SURMOUNT Post Hoc Analysis: What a Composite Endpoint Does and Does Not Show
A SURMOUNT post hoc analysis in PLOS ONE, published 13 August 2026, goes after something the original trials never set out to test: how often people managed to hit three separate targets all at once. The headline result is notable. Still, the way the authors built the question might matter more, because that framing sets the limits on what the result is actually worth.
This isn’t just a tirzepatide story. After a trial is out in the world, two moves show up again and again: post hoc re-reads of the dataset, and composite endpoints that roll multiple outcomes into one. Both can be informative, and both can also invite people to read too much into a clean-looking number. When a paper combines the two, you get a clear view of the mechanics.
What the SURMOUNT post hoc analysis actually counted
Pulling from SURMOUNT-1 through SURMOUNT-4, the authors included participants randomised to tirzepatide or placebo who had a valid baseline and at least one post-baseline measurement for three measures: body weight, systolic blood pressure, and non-high-density lipoprotein cholesterol.
From there, they asked a simple counting question: how many people met all three thresholds at the same time?
The three thresholds were:
- body weight loss, evaluated in three separate versions, at least 5%, at least 10%, or at least 15%
- a systolic blood pressure drop of at least 5 mmHg
- non-HDL cholesterol under 130 mg/dL
They assessed this at each study’s primary endpoint timepoint, reported as week 72 for SURMOUNT-1, -2, and -3, and week 88 for SURMOUNT-4.
Their stated logic is straightforward: modest weight loss is often linked to improvements across several health markers, so they wanted to know how often those changes show up together in the same participants instead of being spread out across different people.
What numbers they reported
Across the four trials, the paper reports that 32% to 38% of participants on tirzepatide reached the “all three targets” combination when the weight-loss bar was set at 5%, compared with 2% to 8% on placebo.
When the weight-loss requirement was 10%, the figures were 28% to 37% on tirzepatide versus 1% to 5% on placebo. At the 15% weight-loss level, they report 22% to 34% versus 1% to 3%.
They also report statistically significant between-treatment odds ratios across trials, with p values under 0.001. The conclusion is what those gaps would suggest: more participants met the triple endpoint on tirzepatide than on placebo. Then comes a key positioning line: they point out that studies looking at possible cardiovascular benefits are still underway.
That one clause places the whole analysis. It tallies risk factors improving in parallel. It does not tally cardiovascular events, and the authors acknowledge that the trials meant to answer that question have not reported yet.
What the original SURMOUNT trial was built to show
A post hoc analysis does not float free. It inherits the design and priorities of the trials that produced its data.
SURMOUNT-1, published in the New England Journal of Medicine in July 2022, was a phase 3, double-blind, randomised controlled trial. It assigned 2,539 adults with BMI 30 or more, or 27 or more with at least one weight-related complication and excluding diabetes, in a 1:1:1:1 split to once-weekly subcutaneous tirzepatide at 5 mg, 10 mg, or 15 mg, or placebo, over 72 weeks, including a 20-week dose-escalation period.
Its coprimary endpoints were prespecified: percent change in body weight from baseline, and the proportion achieving at least 5% weight loss. That prespecification is the difference between a primary endpoint and something observed later.
The trial reported mean weight change at week 72 of -15.0% with 5 mg, -19.5% with 10 mg, and -20.9% with 15 mg, versus -3.1% with placebo, with p below 0.001 for each comparison. It also reported that the most common adverse events on tirzepatide were gastrointestinal, generally mild to moderate, and most often during dose escalation.
So the 2026 SURMOUNT post hoc analysis is re-counting outcomes produced inside that framework, plus the other trials in the programme.
Why post hoc and prespecified are not interchangeable
The composite endpoint issue
Rolling three targets into a single “met all three” figure can be perfectly reasonable, especially if the clinical question is whether improvements stack up in the same person. But the structure brings a catch worth saying plainly: the composite is only as helpful as its weakest piece.
A non-HDL cholesterol target under 130 mg/dL, a 5 mmHg systolic blood pressure drop, and a 5% body weight loss are not the same kind of achievement, and they do not map onto long-term outcomes in identical ways. Combine them and you get one easy-to-share percentage, but it is harder to interpret, because that single number does not tell you which component is driving the result.
That isn’t an accusation against the authors. They describe the composite clearly, and they run the analysis separately for each weight-loss threshold. The warning is about what happens after publication, when a composite percentage travels without its internal wiring.
