CEBM (Centre for Evidence-Based Medicine) Levels of Evidence

Introduction (read before using the new Levels Table!)

Hierarchies of evidence have been somewhat inflexibly used, and criticised, for some decades. The CEBM "levels of evidence" were first produced in 1998 for Evidence-Based On Call to make the process of finding appropriate evidence feasible and its results explicit. We have revised the "levels" in light of new concepts and data, and we would like to hear your feedback.

The "levels" are essentially a heuristic, or short-cut to finding the likely best evidence. While ideally we should look at "all the evidence" that might help to answer a question, doing so is often impractical or unfeasible. For example, suppose for a journal club next week you needed to look at the evidence for the use of warfarin in atrial fibrillation. A PubMed search of the words "atrial fibrillation AND warfarin" yields thousands of hits. As you will not have time to read them all, you will need some guidance as to which articles might be the most valid and useful for purpose. The table below shows a detailed breakdown, filtered by different MeSh terms for study types.

For example, if your main interest is the size of the benefit of warfarin and its common harms, it would be reasonable to focus on systematic reviews or trials. If no trials had been shown, you might look at the cohort studies, etc. Since there are several trials, it would be sensible to find out if a systematic review is being planned, and recommend or subsequently undertake one if there is not.

 

Click here to download the new Levels of Evidence Table (PDF Version)

 

Table: results of a PubMed search for "atrial fibrillation AND warfarin" with some filters

Type	Term used	Number of articles
All articles	(no filter)	2175
RCT	"random allocation" [MeSH]	7
cohort	"cohort studies" [MeSH]	366
Case-control	"Case-Control Studies"[Mesh]	234
Case report	Case Reports [Publication Type]	196

(search done Jan 7th 2010)

 

The reformulated Levels of Evidence table has been set out as a series of searching "steps". While the steps (columns) give the likely "level" of evidence, the level selected will also depend on (i) the nature of the question, and (ii) the quality, quantity, and consistency of primary evidence found at that step. Poor quality or consistency of evidence may mean the "level" is downgraded; large effects or clear dose-response relationships may upgrade the level. The GRADE process [2] spells out in detail what these elements are for issues concerning interventions, and provides a "level" of evidence of high, moderate, low or very low evidence. The background document (REF) provides more detail and examples illustrating these issues.

 

In this revised table several features are worth noting.

1. The rows indicate what type of research is likely to be best for each type of question.

2. The columns indicate the sequence of steps you might take in searching. How many steps you take will be constrained by your time and resources.

3. Evidence for different outcomes may have different levels and even be obtained from different studies. While all outcomes may be covered in one study or review, it may be necessary to assess several studies with different designs, e.g, to find the benefits and harms of a treatment.

Alternatively, suggest:

4. One review or study report may supply all the outcomes sought/needed/need to be considered. But it may be necessary to seek out several studies with different designs to find all the information relevant for purpose. Care is needed to recognise that `quality of evidence` is not necessarily synonymous with `strength of recommendation`, and vice versa. Judgement is necessary. Only studies seeking outcomes relevant to patients` needs should be used.

 

Click here to post your feedback

 

References

Ball CM, Phillips RS. Evidence-Based On-Call. Churchill Livingstone: Edinburgh, 2001.

Guyatt, G.H., et al., GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ, 2008. 336(7650): p. 924-6.

Page last edited: 10 June 2010

• Login

Designed and updated by Cache
Hosted and content managed by Minervation