Causal mediation analysis with multiple causally non-ordered and ordered mediators based on summarized genetic data.

Area of Science:

• Epidemiology
• Statistical Genetics
• Causal Inference

Background:

• Causal mediation analysis is crucial for understanding exposure-outcome mechanisms.
• Unobserved confounding and multiple mediators complicate path-specific effect estimation.
• Existing methods struggle when the sequential ignorability assumption is violated.

Purpose of the Study:

• To propose and validate PSE-MR, a method for estimating path-specific effects with multiple mediators under violated sequential ignorability.
• To address challenges posed by unobserved confounders in mediation analysis.
• To apply PSE-MR to the education-osteoarthritis risk pathway via BMI and smoking.

Main Methods:

• Developed the PSE-MR method utilizing summarized genetic data.
• Established a rank condition: number of instrumental variables must exceed the number of mediators.
• Conducted simulations to assess the method's performance and derived power calculations.

Main Results:

• PSE-MR effectively identifies and estimates path-specific effects with multiple, potentially non-ordered mediators.
• Simulation results show unbiased causal estimates, good coverage, and controlled Type I error rates.
• Determined the minimum instrumental variables needed for 80% power based on mediator count.

Conclusions:

• PSE-MR offers a robust approach to causal mediation analysis with multiple mediators and unobserved confounding.
• The method is applicable even when traditional assumptions like sequential ignorability do not hold.
• Provides practical guidance and demonstrates utility in a real-world example concerning education, BMI, smoking, and osteoarthritis.