Two sentences can describe one measurement. “Semaglutide improved blood vessel function, p=0.03.” And: “Semaglutide did not significantly improve blood vessel function versus placebo, p=0.16.” Both are accurate descriptions of the same number in the same study, and the difference between them is the entire subject of choosing what to compare against.
The study
A post hoc analysis of a 32-week randomized trial that had been designed to look at cardio-renal organ damage. One hundred and twenty participants with type 2 diabetes, aged 50 or over, were randomized to semaglutide, empagliflozin, both together, or placebo. [1] The analysis looked at a measure of how well blood vessels widen in response to demand, and at four adhesion molecules the endothelium releases.
The same estimate, twice
In the semaglutide group the vessel-response index rose 0.11 from each participant’s own starting value, with a 95% confidence interval of 0.008 to 0.21 and p=0.03. Compared with the placebo group it also rose 0.11 — with a confidence interval of −0.04 to 0.24 and p=0.16.
Nothing about the measurement changed. What changed is what it was held up against. People in a trial tend to improve for reasons that have nothing to do with the drug: closer monitoring, changed habits, regression to the mean, the season. A placebo group exists to absorb all of that, and when you subtract it the effect here stopped being distinguishable from noise.
The markers disagreed with each other
Four adhesion molecules were measured and they did not move together. E-Selectin fell against placebo, by 9 in both the semaglutide group, 95% CI −14.1 to −5.1, and the combination group, 95% CI −14.3 to −5.2, both at p<0.01 — a change in the direction usually read as healthier.
VCAM-1 went the other way. It rose against placebo, by 12.3 in the semaglutide group, 95% CI 2.8 to 20.8, p=0.01, and by 16.2 in the combination group, 95% CI 7.2 to 24.3, p<0.01. P-Selectin and ICAM-1 did not move significantly in any group. The authors describe the responses as heterogeneous and suggest the molecules play distinct roles, which is an honest way of saying nobody knows what the pattern means.
Note what a selective summary could do with that. Quote E-Selectin, omit VCAM-1, and this becomes a study showing improved endothelial health. It is the same move a composite endpoint makes structurally, performed by hand.
Why any of this matters to a reader
Not because endothelial biomarkers should influence a purchase. They should not — these are laboratory measures in 120 people, not events in anybody.
It matters because “improved from baseline” is everywhere. It appears in clinic testimonials, in single-arm studies, in before-and-after marketing, and in summaries of real trials that had a placebo group available and quoted the baseline comparison anyway. Once you notice the phrase you cannot stop seeing it, and it is the same weakness that makes an uncontrolled study unable to show what it appears to show.
The test is one question: compared to what? If the answer is “to how they were before,” you have been told how a group of people changed over time, not what the drug did. Almost nothing a seller publishes survives that question, which the disclosure scorecard measures from the other direction.