OBJECTIVES: Evaluating the effectiveness and safety of an intervention is a major component of health technology assessments (HTAs) and network meta-analysis (NMA) has emerged as the standard approach to compare three or more interventions in a single model. However, the connections between interventions within a given network are frequently sparse (few or no direct comparisons), which often results in highly uncertain estimates that have limited utility for HTA decision-making. Here, we investigate one approach that has the potential to solve this problem: incorporating multiple outcomes within a multivariate NMA (M-NMA).

METHODS: We identified dense networks of evidence based on randomized controlled trials (RCTs) that reported at least two outcomes that were likely to be correlated. We implemented standard Bayesian NMAs on each outcome separately with the full networks of evidence. We then removed RCTs from the networks to simulate the common scenario of sparse networks of evidence. Using these sparse networks, we implemented both standard NMAs and M-NMAs that incorporate two outcomes simultaneously. We assessed the precision of the models by comparing results across models for the sparse networks and evaluated the accuracy of the models by comparing model results for the sparse networks with results for the full networks of evidence.

RESULTS: M-NMA models generally produced more precise estimates than standard NMAs (narrower 95% credible intervals) for the sparse networks of evidence. Additionally, for sparse networks, M-NMAs produced estimates closer to the results for the full networks of evidence than did standard NMAs.

CONCLUSIONS: Under some circumstances, accounting for correlations between outcomes in an M-NMA of multiple outcomes can improve the precision of effect estimates and potentially produce more accurate findings. Consequently, M-NMA may be capable of producing estimates that are useful for HTA decision-making when the value of a standard NMA is limited.