Selecting the primary endpoint in a randomized clinical trial: The ARE method

Abstract

The decision on the primary endpoint in a randomized clinical trial is of paramount importance and the combination of several endpoints might be a reasonable choice. Gómez and Lagakos (2013) have developed a method that quantifies how much more efficient it could be to use a composite instead of an individual relevant endpoint. From the information provided by the frequencies of observing the component endpoints in the control group and by the relative treatment effects on each individual endpoint, the Asymptotic Relative Efficiency (ARE) can be computed. This paper presents the applicability of the ARE method as a practical and objective tool to evaluate which components, among the plausible ones, are more efficient in the construction of the primary endpoint. The method is illustrated with two real cardiovascular clinical trials and is extended to allow for different dependence structures between the times to the individual endpoints. The influence of this choice on the recommendation on whether or not to use the composite endpoint as the primary endpoint for the investigation is studied. We conclude that the recommendation between using the composite or the relevant endpoint only depends on the frequencies of the endpoints and the relative effects of the treatment.

Selection of the primary end point in an observational cohort study

“…The dependence between the times to E 1 and to E 2 is also needed. However, it has been proven that different dependence structures lead to similar results12 and different degrees of association, given by Spearman's r, yield, in many situations, to similar recommendations on the choice of the end point. A free web platform CompARE (https://cinna.upc.edu/compare) has been developed to choose between E* and E 1 , based on the ARE value, in the framework of an RCT.…”

Composite endpoints in clinical trials : computational tools, practical guidelines and methodological extensions

“…A ), we consider Frank's Archimedean survival copula, again as Gómez and Lagakos did. Other choices of copulas would be possible, although main conclusions and recommendations will not differ (Plana-Ripoll and Gómez, 2015). 51 Frank's copula depends on the association parameter θ ( j ) between…”

“…We assume equal association parameter θ = θ (0) = θ (1) for both groups 0 and 1, which is biunivocally related to Spearman's rank correlation ρ (−1 < ρ < 1) (see Appendix for more details). Other copulas can also be used (Plana-Ripoll O. and Gómez G., 2015). 51 • Marginal laws: We assume Weibull distributions for T ( j ) 1 and T ( j ) 2 ( j = 0, 1) with scale and shape parameters b…”

“…Other copulas can also be used (Plana-Ripoll O. and Gómez G., 2015). 51 • Marginal laws: We assume Weibull distributions for T ( j ) 1 and T ( j ) 2 ( j = 0, 1) with scale and shape parameters b…”

“…We have also implemented an extension of the methodology to allow to other copulas binding the relevant and the additional endpoint. The methodology was extended by Plana-Ripoll O. and Gómez G. 51 That is, the joint distribution of (T R , T A ) can be derived in C omp ARE by means of the following copulas (see Figure 6.2, bottom):…”