How to read health evidence

Judge a health claim in this order: what kind of study it is, who was studied, how large the effect was, whether it has been repeated, and who funded it. A large effect in a small unrepeated study is weaker evidence than a small effect confirmed across several trials.

A plain-language guide to judging health claims: study design, population, effect size, who paid, and what the claim actually says.

Key takeaways

Start with the study design

Design sets the ceiling on what a study can claim. Randomised controlled trials can support statements about cause, because randomisation distributes unknown differences between groups. Observational studies - cohorts, case-control studies, surveys - can show that two things travel together, and are often the only ethical option, but they cannot on their own establish that one caused the other. Laboratory and animal work sits earlier still: it explains mechanisms and generates hypotheses rather than guiding what a person should do.

Ask who was actually studied

A result applies most confidently to people like those in the study. Sample size, age, sex, health status and country all matter. Findings from young healthy volunteers frequently shrink or disappear in older adults or in people with the condition the intervention targets.

Look at the size of the effect, not only whether it is significant

Statistical significance means a result is unlikely to be chance under the study's assumptions. It says nothing about whether the difference is large enough to matter. With enough participants, trivial differences become significant. Look for the actual numbers - minutes of sleep gained, points on a scale, absolute rather than relative risk.

Check whether it has been repeated

Single findings are provisional by design. Independent replication, or a systematic review pooling several studies, carries far more weight than a striking one-off. When a claim rests on a single trial, the honest description is 'emerging', not 'proven'.

Read the funding and the conflicts

Industry funding does not make a study wrong, but reviews consistently find that funded studies report favourable results more often. Declared conflicts, undisclosed relationships and who chose the outcome measures are all part of reading a claim fairly.

Then read what the claim actually says

Headlines routinely overstate. A study showing a supplement changed a blood marker does not show it improves health. A study showing an intervention improved a questionnaire score after four weeks does not show it works over a year. Match the claim to the outcome that was actually measured.

Frequently asked questions

What is the strongest kind of evidence?

For questions about whether something works, a systematic review of well-conducted randomised trials sits at the top, followed by individual large trials. For questions about rare harms or long-term patterns, large observational studies are often the only practical source.

Is a study in mice useless?

No, but it answers a different question. Animal and laboratory work explains mechanisms and identifies candidates worth testing in people. Most findings that look promising in animals do not replicate in humans, so preclinical results should never be described as though they show what happens in a person.

How do I check a claim quickly?

Find the original study rather than the article about it, read the abstract's methods and numbers, note the design and sample, and see whether a systematic review already exists on the same question.

Related on Beiing

Sources

  1. MedlinePlus - Evaluating Health Information - U.S. National Library of Medicine
  2. NIH - Clinical Research Trials and You - National Institutes of Health
  3. Cochrane Handbook for Systematic Reviews of Interventions - assessing certainty of evidence - Cochrane
  4. Guyatt et al. (2008), GRADE: an emerging consensus on rating quality of evidence and strength of recommendations - BMJ / PubMed
  5. PubMed - National Library of Medicine - NIH

Written by Beiing Editorial Team · Editorially reviewed by Research Desk · Published · Last reviewed . Editorial and research review by Beiing writers and researchers. This is not medical or clinical review, and no clinician signs off on our content.

Educational content only - not medical advice. In a crisis, call or text 988 (US) or your local emergency number. See crisis resources.