Estimate an Exposure Response Function with Negative Controls: A Bayesian Nonparametric Approach.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Unmeasured confounding bias is a major threat to observational study validity.
• Existing methods for bias adjustment often do not utilize auxiliary data.
• Negative controls offer a promising avenue for reducing unmeasured confounding bias.

Purpose of the Study:

• To develop a Bayesian nonparametric method for estimating causal exposure-response functions (CERFs).
• To incorporate information from negative controls to adjust for unmeasured confounding in continuous exposures.
• To provide a computationally efficient and accurate method for causal inference.

Main Methods:

• A Bayesian nonparametric approach modeling CERFs as a mixture of linear models.
• Utilizing negative controls from auxiliary data to adjust for unmeasured confounders.
• Simulation studies to assess method performance under various confounding scenarios.

Main Results:

• The proposed method successfully recovers the true CERF shape in the presence of unmeasured confounding.
• The method demonstrates practical utility in a real-world application.
• The approach maintains computational efficiency while capturing potential nonlinearities.

Conclusions:

• The developed Bayesian method effectively adjusts for unmeasured confounding using negative controls.
• This approach enhances the validity of observational studies, particularly in environmental epidemiology.
• The open-source implementation promotes reproducibility and wider adoption of the methodology.