Clicked Gallery

What are Levels of Evidence? (The Evidence Pyramid)

Highlighted from a real textbook passage. Explained by Clicked.

Used in a sentence

University Course Reader · STEM

The guideline rated the claim at the second-highest level of evidence, since it rested on one randomised trial rather than a systematic review.

The reader highlighted one term β€” in an article or a paper PDF. Clicked made the statistics term β€œlevel of evidence” easy to understand:

Explained in three depths

Same facts, different vibe β€” Slang mode 😎

The Clicked way

●○○

Overview

Levels of evidence are a ranking of the ways a claim can be checked, by how well each one protects the answer from being wrong. At the bottom sits reasoning from how something ought to work, then one patient's story, then a set of such stories. Above those is a comparison of groups that formed on their own (an observational study). Above that is a trial where a random draw decides who gets the treatment (a randomised controlled trial). At the top is a review that gathers every such trial and pools the results (a systematic review). Each step rules out one way of being fooled that the step below let in. The level ranks the design, not the result, so a small careless trial is marked down and a careful lower-level study with a huge effect is marked up. The top is often empty: in Oxford's own example, 2,175 papers on one drug question held 7 randomised trials. Each step up the evidence pyramid takes away one way of being fooled.
●○○

Overview

Levels of evidence are the pecking order of proof: a ranking of how a claim got checked, from "it stands to reason" at the bottom to "every fair trial ever run, pooled" at the top. Bottom rung: someone reasoning that it ought to work. Next: one patient got better. Next: a bunch of patients got better. Next: people who took it did better than people who skipped it, except nobody checked that the two crowds were alike. Next: a random draw picks who gets it, so the two crowds are alike. Top: round up every one of those trials and pool the numbers. Every rung up kills one way of kidding yourself. The rung is a starting weight, not a medal. Oxford's table says a sloppy trial drops down it and a careful study with a huge effect climbs. And the top rungs are often empty: one Oxford search turned up 2,175 papers on a drug question and 7 proper trials among them. Climb the ladder and every rung crosses off one way of being fooled. 😎

A quick take β€” often all you need.

●●○

Detail

Levels of evidence rank study designs by how much each design rules out, so a reader knows what weight a finding starts with. The most cited ranking is the Oxford Centre for Evidence-Based Medicine's 2011 table, which for the question "does this treatment help" has five levels. Level 5, the bottom, is reasoning from how the treatment ought to work. Level 4 is case reports and case series (one patient's story, or a set of them) and case-control studies (people with the disease compared with people without). Level 3 is a cohort study (a group that took the treatment followed alongside one that did not). Level 2 is a randomised controlled trial (a random draw decides who gets the treatment), or an observational study with a dramatic effect. Level 1 is a systematic review of randomised trials (every trial that meets a set bar, gathered and pooled). Each level outranks the one below because it shuts off one way of being fooled. Reasoning can be fooled by the reasoner. One case can be fooled by that case. A series can be fooled by who got counted. Groups that chose themselves can be fooled by a confounding variable (a hidden difference that drives the outcome), and size does not cure that: a million self-selected people can share the same hidden difference. A random draw removes it and leaves chance, and pooling many trials shrinks chance. The level is a starting weight, not a verdict. The table's own footnote grades a level down for poor quality, imprecise results or inconsistency between studies, and up for a very large effect. The top rungs are often empty: Oxford's own search on one drug question returned 2,175 papers, of which 7 were randomised trials, 366 cohort studies, 234 case-control studies and 196 case reports. A review of weak trials is weak, and it pools only what was published (publication bias). Read any study by asking which rung it sits on, what that rules out, and what it leaves in.
●●○

