Real-World Data vs Randomized Trials: Why GLP-1 Research Needs Both

Randomized trials show what a GLP-1 medicine can do when people are assigned at random and followed closely, which makes them the best tool for proving cause and effect. Real-world data (health records, insurance claims, registries) shows what happens in everyday care, with all its copays, shortages and missed refills. Researchers use both because each answers questions the other can't.

Think of it as a test kitchen versus your kitchen. The test kitchen tells you whether the recipe works. Your kitchen tells you whether it works on a Tuesday when you're tired and the store is out of an ingredient.

What does a randomized trial do best?

Randomization is the superpower. When a computer, not a doctor or patient, decides who gets the drug, the groups start out similar in ways you can measure and ways you can't. So if they end up different, the drug is the likeliest reason.

That's why the big efficacy claims for GLP-1 medicines rest on trials such as STEP 1, SURMOUNT-1 and SELECT. But trials have trade-offs:

• They enroll selected people (for example, no diabetes, or existing heart disease only).
• Participants usually get the medicine at no cost, with regular visits and counseling.
• They run for a set time, often 68 to 72 weeks for weight trials.

What does real-world data do best?

Real-world data shines when the question is about everyday life. Two examples from researchers using electronic health records at a large health system in Ohio and Florida:

Why do people stop? In a 2025 study of 288 randomly selected patients who stopped semaglutide or tirzepatide for obesity within a year, the most common reason in the records was cost or insurance (47.6%). Side effects accounted for 14.6%, shortages 11.8%, and unsatisfactory weight loss just 1.7%. No placebo trial could have told us that, because trial participants don't pay for the drug. We dig into this in why people stop GLP-1 medicines.

What happens after stopping? A 2026 study of 7,938 patients who stopped within 3 to 12 months found that, over the following year, 19.6% restarted the same medicine and 35.2% got another obesity treatment. Average weight change in the year after stopping was small, though individual results varied widely. Notice the group: these were people who stopped early, so their average loss while on treatment (−8.4% for those treated for obesity) can't be compared directly with a 72-week trial.

Can real-world studies compare two GLP-1 drugs?

They can try, and here's a great example of how it works and where to be careful.

The real-world study (2024): Researchers used linked health records from US health systems to compare people starting tirzepatide or semaglutide. After matching 18,386 people on their characteristics, they found tirzepatide users were more likely to reach 5%, 10% and 15% weight loss, with a 12-month on-treatment difference of about 6.9 percentage points.

The randomized trial (2025): SURMOUNT-5 then randomly assigned 751 adults with obesity to tirzepatide or semaglutide at their highest tolerated doses. At 72 weeks: −20.2% vs −13.7%.

Same direction! The real-world study pointed the right way. But look at the fine print before treating the two as equal:

Spot the trick

1. "The real-world study proves tirzepatide beats semaglutide."

Show answer

The 2024 study looked at semaglutide and tirzepatide in formulations labeled for type 2 diabetes, used on or off label, and about half the participants had type 2 diabetes. Those aren't the same products or doses tested in SURMOUNT-5. Matching also can't adjust for things the records don't capture, like why a doctor chose one drug. It was the randomized trial that made the comparison solid.

2. "Real-world weight loss is lower, so the trials were rigged."

Show answer

Different setting, different numbers. In the 2024 real-world study, follow-up ended because of discontinuation for more than half the people in each group. In everyday care, people stop for cost, shortages and side effects. Trials measure what the drug can do; real-world data measures what usually happens.

3. "People on drug A had fewer heart attacks than people on drug B, so drug A protects the heart."

Show answer

Maybe, maybe not. People who get a newer, pricier drug may have better insurance, more doctor visits or fewer health problems. That's called confounding, and it's the main reason observational findings need confirmation. Our post on GLP-1s and dementia shows this problem in action.

Scoring: 3 out of 3 means you can read a real-world study without either swooning or scoffing. That's the sweet spot.

How do the two work together on safety?

The clearest example is the question of suicidal thoughts. In July 2023, FDA began looking into postmarketing reports of suicidal ideation in people taking GLP-1 medicines. To check it, FDA used both kinds of evidence:

• Randomized: a meta-analysis of 91 placebo-controlled trials with 107,910 patients, which did not show increased risk.
• Real-world: a study in FDA's Sentinel system, using insurance claims for more than 2.2 million new users, that compared GLP-1 users with people starting another diabetes drug class (SGLT2 inhibitors). It did not find increased risk of intentional self-harm.

On January 13, 2026, FDA asked companies to remove the suicidal-ideation warning from Wegovy, Saxenda and Zepbound labels. Trials gave the clean comparison; real-world data gave the huge numbers needed to look for rare events. Neither alone would have been as convincing.

Quick comparison table

Question	Randomized trial	Real-world data
Does the drug cause this effect?	Strongest	Suggestive only
How big is the effect under ideal use?	Strongest	Usually smaller, harder to isolate
What happens with cost, shortages, real schedules?	Weak	Strongest
Rare side effects in millions of people?	Often too small	Strongest for spotting signals
People excluded from trials (older, sicker, other conditions)?	Often missing	Often included
Speed and cost	Slow, expensive	Faster, cheaper

How should you read a real-world GLP-1 headline?

Ask four things:

1. Where's the data from? Health records, insurance claims or a registry?
2. Who was compared with whom, and why did they get different treatments?
3. Did they adjust for differences, and what couldn't they measure?
4. Has a randomized trial checked this yet?

Then use our five-minute headline routine for the rest. And remember the placebo-trial side of the story in what placebo-controlled trials tell us.

Your own experience is real-world data too, just with a sample size of one. Keeping a steady log of doses, weight and side effects (in a notebook or an app like GLPily) makes your appointments more productive. It's for you and your prescriber, not a study.

Common questions

Is real-world evidence less reliable than clinical trials?

Not less reliable in general, just suited to different questions. It's weaker for proving a drug causes an effect, and stronger for showing how treatments are used and how people do in routine care.

Why is GLP-1 weight loss often lower in real life than in trials?

In everyday care, many people stop or interrupt treatment because of cost, insurance, shortages or side effects, and may not reach the doses used in trials. Trials also give regular counseling and follow-up.

What is FDA's Sentinel system?

It's FDA's network for safety monitoring that uses electronic health data such as insurance claims from multiple data partners. FDA used it to study GLP-1 medicines and self-harm in more than 2.2 million new users.

Can a real-world study ever change medical practice?

Yes, especially for safety questions, rare events and populations excluded from trials. Often it raises a question that a trial then confirms or refutes.