Criteria for the validation of surrogate endpoints in randomized experiments

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Epidemiology

Background:

• Existing methods for validating surrogate endpoints, such as those by Prentice and Freedman, have limitations.
• These limitations include stringent criteria that are rarely met in practice, particularly when treatment effects on the true endpoint are not massive.
• The need for more practical and robust validation criteria for surrogate endpoints is evident.

Purpose of the Study:

• To extend existing proposals for surrogate endpoint validation.
• To introduce and evaluate new metrics, relative effect (RE) and adjusted association (gamma Z), for surrogate endpoint validation.
• To assess the practical requirements, such as sample size, for reliable surrogate endpoint validation.

Main Methods:

• Extension of Prentice's and Freedman's proposals for binary and normally distributed endpoints.
• Introduction of RE (relative effect of treatment on true vs. surrogate endpoint) and gamma Z (association between surrogate and true endpoint adjusted for treatment).
• Application of Fieller's theorem for estimating proportion explained (PE) and RE, along with confidence intervals.
• Utilized logistic regression and global odds ratio models for binary endpoints; linear models for continuous endpoints.

Main Results:

• Proposed criteria for perfect surrogate endpoints at individual (high gamma Z) and population (RE=1) levels.
• Demonstrated that validation requires large sample sizes for practical utility.
• Fieller's theorem is applicable for estimating key validation metrics and their confidence intervals.

Conclusions:

• The proposed metrics RE and gamma Z provide a more comprehensive framework for surrogate endpoint validation.
• A perfect surrogate endpoint exhibits a deterministic relationship with the true endpoint, adjusted for treatment.
• Robust validation of surrogate endpoints necessitates substantial observational data.