Di Shu (University of Pennsylvania Perelman School of Medicine)
Seminar Epidemiology, Biostatistics, & Occupational Health
Title: Survival Analysis in Multi-Site Studies Using Summary-Level Risk Set Tables.
Dr. Shu is an assistant professor of Biostatistics in the Department of Biostatistics, Epidemiology and Informatics in the Perelman School of Medicine at the University of Pennsylvania. She is also the associate director for Biostatistics at the Center for Pediatric Clinical Effectiveness (CPCE) and a Biostatistics faculty member at the Children’s Hospital of Philadelphia. Dr. Shu is interested in developing and applying suitable statistical methods to assess comparative safety and effectiveness of medical products. Her research areas include causal inference, measurement error, pharmacoepidemiology, and privacy-protecting methods. She completed her MSc in Statistics (with specialization in Biostatistics) at the University of Western Ontario, and her PhD in Statistics at the University of Waterloo.
Medical research often analyzes data from multiple sources to increase statistical power and generalizability. A growing number of studies are now conducted within multi-site distributed data networks. For example, the Sentinel System, funded by the U.S. Food and Drug Administration, monitors the safety of approved medical products using data from multiple data partners. Within these networks like the Sentinel System, each data partner maintains physical control of their data and may not always be able or willing to share individual-level data for analysis. In this talk, I will first provide a brief overview of existing methods that enable valid survival analysis without pooling individual-level data across sites. Then, I will introduce a one-step method that allows data partners to share only summary-level risk set tables to estimate overall and site-specific hazard ratios. Finally, I will discuss how to apply risk set tables to other important measures such as Kaplan–Meier curves as well as some future topics. I will justify the method theoretically, illustrate its use, and demonstrate its statistical performance using both real-world and simulated data.
Ìý
Ìý