The validation of different surrogate endpoints in clinical trials with some newly proposed criteria.

ABSTRACTThe Validation of Different Surrogate Endpoints in Clinical Trials with SomeNewly Proposed Criteria.New treatments for cancer or other chronic diseases which are tested in clinicaltrials should be given approval as soon as they show effectiveness. If an earliermeasurable surrogate for survival can be used, the treatment under research can beapproved earlier if it?s effective. A new surrogate needs to be validated before it can beused instead of the true endpoint. Prentice (1989), proposed criteria for the validation ofsurrogate endpoints. Later, other criteria like Proportion explained (PE), Relative Effect(RE) and Adjusted Association (AA) were proposed. Buyse et.al.(1998), additionallyproposed two new criteria R-sqr(Trial) and R-sqr(Indiv) which link RE and AA to themeta-analytic approach.This Thesis aims to analyze the criterias proposed for validation of discrete,continuous and censored time variables from NSABP trial protocols (C-01, C-02, C-03and C-04) as surrogate endpoints. Additionally, Bootstrap confidence intervals for theproposed validation criteria?s were compared with those obtained from Fieller andDelta methods. In the Copula approach, likelihood and nonparametric methods wereapplied for the estimation of R-sqr(indiv) criteria.Data sets from NSABP trials were used for the validation of ?Relapse FreeSurvival? (RFS) as a surrogate for ?Overall Survival? (OS). In the one trial case forwhich PE, RE and AA were proposed, it was observed that these criteria are not veryuseful for the validation of surrogate endpoints, unless the treatment effect is highlysignificant.The validation of RFS as a surrogate for OS was tested for all protocols (C-01 toC-04). But RFS could be validated only for data set C-03 where R-sqr(Trial) was 0.761.By inspecting R-sqr(Trial) and R-sqr(Indiv) RFS was ranked as a medium surrogatefor OS.Using data from published clinical trials on colon cancer, Disease Free Survival(DFS) was ranked as a good to very good (R-sqr(Trial)=0.904) surrogate for OS, at leastat the trial level.As a conclusion, we propose a-the validation criteria?s R-sqr(Trial) and R-sqr(Indiv) in the meta-analytic approach is to be preferred to the other measures forvalidity of Surrogate and b- when determining the validity of variables as surrogate endpoint, data from clinical trial in which treatment effect is highly significant should beused.Key words: Surrogate Endpoint, Validation, Copula, Censored Variable, Cancer