Robust causal inference for point exposures with missing confounders

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Observational studies frequently encounter missing data, complicating causal inference.
• Existing causal inference methods often struggle to address both confounding and missingness simultaneously.
• There is a need for robust statistical methods to handle these challenges in real-world data.

Purpose of the Study:

• To develop an efficient and robust estimator for the causal average treatment effect in cohort studies.
• To address the intersection of confounding and missing data in causal inference.
• To provide a reliable method for analyzing observational data with missing confounders.

Main Methods:

• Proposed a novel likelihood factorization for efficient estimation.
• Enabled flexible modeling of nuisance functions using machine learning.
• Developed an estimator for the causal average treatment effect with missing confounders at random.

Main Results:

• The proposed estimator demonstrates robustness in finite samples through simulations.
• The method facilitates flexible modeling of complex relationships.
• Achieved nominal convergence rates for the final estimators.

Conclusions:

• The novel approach provides an efficient and robust method for causal inference with missing data.
• This estimator can serve as a benchmark for evaluating other methods.
• Applicable to cohort studies and electronic health record data analysis.