Articles / Methods
Two numbers from one trial: why obesity drugs report different weight-loss figures
On treatment or everyone randomised? The same dose of the same drug can show a gap of three points depending on the question being asked. How to read both, and why the gap itself tells you something.
Published Oct 23, 2026 ยท 4 min read
Open any obesity trial press release and you will usually find two weight-loss figures for the same dose. Eli Lilly reported 22.5% for tirzepatide 15 mg in SURMOUNT-1, and also 20.9%. Novo Nordisk reported 16.6% for the Wegovy pill in OASIS 4, and also 13.6%. Neither company is hiding anything. The two numbers answer two different questions, and knowing which is which changes how the drugs compare.
Two questions, two estimands
Statisticians call the question a trial answers its estimand. Obesity trials now report two.
On treatment (the efficacy or trial-product estimand) asks: how much weight would people lose if they all stayed on the drug as prescribed? Measurements taken after someone stops treatment are set aside, and the analysis estimates what continued treatment would have done.
Everyone randomised (the treatment-policy or treatment-regimen estimand) asks: what happened to the whole group that was assigned the drug, including people who stopped early, skipped doses or switched? It is closer to what a clinic sees across all its patients.
Because some people always stop, the everyone-randomised figure is lower. How much lower depends on how many stop and how much weight they regain.
The gap, trial by trial
| Trial and dose | On treatment | Everyone randomised | Gap | Placebo (on treatment / everyone) |
|---|---|---|---|---|
| TRIUMPH-1, retatrutide 12 mg, 80 wk | 28.3% | 25.0% | 3.3 | 2.2% / 3.9% |
| OASIS 4, Wegovy pill 25 mg, 64 wk | 16.6% | 13.6% | 3.0 | 2.2% (one figure reported) |
| REDEFINE 1, CagriSema, 68 wk | 22.7% | 20.4% | 2.3 | 2.3% / 3.0% |
| STEP 1, semaglutide 2.4 mg, 68 wk | 16.9% | 14.9% | 2.0 | 2.4% / 2.4% |
| SURMOUNT-1, tirzepatide 15 mg, 72 wk | 22.5% | 20.9% | 1.6 | 2.4% / 3.1% |
| ATTAIN-1, orforglipron 36 mg, 72 wk | 12.4% | 11.2% | 1.2 | 0.9% / 2.1% |
| GLORY-1, mazdutide 6 mg, 48 wk | 14.8% | 14.0% | 0.8 | 0.5% / gained 0.3% |
The Wegovy pill shows one of the widest gaps. That fits its dosing rules: it is taken on an empty stomach with a short wait before eating, which is harder to follow every day than a weekly shot. Foundayo (orforglipron), a small-molecule pill without those rules, shows one of the narrowest gaps. On the on-treatment figure the Wegovy pill leads Foundayo by 4.2 points; on the everyone-randomised figure the lead shrinks to 2.4 points. Which number you quote decides how big the difference looks.
Placebo moves the other way
In most of these trials the placebo group loses more weight on the everyone-randomised basis than on the on-treatment basis: 3.9% against 2.2% in TRIUMPH-1, 2.1% against 0.9% in ATTAIN-1. The trials do not report why. The practical effect is that subtracting placebo widens the gap further. Retatrutide's placebo-adjusted result is 26.1 points on treatment but 21.1 points for everyone randomised; orforglipron's is 11.5 against 9.1.
Which number should you use?
Neither is the true one. The on-treatment figure is the better measure of what a drug can do biologically, and it is the one most consistently reported in early trials, so it is the figure our GLP-1 Performance Index uses. The everyone-randomised figure is the better guide to what a typical patient, who may stop or miss doses, can expect. A large gap between the two is itself information: it usually points to tolerability or convenience problems that matter outside a trial.
Some sources do not name the estimand at all. The headline results we track for survodutide (SYNCHRONIZE-1), petrelintide (ZUPREME-1), amycretin and injectable VK2735 fall into this group, and we flag them wherever they appear.
See it for every trial arm
Our evidence maps let you switch every chart between the two estimands, and between raw and placebo-adjusted results, across 25 trial arms. Switching back and forth is the fastest way to see which comparisons hold up and which depend on the choice of number.
General information, not medical advice. Sources: Eli Lilly, SURMOUNT-1 results; Wilding et al., NEJM 2021 (STEP 1); Wharton et al., NEJM 2025 (OASIS 4); Eli Lilly, ATTAIN-1 results; Eli Lilly, TRIUMPH-1 results; Novo Nordisk, REDEFINE 1; Innovent, GLORY-1.