Bayesian network meta-analysis for biologic therapies in rheumatoid arthritis

Meta-analysis can only compare studies with the same interventions, while a network meta-analysis can analyze studies with different interventions. Without medical data for direct comparisons, network meta-analysis can utilize existing trials to assess the relative efficacy of competing treatments. Three classes of statistical models are proposed to perform a network meta-analysis: fixed-effects, random-effects, and mixed-effects (meta-regression). The most appropriate model should be selected with the aid of a series of statistical tests including I2 statistic for heterogeneity and DIC for model fitness. Bayesian network meta-analysis provides pooled effect sizes (odds ratio for dichotomous outcome) for each treatment and their 95% probability credible intervals. After a systematic literature review, 20 randomized clinical trials of biologic anti-rheumatic therapies in combination with methotrexate in rheumatoid arthritis patients were identified. Random-effects model was used for ACR20 and ACR70 criteria treatment outcome whereas the mixed-effects model was used for ACR50 treatment outcome. Based on the analysis, we found that all biologics DMARDs were superior to placebo except for ANA in all datasets and RTX in ACR70 dataset. ETN was had the highest probability to be the best treatment in all three datasets. CTZ had the highest probability to be the second best option in ACR20 and ACR50 datasets, and TCZ held the second place in the ACR70 dataset. The rest of the rank probabilities vary by dataset but placebo was the lowest ranked option in all datasets. Therefore, despite the limitations of this study, the results are consistent with current knowledge that biologic DMARDs are superior to placebo and although more research remains to be done, ETN may be the most effective option for rheumatoid arthritis