Sharp bounds on causal effects using a surrogate endpoint.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Evaluating treatment efficacy often relies on surrogate endpoints when true outcomes are difficult to measure.
• Unmeasured confounders between surrogate and true endpoints complicate causal effect estimation.
• Existing methods struggle with identifiability issues in the presence of unmeasured confounding.

Purpose of the Study:

• To derive methods for estimating the causal effect of a treatment (X) on a true endpoint (Y) using a surrogate endpoint (S).
• To address the challenge of unmeasured confounders affecting the relationship between S and Y.
• To provide practical tools for researchers and clinicians to assess treatment effects.

Main Methods:

• Derivation of closed-form formulas for sharp bounds on the causal effect of X on Y.
• Utilizing the causal effect of X on S and the joint probability of S and Y.
• Investigating conditions where observing Y is not necessary for testing null causal effects under monotonicity.

Main Results:

• Sharp bounds for the causal effect of X on Y were derived under various assumptions.
• Demonstrated that observing the true endpoint Y may not always be required for hypothesis testing.
• The derived bounds offer a computationally efficient approach for causal effect assessment.

Conclusions:

• The proposed bounds facilitate the evaluation of treatment effects on true endpoints using surrogate endpoints, even with unmeasured confounding.
• These methods enhance causal inference in clinical research by providing practical estimation tools.
• The findings reduce the reliance on observing true endpoints in certain causal effect testing scenarios.