Objective: 
Marketing Authorization should be granted based on clinical criteria. Surrogate criteria whose validity is often questioned is still used to speed up clinical development. We assessed the implementation and results of a grading approach to test substitution criteria validity.
Methods: 
The validity of Primary Criteria considered as Surrogate Endpoints in 21 clinical trials was checked using 3 methods. These criteria were used in various therapeutics areas. Finally, their level were compared with the Biomarker-Surrogacy Evaluation Schema (BSES3).
Results: 
Surrogates must fulfil several criteria, called Prentice criteria: correlation, capture and prediction. To test them, several methods are available: meta-regression of clinical outcome alteration as a function of surrogate outcome alteration due to treatment on summarized data. It allows to calculate R-squared value (coefficient of determination), Surrogacy Threshold Effect (STE) and Surrogacy Threshold Effect Proportion (STEP). STE is defined as the minimum treatment effect on the surrogate necessary to predict a non-zero effect on the true endpoint. STEP is the absolute value of the STE defined as 1- (STE/(x-axis values)). The other method used is regression on individual data which models clinical criterion alteration as a function of treatment exposure and of surrogate criterion alteration. It allows to calculate the proportion of treatment effect (PTE). This PTE value should tend to 1 to validate a surrogacy criterion. Our literature search found articles which we organised according to statistical methods used to test the surrogate criteria validity. We then analysed the conclusions of these studies separately depending on whether or not surrogate criteria tested were validated. Among 15 criteria, 11 (73%) were validated in accordance with statistical tests employed BUT none of these articles tested the three Prentice criteria together. Criteria tested using the STE/STEP approach or both STE and meta-regression (R²) were always validated. The Biomarker-Surrogacy Evaluation Schema (BSES) was applied to score the surrogate criteria assessed. Only 7 criteria (47%) were possible to classify according to this schema, due to lack of information, since STEP and R² are both needed to define this score.
Conclusion: 
Surrogate criteria must satisfy the three Prentice criteria (capture, correlation and prediction) which is the reference method recommended by the French Haute Autorité de Santé. Surrogate criteria are usually considered validated, even if not all the three criteria be verified. Among all therapeutic areas we explored, oncology studies are the most frequent and use meta-regression (STE, R²) to evaluate criteria and thus make it possible to score them with the BSES. Indeed, two variables can be correlated without having any causality link.  Furthermore, the statistical methods described present limits. When meta-regression is used, attention must be paid to linearity. A good correlation can be evaluated but prediction will be weak due to the lack of linearity. Moreover, meta-regression is applied “inter-study” and may present ecological biases. STE approach can be interrogated: its validation depends on subjectivity and its only calculation may not be sufficient to validate a surrogate criterion. Regarding PTE, there is no threshold determined to validate a criterion according to its value. In 2019 a new approach will be presented by Stuart G. Baker. It would be interesting to verify if this new approach would validate differently surrogate criteria and surrogacy.

Surrogacy : myth or reality ? A review of its testing methods and their applications

DIA Europe 2019

A global, member-driven non-profit organization mobilizing life sciences and healthcare. DIA fosters innovation to improve health and well-being worldwide. Professionals from 80 countries continue to engage with DIA through our unparalleled membership network, educational offerings, and professional development opportunities.