Hypothesis Tests of Direct and Indirect Effects Under Various Semicompeting Risks Models

Area of Science:

• Biostatistics
• Causal Inference
• Survival Analysis

Background:

• Semicompeting risks involve two time-to-event outcomes, with potential censoring of intermediate events by primary events.
• These risks can be modeled using mediation frameworks, allowing study of direct effects (DE) and indirect effects (IE).

Purpose of the Study:

• To propose unified causal testing procedures for evaluating direct and indirect effects in semicompeting risks.
• To establish testing rules and analyze the correspondence between effects and parameters across three classic models.

Main Methods:

• Formulating semicompeting risks as a mediation model.
• Applying causal inference approaches for testing direct and indirect effects.
• Utilizing U-statistic for Clayton copula and nonparametric maximum likelihood estimation for gamma frailty and multistate models.

Main Results:

• The Clayton copula model provides the highest statistical power when assumptions hold but is sensitive to misspecification.
• The gamma frailty model demonstrates robustness at the cost of efficiency.
• The multistate model offers a balance between efficiency and robustness.

Conclusions:

• The study provides a unified framework for analyzing direct and indirect effects in semicompeting risks.
• Application to a hepatitis study indicates that hepatitis B and C increase liver cancer risk via liver cirrhosis.
• The choice of model impacts statistical power, robustness, and efficiency in semicompeting risks analysis.