Why One Trial Reports Two Different Results
A single retatrutide trial is correctly reported as both 28.3% and 25.0%. Here is why neither number is wrong, and why the difference matters more than almost anything else in reading this evidence.
In May 2026, Eli Lilly announced that participants taking retatrutide 12 mg in its TRIUMPH-1 trial lost an average of 28.3% of their body weight over 80 weeks. In September 2026, the same trial was published in the New England Journal of Medicine, and its abstract reported 25.0%.
Same trial. Same dose. Same participants. A gap of 3.3 percentage points — which on a 112 kg starting weight is roughly 3.7 kg.
Nobody made an error, and nobody was being dishonest. The two numbers answer two different questions, and the name for that distinction is the estimand.
The problem every long trial has to solve
Imagine running an 80-week trial. You randomise 2,339 people. Over the following year and a half, real life happens:
- some stop treatment because of side effects;
- some stop for reasons unrelated to the trial — they move, they lose interest, they get ill;
- some start a different weight-management treatment that the protocol prohibits;
- some reduce their dose and continue;
- some take every dose exactly as planned.
At week 80 you weigh everyone. Now: what number do you report?
That is not a statistical technicality. It is a genuine question about what you are even trying to measure, and different answers are appropriate for different purposes.
The two answers
The efficacy estimand
Eli Lilly's definition, quoted from its own release:
efficacy had all randomized participants remained on study intervention (with possible dose interruptions and modifications) without initiating prohibited weight management treatments Eli Lilly, 21 May 2026
In plain terms: what does this treatment do when people actually take it? It estimates the effect under the assumption of continued treatment. This is the question a clinician asks when deciding whether a treatment is worth trying for a patient who will take it.
The treatment-regimen estimand
the average treatment effect regardless of adherence to study intervention or initiation of prohibited weight management treatments Eli Lilly, 23 July 2026
In plain terms: what happens to everyone who starts this treatment? People who stopped at week 12 are still counted, with whatever weight they had at week 80. This is closer to the question a health system asks: across a whole population assigned this treatment, what is the average result?
What this does to the numbers
| Group | Efficacy estimand | Treatment-regimen estimand | Gap |
|---|---|---|---|
| Retatrutide 4 mg | −19.0% | −17.6% | 1.4 |
| Retatrutide 9 mg | −25.9% | −23.7% | 2.2 |
| Retatrutide 12 mg | −28.3% | −25.0% | 3.3 |
| Placebo | −2.2% | −3.9% | −1.7 |
Two things in that table are worth sitting with.
First, the gap widens with dose. It is 1.4 points at 4 mg and 3.3 at 12 mg. That is not a coincidence: discontinuation due to adverse events also rose with dose, from 4.1% at 4 mg to 11.3% at 12 mg. More people leaving the highest-dose arm means a bigger difference between “those who stayed” and “everyone.”
Second, the placebo group moves the other way. Placebo looks better under the treatment-regimen estimand (−3.9%) than the efficacy estimand (−2.2%). Which makes sense: a participant who left the placebo arm and started something else that worked would carry that weight loss into the treatment-regimen figure.
So the difference between retatrutide and placebo shrinks from 26.1 points to 21.1 points depending on which estimand you use. Both differences are large. They are not the same number.
Why neither is the “real” one
It is tempting to decide that one figure is honest and the other is marketing. Resist that.
If you want to know what a treatment is capable of for someone who takes it as intended, the efficacy estimand is the right question and the treatment-regimen estimand understates it by mixing in people who stopped early.
If you want to know what will happen across everyone who starts it — including the proportion who will not tolerate it — the treatment-regimen estimand is the right question and the efficacy estimand overstates what you will see in practice.
The useful instinct
A well-reported trial gives both, as TRIUMPH-1 did. A well-reported number says which one it is. When you see a weight-loss percentage for any medicine in this class and the source does not state the estimand, treat the number as incomplete — not false, but missing the context that makes it mean something.
Why this breaks cross-trial comparisons
This is where the idea earns its keep. Suppose you want to compare retatrutide with another molecule in this class, and you pull a percentage from each.
If one figure is an efficacy estimand and the other a treatment-regimen estimand, your comparison is broken before population differences, baseline weight, trial duration or era are even considered. Within TRIUMPH-1 alone, that single analytical choice moves the headline by 3.3 points — which is larger than many of the differences people argue about between medicines.
This is one of several reasons we do not publish head-to-head numerical comparisons between retatrutide and tirzepatide or semaglutide. The trial built to answer that question properly, TRIUMPH-5, randomises participants between the two molecules and analyses both arms under the same estimand. Until it reports, the comparison does not exist in a reliable form.
A short checklist
When you next meet a weight-loss figure from a trial, ask:
- Which estimand? If unstated, the number is incomplete.
- At what time point? 68 weeks and 80 weeks are different measurements.
- In which population? Diabetes status alone shifts results substantially — TRIUMPH-1 reported 28.3% and TRIUMPH-2 reported 20.8%, both at 12 mg.
- Versus what? A percentage without its placebo comparator tells you very little.
- Pre-specified or post-hoc? Analyses planned in advance carry more weight.
Five questions. They will let you read almost any trial report in this field more carefully than most of the coverage written about it.
A reminder about what these numbers are
Everything above concerns how to read trial results accurately. It is not advice about using anything. Retatrutide is an investigational molecule with no marketing approval in any country, and its developer states it cannot be legally sold for human use. Full disclaimer →