Mendelian randomization analysis of case-control data using structural mean models.

Area of Science:

• Epidemiology
• Biostatistics
• Genetic Epidemiology

Background:

• Instrumental variable (IV) methods are crucial for estimating causal effects of exposures on disease risk.
• Mendelian randomization studies often use genetic information as IVs and are typically analyzed with case-control data.
• Existing IV analyses on case-control data are susceptible to ascertainment bias, potentially invalidating results.

Purpose of the Study:

• To explain the failure of standard IV analyses in the presence of ascertainment bias.
• To propose adjusted IV estimators for improved causal effect estimation.
• To assess the causal relationship between C-reactive protein (CRP) and coronary artery disease (CAD) risk.

Main Methods:

• Development of consistent estimators for causal relative risk and odds ratio under specific prior knowledge (disease prevalence or IV distribution).
• Derivation of approximate estimators assuming a rare disease.
• Application of methods to matched case-control data from the EPIC study.

Main Results:

• Proposed estimators provide consistent estimates of causal effects when prior information is available.
• Approximate estimators are viable under the rare disease assumption.
• The study illustrates the application of these adjusted methods to real-world data.

Conclusions:

• Adjusted instrumental variable methods can overcome ascertainment bias in case-control studies.
• Accurate estimation of causal effects is achievable with appropriate prior knowledge or assumptions.
• The findings offer a more reliable approach to investigating exposure-disease relationships, such as CRP and CAD.