A flexible rank-based framework for individualized treatment selection with mixed-type multivariate outcomes

Area of Science:

• Biostatistics
• Clinical Informatics
• Personalized Medicine

Background:

• Complex diseases often require monitoring multiple, heterogeneous outcomes, not just a single primary endpoint.
• Current individualized treatment selection methods may not adequately handle diverse outcome types.

Purpose of the Study:

• To develop a rank-based framework for individualized treatment selection that accommodates mixed collections of continuous, ordinal, binary, and right-censored survival outcomes.
• To provide a method for summarizing and comparing treatment effectiveness across multiple heterogeneous endpoints.

Main Methods:

• Developed a rank-based framework using Mahalanobis-type distance to quantify discrepancies between outcome-specific rankings.
• Summarized conditional outcome distributions using parameters like quantiles and survival probabilities.
• Employed flexible tree-based ensemble methods for outcome summary estimation.

Main Results:

• Monte Carlo simulations demonstrated reasonably high probabilities of correct treatment selection.
• Method performance improved with increased signal-to-noise ratio and was robust to outcome weight choices.
• Illustrated application in an AIDS clinical trial for individualized antiretroviral regimen recommendations.

Conclusions:

• The proposed rank-based framework offers a robust approach for personalized treatment selection with multiple heterogeneous outcomes.
• This method can balance diverse clinical endpoints to derive optimal, individualized treatment strategies.
• The framework is adaptable and demonstrated effectiveness in a real-world clinical scenario.