Surrogacy assessment using principal stratification when surrogate and outcome measures are multivariate normal

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Causal Inference

Background:

• Surrogate outcome variables (S) can predict treatment effects on true outcomes (T) earlier.
• Validating surrogates is crucial for efficient clinical trial design and interpretation.
• Existing methods may lack causal interpretability or robust estimation strategies.

Purpose of the Study:

• To develop a causally interpretable Bayesian framework for surrogate validation.
• To propose novel surrogate validation measures based on potential outcomes.
• To compare proposed measures with existing criteria, such as Prentice's criteria.

Main Methods:

• Utilizing the principal surrogacy framework for causal interpretation.
• Developing a Bayesian estimation strategy assuming a multivariate normal distribution for potential outcomes.
• Proposing validation measures derived from the joint conditional distribution of potential outcomes.
• Employing informative prior distributions for weakly identified parameters.

Main Results:

• The proposed Bayesian approach offers a causally valid method for surrogate validation.
• New validation measures are derived and explored in relation to established criteria.
• The methodology is demonstrated using real-world data from macular degeneration and ovarian cancer studies.

Conclusions:

• The developed Bayesian framework provides a robust and causally interpretable method for surrogate endpoint validation.
• This approach can improve the assessment of treatment effects in clinical trials.
• The study highlights the utility of principal stratification in advancing surrogate endpoint methodology.