Bayesian Network Meta-Analysis for Estimating Drug Class Effects, with Applications to Primary Open Angle Glaucoma
*Hwanhee Hong, Johns Hopkins Bloomberg School of Public Health
Keywords: Network meta-analysis, drug class effect, Bayesian hierarchical modeling
Network meta-analysis (NMA) gains its popularity for comparing multiple interventions simultaneously in a single analysis by combining evidence within studies (direct evidence) and evidence across studies (indirect evidence). However, relatively few NMA methodologies have been developed to investigate drug class effects. Decision makers including physicians are interested in knowing the class effects as well as the specific drug effects. For example, physicians usually first consider the best drug class for a patient and then decide a specific drug within that class to treat the patient. This would be a more efficient and natural decision-making process than finding the best treatment among all possible treatment candidates. We propose a Bayesian NMA modeling framework for estimating both class and drug effects simultaneously, where the drug effects are nested to the class effects. We can order classes and drugs based on ranking probabilities, where drugs are ranked within each class. We apply our methods to a recent systematic review study for open angle glaucoma and show practical inferences of our findings by incorporating external safety information.