Instrumental variable analysis of multiplicative models with potentially invalid instruments.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Instrumental variable (IV) methods are used to estimate causal effects in the presence of unmeasured confounding.
• IV validity relies on assumptions, including that the instrument's effect on the outcome is solely mediated by the exposure.
• This assumption is often untestable in observational data.

Purpose of the Study:

• To develop a method for estimating causal effects using instrumental variables when the instrument may have direct effects on the outcome.
• To address unmeasured confounding in binary outcomes using multiplicative structural mean models.
• To extend the method for mediation analysis and other complex scenarios.

Main Methods:

• Proposed a novel method to estimate multiplicative structural mean models for binary outcomes.
• Adapted the asymptotically efficient G-estimation approach.
• Utilized generalized method of moments (GMM) for implementation, compatible with standard statistical software.

Main Results:

• Monte Carlo simulation studies demonstrated that the proposed method exhibits low bias.
• The simulations confirmed accurate coverage probabilities for the estimated causal effects.
• The method was successfully applied to a real-world case study on vitamin D and depression.

Conclusions:

• The developed method provides a robust approach to estimate causal effects under relaxed instrumental variable assumptions.
• It is applicable in various settings, including randomized trials and Mendelian randomization studies.
• The method offers a valuable tool for researchers dealing with potential unmeasured confounding and complex instrument-outcome relationships.