints that depend on what a participant decides rather than on what a scale reads.
The endpoint that was actually tested
The primary endpoint was percentage change in body weight at week 68. Mean change was −15.8% with semaglutide against −6.4% with liraglutide, a difference of 9.4 percentage points (95% CI, −12.0 to −6.8; P < .001), with the pooled placebo arm at −1.9%.[1] The confirmatory secondary endpoints — the shares crossing 10%, 15% and 20% — were the only other outcomes controlled for multiple comparisons, and they are set out in the semaglutide weight article.
Everything else in the paper is exploratory by the authors’ own statement. They write that findings outside the primary and confirmatory secondary endpoints should be interpreted as exploratory, and that the data on permanent treatment discontinuations, on reaching 5%, on adverse events and on pulse were summarized by descriptive statistics only.[1]
The adherence column carries no interval
Overall, 19.8% of participants (n = 67) permanently stopped treatment. Split by arm, that was 27.6% on liraglutide, 17.6% on placebo and 13.5% on semaglutide.[1] No confidence interval attaches to the gap between any two of those numbers, and no P value, because none was calculated. A contrast repeated as though it were a result of the trial is a row in a descriptive table.
The placebo column is the one that should stop a reader. The arm taking nothing quit at a higher rate than the arm taking the drug that produced a 15.8% weight reduction. Whatever drives people out of an obesity trial, in STEP 8 it was not simply having an active drug in the syringe.
The exits were not the same exit
Semaglutide was started at 0.25 mg and escalated to 2.4 mg over sixteen weeks, with a 1.7 mg maintenance dose permitted if 2.4 mg proved intolerable and at least one attempt at reescalation advised. Liraglutide was started at 0.6 mg and escalated to 3.0 mg over four weeks, and — matching its prescribing information — treatment was discontinued if 3.0 mg was not tolerated, with any restart beginning the four-week climb again.[1]
One arm had a rung to step down to. The other had a door. The trial report lists this first among its limitations and states plainly that the difference could have led more participants to stop liraglutide permanently after an adverse event, and that liraglutide’s weight result could itself have been affected because participants may have spent less time on treatment.[1] That second clause is the less-quoted half: the asymmetry touches the efficacy column too, not only the adherence one. How a dose ladder is climbed, and what a step down costs, is covered in the titration article.
Severity ran the wrong way
Adverse events of any kind were reported by 95.2%, 96.1% and 95.3% of the semaglutide, liraglutide and placebo arms respectively — three numbers within a point of each other. Severe gastrointestinal events occurred in 3.2% on semaglutide, 2.4% on liraglutide and 3.5% on placebo, which puts the inert arm at the top of that list. Serious adverse events ran 7.9%, 11.0% and 7.1%.[1]
Those numbers are small and the trial was not powered to separate them. The useful inference is negative rather than positive: a background rate of severe gut symptoms exists in a population of this kind before any drug is added, and a page attributing every reported symptom to the injection is describing a comparison the trial did not find.
Liraglutide was not having a bad day
A reader could reasonably suspect the comparator was set up to lose. Its own registration trial says otherwise. SCALE randomized 3,731 adults without diabetes 2:1 to liraglutide 3.0 mg or placebo for 56 weeks and reported a mean loss of 8.4 ± 7.3 kg against 2.8 ± 6.5 kg, a difference of 5.6 kg (95% CI, −6.0 to −5.1), with 63.2% against 27.1% losing at least 5% of body weight.[2] STEP 8’s liraglutide arm performed in the range its own program had established. The gap is a real difference between two products at their own labeled maximums.
What one in eight looks like outside a trial
A retrospective cohort followed 125,474 US adults with overweight or obesity who newly started liraglutide, semaglutide or tirzepatide between 2018 and 2023, drawn from electronic health records. One-year discontinuation was 64.8% (95% CI, 64.4% to 65.2%) among those without type 2 diabetes and 46.5% (95% CI, 46.2% to 46.9%) among those with it. Moderate or severe gastrointestinal events raised the hazard of stopping (HR 1.19; 95% CI, 1.12 to 1.27 without diabetes), and each additional 1% of weight lost lowered it by 3.3%.[3]
Set 13.5% over 68 weeks beside 64.8% over 52. The trial figure describes people with scheduled visits, a study team managing escalation, free drug and a protocol-defined fallback dose. The cohort figure describes people paying, refilling and deciding alone. Neither is wrong; they are answers to different questions, and the one a cash-pay buyer resembles is the second. What a year of that costs is estimated by the cost calculator.
What STEP 8 could not settle
It counted no cardiovascular events, measured nothing past week 68, and followed nobody after treatment stopped — the question handled in the stopping article. It excluded diabetes, enrolled at United States sites only, and ran a cohort roughly four-fifths female, so it says nothing about how the two drugs compare in a man with type 2 diabetes.
It also no longer names the strongest available option. SURMOUNT-5 later randomized 751 adults with obesity and without diabetes, open-label, to the maximum tolerated dose of tirzepatide (10 or 15 mg) or of semaglutide (1.7 or 2.4 mg) for 72 weeks, and reported a least-squares mean weight change of −20.2% (95% CI, −21.4 to −19.1) against −13.7% (95% CI, −14.9 to −12.6), with waist circumference falling 18.4 cm against 13.0 cm.[4] STEP 8’s winner has since been the comparator in a trial it lost.
One more limit is structural. The placebo estimate rests on 85 people assembled from two separate placebo groups with two different injection schedules, and it landed at −1.9% — close to, but not the same as, the −2.4% seen in the much larger blinded placebo arm of STEP 1.[5] A pooled placebo of that size supports the comparison it was built for and very little else.
What was in the syringe
Both active arms held branded, FDA-approved product at a labeled dose, escalated under supervision, alongside counseling every four to six weeks on a 500 kcal/day deficit and a physical activity target.[1] Most sellers on the semaglutide board dispense compounded preparations instead. Compounded drugs are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed, a distinction set out in the compounded-versus-brand article.
That matters more here than in most trials, because the thing being quoted is a tolerability comparison. STEP 8’s semaglutide arm had a 1.7 mg rung written into the protocol and a study team to authorize it. A compounded vial dosed in units, from a seller with no labeled maintenance dose to fall back to, reproduces neither the ladder nor the fallback, and the 13.5% is a figure produced by both.