Model-based hypothesis tests for the causal mediation of semi-competing risks.

Area of Science:

• Biostatistics
• Epidemiology
• Medical Research

Background:

• Semi-competing risks are common in medical research, where an intermediate event can be censored by a primary event.
• Causal mediation analysis is crucial for understanding exposure effects on outcomes through intermediate pathways.

Purpose of the Study:

• To propose a model-based testing procedure for examining the indirect causal effect of an exposure on a primary event through an intermediate event.
• To develop and evaluate statistical tests for assessing causal mediation in semi-competing risks scenarios.

Main Methods:

• Defined causal mediation effect using a counterfactual outcome framework and counting processes.
• Proposed two statistical tests: an intersection-union test (IUT) and a weighted log-rank test (WLR).
• Developed test statistics from semi-parametric estimators using Cox proportional hazards and logistic regression models.

Main Results:

• The intersection-union test (IUT) was found to be a size [Formula: see text] test and statistically more powerful than the weighted log-rank test (WLR).
• Numerical simulations confirmed that both IUT and WLR properly adjust for confounding covariates, with well-protected Type I error rates.
• The IUT demonstrated superior power compared to the WLR in simulations.

Conclusions:

• The proposed methods, particularly the IUT, offer a robust approach to analyzing causal mediation in semi-competing risks.
• Demonstrated significant effects of hepatitis B or C on liver cancer risk mediated by liver cirrhosis.
• The methodology is applicable to surrogate endpoint analyses in clinical trials.