Hierarchy of Evidence

A traditional pyramid ordering designs by internal validity / risk of bias for therapy questions — from systematic reviews & meta-analyses of RCTs at the top, down to expert opinion at the base.

The pyramid tiers: systematic reviews & meta-analyses → RCTs → cohort → case-control → case series & reports → expert opinion. Narrow and stronger at the top; wide and weaker at the base.

Higher generally means lower risk of bias — but the best design depends on the question (RCT for therapy, cohort for prognosis/harm, cross-sectional for diagnosis), and quality within a tier varies enormously.

It is a fast triage of evidence strength — but applied blindly it misleads. A flawed RCT can be weaker than a strong cohort, and the wrong design for the question ranks poorly regardless of tier.

• Higher tier = lower risk of bias (in general)
• Best design depends on the question
• GRADE rates certainty of evidence, not just design

Therapeutic hypothermia after cardiac arrest — watch one question climb the pyramid. Two small RCTs (HACA and Bernard, both NEJM 2002) cooled comatose survivors to ~33°C and reported better neurological outcomes vs no temperature control → cooling entered guidelines. The larger TTM trial (NEJM 2013) found 33°C no better than 36°C, and TTM2 (NEJM 2021, ~1900 patients) found targeted hypothermia no better than normothermia for death or function. The early “positive” signal sat near the top of the pyramid (RCTs) yet was overturned by bigger, better RCTs and pooled analyses — tier alone never settles a question; size, quality and the comparator do.

• Treating the hierarchy as absolute — tier rank is not a verdict.
• Ignoring within-tier quality — a small, biased RCT can sit below a rigorous cohort.
• Using the wrong design for the question (e.g. an RCT where a cohort answers prognosis/harm).