Confidence Intervals in GLP-1 Studies, Explained Without the Math

A confidence interval (CI) is the range in parentheses after a study result, such as "-12.4 percentage points (95% CI, -13.4 to -11.5)". It shows how precisely the study pinned down the average effect: a narrow range means a precise estimate, a wide range means more uncertainty, and for ratios, a range that includes 1.0 means the data are compatible with no difference at all.

That is the whole idea. You do not need to calculate anything; you just need to read the parentheses. Let's practice on real GLP-1 trials.

What does a 95% confidence interval actually mean?

A trial measures a sample of people, not everyone on Earth. If you could rerun the same trial many times, you would get slightly different answers each time. A 95% CI is built so that, across many such repeats, 95% of the intervals calculated this way would contain the true average effect.

In plain English: it is the range of average effects that fit the data reasonably well. It is not:

• a guarantee the true value is inside this particular range
• the range of results individual people got
• a measure of whether the study was well designed

Keep that last point in mind; we will come back to it.

Step 1: Find the point estimate and the range

Take STEP 1, the semaglutide 2.4 mg obesity trial. Average weight change at 68 weeks was -14.9% with semaglutide versus -2.4% with placebo. The difference was -12.4 percentage points (95% CI, -13.4 to -11.5).

• Point estimate: -12.4. The single best guess.
• Range: -13.4 to -11.5. Every value in there is a plausible average difference. All of them are big.

Step 2: Check the width

Width tells you precision, and precision mostly comes from how many people (or events) a study had.

Trial	People	Difference vs placebo (95% CI)	Width
STEP 1 (68 weeks)	1,961	-12.4 points (-13.4 to -11.5)	about 2 points
STEP 5 (104 weeks)	304	-12.6 points (-15.3 to -9.8)	about 5.5 points

Same drug, almost the same point estimate, but STEP 5's range is nearly three times wider because it enrolled about a sixth as many people. Both say semaglutide beats placebo substantially. STEP 1 just says it with more precision.

Step 3: Check whether the range crosses "no difference"

For a difference in percentage points, "no difference" is 0. For a ratio (hazard ratio, odds ratio, relative risk), "no difference" is 1.0. A hazard ratio below 1 means fewer events in the drug group.

Now look at these results from the SELECT heart outcomes trial, as reported in the Wegovy prescribing information:

Outcome	Hazard ratio (95% CI)	Range includes 1.0?
Major cardiovascular events (primary endpoint)	0.80 (0.72 to 0.90)	No
Fatal or non-fatal heart attack	0.72 (0.61 to 0.85)	No
Cardiovascular death	0.85 (0.71 to 1.01)	Yes, just
Fatal or non-fatal stroke	0.89 (0.72 to 1.11)	Yes
Heart failure hospitalization or urgent visit	0.79 (0.60 to 1.03)	Yes

The primary endpoint's whole range sits below 1: compatible with anything from a 28% to a 10% lower risk, and not compatible with no effect. Cardiovascular death's range runs from 0.71 (meaningfully lower) to 1.01 (a hair higher). That does not prove semaglutide has no effect on cardiovascular death. It means this trial could not distinguish a benefit from no difference with confidence. The label says exactly that: superiority for cardiovascular death "was not confirmed," and an effect on heart failure "has not been established."

"Not proven" and "proven not to work" are completely different statements. The CI is how you tell them apart.

What about very dramatic results?

Here is one from SURMOUNT-1's three-year analysis in people with prediabetes. Type 2 diabetes developed in 1.3% of the tirzepatide groups versus 13.3% on placebo: hazard ratio 0.07 (95% CI, 0.0 to 0.1). A range that far from 1 says the difference is very unlikely to be chance.

But the CI cannot tell you how much that matters to a person. For that you need the absolute numbers (1.3% versus 13.3%), the population (adults with obesity and prediabetes) and the time frame (176 weeks). Pair this post with relative vs absolute risk and you have the whole toolkit.

Three common CI mistakes

1. Treating the interval as your personal range. The STEP 1 CI of -13.4 to -11.5 describes the average difference between groups. Individuals were far more spread out: about half of the semaglutide group lost 15% or more, while some lost very little. (See average weight loss vs responder rates.)

2. Comparing two drugs by eyeballing their separate intervals. SURMOUNT-1's 15 mg dose had a 72-week change of -20.9% (CI -21.8 to -19.9); STEP 1's semaglutide group lost -14.9%. Those numbers came from different trials with different people. To compare drugs fairly you need a head-to-head trial with its own CI for the difference, as in SURMOUNT-5. Even within one trial, whether two intervals overlap is not a proper test of whether they differ. (Why studies of the same drug differ.)

3. Thinking a narrow CI means a trustworthy study. A CI only measures random error. If the study was biased, measured the wrong thing or studied people unlike you, a tight CI just means it is precisely answering a different question. Always check the endpoint too (GLP-1 trial endpoints explained).

A 20-second CI check you can use anywhere

When you see "X (95% CI, A to B)":

1. Is the whole range on the helpful side of 0 (or 1)?
2. Is it narrow or wide relative to the effect?
3. Would the low end of the range still matter to you?
4. Is this the primary endpoint, and in people like you?

That last question is the one statistics can't answer for you. Your prescriber can help with that part.

Common questions

What does "95% CI" stand for?

95% confidence interval: a range calculated so that, over many repeated studies, 95% of such ranges would capture the true average effect.

What does it mean if a hazard ratio's confidence interval includes 1?

The data are compatible with no difference between groups. It does not prove there is no effect; it means the study could not confidently show one.

Why do small GLP-1 trials have wider confidence intervals?

Fewer people (or fewer events) means more room for chance, so the range of plausible average effects is wider. STEP 5, with 304 people, had a range nearly three times wider than STEP 1 with 1,961.

Is a p-value the same as a confidence interval?

They are related but not the same. A p-value tells you how surprising the data would be if there were no effect; a CI shows the range of plausible effect sizes, which is usually more useful.