Rule ensemble method with adaptive group lasso for heterogeneous treatment effect estimation.

Area of Science:

• Computational biology
• Data science
• Medical informatics

Background:

• Precision medicine increasingly relies on real-world data to understand treatment effects.
• Estimating heterogeneous treatment effects (HTE) is challenging due to data complexity and inherent individual variability.
• Existing machine learning (ML) methods for HTE often lack interpretability due to their 'black-box' nature.

Purpose of the Study:

• To develop an interpretable machine learning method for estimating heterogeneous treatment effects (HTE).
• To address the limitations of current ML models that hinder direct interpretation of treatment effect drivers.
• To improve the accuracy and interpretability of HTE estimation in precision medicine.

Main Methods:

• Modification of the RuleFit algorithm to estimate HTE within the potential outcome framework.
• Development of a novel ML approach combining accuracy and interpretability for HTE analysis.
• Application of the proposed method to the ACTG 175 HIV clinical dataset.

Main Results:

• The proposed modified RuleFit method demonstrated high prediction accuracy compared to existing approaches.
• The method successfully generated an interpretable model, revealing relationships between patient characteristics and treatment effects.
• Ensemble of rules provided direct insights into individual feature impacts on treatment outcomes.

Conclusions:

• The developed ML method effectively estimates HTE with both high accuracy and interpretability.
• This approach facilitates a deeper understanding of treatment effect variations in real-world data.
• The findings support the application of interpretable ML in advancing precision medicine.