Online Program

Comparison of Binomial and Multinomial Network Meta-Analysis Model by Simulation

*Ruoshui Zhai, Graduate Student at Brown University Biostatistics Department 
Christopher Schmid, Brown University  
Thomas Trikalinos, Brown University 

Keywords: Multivariate Network Meta-Analysis, Simulation, Fractional Factorial Design

Network meta-analysis (NMA) simultaneously compares multiple treatments from a series of clinical studies. We have previously proposed an NMA method based on multinomial distributions for unordered categorical outcomes. In contrast to the usual methods that compare outcome categories in pairs using binomial likelihoods, the multinomial method incorporates the correlations between the outcomes and allows for simultaneous comparison of all treatments on all outcomes. Although the multinomial model should theoretically give better answers than separate binomial models, it may be sensitive to how well the correlations can be estimated. We compare the two approaches in an extensive adaptive simulation study varying the total number of studies, relative proportions of each treatment comparison, underlying probabilities in the reference treatment and outcome groups, and odds ratios between the groups. The adaptive scheme is based on a series of fractional factorial designs that enable us to focus on the key factors that differentiate the two approaches.