This isn’t the first post hoc output from SURMOUNT
The 2026 analysis also sits in a broader pattern. A January 2024 post hoc paper in Diabetes, Obesity and Metabolism looked at predicted 10-year atherosclerotic cardiovascular disease risk among SURMOUNT-1 participants without prior ASCVD. In 2,461 participants, the authors reported a low baseline median risk score across groups, about 1.5% to 1.6%, and a larger relative change in predicted risk from baseline to week 72 with tirzepatide, -23.5% to -16.4%, than with placebo at 12.7%, again with p below 0.001.
Two points from that earlier work matter when you read the newer one. First, the outcome was a model output, a predicted risk score, not observed cardiovascular events. Second, baseline risk was low, as the authors noted, which means even large relative shifts can sit on small absolute numbers.
A trial programme producing multiple post hoc papers isn’t inherently suspect. It is often exactly what big datasets are meant to support. The burden shifts to the reader: keep straight what was specified before the data existed, and what was asked after the fact.
Who did the work, and why it belongs next to the result
Both the 2026 SURMOUNT post hoc analysis and its 2024 predecessor include disclosures that should be read alongside the statistics.
In the 2026 paper, several authors are employees and shareholders of Eli Lilly and Company, and the first author reports consulting or speaker fees from many pharmaceutical companies including Eli Lilly, plus institutional grant support from others. In the 2024 paper, most authors are directly affiliated with Eli Lilly, with co-authors at a contract data services firm and the University of Alabama at Birmingham.
The original SURMOUNT-1 publication also lists Eli Lilly authors alongside academic investigators, which is typical for industry-sponsored phase 3 trials.
None of this automatically breaks the math. A peer-reviewed, open-access analysis with clear conflict statements is far more informative than vague claims without any disclosures. The point is narrower: sponsorship influences which questions get asked in the first place, and in post hoc work the question is much of the story.
Why this matters to people at the bench
These papers evaluate a prescription drug in human participants under regulatory oversight. That context is not the same as a compound handled as lab material, and it is easy to blur the two if you are not careful.
If someone is working with tirzepatide as a research compound, they are not the SURMOUNT population, and nothing here converts into a bench protocol. What does carry over is a way of reading: check whether an endpoint was prespecified, check whether a composite is masking variation among its components, and check whether the outcome is directly measured or produced by a model.
Those habits apply just as much in preclinical work, where composite scores and retrospective subgroup comparisons are routine. Anyone trying to map which compounds have the deepest literature base will find our ranking of the most-studied research peptides of 2026 a more useful starting point than any single trial readout, and anyone new to the terms will find the research use only designation defined in our glossary.
What to take away
The 2026 SURMOUNT post hoc analysis reports a real, sizable separation between tirzepatide and placebo on a combined metabolic target, and it shows that separation across four trials.
It still isn’t a report of cardiovascular events. It also isn’t the question those trials were designed around, and the composite format means the headline percentage is something you should break apart, not treat as a single clean claim. The authors signal that themselves by pointing to cardiovascular outcome trials that are still in progress.
Kept inside those boundaries, the analysis adds useful context. Treated as standalone proof of cardiovascular benefit, it goes beyond what the data here can support.
References
- Sattar N, Srinath R, Garcia-Perez LE, et al. Achieving the triple endpoint of body weight reduction thresholds, systolic blood pressure reduction ≥5 mmHg and non-HDL cholesterol <130 mg/dL with tirzepatide in people with obesity: A post hoc analysis from the SURMOUNT trials. PLoS One. 2026;21(8):e0345032. https://pubmed.ncbi.nlm.nih.gov/42594122/ · https://doi.org/10.1371/journal.pone.0345032
- Jastreboff AM, Aronne LJ, Ahmad NN, et al. Tirzepatide Once Weekly for the Treatment of Obesity. N Engl J Med. 2022;387(3):205-216. https://pubmed.ncbi.nlm.nih.gov/35658024/ · https://doi.org/10.1056/NEJMoa2206038
- Hankosky ER, Wang H, Neff LM, et al. Tirzepatide reduces the predicted risk of atherosclerotic cardiovascular disease and improves cardiometabolic risk factors in adults with obesity or overweight: SURMOUNT-1 post hoc analysis. Diabetes Obes Metab. 2024;26(1):319-328. https://pubmed.ncbi.nlm.nih.gov/37932236/
The products referenced on this site are supplied for laboratory research use only. They are not medicines and are not intended for human or veterinary use. This article summarises published research for informational purposes and is not medical advice. Statements about third-party studies belong to their authors.
