The biggest advantages of network meta-analysis (NMA) are to compare the effectiveness of different interventions about one conditions using a quantitative way, pool the results of direct comparison and indirect comparison, and rank the effectiveness based on outcomes, so as to select the best decision for patients. In the paper we introduce the methods of applying GRADE system to NMAs based on the papers published by GRADE working group and other relative studies. The steps of using GRADE to NMAs are mainly based on four aspects: firstly, presenting direct and indirect effect estimates and 95% CI; secondly, rating of quality of direct and indirect estimates; thirdly, presenting the results of NMAs; and the last step is to rating the quality of NMA effect estimates. The methods of rating the quality of direct comparison are the same to use GRADE in traditional meta-analysis. The rating of the quality of the indirect estimates is based on the ratings of the two pairwise estimates that contributes to the indirect estimate of the comparison of interest. The lower confidence rating of the two direct comparisons constitutes the confidence rating of the indirect comparison. When both direct and indirect evidence are available, we suggest using the higher of the two quality ratings as the quality rating for NMA estimate. The four steps of rating the quality of NMA from GRADE working group have promoted the theoretical system of NMA. But the process requires the evaluators to be familiar with GRADE system, and conduct pilot test to make sure the evaluators had understood the items of GRADE system correctly. In addition, we also need to concern that the non-transitivity among different groups and the inconsistency between direct comparison and indirect comparison.
Evidence synthesis serves as a bridge between clinical practice and the best available evidence. Evidence synthesis based on high-quality randomized controlled trials is generally considered the highest level of evidence, but its external validity is limited. In some scenarios, the inclusion of non-randomized intervention studies (NRSI) in evidence synthesis may further supplement or even replace randomized controlled trial evidence, such as assessing intervention effectiveness and rare events in a broader population to provide more information for health care decision-making. With the rapid development of real-world data and the improvement of statistical analysis methods, real-world evidence, as an important source of evidence for NRSI, has accelerated the development of high-quality NRSI. However, there are numerous challenges in integrating evidence from randomized and non-randomized intervention studies due to selection and confounding biases caused by the lack of randomization. Based on previous studies, this paper systematically examines the current status of integrated randomized and non-randomized intervention studies, including integration premise, timing, methods, and result interpretation, in order to provide references for researchers and policy-makers to correctly use non-randomized research evidence and further promote optimal evidence generation and clinical practice translation.