Bayesian meta-analysis of the binary outcomes of randomized clinical trials

Abstract

The study presented the main aspects of conducting Bayesian network meta-analysis as a method of indirect comparisons using mathematical models. To describe the performance of the Bayesian network meta-analysis, codes for the random-effects and fixed-effects models were included. The models were written in Component Pascal and run in the JAGS program. To allow cross-reference of results, data used for modeling purposes were generated data from the article by D. Hu, A.M. O'Connor, S. Wang et al. [6]. The R language and rjags package were used to load data into the model and run the program. To determine the best model, model adequacy indicators were used, such as total residual deviation, and leverage and deviation information criterion were calculated using the original R code. A graphical method was also used to determine the adequacy of the models using the ggplot2 package. An example of constructing a network of evidence based on available results on the effectiveness of drugs from clinical trials was considered, taking into account the assumptions of transitivity and heterogeneity. Indirect and direct comparisons to determine the true estimates of drugs were possible. The elements of Bayesian statistics, such as prior and posterior probabilities and likelihood, and the advantages of using them in meta-analysis were explained. The mathematical apparatus of the generalized linear model was presented in both general and specific forms, using binomial output data to obtain relative estimates of the effects of therapies. An explanation of how the models work is presented. The random-effects model showed superiority over the fixed-effects model in the comparison of adequacy metrics. To achieve better adequacy, data must be carefully downloaded from publications, and informative priors selected. In general, Bayesian synthesis is a distinct and important type of network meta-analysis. It is unique because it uses a probabilistic approach to data analysis. Understanding the basic principles of Bayesian statistics is also an important aspect of the successful use of this method in various research fields. However, for effective application of this method, attention must be paid to careful data preparation and the choice of priors. With informative prior distributions and proper implementation, Bayesian synthesis can produce more accurate and reliable results than other meta-analysis methods. Bayesian synthesis is a method of statistical data analysis recognized worldwide and in the Russian Federation..