Individualized treatment effects with censored data via fully nonparametric Bayesian accelerated failure time models.

Area of Science:

• Biostatistics
• Personalized Medicine
• Clinical Trial Analysis

Background:

• Patient responses to identical treatments vary significantly, necessitating methods to characterize heterogeneous treatment effects (HTE).
• Personalized medicine aims to tailor treatments based on individual patient characteristics and predicted responses.
• Time-to-event data is common in clinical trials, requiring specialized analytical approaches.

Purpose of the Study:

• To present a novel nonparametric accelerated failure time model for analyzing HTE in time-to-event data.
• To develop a flexible statistical framework that accommodates complex treatment effect variations.
• To provide a method that yields natural estimates of individual treatment effects and requires minimal user input.

Main Methods:

• Utilized Bayesian additive regression trees (BART) for flexible regression modeling.
• Employed a mean-constrained Dirichlet process mixture model to handle baseline hazard flexibility.
• Developed a nonparametric approach for estimating individual treatment effects and assessing HTE.

Main Results:

• The proposed method demonstrated strong predictive performance in analyzing clinical trial data.
• The approach exhibited good frequentist properties, including accurate parameter coverage and mitigation of spurious HTE findings.
• Analysis of two large clinical trials (N=6769) for congestive heart failure revealed substantial evidence of HTE.

Conclusions:

• The nonparametric accelerated failure time model effectively identifies and quantifies HTE in time-to-event data.
• The method's flexibility and minimal input requirements make it suitable for various HTE assessment goals.
• Findings from clinical trials underscore the importance of considering individual treatment effect variations in medical practice.