This network meta-analysis was carried out in accordance with the previously submitted protocol and is consistent with the statements of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). We followed good research practices as outlined in the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) report on interpreting indirect treatment comparisons and network meta-analysis for healthcare decision making.
Inclusion Criteria
The included studies must be an RCT, and the observation time should not be less than 6 months. In addition, the following points must be met: 1) The patients must be adults (older than 18 years) who were diagnosed with IMN by renal biopsy; 2) The interventions in treatment group must be CTX, Chlorambucil, MMF, TAC, CsA, rituximab or steroids, of which the first five immune regiments can be combined with steroids or used alone. The measures in control groups included placebo/no treatment, conservative treatments (ACEI/ARB or other supportive treatment), steroids, CTX, chlorambucil or CsA. The latter three immunotherapies can be combined with steroids or used alone. 4) Outcomes must include remission rate (including complete remission and partial remission) or incidence of side effects, or both.
Exclusion Criteria
Literature was excluded according to the following criteria: 1) Studies which were non-randomized trials or observational studies; 2) Studies with patients of secondary membranous nephropathy (e.g. hepatitis B-related nephropathy); 3) Studies which included patients younger than 18 years; 4) Studies with an observation time less than 6 months; 5) Studies which two or more immunosuppressive drugs were used in one intervention or; 6) used drugs which are not widely used such as ACTH and leflunomide; 7) Chinese medicine was excluded; 8) Azathioprine and mizoribine were also excluded due to their documented poor efficacy, and serious side profile.
Retrieval Strategy
Two researchers (NB and XY) performed literature retrieval independently and only articles written in English were included. The retrieval databases include Pubmed, Embase, Cochrane Database and Medline. The timeline for retrieval was from the establishment of each database to March 2019. In the case of a duplicate, each case will be resolved through negotiation. The details of the retrieval strategy are shown in supplement 2.
Baseline characteristics and quality assessment
Detailed information such as the study design, sample sizes, specific dosage of therapeutic drugs and specific characteristics of patients included in the study will be extracted separately by two researchers (M-JG and XY). Outcomes were extracted at the end of each study. When multiple time points were reported, the last time-point was used. We used the Cochrane Risk of Bias assessment tool to assess the quality of individual literature. According to the Cochrane Risk of Bias assessment, remission rate and the incidence of side effects were objective indicators and thus, could not be affected by performance bias and detection bias. Therefore, while most of the included studies were not double-blinded, the aforementioned sources of bias can be assumed to be low. Any doubts were resolved by direct discussion with the group through consultation.
Statistic Methods for Outcomes
This study primarily analyzed the remission rate and the incidence of adverse effects such as leukopenia and infection. We used the intention-to-treat (ITT) approach to calculate the remission rate. Specifically, all patients in treatment groups were analyzed using statistical analysis. All patients lost to follow-up were considered as failures of treatment. While the ITT method is well suited for the remission rate, the per-protocol (PP) method was better suited to measure the incidence of adverse events. Specifically, only patients demonstrating a high level of compliance that completed treatment were included.
Data Extraction and Synthesis
In this study, the rate of remission and the incidence of adverse events can be expressed by dichotomous variables. A direct comparison was made using the random effect model to determine the odds ratio (OR) with Revman 5.3 (Cochrane Collaboration, Copenhagen, Denmark). The confidence interval (CI) is was set at 95%. Heterogeneity was determined by analyzing the value of I2. If I2 > 50%, the heterogeneity was said to be high, otherwise it was low. Publication bias will be tested through the Egger's regression test11. If the P-value > 0.1, then no publication bias was said to have occurred between these papers; otherwise publication bias was suggested. Stata 15 (Stata Corp. 2017. Stata Statistical Software) was used to carry out network meta-analysis, including network diagram, contribution plots, and inconsistency tests. In addition, we used R software x64 3.5.0 (Department of Statistics, University of Auckland, New Zealand) to indirectly evaluate the size between these interventions with OR within the random effect model and the confidence interval was set at 95%. To make an indirect comparison between the two interventions, it is necessary to find a reference which performs a direct comparison with both and merge the effects. If there were both direct and indirect comparisons between two interventions, the node splitting method was used to calculate the value for direct and indirect comparisons respectively. Then the Markov chain Monte Carlo model was used to diagnose convergence and draw trajectory density maps with 20,000 simulated draws after a burn in of 5,000 iterations. Finally, the GEMTC package was used to rank each intervention in terms of alleviation rate and side effects respectively. The higher the percentage, the greater the possibility of ranking ahead. If the percentage was 100%, the drug ranks first.
Evidence Quality Assessment
We evaluated the quality of the included direct comparisons based on the RCT evidence quality assessment method published by the GRADE working group12-16. At the beginning of the assessment, all included RCTs were set at a high level of quality. Then the assessment was performed according to five aspects including study limitations, publication bias, imprecision, inconsistency, and indirectness. For each criterion a study failed to meet, the level of guidance was lowered by one. For serious cases, two points were removed for the infraction. Results were summarized after the assessment of five items was completed.
The quality assessment for indirect comparisons was carried out according to the method described by Puhan et al17. First of all, the best comparable path needs to be selected. The lower the number of interventions in the indirect comparison path, the more credible the results will be. After the best indirect comparison path was determined, the quality of evidence for a single direct comparison in the path was assessed according to the method mentioned above. The lowest level of evidence was chosen to reflect the final quality of evidence for this group of indirect comparisons. If there are both direct and indirect effects in a group of comparisons, both comparisons were assessed individually, and the higher level of evidence was selected to reflect the final result. The final step was to evaluate the inconsistencies in results, which included differences in baseline characteristics, common references, and measurements for outcomes among different groups. If the differences were significant, the final quality level will be further reduced by one. This work was performed on Stata.