Links between causal effects and causal association for surrogacy evaluation in a gaussian setting

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Causal Inference

Background:

• Surrogate markers (S) are used to predict true outcomes (T) in clinical trials.
• Two main paradigms exist for evaluating surrogate markers: causal effects and causal association.
• Both paradigms require specific assumptions for estimation and validation.

Purpose of the Study:

• To relate quantities used for surrogacy evaluation under both causal paradigms.
• To examine the implications of assumptions made within one framework on the other.
• To compare conditions for identifiability and correspondence of surrogacy parameters.

Main Methods:

• Considered Gaussian surrogate markers (S) and true outcomes (T) generated from structural models with unobserved confounders.
• Related key quantities for surrogacy evaluation within both causal effects and causal association frameworks.
• Reviewed common assumptions and analyzed their impact across frameworks.

Main Results:

• Demonstrated similarity but not exact correspondence between evaluation quantities in the two paradigms.
• Showed that assumptions in one framework can impose strong assumptions in the alternative framework.
• Identified distinct conditions for the identifiability of surrogacy parameters versus parameter correspondence.

Conclusions:

• The causal effects and causal association paradigms for surrogate marker evaluation have related but distinct quantities and assumptions.
• Careful consideration of assumptions is crucial when applying either framework.
• Identifiability and correspondence conditions differ, impacting the validation of surrogate markers.