There are two ways to run a drug trial and they are not interchangeable. In one, every participant gets the same thing and a group gets nothing that looks like it. In the other, participants are split between two real treatments and nobody gets a placebo at all. The second design looks, to a headline, like a stronger version of the first. It is a different experiment, and the difference shows up in exactly the places that matter when you read a weight number.
Both designs produced big numbers in the obesity drug literature. We have covered each drug’s own trial data on its own terms. What follows is the comparison between the two designs, using one trial of each.
What a placebo-controlled trial can and cannot tell you
The 2023 retatrutide trial enrolled 338 adults with obesity and ran for 48 weeks. It randomised people across five active amount levels and one placebo arm in fixed proportions. The headline was a clean staircase: roughly seven percent mean body weight change at the lowest level, roughly thirteen at the middle, roughly twenty-two at the highest after 48 weeks, against roughly two percent on placebo.
That design is excellent at answering “does this work at all, and does more of it do more?” The placebo arm is doing work here that no comparison arm could do. Without it you cannot tell whether an eight-percent drop came from the drug or from the fact that everybody in the study suddenly ate differently, weighed themselves more often, and expected to lose weight. Placebo response in weight trials is not a rounding error, which is why it gets an arm of its own.
What the comparison trial adds
The 2025 trial randomised 751 adults with obesity and without diabetes to tirzepatide or semaglutide, each at the highest level each participant could tolerate, for 72 weeks. Mean weight change was just over twenty percent on tirzepatide against just under fourteen on semaglutide, a gap of roughly six and a half percentage points that cleared the threshold for statistical significance by a wide margin. Waist circumference moved by about eighteen centimetres against about thirteen. Both endpoints favoured one drug clearly.
What the comparison arm buys you is exactly the thing a placebo arm cannot produce: a statement about ranking. Each drug has its own placebo-controlled evidence base, so in principle you could subtract. In practice you cannot, because the populations differ, the background diets differ, the sites differ, and the two trials were run at different times with different comparators. So the comparison trial answers the ranking question directly instead of by arithmetic. The head-to-head numbers are set out in full in the Polish version of this piece’s research summary.
And here is the thing the headline leaves out
The comparison trial was open-label. Nobody was blinded to which drug they were getting, and for a drug that requires a self-administered weekly injection, with a visible syringe, a step-up schedule and effects you feel in your gastrointestinal tract within days, that is not a technicality.
Expectation is a live variable here. People who know they received the stronger regimen may report side effects differently, may persist longer, may change adjacent behaviours more. The placebo arm is immune to this by construction, which is one of the reasons placebo arms exist at all. The gap would very probably be smaller under blinding. The gap was still large enough to survive that objection in an open-label design, which is worth saying in the other direction too: this was not a fragile result.
Phase 2 is a mapping exercise: find the plausible range, map the side effects, do not yet prove anything at scale. Reading it as a preview of a drug is fair. Reading it as a comparison against semaglutide is not, because there is none in that dataset.
What the data do not show
Neither design shows that a six-and-a-half-point difference in mean weight change translates into a six-and-a-half-point difference in how a person feels, functions or lives. Mean changes hide the spread: without the full distribution you cannot see what happened to the participants in the worst decile, and in any drug with gastrointestinal side effects that is the decile people remember. Neither design shows durability. Both trials end when the trial ends, and obesity is explicitly a condition that regains what it loses, so the weight on the last study day is the least certain number in the whole dataset. And an open-label comparison, as above, cannot claim to be free of expectation effects.
The part a shop should not tell you
If a page selling either of these drugs leads with a percentage, ask which design produced it. A twenty percent number from a placebo-controlled trial and a twenty percent number from an open-label comparison answer different questions, and the difference between them is not a detail of presentation. It is the difference between “this works” and “this works better than the other one”, and only one of those is a claim you can make about a product you are holding.
This article summarises two published trials. It is not medical advice and it is not a recommendation for any individual.
