MULTIPLE OUTCOME META-ANALYSES

Introduction Commonly, the main aim of a meta-analysis is to provide, when possible, a clear and definite evidence on, for example, the best treatment available for a disease. However, when multiple potentially correlated outcomes are of interest in evidence synthesis, they should be jointly analyzed, taking into account their correlation. Standard meta-analysis combines estimates of one parameter over several studies, but it is not appropriate when there is the necessity to consider multiple potentially correlated outcomes. Thus, the Multiple Outcome Meta-Analysis (MOMA) was thought as an extension of the standard meta-analysis in order to combine estimates of several related parameters. Methods In order to evaluate the feasibility, advantages and limitations of the MOMA compared to the standard meta-analysis, data on azathioprine use in multiple sclerosis treatment, in terms of number of patients with relapses, number of dropouts and number of patients with a disease progression over two years, were analyzed through the use of these two methods. Results Using a method proposed by Riley at al. in 2008 in the context of the MOMA, which showed to be a good method especially when the within-study correlations are not known, compared to the standard meta-analysis, different results were observed in terms of Odds Ratios (ORs), corresponding 95% Confidence Intervals (95% CIs), explained and residual variances, and between-studies correlations. In particular, considering the two outcomes of efficacy, in terms of number of patients with relapses, and of safety, in terms of number of dropouts, over two years, and comparing results obtained through the use of the MOMA with those derived from the standard meta-analysis, the ORs and the corresponding 95% CIs were similar, but the corresponding standard errors (SE) appeared to be very different. Accordingly, using the MOMA methods, SE reductions of 18% and 5% for the two outcomes were observed, respectively. Moreover, when the two outcomes representing two different efficacy measures in terms of number of patients with relapses and number of patients with a disease progression over two years were considered, a relapse risk reduction of over 20% (i.e., an OR of 0.50 from the MOMA vs 0.64 from the standard meta-analysis), and a progression risk reduction of almost 15% (i.e., a significant OR of 0.66 vs a non-significant OR of 0.77) were observed. In this second application, however, the small number of studies could have led to a very high between-studies correlation estimate, which is associated with unstable pooled estimates and SE. Thus, in this case the MOMA could not represent the better choice. Discussion Unless the variation observed was very large, or the number of studies in the meta-analysis was small, and when there is the necessity to combine estimates of several potentially correlated outcomes, the MOMA methods appear to produce appropriate pooled estimates, which show better statistical properties than those from the standard meta-analysis. Finally, how to combine the methods of the MOMA with those of the Multiple Treatment Meta-Analysis (MTMA), in order to jointly analyzed multiple outcomes and including both direct and indirect comparisons, was presented.