Deconfounded Warm-Start Thompson Sampling with Applications to Precision Medicine

2025-05-22

AI Summary

Large-scale randomized clinical trials (RCTs) are resource-intensive, while observational data suffer from confounding and latent-variable bias, limiting their utility in adaptive experimentation. To address this, we propose Deconfounded Warm-Started Thompson Sampling (DWTS), the first method to integrate doubly robust debiased LASSO estimation into the prior initialization of Linear Thompson Sampling (LinTS). DWTS explicitly separates observed covariates from latent confounders, yielding a compact, causally reliable contextual representation that enables safe warm-starting of adaptive trials using observational data. Evaluated on synthetic benchmarks and a virtual trial environment built from real-world cardiovascular risk data, DWTS significantly reduces cumulative regret, enhances statistical efficiency, and improves personalized treatment decision-making. Our approach establishes an interpretable, empirically verifiable paradigm for causal adaptive experimental design.

Technology Category

• Machine Learning: Causal Learning
• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Randomized clinical trials often require large patient cohorts before drawing definitive conclusions, yet abundant observational data from parallel studies remains underutilized due to confounding and hidden biases. To bridge this gap, we propose Deconfounded Warm-Start Thompson Sampling (DWTS), a practical approach that leverages a Doubly Debiased LASSO (DDL) procedure to identify a sparse set of reliable measured covariates and combines them with key hidden covariates to form a reduced context. By initializing Thompson Sampling (LinTS) priors with DDL-estimated means and variances on these measured features -- while keeping uninformative priors on hidden features -- DWTS effectively harnesses confounded observational data to kick-start adaptive clinical trials. Evaluated on both a purely synthetic environment and a virtual environment created using real cardiovascular risk dataset, DWTS consistently achieves lower cumulative regret than standard LinTS, showing how offline causal insights from observational data can improve trial efficiency and support more personalized treatment decisions.

Problem

• Utilizing observational data to enhance clinical trial efficiency
• Reducing bias in adaptive trials via deconfounded Thompson Sampling
• Improving personalized treatment decisions with offline causal insights

Innovation

• Deconfounded Warm-Start Thompson Sampling (DWTS)
• Doubly Debiased LASSO (DDL) for covariate selection
• Combines measured and hidden covariates efficiently