Levels of Evidence

Recommendations from clinical trials are often ranked by the level of the evidence. This is a method of classifying data based upon the degree to which bias has been eliminated, typically with systematic reviews of randomised controlled trials as the 'top' level and expert recommendation as the 'bottom' level. There are several systems in place.



The NHMRC ranks evidence for an intervention on four levels, with level III divided into three 'sub-levels':

  • Level I: Systematic review of randomised controlled trials
  • Level II: Single randomised controlled trial
  • Level III:
    • Level IIIA: Pseudo-randomised controlled trial (eg. alternate allocation instead of true randomisation)
    • Level IIIB: Non-randomised study with concurrent control (eg. case cohort study)
    • Level IIIC: Non-randomised study with no concurrent controls (eg. historical cohort study)
  • Level IV: Case series

The NHMRC also ranks evidence for other research questions, including prognostic studies.


The National Cancer Care Network publishes guidelines on many topics, and includes a four tier ranking of levels of evidence that also includes whether there is consensus opinion amongst members of the NCCN.

  • Category 1: Uniform consensus based on high level evidence
  • Category 2A: Uniform consensus based on low level evidence
  • Category 2B: Non-uniform consensus based on low level evidence
  • Category 3: Based upon any level of evidence, there is major disagreement within the NCCN that the intervention is appopriate

Eg: For DCIS, NCCN guidelines state that lumpectomy + RT is category 1, mastectomy is category 2A, and lumpectomy alone is category 2B.



Grading research by the methods used to reduce bias allows the reader to quickly determine the strength of an article or guideline's recommendation, rather than taking all recommendations at the same face value or looking up each recommendation individually.


Ranking evidence based upon the means used to reduce bias does not necessarily mean that the evidence is solid; there may be biases that were not disclosed or the population studied may not be relevant to a particular patient (eg. elderly patient with breast cancer). For some rare tumours, evidence may be limited and therefore randomised controlled trials may be impossible; this may make any recommendations of low level but still the best available given the limitations.