Case study in rheumatology

This chapter illustrates the use of Bayesian hierarchical models for network meta-analysis using an application in rheumatoid arthritis. A number of biologic agents are licensed for first line treatment in rheumatoid arthritis. All of these have demonstrated considerable effectiveness in placebo controlled trials; however, few head to head comparisons exist. We aimed to combine evidence from randomized controlled trials and observational studies to determine the relative efficacy between these agents. A systematic review first identified relevant trials comparing the efficacy of biologic agents licensed for first line use to placebo or one another at 24 weeks in patients who have had an inadequate response to methotrexate (MTX). Randomized controlled trial data was analyzed using Bayesian network meta-analysis models estimating the relative efficacy for three alternative outcome measures: American College of Rheumatology (ACR) 20 and ACR50 response and percentage improvement in Health Assessment Questionnaire (HAQ) score. Odds ratios were estimated for binary outcomes; the mean difference is estimated for the continuous measure. The model was extended to a three level hierarchical model to allow for the systematic inclusion of observational evidence. The model estimates trial type level as well as overall estimates and allowed for bias adjustment while also accounting for heterogeneity between trial designs. The analysis confirmed the considerable effectiveness of all agents over placebo and detects some differences between agents. While a general tendency was found, the analysis highlighted the dependency of results on the chosen measure of efficacy. Incorporating observational data strengthened the overall results and outcomes were robust against bias adjustments. Most network meta-analyses only consider evidence from randomized controlled trials, while information from other trial designs is ignored. In this chapter we illustrate how a flexible methodology can be used to synthesize data from different study designs. This should allows for informed decision making based on all available evidence.