TITLE: Subgroup Analysis and Causal Inference with Application in Medical Research
SPEAKER: Prof. Menggang Yu, University of Wisconsin-Madison
MODERATOR: William Wang
Abstract:
Modern therapeutic studies typically involve subgroup analysis, whether it is pre-specified or post hoc. A key aspect of subgroup analysis is to identify clinically relevant patient characteristics for estimation of treatment effect heterogeneity and for treatment recommendation that may optimize treatment effectiveness or generate interesting research questions.
In this short course, we will introduce two general frameworks that encompass many recent statistical methods for subgroup analysis. These methods focus on modeling treatment effect modification, instead of the potential outcomes. The causal effect of treatment effect modifiers will be examined mainly based on the no-unmeasured confounders assumption. However instrumental variable approach will also be discussed to deal with possible unmeasured confounders. The proposed methods are quite flexible and can be used for analysis of both randomized clinical trials and observational studies. They also link nicely with the estimand framework laid out in the International Council for Harmonisation (ICH) E9 (R1) document that aims to provide guidance for statistical analysis for clinical trials.
We will examine the empirical performance of several procedures belonging to the proposed framework through numerical studies and real data analyses. In particular, we will discuss our experience in intervention recommendation for a transitional care program at the University of Wisconsin Health system and in a breast cancer screening study where we use health behavioral constructs, comorbidity variables, and social economic factors as treatment effect modifiers.
Prerequisites
The course is accessible to anyone with a knowledge of statistical inference at the level of introductory graduate level courses in mathematical statistics and probability. Exposure to causal inference (based on the potential outcomes), statistical learning theory (e.g. regularization method) can be helpful, but is not required.
Instructors’ Biography:
Dr. Menggang Yu is a professor at the Department of Biostatistics and Medical Informatics, University of Wisconsin – Madison and Director of the Biostatistics Shared Resources at the UW Carbone Cancer Center. He is an elected fellow of the American Statistical Association (ASA).
Dr. Yu conducts broad statistical methodology research, all motivated by his daily collaborative experience with medical investigators. His methodological publications cover extensive topics including joint modeling of longitudinal and survival data, missing data, clinical trial design and analysis, causal inference, and personalized medicine.
Dr. Yu is also a devoted statistical collaborator. One of his career goals is to make integral contributions to scientific research that has direct impact on human health. His scientific collaboration is mainly in the areas of cancer and health care research.
He has also co-authored over 80 collaborative medical papers, among which over 25% are in highly impactful medical journals (with 2017 impact factors range from 10.0 to 244.6). These publications introduce ground-breaking and practice-changing results to many areas of oncology and health services.