Fan Li (Yale University School of Public Health)
Title:听Propensity Score Weighting Analysis of Survival Outcomes Using Pseudo-observations.
Dr. Fan Li is an assistant professor in the Department of Biostatistics at Yale University School of Public Health. He is also faculty member at the Center for Methods in Implementation and Prevention Science (CMIPS) and the Yale Center for Analytical Sciences (YCAS). Dr. Li receives his PhD in Biostatistics from Duke University in 2019. His research interests include developing methods for comparative effectiveness research with randomized trials and observational studies. He is also an expert in the design, monitoring and analysis of pragmatic cluster randomized trials, and is currently Principal Investigator of a Patient-Centered Outcome Research Institute (PCORI) methods award 鈥淣ew methods for planning cluster randomized trials to detect treatment effect heterogeneity鈥�. Website:
Survival outcomes are common in comparative effectiveness studies. A standard approach for causal inference with survival outcomes is to fit a Cox proportional hazards model to an inverse probability weighted (IPW) sample. However, this method can be subject to model misspecification and the resulting hazard ratio estimate lacks causal interpretation. Moreover, IPW often corresponds to an inappropriate target population when there is lack of covariate overlap between the treatment groups. To address these limitations, we propose a general class of model-free causal estimands with survival outcomes on user-specified target populations, and develop a class of propensity score weighting estimators via the pseudo-observation approach. As the pseudo-observations are constructed by jackknifing, re-sampling-based inference are generally computationally intensive. To circumvent the computational intensity, we develop new asymptotic variance expressions for the class of weighting estimators based on the functional delta-method and von Mises expansion of pseudo-observations. We show that the overlap weights developed for non-censored outcomes still lead to the most asymptotically efficient causal comparisons based on pseudo-observations, expanding the theoretical underpinnings of overlap weights. Extensive simulations are carried out to examine the operating characteristics of the weighting estimators based on pseudo-observations. Finally, we apply the proposed methods to study the treatment effect of radiotherapeutic or surgical approaches for patient with high-risk localized prostate cancer.
听