Mediation analysis with case-control sampling: Identification and estimation in the presence of a binary mediator

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Stratified case-control (SCC) designs are common in epidemiological research.
• Mediation analysis with binary outcomes and mediators in SCC data presents unique statistical challenges.
• Existing methods may not fully account for the sampling design's impact on mediation parameter estimation.

Purpose of the Study:

• To derive and evaluate methods for estimating causal mediation effects in the presence of sampling design distortion in SCC data.
• To compare the performance of proposed maximum likelihood (ML) and M-estimators against existing weighting-based methods.
• To provide a general strategy applicable to both parametric and nonparametric mediation analysis.

Main Methods:

• Derivation of the sampling design's distortion on logistic model parameters for secondary variables in SCC data.
• Development of maximum likelihood (ML) and M-estimators for the joint outcome-mediator parameter vector.
• Simulation studies to assess the accuracy of natural effect estimation.
• Reanalysis of a German case-control dataset as an illustrative example.

Main Results:

• The proposed methods provide accurate estimation of causal mediation quantities, even with complex sampling designs.
• Simulation results demonstrate the superiority of ML and M-estimation over traditional weighting methods in specific scenarios.
• The reanalysis identified a potential mediation pathway for reduced immunocompetency on listeriosis onset.

Conclusions:

• The developed statistical framework effectively corrects for sampling design biases in SCC mediation analysis.
• ML and M-estimation offer robust alternatives for estimating natural effects in binary outcome/mediator settings.
• This approach enhances the reliability of causal inference from stratified case-control studies.