New proposal to address mediation analysis interrogations by using genetic variants as instrumental variables

Area of Science:

• Epidemiology
• Genetic Epidemiology
• Biostatistics

Background:

• Causal mediation analysis (CMA) is increasingly used in epidemiology but relies on strong assumptions regarding confounding bias.
• Existing methods may struggle with confounding bias, limiting the reliability of mediation effect estimations.
• There is a need for robust analytical approaches that can account for confounding in mediation analyses.

Purpose of the Study:

• To propose and validate a novel extension of causal mediation analysis (CMA) by integrating Mendelian randomization (MR).
• To introduce Mendelian randomization in causal mediation analysis (MRinCMA) to address limitations of confounding bias.
• To evaluate the causal effects of obesity and diabetes on pancreatic cancer using the MRinCMA approach.

Main Methods:

• Developed Mendelian randomization in causal mediation analysis (MRinCMA), combining CMA with MR principles.
• Applied MRinCMA to investigate the mediation pathways of obesity and diabetes on pancreatic cancer.
• Validated MRinCMA performance against structural equation models using simulated datasets with varying variable types (continuous and non-continuous).

Main Results:

• MRinCMA demonstrated comparable performance to structural equation models for continuous variables, yielding unbiased estimates.
• MRinCMA exhibited lower bias than structural equation models when analyzing non-continuous variables.
• The application of MRinCMA did not reveal evidence supporting a causal relationship between obesity or diabetes and pancreatic cancer.

Conclusions:

• MRinCMA offers a robust methodology for causal mediation analysis, effectively mitigating confounding bias across different study conditions.
• The developed approach allows researchers to rigorously test mediation hypotheses while appropriately handling confounding.
• The study's findings suggest that obesity and diabetes may not be causal factors for pancreatic cancer, based on the MRinCMA analysis.