Every article on this site ends with a list of sources, and the point of that list is that you can check us. This article is about how.
You do not need to understand the statistics. You need to know where to look and which four or five questions expose most of the problems. That is a skill you can acquire in an afternoon, and it is more useful than any individual health fact you will ever read.
Where to find the paper
Start with PubMed (pubmed.ncbi.nlm.nih.gov), the free index of biomedical literature. Search the authors and a few words from the title.
Every entry has an abstract — a structured summary, always free. Many have a link to PMC (PubMed Central), which is the full text, also free. If the paper is paywalled, search the title plus “PDF”; authors frequently post copies, and many are legitimately open access.
The DOI is the permanent identifier. Anything we cite with a DOI can be found even if the link rots.
Read it in this order
Not front to back. Papers are written for other researchers, and the useful parts for you are not where you would expect.
1. The abstract, to see what they claim.
2. The methods, which is where the answer to “should I believe this” actually lives. Most people skip it. It is the most important section for a non-specialist, because study design determines what a result can possibly mean.
3. The results, specifically the numbers, not the words describing them.
4. The limitations, near the end of the discussion. Authors are required to state the weaknesses of their own work, and they usually do so honestly, in language nobody quotes.
5. The funding and conflict of interest statement, at the very end.
The discussion section, in between, is where authors interpret their own findings and is the most optimistic part of any paper. Read it last and with the results already in mind.
The five questions
Was it done in humans?
Look for the species in the methods. A great deal of longevity research is done in mice, worms, flies or cells. That work is valuable and it is where most exciting findings begin — and stop. Most interventions that extend life in animals have never shown anything comparable in people.
What kind of study was it?
This determines what the result can mean, and there are three main types.
Randomised controlled trial — participants assigned by chance to intervention or control. This is the only design that can establish causation, because randomisation balances everything else, known and unknown.
Cohort or prospective observational study — people are followed and observed. It can show that two things travel together. It cannot show that one causes the other, because the groups differ in ways beyond the exposure.
Cross-sectional study — a snapshot at one moment. Weakest of the three, since you cannot even establish which came first.
The single most common error in health reporting is treating a cohort finding as though it were a trial finding.
How many people, and for how long?
Thirty participants over eight weeks is a pilot. Fifty thousand over fifteen years is evidence about ageing. Both get reported with the same confidence.
Also check who: a study in 55-year-old men with high cardiovascular risk tells you little about a healthy 40-year-old woman.
What is the actual number?
Find the effect size and the confidence interval. You do not need to calculate anything.
A hazard ratio or odds ratio of 1.0 means no effect. Above 1.0 is increased risk, below is decreased. So 1.22 is a 22% higher rate.
The confidence interval in brackets is the range within which the true value probably sits. If it crosses 1.0 — say, 0.98 to 1.57 — the result is compatible with no effect at all, and no firm conclusion follows. This one check catches an enormous amount of overstatement.
And relative versus absolute: a 50% risk reduction sounds dramatic and may mean going from two in a thousand to one in a thousand. Ask what the underlying rate was.
Who paid?
Funding does not make research wrong. It shifts the odds — industry-funded studies produce results favourable to the funder more often than independent ones. Worth knowing, not worth treating as disqualifying.
Three habits that do most of the work
Read the limitations section. It is the highest-value paragraph in any paper for a non-specialist. The authors will tell you exactly what is wrong with their study, plainly, because their peers would catch it otherwise.
Check whether it is one study or many. A single striking result is a hypothesis. Confidence should rise when independent groups using different methods arrive at the same place, and fall when they do not. This is why systematic reviews and meta-analyses sit near the top of the evidence hierarchy.
Find the original. News coverage compresses, press releases from universities overstate, and each retelling drifts. Going from a headline to the abstract takes two minutes and frequently reveals that the paper says something narrower than the article about it.
The vocabulary that matters
Peer review — other researchers checked it before publication. Imperfect, better than nothing.
Preprint — posted before peer review, on servers like bioRxiv or medRxiv. Not worthless; unchecked. Say so when citing one.
Systematic review — all studies on a question, found by defined criteria. Meta-analysis — their results statistically combined.
p < 0.05 — the conventional threshold for calling a result statistically significant. It says the result is unlikely to be chance. It says nothing about whether the effect is large enough to matter.
Statistically significant ≠ important. With a big enough sample, trivial differences become significant. Always look at the size of the effect, not just whether it cleared the threshold.
Why this is the most useful thing on this site
Health information will keep arriving for the rest of your life, from sources with varying motives and competence. Individual facts go stale. Guidance changes. The compounds being sold today will be replaced by different ones.
The ability to look at a claim and ask in what species, in how many people, for how long, and how big was the effect does not go stale. It works on our articles as well as anyone else’s, which is the point.
We tell you when a study was done in mice, when a confidence interval crosses one, and when the primary outcome was null. If we ever stop, you now have what you need to notice.
Further reading
- PubMed — pubmed.ncbi.nlm.nih.gov — the free index of biomedical literature
- Cochrane Library — systematic reviews, considered among the most rigorous evidence syntheses available
- Bad Science, Ben Goldacre — how health claims go wrong, written for non-specialists
- Testing Treatments, Evans, Thornton, Chalmers and Glasziou — free to download, on why fair trials matter
- NHS Behind the Headlines — appraisals of health stories in the news against the studies behind them
This article is for general information and is not medical advice. If a health problem is affecting your daily life, speak to your GP.
Comments
We read every comment. Be kind, stay on topic, and please don't ask for medical advice — we can't give it.