Network Meta-Analysis for Decision-Making

11*Multiple Outcomes

11.1 Introduction

In analyses of which treatment is most effective or most cost-effective, there are, more often than not, several outcomes to consider. Some of these evidence structures, those relating to ordered categories and competing risks, have already been touched on in Chapter 4. In this chapter we turn to structures where the relationships between the outcomes are more complex. These might be multiple observations taken at different follow-up times, different outcomes measured at the same time or simply different endpoints such as stroke, bleed and death. Almost invariably we find that while many trials report more than one outcome, each trial reports a different set of outcomes. Evidence structures are therefore always incomplete and sometimes very sparse.

Separate analysis of each outcome is certainly an option and may serve as a useful preliminary to some of the more sophisticated analyses described in this chapter. But a combined analysis of multiple outcomes is likely to be preferable for several reasons. Firstly, the outcomes are likely to be correlated, both within and between trials, so that separate analyses would involve a degree of double counting. In addition, if several outcomes are used in a cost-effectiveness analysis or in a multi-criterion decision analysis, it is essential that correlations are propagated into the decision model.

Secondly, it may produce a more richly connected network that includes comparisons between treatments ...