Validation of Surrogate Endpoints by Bayesian Equivalence Testing

Surrogate endpoints are often used in clinical trials to serve as a substitute for a hard to achieve clinically meaningful endpoint. The surrogate endpoint is expected to predict the effect of the treatment on the true endpoint. Prentice (1989, Stat Med) proposed a set of criteria for surrogate endpoint validation. The main criterion is to show the conditional independence of the treatment and the true endpoint in the presence of a surrogate endpoint, and thereby showing that the surrogate endpoint captures the full effect of the treatment. To prove this criterion, one has to show the failure to reject the null hypothesis (acceptance of null) of conditional independence between treatment and the true endpoint, given a surrogate. In such settings, one could use an equivalence testing approach.;In this research, an equivalence testing approach is studied for validation of surrogate endpoints using a Bayesian framework. Both normal endpoints and binary endpoints are evaluated by means of Bayes factors. The main idea of this research is to obtain an appropriate Bayes factor cutoff value or a rejection threshold, which can then be used to test for equivalence of true and surrogate endpoint. A calibrated Bayes approach is adopted to evaluate the Bayesian equivalence testing procedure. The validity of surrogate endpoints is assessed using different set of priors. The methodology is applied to a simple dataset. An equivalence testing criteria is also proposed for Relative Effect, an alternate measure of surrogacy (Buyse etal, 1998). In addition, exploratory investigation into non-parametric validation of surrogate endpoint is performed.