Robust double machine learning model with application to omics data.

Area of Science:

• Causal Inference
• Machine Learning
• Robust Statistics

Background:

• Growing interest in combining causal inference with machine learning.
• Double Machine Learning (DML) models are effective for high-dimensional data but sensitive to outliers.
• Need for robust methods when outcome distributions have outliers or heavy tails.

Purpose of the Study:

• Propose the Robust Double Machine Learning (RDML) model.
• Achieve robust estimation of causal effects in the presence of data outliers or heavy-tailed distributions.
• Improve causal inference accuracy in challenging data scenarios.

Main Methods:

• Employ median machine learning algorithms for robust predictions of treatment and outcome variables.
• Establish a median regression model for prediction residuals.
• Utilize these median-based approaches for robust causal effect estimation.

Main Results:

• RDML shows comparable performance to DML with normal distributions.
• RDML significantly outperforms DML with mixed normal and t-distributions (smaller RMSE).
• Applied RDML to Alzheimer's disease data to study CSF A$$42 impact on AD severity.

Conclusions:

• The RDML model provides robust causal effect estimation.
• RDML is effective even with outcome distributions affected by outliers or heavy tails.
• Demonstrates the utility of RDML in real-world applications like Alzheimer's disease research.