Relative vs Absolute Risk: The GLP-1 Headline Math to Know

Relative risk tells you how much a risk changed compared with another group ("20% lower"). Absolute risk tells you how many people out of 100 that change actually affected ("1.5 fewer per 100"). GLP-1 headlines usually lead with the relative number because it is bigger. To understand a result, you need both.

This is the most useful piece of math in health news, and it fits on a sticky note. Let's do it with real GLP-1 trial numbers.

What is the difference between relative and absolute risk?

Picture two jars of 100 marbles. In jar A, 8 marbles are pink (pink = something bad happened). In jar B, 6.5 are pink. (Yes, half a marble. Stay with us.)

• Absolute risk in each jar: 8% and 6.5%.
• Absolute risk reduction: 8 − 6.5 = 1.5 percentage points.
• Relative risk reduction: 1.5 ÷ 8 = about 19%, which rounds nicely into a headline.

Same jars, two very different-sounding sentences. Neither is wrong.

How does this work with the SELECT trial?

These are not made-up jars. They are roughly the SELECT trial, which tested semaglutide 2.4 mg in 17,604 adults with existing cardiovascular disease, overweight or obesity, and no diabetes.

Over a mean follow-up of 39.8 months (about 3.3 years):

	Placebo	Semaglutide
Had a heart attack, stroke or cardiovascular death	701 of 8,801 (8.0%)	569 of 8,803 (6.5%)

The paper reported a hazard ratio of 0.80, which most coverage turned into "20% lower risk." (The hazard ratio is a time-based measure, which is why it comes out a little different from the simple 19% above.)

Now the absolute view: 1.5 fewer people out of every 100 had one of those events over about three years.

Flip that around and you get the number needed to treat (NNT): 1 ÷ 0.015 ≈ 67. Roughly 67 people like those in SELECT would need to take semaglutide for about three years for one fewer of them to have a major heart event.

Is 67 "good"? That depends on what's being prevented, how serious it is, and the costs and side effects. The point is not that 20% is a lie. It's that "20%" alone can't tell you whether we're talking about 20 fewer events per 100 people or 1.5. For the full trial story, read what SELECT actually found.

Does the same math apply to side effects?

Yes, and this is the step most people skip. Relative numbers make harms sound scary in exactly the same way they make benefits sound impressive.

SELECT, side of the ledger nobody quotes: 16.6% of people on semaglutide permanently stopped the study drug because of adverse events, compared with 8.2% on placebo.

• Relative: "twice as many people quit because of side effects."
• Absolute: 8.4 more people per 100 stopped.
• Number needed to harm: 1 ÷ 0.084 ≈ 12.

STEP 1, stomach edition: 4.5% of people on semaglutide stopped treatment because of gastrointestinal events, versus 0.8% on placebo.

• Relative: "more than five times as likely to quit due to stomach problems."
• Absolute: about 3.7 more per 100.

Both statements are true. A headline that picks the relative number for harms and the absolute number for benefits (or the other way round) is choosing a mood for you.

Spot the trick

1. "GLP-1 users had double the risk of [rare condition]."

Show answer

Ask what it doubled from. Going from 1 in 10,000 to 2 in 10,000 is "double," and it is also one extra case per 10,000 people. Then ask what kind of study it was. A doubling in an observational study can come from who got prescribed the drug, not from the drug.

2. "The drug cut risk by 20%, so my risk is now 20% lower."

Show answer

Not necessarily. A trial's relative effect is an average for the people in that trial. Your own starting risk depends on your age, health history and other factors, so the absolute benefit for you could be bigger or smaller. That conversation belongs with your clinician.

3. "Only 1.5% benefit? That's nothing."

Show answer

This is the reverse trick. Small absolute differences can matter a lot when the outcome is serious (like a stroke) and the treatment is used by many people. Absolute numbers give you scale. They don't automatically mean "worthless."

Scoring: If you got all three, you already think like a trial statistician. Treat yourself to a coffee.

What questions should you ask when you see a percentage?

Keep this checklist for any GLP-1 headline with a percent sign in it:

1. Percent of what? Is it a change in risk, or a share of people?
2. Out of how many? Find the raw event counts in each group.
3. Over how long? 1.5 per 100 over three years is not the same as 1.5 per 100 per year.
4. Who was studied? SELECT's numbers apply to people with existing heart disease.
5. How certain? SELECT's 95% confidence interval for the hazard ratio was 0.72 to 0.90. Our sibling post on confidence intervals explains what that range means.
6. Benefits and harms in the same units? Put both on the "per 100 people" scale.

A simple formula card

You want	Do this	SELECT example
Absolute risk reduction	Control % − treatment %	8.0 − 6.5 = 1.5 points
Relative risk reduction	ARR ÷ control %	1.5 ÷ 8.0 ≈ 19%
Number needed to treat	100 ÷ ARR (in points)	100 ÷ 1.5 ≈ 67
Number needed to harm	100 ÷ extra harm (in points)	100 ÷ 8.4 ≈ 12 (stopped for adverse events)

This math describes groups in a trial. It is not a personal risk calculator, and it can't tell you whether a medicine is right for you. That's a decision to make with your prescriber, who knows your history.

Where does this fit in reading GLP-1 research?

Relative vs absolute risk is minute five of our five-minute headline routine: finding the number that's missing. When the result is a weight change rather than an event, the related trap is averages, covered in average weight loss vs responder rates.

Common questions

Is relative risk or absolute risk more important?

You need both. Relative risk shows how strong the effect is; absolute risk shows how many people it actually affects. The CONSORT 2025 reporting guideline for randomized trials asks authors to present both absolute and relative effect sizes for yes/no outcomes like heart attacks.

What does "number needed to treat" mean?

It is how many people would need the treatment, over the trial's time frame, for one extra person to benefit compared with the control group. Lower means more people benefit per person treated.

Why did SELECT say 20% when my math gives 19%?

SELECT reported a hazard ratio of 0.80, which accounts for when events happened over follow-up. The simple proportion calculation gives about 19%. They are close but not identical measures.

Can I use these numbers to work out my own risk?

No. Trial numbers describe groups with specific characteristics. Your clinician can estimate your personal risk using your own health information.