Scientific Writing

Levels of Evidence: Where Your Study Sits and How to Say It

DE By Directive Editorial Team, Directive Publications ·12 Nov 2026 ·7 min read
Levels of Evidence: Where Your Study Sits and How to Say It

Levels of evidence rank study designs by how reliably each can answer one type of clinical question. In the 2011 Levels of Evidence from the Oxford Centre for Evidence-Based Medicine (CEBM), a systematic review of randomised trials is Level 1 for treatment benefit, but the order changes for diagnosis, prognosis and harms. State the scheme, the question and the level together.

This guide shows where your study sits and how to state its level without overstating it. How strongly to word findings for each design is covered in our guides to writing a research paper conclusion that matches its design and how to write a Discussion section.

What a levels of evidence scheme is, and why several exist

A levels of evidence scheme is a ranked list of study designs for a stated type of question. Designs near the top are less prone to bias for that question. The idea is often drawn as an evidence pyramid, with systematic reviews at the apex and expert opinion at the base, but a single ranking cannot show that the best design depends on the question.

Three kinds of scheme appear in manuscripts and journal instructions:

  • The Oxford CEBM 2011 Levels of Evidence. A table that ranks designs separately for each clinical question, numbered Level 1 to Level 5.
  • Journal-specific tables. Some journals print their own levels table in their instructions for authors and ask authors to state a level. When a journal asks, use that journal's table.
  • Generic hierarchies and pyramids. Diagrams in textbooks and teaching slides, often with no version, no source and their own numbering.

Numbering is not shared between schemes. Some use Arabic numerals and others Roman numerals, and the designs behind each number differ. "Level IV evidence" means nothing until the reader knows which scheme assigned it.

How the Oxford CEBM 2011 levels of evidence rate each question

The CEBM table was written by the centre's Levels of Evidence Working Group. It has one row per clinical question and five steps across. Step 1 is the likely best evidence for that question, and later steps are the next best when nothing better exists. The table below paraphrases six of its rows. Check the original levels table on the CEBM website before you cite a level, because the exact wording carries conditions this summary shortens. That page now asks for the table to be cited as "The Oxford Levels of Evidence 2", although the table itself is headed 2011.

QuestionLevel 1Level 2Level 3Level 4Level 5
How common is the problem? (prevalence)Local and current random-sample surveys or censusesSystematic review of surveys that allow matching to local circumstancesLocal non-random sampleCase seriesNot applicable
Is this diagnostic or monitoring test accurate? (diagnosis)Systematic review of cross-sectional studies with a consistently applied reference standard and blindingIndividual cross-sectional study with a consistently applied reference standard and blindingNon-consecutive studies, or studies without a consistently applied reference standardCase-control studies, or a poor or non-independent reference standardMechanism-based reasoning
What will happen without added therapy? (prognosis)Systematic review of inception cohort studiesInception cohort studiesCohort study, or the control arm of a randomised trialCase series, case-control studies, or a poor-quality prognostic cohortNot applicable
Does this intervention help? (treatment benefit)Systematic review of randomised trials or n-of-1 trialsRandomised trial, or observational study with a dramatic effectNon-randomised controlled cohort or follow-up studyCase series, case-control or historically controlled studiesMechanism-based reasoning
What are the common harms? (treatment harms)Systematic review of randomised trials or of nested case-control studies, n-of-1 trial in the patient concerned, or observational study with a dramatic effectIndividual randomised trial, or (exceptionally) observational study with a dramatic effectNon-randomised controlled cohort or follow-up study, with enough patients to rule out a common harmCase series, case-control or historically controlled studiesMechanism-based reasoning
Is this early-detection test worthwhile? (screening)Systematic review of randomised trialsRandomised trialNon-randomised controlled cohort or follow-up studyCase series, case-control or historically controlled studiesMechanism-based reasoning

A seventh row covers rare harms and lists only the first two steps. A footnote adds that a systematic review is generally better than an individual study.

Why the same study design sits at different levels

In the CEBM table, a study design has no fixed level. Its level depends on the question the study answers.

  • An inception cohort enrols patients at a common early point in their illness. It is Level 2 for prognosis. A non-randomised comparative cohort is Level 3 for treatment benefit, because the groups may differ in ways that also affect the outcome.
  • A cross-sectional study comparing a test with a consistently applied reference standard, with blinding, is Level 2 for diagnosis. For test accuracy it is the right design, not a weak substitute for a trial.
  • A local random-sample survey is Level 1 for how common a problem is. No randomised trial answers that question.
  • Case-control studies sit at Level 4 for diagnosis, treatment benefit and screening.

So name the question before you look for your level. A retrospective cohort that reports both outcome over time and response to treatment answers two questions, and it may sit at a different level for each.

