Variability of Definitions for Survival Endpoints and Surrogate Properties for OS in Sarcoma Trials: a Meta-analysis

Background: In randomized phase III cancer clinical trials, the validated and most objectively defined evaluation criterion is overall survival (OS). Therapeutic progress, which in certain contexts has significantly reduced overall mortality, the development of new types of cytostatic treatments (as opposed to cytotoxic treatments), the current context of strategic trials and the multiplication of lines of treatment have resulted in the necessity of creating new evaluation criteria measuring treatment efficacy sooner and more precisely: for example, progression-free survival in second line treatment, duration of local control, and time until treatment failure. These types of surrogate endpoints are commonly used in phase II trials but are increasingly being used to replace overall survival in phase III trials. Their development is strongly influenced by the necessity of reducing clinical trial duration, cost and number of patients.

However while these survival endpoints are frequently used, they are often poorly defined and when they are, the definition can vary between trials. The lack of standardized definitions constitutes a clear limitation to their use as primary endpoints. Furthermore, this variability of definitions can have an important impact on trials' results by affecting power and estimation.

The ongoing DATECAN-1 project is aimed at providing guidelines for the definitions of survival endpoints in cancer trials (2009-2011 grant from the French League Against Cancer). Standardized recommendations will be available for survival endpoints commonly used for various cancer sites including pancreas, sarcomas and GIST (gastro-intestinal stromal tumors), breast, stomach.

Objectives: Following these guidelines, one can wonder how sensitive are these definitions? Or similarly, how do survival endpoints' definitions impact the conclusions of clinical trials? The objective of the DATECAN-2 study is to assess the impact of survival endpoints' definitions, as defined by the consensus guidelines, on trials' results and conclusions. The second objective is to study the surrogate properties for OS of these survival endpoints.

Methods: The evaluation of the impact of the variability of the definitions of survival endpoints on the results of clinical trials will be evaluated using individual data from datasets collected in the context of published (academic) clinical trials, as well as simulated datasets.

After approval by the sponsors, data will be analyzed using various endpoints' definitions including (i) the definition provided in the publication and (ii) the definition provided by the guidelines. We will identify which survival endpoint, as defined by the guidelines, was reported in the publication. Next, realistic data sets will be simulated that mimic data that could be observed in randomized cancer trials. We will generate data sets with varying proportions of events (number of deaths, progression, etc) depending on the survival endpoints of interest. Survival endpoints will be compared across treatment arms using the definition provided by the guidelines, and based on various scenarios (different proportions of events, length of follow-up, etc).

Using data from published data sets, we will evaluate survival endpoints in terms of surrogate candidates for OS. A hierarchy of the survival endpoints will be proposed according to their surrogate properties based on two criteria: the Fleming classification and the R2 value for validated surrogates. Depending on the number of clinical trials, single-trial or multiple-trials method will be employed. Single-trial methodology relies on Prentice criteria and Freedman's proportion of treatment effect (PTE) explained by the surrogate. In case of multiple trials, and when meta-analysis of clinical trials is feasible, surrogacy of candidate endpoints for OS will also be explored using weighted linear regression, which jointly estimates the level of association between endpoints and the trial-level association (R²) between treatment effects on the candidate surrogate and the final endpoint. Based on these results (R² and PTE), survival endpoints will be ranked according to their surrogate capabilities for OS.

Expected results: Analysis of the sensitivity of clinical trials' results to the survival endpoints' definitions and surrogacy properties are key features when designing and conducting clinical trials. Based on our results, we will be able to anticipate the expected impact of the definitions on effect size, sample size and power. We will be able to estimate these parameters more precisely and as such provide more efficient estimations. Similarly our assessment of the surrogate properties of the survival endpoints should help us for the selection of the best surrogate marker for OS, and thus limit biases. Overall, by producing less biased results and more efficient designs, our project should have an important role in the design and conduct of future randomized trials.