This paper presents a generalized linear model (GLM) framework for the synthesis of data from randomized controlled trials (RCTs), applicable to both pairwise and network meta-analysis. The framework uses a linear regression approach with various likelihoods (normal, binomial, Poisson, multinomial) and link functions (logit, log, complementary log-log, probit) to handle different types of outcomes, including binary, continuous, rate, and ordered categorical data. A Bayesian approach is used, with WinBUGS code for Markov chain Monte Carlo (MCMC) simulation. The framework allows for the comparison of models using the deviance information criterion (DIC) and assessment of goodness of fit using residual deviance. It is illustrated through worked examples for common evidence formats. The approach is applicable to pairwise meta-analysis, indirect comparisons, synthesis of multiarm trials, and mixed treatment comparisons. The paper also discusses the use of shared parameter models, where different trials report outcomes in different formats but from a common underlying model. The framework is extended to handle competing risks, continuous data, and other data types. The paper emphasizes the importance of model comparison and the use of DIC to select the most parsimonious model. It also addresses issues related to consistency in evidence and the choice of prior distributions in Bayesian analysis. The paper concludes with a discussion of the conceptual and practical advantages of Bayesian MCMC in probabilistic decision making, and the need for careful attention to technical issues such as convergence, Monte Carlo error, and parameterization.This paper presents a generalized linear model (GLM) framework for the synthesis of data from randomized controlled trials (RCTs), applicable to both pairwise and network meta-analysis. The framework uses a linear regression approach with various likelihoods (normal, binomial, Poisson, multinomial) and link functions (logit, log, complementary log-log, probit) to handle different types of outcomes, including binary, continuous, rate, and ordered categorical data. A Bayesian approach is used, with WinBUGS code for Markov chain Monte Carlo (MCMC) simulation. The framework allows for the comparison of models using the deviance information criterion (DIC) and assessment of goodness of fit using residual deviance. It is illustrated through worked examples for common evidence formats. The approach is applicable to pairwise meta-analysis, indirect comparisons, synthesis of multiarm trials, and mixed treatment comparisons. The paper also discusses the use of shared parameter models, where different trials report outcomes in different formats but from a common underlying model. The framework is extended to handle competing risks, continuous data, and other data types. The paper emphasizes the importance of model comparison and the use of DIC to select the most parsimonious model. It also addresses issues related to consistency in evidence and the choice of prior distributions in Bayesian analysis. The paper concludes with a discussion of the conceptual and practical advantages of Bayesian MCMC in probabilistic decision making, and the need for careful attention to technical issues such as convergence, Monte Carlo error, and parameterization.