Valid instrumental variable selection method using negative control outcomes and constructing efficient estimator.

Area of Science:

• Epidemiology
• Statistical Genetics
• Biostatistics

Background:

• Instrumental variable (IV) methods are crucial for causal inference in observational studies with unmeasured confounding.
• Mendelian randomization often uses allele scores but risks biased causal effects from invalid instrumental variables (IVs) associated with unobserved factors.

Purpose of the Study:

• To develop a novel strategy for selecting valid IVs and excluding invalid ones in observational studies, particularly Mendelian randomization.
• To improve the accuracy of causal effect estimation by addressing the challenge of unknown invalid IVs.

Main Methods:

• Developed a new strategy employing negative control outcomes (NCOs) as auxiliary variables to identify valid IVs.
• Implemented a novel two-step estimation procedure, proving the semiparametric efficiency of the proposed estimator.
• Validated the method through simulations and application to the UK Biobank dataset.

Main Results:

• The proposed method successfully selects valid IVs and excludes invalid ones without prior knowledge of instrument validity.
• Simulations demonstrated superior performance compared to existing methods.
• Application to UK Biobank data confirmed the utility of NCOs for valid IV selection.

Conclusions:

• The use of NCOs as auxiliary variables provides a robust approach to enhance the validity of IV selection in causal inference.
• This method offers an alternative to previous IV selection strategies, enabling more reliable estimation of causal effects in observational data.