A Simple Nonparametric Least-Squares-Based Causal Inference for Heterogeneous Treatment Effects

Area of Science:

• Causal Inference
• Econometrics
• Biostatistics

Background:

• Estimating treatment effects in observational studies is complex due to unknown outcome and treatment assignment models.
• The potential outcomes framework is a standard approach for causal inference.
• Heterogeneous treatment effects (HTE) analysis is crucial for personalized medicine.

Purpose of the Study:

• To propose a simple nonparametric least-squares spline-based method for estimating HTE.
• To analyze the asymptotic properties of the proposed method using empirical process theory.
• To apply the method to assess anti-rheumatic treatment effects in children with juvenile idiopathic arthritis.

Main Methods:

• Nonparametric least-squares spline regression.
• Empirical process theory for asymptotic analysis.
• Simulation studies for performance evaluation.
• Application to electronic health records (EHR) data.

Main Results:

• The proposed method provides accurate estimation of heterogeneous treatment effects.
• Asymptotic properties of the estimator are theoretically established.
• Simulation studies confirm the method's operational characteristics.
• The method was successfully applied to real-world EHR data.

Conclusions:

• The developed nonparametric spline-based method is effective for estimating HTE from observational data.
• The method allows for robust causal inference in the presence of unobserved confounding.
• This approach has significant implications for clinical decision-making and personalized treatment strategies in pediatric rheumatology.