Estimating individualized treatment rules for ordinal treatments

Area of Science:

• Precision medicine
• Statistical learning
• Clinical decision-making

Background:

• Precision medicine aims to tailor disease treatment and prevention to individual patient characteristics.
• Optimal individual treatment rules (ITRs) are crucial for personalized treatment selection.
• Existing methods like outcome weighted learning (OWL) are primarily for binary treatments, leaving a gap for ordinal treatment settings.

Purpose of the Study:

• To propose a novel statistical method for estimating ITRs in ordinal treatment settings.
• To adapt outcome weighted learning (OWL) for individualized dose finding and similar ordinal treatment scenarios.
• To establish theoretical properties and demonstrate the practical performance of the new method.

Main Methods:

• A data duplication technique is proposed to handle ordinal treatments.
• A piecewise convex loss function is utilized within the outcome weighted learning framework.
• Fisher consistency, convergence, and risk bound properties of the estimated ITR are established.

Main Results:

• The proposed method provides a statistically sound approach for estimating ITRs with ordinal treatments.
• Theoretical properties including Fisher consistency and convergence are established under specific conditions.
• Simulations and a type 2 diabetes mellitus dataset application show competitive performance against existing methods.

Conclusions:

• The developed technique effectively extends outcome weighted learning to ordinal treatment settings.
• This advancement offers a valuable tool for personalized medicine, particularly in individualized dose-finding.
• The method shows promise for improving clinical decision-making in complex treatment scenarios.