Surrogacy validation for time-to-event outcomes with illness-death frailty models

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Causal Inference

Background:

• Clinical trials often use intermediate outcomes due to difficulties measuring primary endpoints.
• Validating these surrogate outcomes is crucial for reliable treatment effect assessment.
• Time-to-event data presents unique challenges in surrogate outcome validation.

Purpose of the Study:

• To develop a causally valid framework for validating intermediate outcomes in time-to-event clinical trials.
• To assess the relationship between treatment effects on surrogate and primary outcomes using a causal paradigm.
• To propose and evaluate statistical models for this validation process.

Main Methods:

• Utilized counterfactual outcomes within a causal association paradigm.
• Proposed illness-death models to handle censored and semicompeting risk survival data.
• Incorporated estimable and counterfactual frailty terms into multistate models.
• Employed a Bayesian method with Markov chain Monte Carlo for estimation.
• Assessed model sensitivity to assumption violations.

Main Results:

• Characterized valid surrogate outcomes using a causal effect predictiveness plot.
• Demonstrated the utility of illness-death models for causally validating intermediate outcomes.
• Evaluated the estimation properties and sensitivity of the proposed Bayesian approach.
• Applied the methodology to a prostate cancer clinical trial dataset.

Conclusions:

• The proposed causal framework provides a robust method for validating intermediate outcomes in survival analysis.
• Illness-death models offer a suitable approach for handling complex survival data structures in surrogate validation.
• The Bayesian estimation method shows good properties for assessing causal surrogate relationships.