Set-valued dynamic treatment regimes for competing outcomes

Area of Science:

• Biostatistics
• Clinical Decision Making
• Pharmacoeconomics

Background:

• Dynamic treatment regimes (DTRs) typically optimize a single outcome.
• Clinical decisions often involve balancing multiple, potentially competing outcomes (e.g., symptom relief vs. side effects).
• Existing methods are insufficient when patient preferences are unknown, uncommunicated, or change over time.

Purpose of the Study:

• To develop a novel method for constructing DTRs that accommodates competing outcomes.
• To address limitations of single-outcome optimization in complex clinical scenarios.
• To provide a framework for DTRs that respects patient preferences even when they are not explicitly defined.

Main Methods:

• Proposed a method for constructing set-valued DTRs, recommending subsets of treatments at each decision point.
• Developed an exact enumeration algorithm by reformulating the problem as a linear mixed integer program.
• Illustrated the methodology using data from the CATIE schizophrenia study.

Main Results:

• The proposed method generates sets of non-inferior treatments for given patient histories.
• Successfully applied an exact enumeration algorithm to solve the underlying optimization problem.
• Demonstrated the practical application of set-valued DTRs in a real-world clinical dataset.

Conclusions:

• The novel approach effectively handles competing outcomes in DTRs, offering a more realistic clinical decision-making framework.
• Set-valued DTRs provide a robust alternative to single-outcome optimization, especially in personalized medicine.
• The linear mixed integer programming approach offers an efficient solution for constructing these complex DTRs.