Multiple Outcomes in Heart Failure Research: Composite Endpoints and Multivariate Modelling

Abstract

Composite endpoints are increasingly popular outcomes in clinical trials of heart failure. Uptake has outpaced guidance on their use and little consistency is seen in their construction. We must consider how best to handle multiple outcomes statistically and clinically, ie in a way that is both cogent for the clinical audience and statistically powerful. The clinical interpretation of composites has been emphasised along with its straightforward analysis and presentation. However there is a loss of information and a more thorough statistical analysis may offer advantages that are not easily dismissed, most obviously a gain in statistical efficiency and power. The modelling approach offers a number of other advantages: 1) adjustment for covariates, 2) a simple test of heterogeneity as the interaction between treatment and outcome, 3) analyses of the individual component endpoints are a consequence of the model, 4) correlations among outcomes are acknowledged, 5) recognises a constellation of risk factors or manifestations of the syndrome without blending them, 6) clinical weights are easily incorporated, and 7) an overall estimate of the effect is obtainable - making it comparable with the results from a composite endpoint. Thus the multivariate modelling approach yields a more powerful and thorough analysis without the loss of information that occurs when multiple outcomes are reduced to a single univariate composite measure. We use data simulations and real clinical trial data to illustrate and evaluate clinical composite endpoints and multivariate modelling. We developed SAS macros for data simulations and analysis methods which we make available.