All learning resources · Evidence and editorial policy
Start with the claim, not the compound name
Write down the claim being made. Is it about a laboratory pathway, a change in a blood measurement, symptom relief, or better daily function? Then find the original paper rather than a shop page or social post. Record the molecule, formulation, route, population, comparison, outcome, and follow-up period. If any of these are missing, the claim is not ready to evaluate.
Ask who or what was studied
Cells, animals, healthy volunteers, and people with a defined diagnosis answer different questions. Human-derived cells are not human trial participants. A study of a naturally occurring substance is not necessarily a trial of administering it. Even a human trial may not apply to a different age group, medical condition, formulation, or proposed use.
Understand the comparison
A randomized trial assigns participants to groups by chance. Blinding can reduce the effect of expectations on reporting and assessment. Observational studies follow associations without assigning the intervention. An uncontrolled before-and-after report cannot reliably separate an intervention from natural recovery, other care, or selection effects. No design label removes the need to inspect how the study was carried out.
Separate biomarkers from outcomes people notice
A biomarker is a measured biological characteristic. It may help investigate a mechanism without demonstrating a practical benefit. In a fictional example, a small trial reports a hormone change after two weeks but does not measure strength or physical function. It cannot establish stronger muscles, faster recovery, or longer life. This example is illustrative, not a report of a particular trial.
Look at the size and uncertainty of the result
Read the results for all comparison groups. Ask how large the difference was and whether uncertainty intervals leave room for a small or unimportant effect. A statistically significant finding is not automatically a meaningful improvement. For a fictional risk example, a change from 2 in 100 to 1 in 100 is a 50% relative reduction but a 1 percentage-point absolute reduction. Both descriptions need the time period and population.
Check missing participants and selected outcomes
How many participants started, finished, and were included in the analysis? Why did people leave? Were outcomes specified beforehand or selected after results were known? Many tests increase opportunities for chance findings. An abstract may emphasize a favorable secondary outcome while the primary outcome was inconclusive. Look for the trial registration, protocol, and full report when available.
Read safety and funding with the same care
Check how adverse events were collected, how long participants were followed, and whether the sample could detect uncommon or delayed harms. No observed serious event in a small study does not establish no risk. Funding or an author conflict does not automatically invalidate results, but transparent disclosures, independent replication, and balanced reporting matter.
Use a short evidence note
Summarize the study in four sentences: what was tested and in whom; what comparison and outcome were used; what was found and how uncertain it was; what remains unknown. Keep the result attached to its exact setting. Use the provider-question guide before discussing whether any finding matters to your own care.