Using multiple primary endpoints in clinical trials with a focus on heart failure

Abstract

The use of multiple primary endpoints in cardiovascular clinical trials could be useful addition to the arsenal of comprehensive evaluations of meaningful clinical outcomes. Particularly, it may be advantageous and more economic to use several primary endpoints, if several useful endpoint alternatives exist and when it is uncertain what degree of benefit a certain intervention to be tested can achieve, i.e. what power a trial has for a given endpoint. However, analysis of multiple endpoints gives rise to issues of multiplicity of outcomes and family-wise error rate. There are statistical adjustment models (single and multistep) that modify the level of significance for each endpoint based on the number of endpoints considered overall to control the family-wise error rate. The Bonferroni method is a single step approach that divides the nominal significance level alpha equally across all endpoints but is considered a conservative approach in cases where the number of endpoints is large and where endpoints are correlated. The most used multistep approaches include the Holm and Hochberg procedures. The Hochberg method is a more efficient, and less conservative approach towards alpha adjustment compared to the Holm procedure. The Bonferroni, Holm and Hochberg test procedures are all considered suitable analysis strategies for multiple primary endpoints with no need to determine a priori the order for the testing to be performed as is needed in all hierarchical test procedures that are most commonly used today. Furthermore, these strategies can also be used to protect the error rate when including secondary endpoints in an extended analytical procedure. The use of any of these methods needs to be specified a priori in the statistical analysis plan to ensure adequate statistical validity. Examples of clinical trials in the heart failure field that have used or are using such multiple primary endpoint approaches are: MIRACLE, ASCEND-HF, EVEREST, FAIR-HF, DELIVER, RESHAPE-HF2 and FAIR-HF2.