May 7, 2022
1:30 PM – 5:30 PM CT

Statistical Methods for Precision Medicine

Lu Tian, Stanford University

The basic idea of precision medicine is to use patient's specific characteristics, such as genetic make-up, biomarker profile, clinical history, environmental exposure, etc., to guide clinical decision making for effective prevention and treatment. Recent advances in high-throughput and information technologies can easily and robustly generate a large amount data to characterize individual patients, offering opportunities to develop and promote precision medicine in daily clinical practice. The development of such a smart and targeted strategy needs to be empirically data-driven and the corresponding challenges in the statistical front are huge, mainly because analyzing heterogeneous treatment effect/associations, i.e., interaction, is much more difficult than analyzing the homogeneous counterparts, i.e., main effect. The goal of this course to introduce recently developed statistical and machine learning techniques for precision medicine. Most of the course will focus on how to optimally assign treatment to patients according to his or her personal characteristics in the context of two-arm randomized clinical trial. However, we will also discuss extensions to multi-arm trials and observational studies. Specifically, we will cover the following topics: the casual inference framework for personalized treatment effect based on counterfactual outcomes; estimation of optimal treatment selection rule including subgroup analysis, treebased method, regression modeling, modified covariates approach, Q-learning and outcome weighted learning; and validation and statistical inference of the optimal treatment selection strategy. We will also discuss the computational perspective of the aforementioned methods including dimension reduction via regularization and applications of machine learning methods. We will provide examples for how to construct and evaluate estimated optimal treatment regimes in R. There is no requirement for prior exposure to precision medicine or machine learning methods, but prior knowledge of basic methods such as regression, interactions, and analysis of randomized clinical trials is expected.


Lu Tian

Stanford University