OBJECTIVES: Clinical studies generally include several endpoints to compare the effects of alternative interventions. Meta-analyses are usually performed on different endpoints separately. We investigated advantages of bivariate meta-analysis models, accounting for the correlation between endpoints, compared to univariate meta-analyses. METHODS: Alternative meta-analysis approaches were applied and compared using simulated datasets of logarithms of odds ratios (OR) for two endpoints.  Several datasets of 20 studies were simulated, with different correlations between endpoints, and with or without missing values. Simulations were based on a bivariate normal distribution with mean log ORs of -0.5, corresponding to ORs of 0.61, and variances of 0.25 for both endpoints. The models used were: 1) random-effects univariate models for each endpoint separately; 2) two-stage approach using univariate model for studies with one endpoint and bivariate model for studies with two endpoints; and 3) bivariate model with prior imputation of the variance of second endpoint for studies with one endpoint only, based on the correlation between variances for the two endpoints. All the models were estimated in a Bayesian framework, using WinBugs. RESULTS: Results of different models were fairly similar in absence of missing data. In a situation with one endpoint missing at random for 10 studies, and a correlation of 0.8, the bias around estimated OR for that endpoint was 0.12, 0.03, and 0.04 with models 1, 2 and 3 respectively, when an informative prior was used for the correlation. The bias was not reduced with uninformative prior. Variance estimates also differed between models, and were very large with model 2 for some simulations. CONCLUSIONS: Bivariate meta-analysis can improve treatment effect estimates when information is collected for two correlated endpoints, especially for an endpoint which is not included in all studies. Furthermore, the model with prior imputation of the variance appeared to be more stable than two-stage model.