A Bayesian approach for two-stage multivariate Mendelian randomization with mixed outcomes.

Area of Science:

• Biostatistics
• Genetic Epidemiology
• Clinical Research Methodology

Background:

• Observational studies on baseline factors and clinical outcomes are often confounded.
• Understanding causal relationships is crucial for personalized medicine and disease management.
• Existing Mendelian randomization (MR) methods typically analyze one outcome at a time, ignoring correlations.

Purpose of the Study:

• To develop a method for jointly analyzing multiple, mixed-type clinical outcomes using genetic instrumental variables.
• To address confounding effects in the estimation of causal relationships between exposures and multiple outcomes.
• To improve upon existing univariate MR approaches by incorporating multivariate modeling.

Main Methods:

• Proposed a Bayesian-based two-stage multivariate Mendelian randomization (MR) method (BMRMO).
• The method is designed to handle mixed outcomes (e.g., toxicities, quality of life) across multiple datasets.
• Employed simulation studies and real-world clinical data (CO.17 and CO.20 studies) for validation.

Main Results:

• The BMRMO method demonstrated superior performance compared to the standard univariate two-stage MR approach.
• The multivariate approach effectively accounts for the correlation structure among multiple outcomes.
• Validated the method's utility in analyzing complex clinical data.

Conclusions:

• The proposed BMRMO method offers a robust framework for causal inference with multiple, mixed clinical outcomes.
• This approach can lead to a better understanding of disease mechanisms and facilitate personalized medicine.
• Highlights the advantage of joint analysis over univariate methods in genetic epidemiology.