Reading a GLP-1 Headline Without Getting Misled

A GLP-1 headline is usually true and usually incomplete. The figure in the headline is real; the figure that would tell you what it means for someone like you sits one layer down, in the trial report. Three things get lost in that gap most often: whether a risk reduction is relative or absolute, whether the endpoint is something people experience or a marker standing in for it, and which participants were counted. Each one is easier to see with a published trial in front of you.

Tray, breakfast, muesli — illustrating Reading a GLP-1 Headline Without Getting Misled

Relative always sounds bigger than absolute

SELECT tested once-weekly semaglutide 2.4 mg against placebo in 17,604 patients aged 45 or older with a BMI of 27 or greater and preexisting cardiovascular disease, but no history of diabetes. Over a mean follow-up of 39.8 months, a first primary event (cardiovascular death, non-fatal myocardial infarction, or non-fatal stroke) occurred in 569 of 8,803 patients on semaglutide (6.5%) and 701 of 8,801 on placebo (8.0%), giving a hazard ratio of 0.80 (95% CI, 0.72 to 0.90).

The same SELECT result, framed two ways	Figure
Relative risk reduction	20%
Event rate, semaglutide	6.5% (569/8,803)
Event rate, placebo	8.0% (701/8,801)
Absolute difference	1.5 percentage points
Treated per event avoided (1 ÷ 0.015)	roughly 67, over ~3.3 years

Both rows describe the same data. “Cuts cardiovascular events by 20%” is accurate. So is “for about 98 of every 100 high-risk patients treated for three years, the event tally was the same either way.” A headline that gives you only the first has not lied: it has chosen the flattering half and made an editorial decision on your behalf.

The relative number tells you how well a drug works. The absolute number tells you how much it changes the odds for one person. If a claim gives you one and not the other, the missing half is the one you need.

Blueberries, berries, fruits — illustrating Reading a GLP-1 Headline Without Getting Misled

What was measured, and in whom

Was the endpoint a hard outcome or a stand-in? SELECT counted deaths, heart attacks and strokes, things that happen to people. Most GLP-1 trials do not. STEP 1 measured percent change in body weight over 68 weeks in 1,961 adults, reporting −14.9% with semaglutide versus −2.4% with placebo, with 50.5% of the semaglutide group losing at least 15% of body weight versus 4.9% on placebo. Those are real, useful results, but weight is a step on the road to cardiovascular events, not the events themselves. SELECT is notable precisely because it went the whole distance. Our piece on why GLP-1 trials use different endpoints goes further into that choice.

Which participants were counted? This is the least visible and most consequential decision in the whole report. SURMOUNT-1 published two answers to the same question. Under the treatment-regimen estimand (everyone randomized, counted regardless of whether they stopped the drug), mean weight change at week 72 was −20.9% on tirzepatide 15 mg versus −3.1% on placebo. The results posted to ClinicalTrials.gov, which exclude data after participants prematurely stopped study drug, put the same arm at −22.5% versus −2.4%.

SURMOUNT-1, week 72, 15 mg arm	Tirzepatide	Placebo
Everyone randomized (treatment-regimen estimand)	−20.9%	−3.1%
Excluding data after stopping study drug	−22.5%	−2.4%

Same trial, same people, 1.6 percentage points of difference from the analysis rule alone. STEP 1’s posted results show the identical pattern: −15.6% for the in-trial observation period against −16.9% on-treatment. The on-treatment figure is always the more flattering one, because it quietly sets aside the people for whom the drug did not work out. In STEP 1, 4.5% of the semaglutide group discontinued because of gastrointestinal events, versus 0.8% on placebo.

Four questions worth asking of any GLP-1 headline

Relative or absolute? A percentage reduction with no baseline risk attached is half a number.
Hard endpoint or surrogate? Weight, HbA1c and blood pressure predict outcomes. They are not outcomes.
Who was counted? Intention-to-treat and on-treatment answers differ, and the size of the gap is itself information about tolerability.
Who was studied? SELECT enrolled people with established cardiovascular disease and no diabetes. That result does not transfer automatically to a healthy 35-year-old.

Strawberries, fruits, food — illustrating Reading a GLP-1 Headline Without Getting Misled

The takeaway

None of this makes GLP-1 coverage untrustworthy. The trials above are large, well-run, and their headline numbers are real. The problem is compression: a single sentence cannot carry a relative risk, its baseline, the endpoint definition and the analysis population at once, so something is always dropped, and what gets dropped is reliably the part that makes the effect look smaller. Reading the abstract, or the posted results record, takes a few minutes and usually changes the size of the claim rather than its direction. For the specific cardiovascular numbers, we cover them in detail in semaglutide’s cardiovascular outcomes.