The evidence pyramid

The core idea

The evidence pyramid is an educational ranking of study types according to their ability to answer questions about effectiveness and safety in humans: at the top are reviews that bring together many studies, and at the bottom are early signals.

Well-planned clinical trials give the clearest information on whether a treatment or lifestyle change is effective and safe in humans, but they are complex, long, involve many participants and can be very expensive, which is why they usually come after smaller preliminary studies.

When many studies reach the same result, confidence increases that the result is reliable; this is why systematic reviews and meta-analyses sit at the top.

Why it matters to you

Much health news is reliable, but some lacks important information or is confusing, conflicting or misleading; knowing what type of study a news item is based on helps you judge its weight before you change anything about your health.

A single study rarely proves anything, and one study is not the last word; repeating the result in other studies is what makes it more reliable.

The layers

The top — systematic reviews and meta-analyses: a systematic review identifies all the evidence that meets pre-specified conditions and assesses and summarises it using explicit methods that reduce bias; if the results of the studies are combined into a single statistical figure, this is called a meta-analysis, and not every review includes a meta-analysis.

Randomised controlled trials (RCTs): they have at least one experimental treatment and one control treatment, participants are enrolled and followed at the same time, and each one's treatment is determined randomly; in GRADE their evidence starts at "high certainty".

Observational studies (cohort, case-control and cross-sectional): the researcher observes exposure and disease without determining the exposure; they reveal associations, and their evidence starts in GRADE at "low certainty".

Early signals: case reports (clinical presentations that may be followed by evaluation studies), and laboratory and animal research whose results are not considered directly relevant to human health until clinical research has tested them.

Outside the pyramid: editorials, commentaries and letters to the editor, which are opinions or discussions of published research and not stand-alone studies.

How to use the pyramid

Start with a question: what type of study is this? You can find the type of almost any published study from its DOI via the Study Reader, which displays publication types according to the US National Library of Medicine's classification.

Then ask: was it tested in humans? Did it include enough people? Was the study long enough to show benefits or risks in the long term?

And remember that a study's position in the pyramid is only a starting point: in GRADE, the certainty of randomised trial evidence may be downgraded because of risk of bias, inconsistency, indirectness, imprecision or publication bias, and the certainty of non-randomised studies may be raised in specific cases.

Common pitfalls

Treating any "review" as a systematic review: an ordinary review is based on examining what has been published on a subject, and its comprehensiveness may vary, unlike a systematic review, which follows explicit conditions and methods to reduce bias.

Judging a treatment from a laboratory or animal study: the results of such research need later testing in humans before they mean anything for your health.

Dismissing observational studies altogether: they reveal patterns and signals and raise new hypotheses, but on their own they do not prove causation.

Limits of this page

The pyramid is an educational simplification; established evidence-grading systems such as GRADE assess certainty for each health outcome across the body of evidence, not by study type alone.

The pyramid does not say that a result applies to you personally; the study participants may differ from you in age, sex or other important matters.

This page explains how to read evidence, and does not evaluate any particular treatment.

Common questions