How to Read a Clinical Trial Like a Skeptic

A headline says a pill "works." Before you change anything you do, here is how to open the actual trial and check whether it earned that word.

A friend forwards you a headline: a new supplement "cuts the risk" of something you're scared of. Your stomach does the thing. You're already half-deciding to buy it. That feeling — hope plus fear, arriving before you've seen a single number — is exactly the moment to slow down. Not because the claim is false, but because a headline is a sales pitch and a trial is a record. The good news: the record is usually public, and you can read it. Here is how to read it like a skeptic, not a sucker.

Find the actual study, not the press release

Most news stories are summarizing a paper, and most serious trials are registered before they start. The free home base is ClinicalTrials.gov, run by the U.S. National Library of Medicine. Search the drug or condition and you'll get a registry entry with an NCT number (the trial's ID), its design, what it measured, who ran it, and often its results. Registration matters for a plain reason: it's a timestamped promise of what the researchers said they'd test before they saw the data. That lets you catch a team that quietly changed the goalposts later.

For the published paper itself, PubMed (pubmed.gov) is the free index. You may only get the abstract, but the abstract plus the registry entry already answers most skeptic questions.

Sample size: how many people, really?

The first number to find is n — how many participants. A result from 18 people and a result from 18,000 are not the same kind of fact, even if both say "significant." Small studies swing wildly on chance and rarely catch side effects that show up in 1 in 1,000 users.

Also ask who the people were. A trial run only on healthy young men may not tell you much about an older woman with three other conditions. And check how many dropped out. If a third of participants quit and you only hear about the ones who finished, that's a flattering edited highlight reel, not the whole game.

Controls: compared to what?

A result means nothing without a comparison. The strongest design is randomized and controlled: people are randomly sorted into a group that gets the treatment and a group that gets a placebo or the current standard care. Randomizing is what stops researchers from (even unconsciously) putting healthier people in the treatment group.

Two more words earn trust. Blinded means participants don't know which group they're in; double-blinded means the researchers handing out the pills don't either, so they can't nudge the results. When you see "single-arm," "open-label," or "observational," lower your confidence — those can be useful early signals, but they can't prove a treatment caused an outcome. The famous trap: a study showing coffee drinkers are healthier doesn't prove coffee helps; maybe healthier people just drink more coffee. That's correlation wearing a lab coat.

Endpoints: did they measure what you care about?

An endpoint is what the trial counted as success. The thing to watch for is a surrogate endpoint — a stand-in that's easy to measure but isn't the thing you actually want. A drug might lower a number on a blood test (surrogate) without anyone living longer or feeling better (the clinical endpoint you actually care about). Lowering a marker is a promising clue, not a promise of real-world benefit.

Find the primary endpoint — the one main thing the trial was built to test. Then watch for outcome switching: a trial that misses its primary endpoint but trumpets some secondary finding it stumbled onto. With enough secondary measures, something will look good by pure luck. That's why the pre-registered plan matters. Compare what they promised to measure against what they're now bragging about.

Who funded it, and who ran the math

Funding doesn't automatically make a study wrong, but it's a thumb on the scale you should know about. Look for the funding source and conflict of interest statements (usually near the end of the paper, or in the registry's sponsor field). A trial of a product, paid for by the company that sells it, with authors on that company's payroll, isn't disqualified — but it deserves a harder look and a wait for independent replication. The strongest evidence is when different teams, with different funders, keep finding the same thing.

A 60-second skeptic's checklist

• How many people, and who were they? Small or narrow means tentative.
• Compared to what? Randomized + placebo + double-blind beats a single group every time.
• Did they measure a real outcome or a stand-in? Living better beats a moved number.
• Did they hit the goal they set in advance? Check the registry; beware switched endpoints.
• Who paid, and who else found the same result? One funded study is a lead; independent replication is a fact.
• Statistical vs. real-world size. "Significant" only means the effect is probably not zero — it can still be tiny. Look for the actual size of the benefit, and the confidence interval if it's given.

The honest limits

Reading a trial well doesn't make you a doctor, and a single study — however clean — is one brick, not the house. Medicine moves on the weight of many studies, summarized in systematic reviews. The point of all this isn't to "debunk" everything or to trust nothing. It's to turn that first jolt of hope-and-fear into a few calm questions you can actually answer. The headline wants you to feel. The record lets you check. Open the record first.