Estimating treatment effects with machine learning.

Area of Science:

• Biostatistics
• Epidemiology
• Health Informatics

Background:

• Estimating average treatment effects is crucial in observational studies.
• Traditional regression methods can suffer from bias when dealing with complex data structures.
• Machine learning (ML) offers advanced tools for causal inference.

Purpose of the Study:

• To evaluate the performance of various ML algorithms for estimating average treatment effects.
• To compare ML-based estimators against traditional regression techniques.
• To provide practical guidance for implementing ML in causal inference.

Main Methods:

• Assessed ML estimators: Targeted Maximum Likelihood Estimation, Bayesian Additive Regression Trees, Causal Random Forests, Double Machine Learning, and Bayesian Causal Forests.
• Applied methods to simulated data and observational data from hospitalized adults.
• Evaluated performance using Monte Carlo studies.

Main Results:

• ML-based estimators demonstrated substantial bias reduction (69%-98%) compared to traditional regression.
• Bayesian Causal Forests and Double Machine Learning were identified as top-performing methods.
• All ML methods showed sensitivity to high-dimensional covariate sets (>150).

Conclusions:

• ML methods are promising for estimating treatment effects, accommodating numerous covariates and automating nonlinearity detection.
• These advanced techniques enhance causal inference by handling complex variable interactions.
• The study offers resources for researchers to adopt ML tools in empirical analyses.