Outcome trajectory estimation for optimal dynamic treatment regimes with repeated measures

Area of Science:

• Statistics
• Biostatistics
• Health Informatics

Background:

• Sequential multiple assignment randomized trials assess long-term impacts of dynamic treatment regimes (DTRs) using repeated outcome assessments.
• Current Q-learning methods require scalar responses, limiting their ability to identify optimal DTRs.
• Inverse probability weighting (IPW) for optimal outcome trajectory estimation is inefficient and prone to model mis-specification.

Purpose of the Study:

• To propose a modified Q-learning method using generalized estimating equations (GEE) to overcome limitations of existing approaches.
• To enhance the estimation of optimal dynamic treatment regimes (DTRs) in the context of longitudinal data.
• To improve the characterization of time-varying treatment effects.

Main Methods:

• Developed a modified Q-learning algorithm incorporating generalized estimating equations (GEE).
• Applied the proposed method to the M-bridge trial, focusing on adaptive interventions for problematic drinking in college freshmen.
• Conducted simulation studies to evaluate the performance of the new method.

Main Results:

• The modified Q-learning with GEE demonstrated improved efficiency compared to traditional methods.
• The proposed method showed enhanced robustness against model mis-specification.
• The approach effectively characterized how treatment effects manifest over time in adaptive interventions.

Conclusions:

• Modified Q-learning with GEE offers a more efficient and robust approach for analyzing sequential multiple assignment randomized trials.
• This method provides a superior tool for identifying optimal dynamic treatment regimes (DTRs) with longitudinal outcomes.
• The findings have implications for the design and analysis of adaptive interventions in public health research.