Using multiple genetic variants as instrumental variables for modifiable risk factors.

Area of Science:

• Genetic Epidemiology
• Biostatistics
• Causal Inference

Background:

• Mendelian randomization (MR) employs genetic variants as instrumental variables (IVs) to infer causal relationships between modifiable risk factors and disease outcomes.
• Traditional MR studies often require large sample sizes due to the limited variability explained by individual genetic variants.
• Genome-wide association studies (GWAS) increasingly identify numerous genetic variants associated with disease outcomes, offering potential for multi-instrument MR.

Purpose of the Study:

• To discuss and evaluate the application of multiple genetic variants as instrumental variables in Mendelian randomization analyses.
• To explore how multi-instrument MR can improve the precision of causal effect estimates and facilitate the examination of instrumental variable assumptions.
• To investigate the impact of weak instruments and missing data on the reliability of MR estimates.

Main Methods:

• The study discusses theoretical aspects of using multiple genetic variants in linear Mendelian randomization models.
• It describes methods for identifying violations of instrumental variable assumptions within multi-instrument analyses.
• An empirical example uses four adiposity-associated genetic variants in 5509 children (ALSPAC cohort) to estimate the causal effect of fat mass on bone density, supplemented by simulation studies.

Main Results:

• Utilizing multiple instruments generally increases the precision of instrumental variable estimates when each instrument independently explains risk factor variability.
• Inclusion of weak instruments can exacerbate finite sample bias, potentially compromising the accuracy of causal effect estimates.
• Missing data across multiple genetic variants can reduce the effective sample size, impacting precision compared to single-instrument analyses.

Conclusions:

• Multi-instrument Mendelian randomization is a powerful approach for estimating causal effects, offering enhanced precision and assumption checking under appropriate conditions.
• Weighted allele scores demonstrated similar properties to multi-instrument estimates in simulations with additive genetic effects.
• Further research into multiple imputation techniques is needed to effectively address missing data challenges in instrumental variable estimation.