Guidance on individualized treatment rule estimation in high dimensions

Area of Science:

• Biostatistics
• Clinical Trials
• Precision Medicine

Background:

• Individualized treatment rules (ITRs) optimize patient outcomes by tailoring treatments based on pre-treatment covariates.
• Existing ITR estimation methods perform well in traditional settings but lack characterization in high-dimensional covariate scenarios common in modern clinical research.

Purpose of the Study:

• To comprehensively compare state-of-the-art ITR estimators in high-dimensional settings.
• To assess estimator performance based on rule quality, interpretability, and computational efficiency.
• To propose and evaluate a novel covariate filtering procedure to enhance ITR estimation.

Main Methods:

• Conducted a simulation study using sixteen data-generating processes with continuous outcomes and binary treatments.
• Compared multiple ITR estimators across diverse randomized and observational study designs.
• Developed and tested a new pre-treatment covariate filtering technique to improve interpretability and rule quality.

Main Results:

• High-dimensional settings pose challenges for current ITR estimation methods, impacting performance.
• The proposed covariate filtering procedure significantly enhances both the quality and interpretability of ITR estimators.
• Simulation results provide practical guidance for researchers estimating ITRs in complex, high-dimensional data.

Conclusions:

• Current ITR estimation methods require careful consideration in high-dimensional clinical trial data.
• The novel covariate filtering approach offers a valuable tool for improving the reliability and understandability of precision medicine strategies.
• Publicly available code facilitates further research and application of these methods.