Highly robust causal semiparametric U-statistic with applications in biomedical studies

Area of Science:

• Biostatistics
• Causal Inference
• Semiparametric Statistics

Background:

• Causal inference is crucial in biomedicine and economics, but existing methods often rely on unverifiable assumptions.
• Doubly robust estimators (DREs) improve robustness, but are not applicable to complex outcome measures estimated via U-statistics.
• Robust modeling is essential for reliable causal inference, complementing sensitivity analysis.

Purpose of the Study:

• To propose a new class of highly robust U-statistic estimators (HREs) for causal inference.
• To extend robust estimation methods to handle complex outcome measures that are functionals of multiple distributions.
• To provide a more robust alternative to existing DREs in situations involving U-statistics.

Main Methods:

• Developed highly robust U-statistic estimators (HREs) using semiparametric specifications for propensity score and outcome models.
• Derived comprehensive asymptotic properties for the proposed HREs.
• Conducted extensive simulation studies to evaluate finite sample performance.

Main Results:

• HREs demonstrated significant advantages over parametric U-statistics and naive estimators in simulations.
• The proposed estimators exhibit enhanced robustness compared to existing DREs for U-statistic-based estimands.
• The method was successfully applied to analyze a clinical trial from the AIDS Clinical Trials Group.

Conclusions:

• The proposed highly robust U-statistic estimators (HREs) provide a valuable and robust tool for causal inference.
• HREs effectively address limitations of existing methods when dealing with complex outcome measures.
• This approach enhances the reliability of causal inference in large-scale data analysis in various scientific fields.