Six Trials Out of Forty-Two

What the analysis set out to do

No trial has randomised all three of tirzepatide, semaglutide and liraglutide against each other, and until recently no trial had randomised tirzepatide against either. A network meta-analysis fills that gap by linking trials through shared comparators: if trial A compares drug 1 with placebo and trial B compares drug 2 with placebo, the two drugs can be compared indirectly through their common placebo arms. This analysis did that for semaglutide 2.4 mg, liraglutide 3 mg and tirzepatide at 5, 10 and 15 mg, in adults without type 2 diabetes with obesity or with overweight plus at least one obesity-related complication.

Six of forty-two, and why that is not automatically a criticism

The systematic review identified 42 randomised controlled trials. Six entered the network: SURMOUNT-1, STEP 1, STEP 5, STEP 8, the O'Neil 2018 dose-ranging trial, and SCALE Obesity and Prediabetes. Excluding 86% of the identified evidence sounds damning and is not, necessarily. An indirect comparison is only valid if the trials being linked are exchangeable - similar populations, similar background therapy, similar outcome definitions. Trials that differ on those dimensions do not add information; they add bias. Discarding them is the correct move. But it is also a decision that determines the answer, made by the analysts, and it is the first thing to interrogate in any network meta-analysis. The paper reports its inclusion reasoning explicitly, which is what allows the interrogation to happen.

What it found

On the efficacy estimand, all three tirzepatide doses produced statistically greater weight reduction than liraglutide, and tirzepatide 10 mg and 15 mg produced statistically greater reduction than semaglutide. Against liraglutide, percentage weight reduction was -12.86% for tirzepatide 10 mg and -13.95% for 15 mg. Against semaglutide 2.4 mg, the corresponding figures were -4.85% and -6.26%. Waist circumference moved in the same direction: -11.79 cm and -12.30 cm against liraglutide, -4.81 cm and -5.32 cm against semaglutide. Safety profiles were reported as generally comparable across all three. Tirzepatide 5 mg was broadly comparable with semaglutide rather than superior to it, which is a detail that tends to disappear in summaries.

The part that makes this analysis unusually interesting

SURMOUNT-5, the open-label head-to-head trial of tirzepatide against semaglutide, was published after this analysis was conducted. The authors say so, and state their reason for not adding it: the maximum-tolerated-dose design meant the response could not be attributed to a specific dose, so it was not translatable to the fixed-dose structure of their network. That makes the comparison that follows genuinely independent. SURMOUNT-5 randomised 751 adults for 72 weeks and reported -20.2% with tirzepatide against -13.7% with semaglutide - a difference of 6.5 percentage points. The network meta-analysis, using no head-to-head data at all, estimated -6.26% for tirzepatide 15 mg against semaglutide 2.4 mg. The indirect estimate and the direct trial agree to within a quarter of a percentage point.

Why that agreement matters, and how far to push it

Indirect comparison is routinely dismissed as a weaker substitute for a real trial, and in the hierarchy of evidence it is. But this is a documented case in which an indirect estimate predicted a subsequent head-to-head result almost exactly, in a field where nobody had run the head-to-head yet. That is worth knowing when a health technology assessment rests on indirect evidence, which many do. The caveats are real and should travel with the claim: SURMOUNT-5 used a mix of 10 mg and 15 mg tirzepatide against a mix of 1.7 mg and 2.4 mg semaglutide, so it is not a clean 15 mg versus 2.4 mg comparison; it was open-label where the network's trials were mostly double-blind; and one agreement does not establish that indirect comparison is generally reliable. It establishes that it was reliable here.

Who paid for it, and why that is stated rather than implied

The study was sponsored by Eli Lilly, which markets tirzepatide, and Lilly also funded the journal's rapid service and open access fees. Five of the nine authors are declared employees and shareholders of Eli Lilly. Three are employees of Costello Medical, which was paid by Lilly for analytical services for this study. The remaining author declares honoraria from AstraZeneca, Boehringer Ingelheim, Lilly and Novo Nordisk. All of that is stated in the paper's own declarations, in full. The correct response is not to discard the analysis - a sponsored study is not a false one, and the SURMOUNT-5 agreement is evidence that this one estimated well - but to hold the funding in mind against the choices only the analysts could make, of which the exclusion of 36 trials is the largest.

Quick reference