Detail

Levels of evidence are a ranking of how a claim got tested, and the rank tells you how much benefit of the doubt a finding begins with. Oxford's 2011 table is the one people cite, and for "does this treatment help" it has five rungs, so take them from the top. Level 1: a systematic review, which hunts down every randomised trial on the question and pools them, so no single trial's luck gets the final say. Level 2: one randomised controlled trial, where chance alone picks who gets the treatment, so the two groups match on everything, including the stuff nobody measured. Level 3: a cohort study, where people who took it are followed next to people who did not, and now a hidden difference between the two crowds (a confounding variable) can sneak in. Level 4: case-control studies, case series and single case reports, which begin from people who already have the outcome and look backwards, so who ended up in the sample can rig it. Level 5: reasoning that it ought to work because of how the body works, checked by nobody but whoever did the reasoning. Each rung down lets one more way of being wrong back in. Two traps. First, big does not beat fair: a cohort the size of a city, all volunteers, can still carry one hidden difference, and a trial of 500 picked by chance cannot. Second, the rung is where a study begins, not where it ends. The table's own small print marks a level down for shoddy quality, fuzzy results or trials that disagree, and up for an enormous effect. The top rungs are mostly bare: Oxford's own search found 2,175 papers on one drug question, 7 of them randomised trials, while nearly 800 sat on the two rungs below. And a review is only as good as its trials, and it can only pool what got published (publication bias). So when a headline says "a study shows", ask which rung, what it rules out, and what it let back in. 😎

Want more? One click digs deeper.

●●●

Analogy

Levels of evidence rank ways of checking a claim by how many ways of being fooled each one shuts off, and hiring someone works the same way. Lowest: the candidate's own account of their work, which only the candidate vouches for. Next: one reference, which is one person's view. Then three references from three different jobs that agree, which can still be fooled by who the candidate chose to list. Then their sales, 20% above the team's over two years, which can be fooled by the patch they were given. Highest: a work trial where two candidates do the same task and the marker does not know whose work is whose, which only a bad day can fool. Each step up closes one door that the step below left open. Where the ranking is a start and not a verdict: a work trial marked by the candidate's friend is worth less than three solid references. A high rung done badly loses to a low rung done well, and that holds for studies too.
●●●

Analogy

Levels of evidence sort the ways of finding something out by how many routes to being wrong each one has blocked, and "will these earbuds break?" runs the whole ladder. One one-star review: one story. A thread where 300 people say theirs died: 300 stories, from the people annoyed enough to post. A warranty firm's return rates by brand: a real comparison, except heavy users tend to buy one particular brand, so the brands are not being used the same way. A testing lab that drops 25 pairs of each brand from the same height: a fair test, where only chance remains. A site that gathers every lab's drop tests: the review at the top. Every step up bolts another exit for error. Now the catch: a lab that dropped two pairs sits below a warranty firm with 50,000 returns, because a high rung botched comes in under a low rung nailed. The rung tells you where the weighing begins, not where it ends. 😎

Unfamiliar concept? A real-world example makes it click β€” fresh analogies on tap.

AI explanations may contain errors · Not professional advice

Formal definition β€” The same term, explained the usual way

Levels of evidence are hierarchies that rank study designs by their expected protection against bias, used to locate the likely best evidence for a clinical question and to assign a starting grade to it. The Oxford Centre for Evidence-Based Medicine (OCEBM) 2011 Levels of Evidence (v2.1) is a grid of seven question types against five levels; for treatment benefits, level 1 is a systematic review of randomised trials or n-of-1 trials, level 2 a randomised trial or an observational study with dramatic effect, level 3 a non-randomised controlled cohort or follow-up study, level 4 case-series, case-control or historically controlled studies, and level 5 mechanism-based reasoning. A level may be graded down for study quality, imprecision, indirectness, inconsistency or a very small absolute effect, and graded up for a large or very large effect. The OCEBM's introductory document states that the Levels are a hierarchy of the likely best evidence and a short-cut for appraisal, not a definitive judgment of quality and not a recommendation; its PubMed example ("atrial fibrillation AND warfarin") returns 2,175 articles, of which 7 are randomised trials, 366 cohort studies, 234 case-control studies and 196 case reports. Other schemes include the Melnyk seven-level pyramid used in nursing, the 6S pyramid of pre-appraised sources, and GRADE, which rates certainty in an effect estimate rather than ranking designs.

Want Clicked to explain terms like “level of evidence” directly in your browser β€” including on PDFs?

Add to Chrome β€” Free

50 free Explanations · No credit card required