OBJECTIVES: In health technology assessment meta-analysis is used to combine evidence from a number of studies to inform the decision-making process. When evaluating new health technologies at early stages of their development, treatment effects on short-term surrogate endpoints may be used to predict the effect on the final outcome that otherwise requires long follow-up time. Meta-analysis of multiple outcomes which takes into account the correlations between them is particularly suitable for modelling surrogate endpoints. The aim of this study was to investigate the choice of distributional assumptions when developing meta-analytic methods for evaluation of surrogate endpoints. METHODS: Two bivariate meta-analytical models are applied to a case study in chronic myeloid leukaemia where overall survival (OS) proportion, is the final outcome and complete cytogenetic response (CCYR) rate at 12 months is a surrogate endpoint. A normal model on log relative risk scale for both outcomes is applied to evaluate CCYR as a surrogate endpoint for OS. This model is then extended by relaxing the assumption of normality by the use of binomial distributions for both outcomes. RESULTS: The effect on CCYR was a significant predictor of the effect on OS. Both models gave similar results for the effect of CCYR on OS. However, the heterogeneity parameter was larger in the binomial model (τ=0.09 with 95% CrI; 0.0 to 0.52) compared to normal case (τ=0.07 with 95% CrI; 0.0 to 0.39). CONCLUSIONS: The results of both models were similar for this case study. However, the choice of distributional assumption can lead to different estimates of the effect on the final outcome in other disease areas when the normality assumption is not suitable and consequently this can impact on HTA decisions.