On Mendelian randomization analysis of case-control study.

Area of Science:

• Epidemiology
• Statistical Genetics
• Biostatistics

Background:

• Mendelian randomization (MR) typically analyzes prospective cohort data to infer causal effects.
• Existing MR methods are limited when applied to case-control study designs.
• Case-control studies present unique challenges, such as ascertainment bias, for MR analysis.

Purpose of the Study:

• To develop a robust Mendelian randomization procedure for binary outcomes within a case-control framework.
• To address ascertainment bias inherent in case-control sampling using a quasi-empirical likelihood approach.
• To provide reliable methods for causal effect estimation and hypothesis testing in this specific design.

Main Methods:

• Developed a novel procedure based on two common MR working models.
• Employed a quasi-empirical likelihood framework to correct for case-control ascertainment bias.
• Derived multiple estimation and hypothesis testing approaches within the empirical likelihood framework.

Main Results:

• The proposed empirical likelihood estimate demonstrated reduced bias compared to existing methods.
• The Lagrange multiplier (LM) test exhibited the highest statistical power among evaluated approaches.
• Confidence intervals from the LM test provided the most accurate coverage rates in simulations.

Conclusions:

• The developed empirical likelihood method offers a less biased and more accurate approach for Mendelian randomization in case-control studies.
• The LM test is recommended for its superior power and accurate confidence interval coverage.
• The method was successfully illustrated using prostate cancer case-control data and vitamin D levels.