Home › Know Your Body

How to Read a Study Without a Science Degree

By the Strength from the Well Editorial Team · Evidence last reviewed September 2026 · Not reviewed by a licensed clinician (why we say that)

The short answer

Read a study in a fixed order rather than front to back: what kind of study it is, who was in it, what it compared the treatment to, and what outcome was actually measured. Most of what you need is in the abstract and the table describing the participants. You can judge whether a result applies to you without ever understanding the statistics used to produce it.

Key points

The study design tells you more than the result does. A strong-sounding finding from a weak design is still a weak finding.
Always ask what the treatment was compared to. No comparison group means the result cannot separate the treatment from time, attention, or chance.
Percentages are usually relative and can make small differences sound enormous. Look for the actual numbers of people.
A surrogate marker, like a lab value, is not the same as an outcome you would feel, like fewer hospital visits.
Small and short studies are common and are not fraudulent, but they can only suggest, never settle.
Check who funded the work and what the authors declared. It is context, not proof of anything by itself.

Read in this order, not front to back

Research papers are written for other researchers, which is why reading one straight through feels like drowning. The fix is to stop reading it as prose and start reading it as a form you are filling out.

Go in this order. First the title and abstract, to find out what was tested and what they claim. Then the methods section, specifically the paragraph describing who took part. Then the first table, which almost always describes the participants. Then the results for the main outcome only. Then the limitations paragraph near the end of the discussion, which is where authors say the quiet part in plain language. The introduction and the rest of the discussion are the least useful parts of the paper for a patient, because that is where authors argue for their own work.

Fifteen minutes spent this way will tell you more than two hours spent reading it like a book.

What kind of study is this?

This is the first question and it constrains everything else. The abstract usually names the design in the first two sentences. If it does not, that is worth noticing.

A randomized controlled trial assigns people by chance to the treatment or a comparison. Randomizing is what lets researchers claim the treatment caused the difference, because chance assignment tends to balance out everything else, including the things nobody thought to measure.

An observational or cohort study watches what happens to people who already made different choices. It can find associations and it cannot establish cause, because the people who chose the treatment usually differ from those who did not in dozens of ways. This is the source of most reversals you have seen in health news, where something appeared protective for years and then did not hold up in a trial.

A case report or case series describes what happened to one person or a few. Valuable as a signal that something is possible or that a side effect exists. Not evidence that a treatment works.

Preclinical work happens in cells or animals. Look for the words 'in vitro', 'murine', 'mice', or 'rats'. Most compounds that work in mice never work in people, so treat these as the beginning of a question.

Who was actually in it?

This is where a study either applies to you or quietly does not. Go to the participants paragraph and the first table and read the specifics: age range, sex, weight, existing diagnoses, and what medications were allowed.

Then read the exclusion criteria, which almost nobody does. Trials routinely exclude people with kidney or liver problems, people on common medications, people over a certain age, and people who are pregnant. If you would have been excluded from the trial, the result is a hint about you at best.

Also check where it was run and in whom. A result in one population does not automatically transfer to another. This is not a political point, it is a biological one: baseline risk, typical diet, genetics, and access to care all differ.

Finally, note how many people finished. Look for the number who dropped out. If a quarter of participants left before the end, ask yourself why they left and whether the ones who stayed were the ones it was working for.

Compared to what?

The comparison group is the whole machinery of the study. Without it there is nothing to attribute the result to, because people often improve on their own, symptoms fluctuate, and attention from researchers is itself a kind of treatment.

Ask what the comparison actually was. A placebo comparison tells you what the treatment adds beyond expectation. A comparison to an existing standard treatment tells you something far more useful: whether this is better than what you could already have. A comparison to nothing at all, or to a waiting list, tells you the least.

Also check whether participants and researchers knew who got what. 'Double-blind' means neither the participants nor the people assessing them knew. Blinding matters most when the outcome is subjective, like pain or mood, because knowing you got the real thing reliably changes how you report feeling. It matters less for outcomes that are hard to influence by expectation, like death or a fracture.

What did they actually measure?

There is a large and frequently exploited gap between an outcome you would care about and a number that stands in for it. The stand-in is called a surrogate marker.

A drug that lowers a cholesterol number has changed a surrogate marker. A drug that reduces heart attacks has changed an outcome. These are not the same claim, and medicine has a long history of treatments that moved the number reliably while doing nothing for patients, and a few that moved the number while causing harm.