A level of evidence is not a judgement of quality

The CEBM levels rank designs, not individual studies. The CEBM introductory document says the levels are not intended to give "a definitive judgment about the quality of evidence". It also says they will not provide a recommendation. A footnote to the table lets a level move:

  • Graded down for study quality, imprecision, inconsistency between studies, or a very small absolute effect.
  • Graded down for indirectness, where the study's population, intervention, comparison and outcome (PICO) do not match those of the question.
  • Graded up for a large or very large effect.

A randomised trial with heavy loss to follow-up can therefore fall below Level 2. The introductory document gives its own example: an observational study with a dramatic effect may provide stronger evidence than a systematic review of a few studies that reached no clear result.

Reviews that judge a whole body of evidence on one outcome use certainty ratings instead. Our guide to risk of bias tools and GRADE certainty explains how Grading of Recommendations Assessment, Development and Evaluation (GRADE) rates certainty from high to very low.

Where expert opinion, mechanism and narrative reviews sit

Mechanism-based reasoning is Level 5 in the 2011 table for diagnosis, treatment benefit, common harms and screening. It argues from how a disease or treatment works rather than from observed outcomes. For prevalence and prognosis, the table marks Level 5 as not applicable.

Expert opinion does not appear as an entry in the 2011 table, although pyramid diagrams often place it at the base. If a scheme you cite lists expert opinion, quote that scheme's wording rather than the pyramid's.

A narrative review is not in the table either, so cite the level of the primary studies it discusses instead. Our comparison of systematic reviews, meta-analyses and literature reviews explains the difference in method.

A systematic review is only as high as the studies it pools. Level 1 for treatment benefit requires a systematic review of randomised trials. Pooling case series gives a more complete summary of Level 4 evidence; it does not turn that evidence into Level 1. What a single case or a series can claim at all is set out in our guide to choosing between a case report, case series and cohort study.

How to state your study's level of evidence in a manuscript

  1. Name the scheme and its version, for example the Oxford CEBM 2011 Levels of Evidence.
  2. Name the question your primary objective answers: treatment benefit, harms, diagnosis, prognosis, screening or prevalence.
  3. Find the row for that question and the cell that describes your design as you ran it, not as you planned it.
  4. Grade down if a reason in the table's footnote applies, and say which reason.
  5. Check that the design named in the statement matches the design in your title and Methods.

Copy this template and fill every slot:

Level of evidence: Level [number], [scheme name and version]. Question type: [treatment benefit / harms / diagnosis / prognosis / screening / prevalence]. Design: [design as named in the Methods]. [Not graded down / Graded down from Level [number] for [reason]].

The study below is invented to illustrate the structure; it is not real data. A single-centre retrospective cohort compares two surgical approaches for one condition, using the records of consecutive patients. Its primary objective is a treatment-benefit question, and its main estimate has a wide confidence interval.

VersionStatementProblem
OverstatedLevel of evidence: Level 1. This large comparative study provides high-level evidence that approach A is superior.No scheme or question; a cohort placed at the level for systematic reviews of randomised trials; a comparative claim added to a label
HonestLevel of evidence: Level 4, Oxford CEBM 2011 Levels of Evidence. Question type: treatment benefit. Design: retrospective non-randomised controlled cohort. Graded down from Level 3 for imprecision.None: scheme, question, design and the reason for grading down are all stated

Do Directive journals ask for a level of evidence statement?

No. Our author guidelines do not ask for a level of evidence statement. They ask you to match your manuscript to the relevant reporting checklist from the EQUATOR Network, such as CONSORT for randomised trials or STROBE for observational studies.

A level is therefore optional; if you add one, use the template above. If you find an error in this guide, report the problem to us and we will look at it.

Frequently Asked Questions

Is a single randomised trial Level 1 evidence?
Not in the Oxford CEBM 2011 Levels of Evidence. For treatment benefit, Level 1 is a systematic review of randomised trials or n-of-1 trials, and an individual randomised trial is Level 2. A trial with serious flaws or imprecise results can also be graded down.
What does Level 4 evidence mean?
It depends on the scheme, because schemes number their designs differently. In the Oxford CEBM 2011 table, Level 4 for treatment benefit covers case series, case-control studies and historically controlled studies. Always name the scheme beside the number.
Is a systematic review always the highest level of evidence?
No. A systematic review sits at the top only when it pools the designs that are strongest for its question, such as randomised trials for treatment benefit. A systematic review of case series summarises Level 4 evidence more completely, but it does not raise that evidence to Level 1.
Do I have to state a level of evidence when I submit to Directive Publications?
No. The Directive author guidelines do not ask for one; they ask you to match your manuscript to the relevant EQUATOR reporting checklist. If you include a level anyway, name the scheme, its version and the question type.
DE
Directive Editorial Team
Directive Publications

The editorial team at Directive Publications — an international open-access publisher of peer-reviewed medical and scientific journals.

Publishing your research?

Directive Publications is an open-access publisher — every article peer-reviewed, Crossref-registered, and free to read under CC BY 4.0.

Submit a manuscript →Read our Scientific Writing policy →