Estimation of indirect effect when the mediator is a censored variable

Area of Science:

• Biostatistics
• Statistical Genetics
• Epidemiology

Background:

• Mediation analysis commonly examines direct and indirect effects but often encounters censored data.
• Existing methods for mediation with censored data primarily address censored outcomes, neglecting censored mediators.
• Censored mediators present unique challenges in accurately estimating indirect effects.

Purpose of the Study:

• To propose and evaluate a novel strategy for mediation analysis incorporating censored mediators.
• To adapt indirect effect measures for mediation models with censored mediators at group and individual levels.
• To compare the performance of the proposed strategy against existing methods using simulation studies.

Main Methods:

• Utilized an accelerated failure time model combined with a multiple imputation approach.
• Developed an adapted measure for indirect effects in mediation models with censored mediators.
• Conducted extensive simulations to assess bias in path and indirect effect estimations with existing methods (naïve, complete-case, Tobit).

Main Results:

• The proposed strategy demonstrated accurate estimation of path coefficients (a, b, c') and indirect effects.
• Simulation results confirmed the proposed method's superiority over existing approaches in handling censored mediators.
• Application to SNP, age at menopause, and fasting glucose data revealed no significant indirect effect mediated by age at menopause.

Conclusions:

• The developed strategy provides a robust and accurate method for mediation analysis involving censored mediators.
• Existing approaches may introduce bias when dealing with censored mediators, highlighting the need for improved methods.
• The study's findings suggest age at menopause does not mediate the association between SNPs and fasting glucose levels.