【数学与统计及交叉学科前沿论坛------高端学术讲座第171场】

报告题目：Causal inference for all: Marginal causal effects for outcomes truncated by death

报 告 人：王林勃副教授  多伦多大学、华盛顿大学

报告时间：2025年11月4日周二10:00-11:00

报告地点：腾讯会议599-2258-0337

报告摘要：In longitudinal studies, outcomes of interest are often truncated by death, meaning they are only observed or well-defined conditional on intermediate outcomes such as survival. Standard causal estimands, such as the survivor average causal effect, focus on a nonidentifiable subgroup and are therefore difficult to interpret, and their extension to longitudinal settings introduces further complications. We address these challenges by introducing a flexible class of marginal causal effect estimands that (i) apply to the entire population and (ii) summarize potential outcomes over time. This framework supports a range of clinically relevant summaries, such as cumulative or last observed outcomes, and can be tailored using weighting schemes to align with different decision making goals. For individuals who would survive longer under one treatment than under the alternative, we further define a class of secondary estimands to evaluate outcomes during the additional survival time. We illustrate the approach through a reanalysis of a prostate cancer trial, highlighting how different estimands may lead to different treatment conclusions.

报告人简介：Linbo Wang is Canada Research Chair in Causal Machine Learning, and an associate professor in the Department of Statistical Sciences and the Department of Computer and Mathematical Sciences, University of Toronto. He is also a faculty affiliate at the Vector Institute and holds affiliate positions in the Department of Statistics at the University of Washington and the Department of Computer Science at the University of Toronto. His research focuses on causality and its interaction with statistics and machine learning.

The field of Statistics and Data Science is increasingly focused on

methodologies for detailed analysis of large scale data sets. In a clinical diagnostic setting, CT/MR and PET scanners now play a key role in the management of stroke, cardiac arrest and cancer. Current scanners are often used to image target volumes dynamically in time. This can lead to 4-D data sets on the order of 30 Gbytes or more. However, patient physiology and dose considerations critically limit the statistical quality of these measurements. Consequently there is a substantial role for development of suitable methodologies for analysis. The evaluation of local biological properties of tissue from the data involves inference about a life-table based on indirectly observed information. The talk describes a multivariate data profiling technique for use in the recovery of monparametric life-tables estimates at the voxel-level. A data-adaptive bootstrapping process, accounting for distributional and spatio-temporal covariance characteristics, is used to generate patient-specific assessment of uncertainty in derived disease biomarkers. The consistency of this technique as a function of dose is discussed. Methods are illustrated by application to data from state-of-the-art scanners.