So find the primary outcome, named as such in the methods, and ask whether it is something you would notice in your life. Then look at how many other outcomes were measured. A study that measured one thing and found it is much stronger than a study that measured thirty things and reported the four that came out well. If a paper's headline finding is not the primary outcome it declared in advance, be skeptical of it.

The numbers that mislead, and the ones that do not

You can dismantle most misleading health statistics with one habit: ask for the raw counts.

A treatment that takes risk from 2 in 1,000 down to 1 in 1,000 has cut risk by 50 percent in relative terms. That is technically true and it is how the press release will put it. In absolute terms it helped 1 person out of every 1,000 treated. Both numbers describe the same study. Only the second tells you what to expect.

'Statistically significant' is another phrase that carries less than people think. It means the result is unlikely to be pure chance under certain assumptions. It says nothing about whether the difference is large enough to matter. With enough participants, a trivial difference becomes statistically significant. Look for the size of the effect, not just whether it cleared the bar.

Confidence intervals are more useful than they look and you do not need the math. They give a range of values the true effect is plausibly within. A narrow range means the study pinned the answer down. A wide range, especially one that includes 'no effect at all', means the study is compatible with the treatment doing very little.

Relative risk reduction: sounds big, needs context. Always find the absolute numbers.
Number needed to treat: how many people must be treated for one to benefit. A very honest number when a paper reports it.
Statistical significance: about chance, not about importance.
Wide confidence interval: the study did not really settle the question.
Post-hoc and subgroup findings: generated after the data came in, and much weaker than they sound.

Funding, and what it does and does not prove

Look at the funding statement and the conflict of interest declarations, usually near the end. Industry funding is extremely common and does not make a study false. Much of the useful research in the world is funded by the companies who make the products, because that is who has the money to run trials.

What it does is give you a reason to check the details more carefully: whether the comparison was a fair one, whether the trial ran long enough to catch problems, whether the outcome chosen favors the product. The honest position is that funding is context. It should raise your attention, not settle your conclusion.

The same applies in the other direction. A study funded by a foundation with a strong position on the topic deserves the same look.

One study is one study

The single most common error in reading research is treating one paper as the answer. Individual studies disagree constantly. That is normal science, not a scandal, and it is why systematic reviews exist.

A systematic review gathers all the decent studies on a question with a defined search method, then weighs them together. A meta-analysis goes further and combines their numbers statistically. When one exists for your question, it beats anything you would piece together on your own. Cochrane reviews are the best known and include plain-language summaries written for patients.

So before you decide a study has told you something, spend five minutes looking for whether anyone has already done the work of pooling it. It is the fastest quality upgrade available to a non-specialist.

A short checklist you can reuse

Keep this somewhere you can find it. Run it on any study you are tempted to act on. If you cannot answer four of these from the abstract and the first table, you do not yet know what the study found.

What design was it, and was it in humans?
How many people, and how many finished?
Would I have been eligible?
What was it compared to, and was anyone blinded?
What was the primary outcome, and is it something I would feel?
How big was the effect in absolute terms?
How long did it run, relative to how long I would use this?
Does a systematic review already exist on this question?

Common questions

Do I need to understand the statistics?

No. Design, population, comparison, and outcome are where most of a study's quality lives, and all four are described in plain language. If the conclusion depends entirely on a statistical method you cannot evaluate, that is itself a reason for caution.

What does peer review actually guarantee?

That a few people in the field read the manuscript and did not find a fatal problem. It is a filter, not a certification. Peer-reviewed papers are retracted regularly, and some journals review very lightly. It raises the floor rather than proving the result.

How can I tell if a journal is reputable?

Check whether the journal is indexed in PubMed, which is a reasonable first filter. Be wary of journals that solicit submissions by email, promise publication within days, or whose names closely imitate well-known titles. Those are the marks of predatory publishing.

What if two good studies disagree?

Look at how they differ. Different populations, doses, durations, and outcome measures explain most disagreements. When they cannot be reconciled, the honest reading is that the question is unsettled, and saying so is more accurate than picking the study you prefer.

Is a big study always better than a small one?

Usually, but not always. A large observational study can still be confounded in ways a small randomized trial is not. Size improves precision, it does not fix a flawed design.