The Validation of a candidate surrogate of a Survival Endpoint in Leukemia

Introduction The selection of the primary “endpoint” is a very important step in the design of clinical trials. Typically, the goal of a clinical trial is to assess the effect of treatment on this endpoint. It often appears, however, that the most sensitive and relevant clinical endpoint, which will be called the “true” endpoint, might be difficult to assess and to overcome this problem, a seemingly attractive solution is to replace the true endpoint with another one, which is measured earlier, more conveniently or more frequently. Such “replacement” endpoint is termed “surrogate” endpoint and has the purpose of evaluating the effect of a specific treatment for a specific disease. Once a candidate surrogate is identified according to some specific properties, several formal methods are available for the validation of the surrogate endpoint, depending on the number of the trials performed. The first formal statistical approach dates back to 1989, when Prentice proposed a definition of surrogate endpoint and four criteria to validate it. The most important criterion among these is called “The Prentice’s Criterion”, which implies that “the full effect of treatment upon the true endpoint is captured by the surrogate” (Prentice 1989). More recently, a meta-analytic approach to the validation of a surrogate endpoint was proposed for a multi-trial context (Burzykowski et al. 2005). It consists in estimating associations at two different levels: the association between the surrogate and the clinical endpoint, called the “individual-level association”, and the association between the effect of treatment on the surrogate and on the clinical endpoint, called “trial-level association”. A good surrogate is one that has biological/clinical plausibility and is shown, statistically, to have strong individual-level and trial-level associations with the true endpoint. We face here the evaluation of Minimal Residual Disease (MRD) as a candidate surrogate endpoint in acute lymphoblastic leukaemia (ALL). In fact MRD, that quantifies the small numbers of leukemic cells circulating in the patient, has not yet been formally validated as a surrogate endpoint, whilst it is a well-established prognostic biomarker in ALL. The challenge has now evolved to the qualification of early MRD as an efficacy-response biomarker in the assessment of new drugs for the treatment of ALL. Objectives The main goal of the present work is to assess whether MRD, evaluated at the end of the induction treatment, can be considered a surrogate for Event Free Survival (EFS) and Overall Survival (OS) in childhood B-lineage ALL patients treated (after randomization) with Dexamethasone or Prednisone. Methods The meta-analytic approach was used on the data of the 1066 patients enrolled in the AIEOP-ALL 2000 protocol, who were B-lineage, were randomly assigned to receive Dexamethasone or Prednisone as front line treatment and were assessed for MRD at the end of induction phase (day +33 from start treatment). In this application on a single trial, treating centers were considered as “trial-units” (Burzykowski et al. 2005). Since it was mandatory that at least one event was present in each treatment group within hospital, centers were grouped according to size and geographical area and a total of 17 units were identified for the analysis (patients in the trial-units ranged from 43 to 108). The standard three categories classification of MRD (0, 0-1^10-3 and ≥1^10-3) was assessed as surrogate for the two time to event endpoints and contrasted with a binary MRD stratification (0 vs ≥1^10-3). In each trial-unit we estimated the relative treatment effect on both EFS and OS (βi, i=1,..,17), through proportional hazard model adjusted for age and WBC, and on the MRD endpoints (αi, i=1,..,17), with a proportional odds model or a logistic model for the 3 or 2 ordered categories, respectively. The quality of the surrogate at the individual trial level was assessed on the basis of the bivariate Plackett copula model (1965), with the parameter θ representing the (constant) global odds ratio. The quality of the surrogate at the trial level was assessed on the basis of the coefficient of determination R^2 trial that pertains the distribution of βi conditional on the set of trial specific parameters αi.We assumed here that the vector of parameters (αi,βi ) follows a normal distribution with mean (α,β) and with an unstructured covariance matrix. Results We found that MRD, classified both in three and two categories, is a poor surrogate for the treatment effect on both EFS and OS at the trial level (with values of R^2 trial<0.20), while, at the individual level, it shows a considerable prognostic association with EFS and OS, after adjusting for treatment (with values of θ ranging from 0.20 to 0.28 and statistically significant). Conclusion Using the meta-analytic approach on the data of the AIEOP-ALL 2000 protocol, MRD cannot be considered a valid surrogate for the prediction, at the trial level, of the treatment effect on EFS and OS. However, it seems to perform better in the prediction of the two time to event clinical endpoints at the individual level. Noteworthy, these are only preliminary results based on partial data available from one of the four clinical trials performed in different European countries and in the US, which will contribute to a final evaluation based on a large number of patients. Moreover, there are a lot of issues that will deserve more work in the future and that will be considered in the final analysis. The plan is to explore the impact on the results of a more refined model formulation, both in the direction of adding other important covariates and in a better specification (i.e. accounting for non-proportionalities of the hazards in the Cox model) and of different copula bivariate models.