Background: The World Health Organization (WHO), like many other organisations around the world, has recognised the need to use more rigorous processes to ensure that health care recommendations are informed by the best available research evidence. This is the seventh of a series of 16 reviews that have been prepared as background for advice from the WHO Advisory Committee on Health Research to WHO on how to achieve this.
Objectives: We reviewed the literature on what constitutes "evidence" in guidelines and recommendations.
Methods: We searched PubMed and three databases of methodological studies for existing systematic reviews and relevant methodological research. We did not conduct systematic reviews ourselves. Our conclusions are based on the available evidence, consideration of what WHO and other organisations are doing and logical arguments. KEY QUESTION AND ANSWERS: We found several systematic reviews that compared the findings of observational studies with randomised trials, a systematic review of methods for evaluating bias in non-randomised trials and several descriptive studies of methods used in systematic reviews of population interventions and harmful effects. What types of evidence should be used to address different types of questions? The most important type of evidence for informing global recommendations is evidence of the effects of the options (interventions or actions) that are considered in a recommendation. This evidence is essential, but not sufficient for making recommendations about what to do. Other types of required evidence are largely context specific. The study designs to be included in a review should be dictated by the interventions and outcomes being considered. A decision about how broad a range of study designs to consider should be made in relationship to the characteristics of the interventions being considered, what evidence is available, and the time and resources available. There is uncertainty regarding what study designs to include for some specific types of questions, particularly for questions regarding population interventions, harmful effects and interventions where there is only limited human evidence. Decisions about the range of study designs to include should be made explicitly. Great caution should be taken to avoid confusing a lack of evidence with evidence of no effect, and to acknowledge uncertainty. Expert opinion is not a type of study design and should not be used as evidence. The evidence (experience or observations) that is the basis of expert opinions should be identified and appraised in a systematic and transparent way.