Hierarchy of Designs by Question Type

There is no single evidence pyramid that fits every question. The optimal design changes with the question type: therapy → RCT (or a systematic review of RCTs); diagnosis → cross-sectional study comparing the test against a reference standard; prognosis → cohort study; harm / aetiology → cohort or case-control (an RCT only where exposure can be ethically allocated); cost → economic evaluation.

The familiar pyramid (systematic reviews → RCTs → cohort → case-control → case series → expert opinion) is the hierarchy for therapy questions. For a prognosis question the cohort study becomes the appropriate top tier; for diagnostic accuracy it is a cross-sectional study against a reference standard; for harm a cohort or case-control study often leads, because randomising patients to a suspected harm is usually unethical.

First classify the question (therapy, diagnosis, prognosis, harm/aetiology, cost), then pick the hierarchy that matches it. Only therapy questions sit naturally with RCT-on-top. Reaching for an RCT when the question is prognostic or about harm is choosing the wrong tool, not raising the standard of evidence.

Examiners test whether you can match design to question rather than reflexively crowning the RCT. A well-conducted cohort can be the strongest possible evidence for a prognosis or harm question — here an RCT may be impossible, unethical, or simply less relevant.

• Therapy → RCT / SR of RCTs
• Diagnosis → cross-sectional vs reference standard
• Prognosis → cohort · Harm → cohort / case-control
• Cost → economic evaluation

Choosing a design for an ED question — suppose you want to know why patients leave the emergency department before being seen. No RCT can answer that: you cannot randomise people to feel ignored. The fit-for-purpose design is qualitative — semi-structured interviews or focus groups exploring patient experience — possibly feeding a later cohort study to quantify which factors predict leaving. For a “why / what is the experience” question a qualitative or observational design sits at the top; an RCT here would be both unfeasible and the wrong tool entirely.

• Choosing the wrong design for the question type (e.g. an RCT for a prognosis question).
• Assuming the RCT is always the answer — it is only the therapy top tier.
• Ignoring feasibility and ethics — some exposures cannot be randomised.