ABSTRACT
Validation of surrogate endpoint has been an extensive area of research. Currently more and more intermediate endpoints are used and accepted by regulatory agencies to approve new treatments. However how to validate a surrogate endpoint is still a matter of intense debates. Interestingly, both the concepts as requested evidence have evolved, which in turn has stimulated the development of novel statistical methods. In particular, there have been important developments based on modeling of joint distributions with or without the prism of causal inference. Using several examples and R packages specifically developed to assess the value of surrogate endpoints, we review some of the most recent statistical methods, introduce the main formulas and properties, and describe their implementation. We discuss their limits and pitfalls.
