Abstract
The GRADE method (Grading of Recommendations, Assessment, Development, and Evaluation) provides a tool for rating the quality of evidence for systematic reviews and clinical guidelines. This article aims to analyse conceptually how well grounded the GRADE method is, and to suggest improvements. The eight criteria for rating the quality of evidence as proposed by GRADE are here analysed in terms of each criterion's potential to provide valid information for grading evidence. Secondly, the GRADE method of allocating weights and summarizing the values of the criteria is considered. It is concluded that three GRADE criteria have an appropriate conceptual basis to be used as indicators of confidence in research evidence in systematic reviews: internal validity of a study, consistency of the findings, and publication bias. In network meta-analyses, the indirectness of evidence may also be considered. It is here proposed that the grade for the internal validity of a study could in some instances justifiably decrease the overall grade by three grades (e.g. from high to very low) instead of the up to two grade decrease, as suggested by the GRADE method.