Interactive Q-learning for Quantiles.

Area of Science:

• Biostatistics
• Clinical Informatics
• Health Services Research

Background:

• Dynamic treatment regimes (DTRs) guide sequential medical decisions.
• Current DTR methods often optimize the mean response, which may not capture the full picture.
• Alternative performance metrics are needed for personalized medicine.

Purpose of the Study:

• To develop novel estimators for DTRs that optimize response probabilities and quantiles.
• To enable DTRs that target specific points or quantiles (e.g., median) of the response distribution.
• To provide more flexible and comprehensive optimization strategies for DTRs.

Main Methods:

• Derived estimators for decision rules in two-stage, binary treatment settings.
• Focused on optimizing probabilities and quantiles of the response distribution.
• Utilized simulation experiments to evaluate performance.

Main Results:

• Proposed methods demonstrated favorable performance in simulation studies.
• The approach allows optimization of the cumulative distribution function at specific points.
• Enabled estimation of DTRs targeting specific response quantiles, like the median.

Conclusions:

• The developed methods offer a valuable alternative to mean-based optimization for DTRs.
• These techniques enhance the ability to personalize treatment strategies based on desired outcomes.
• Applied to a depression symptom remission trial, showing practical utility.