Bayesian evaluation and adaptive trial design for surrogate time-to-event endpoints in clinical trials.
Access rightsWorldwide access.
Access changed 9/12/13.
Renfro, Lindsay A.
MetadataShow full item record
Surrogate endpoints are desirable in clinical trials when primary endpoints are costly to obtain, difficult to measure, or require lengthy follow-up to observe. Despite legitimate concerns, evaluation of potentially beneficial treatments in some settings remains impossible or implausible without the use of surrogates. Furthermore, strong evidence based on a collection of trials, rather than a relationship observed within a single trial, is required to validate a surrogate endpoint for future primary use. We present a Bayesian approach to evaluating surrogacy using patient data from multiple trials with time-to-event endpoints that accounts for estimation error of treatment effects and offers greater computational stability than existing methods. Once a surrogate endpoint has been deemed valid for use in a future trial, a healthy skepticism should remain regarding its ability to reflect the true treatment effect that would have been observed on the primary endpoint. Despite the surrogate's intended role, few (if any) efforts have been made to formalize existing knowledge and uncertainty in the design of such a trial. We propose a Bayesian adaptive design that uses the validated surrogate as the primary endpoint, while acknowledging that this endpoint is really a surrogate, and perhaps only a recently- validated one. At prospectively-defined checkpoints, we assess the performance of the surrogate and decide whether to continue its use or switch consideration to the primary endpoint. Furthermore, our design incorporates other favorable aspects of Bayesian adaptive trials, including the ability to stop a trial early for treatment efficacy, inferiority, or trial futility. Flowgraphs are useful for modeling diseases that are well-described by multi- state models, but for which Markov assumptions are inadequate and returns to previous states are possible. Furthermore, censoring and covariates may influence the distribution of waiting times between any two states, and to a differing degree for separate transitions within the same system. We discuss the construction and advantages of flowgraph models when used to describe cancer progression within two clinical trials, where our goal is improved modeling of treatment effects and prediction of patient outcomes for the purpose of more realistic surrogacy evaluation.