Speaker: Yanxun Xu, Jenniches Faculty Scholar, is a professor of applied mathematics and statistics, an adjunct assistant professor in Biostatistics and Bioinformatics at the Johns Hopkins Sidney Kimmel Comprehensive Cancer Center, and a member of the Data Science and AI Institute. Her research develops Bayesian statistical methods and computational tools for complex, heterogeneous, large-scale data, with applications in clinical trials, electronic health records, cancer genomics, network data, precision medicine, and HIV research. Funded by the NSF, NIH, Johns Hopkins Center for AIDS Research, Institut National du Cancer, and industry partners, Xu has published more than 80 journal articles and three book chapters and has received honors including the 2016 ISBA Mitchell Prize, the Johns Hopkins Center for AIDS Research Faculty Development Award, and the Hopkins in Health Discovery Program Award.
Timing: Tuesday, September 8 from 11:00am - 1:00pm ET
Format: Online (Via Zoom)
Description: Survival analysis plays a central role in clinical research and biopharmaceutical
decision-making, but much of the evidence needed for modern comparative analyses is not
available as clean, patient-level data. Instead, it is often scattered across published Kaplan–
Meier curves, risk tables, subgroup summaries, baseline characteristic tables, trial reports, and
free-text eligibility criteria. These sources contain rich information about treatment effects,
disease progression, trial populations, and study design, but they are typically presented in
formats that are easy for humans to read and difficult for statistical models to use directly.
This webinar will discuss how recent advances in generative AI, multimodal reasoning, and agentic workflows can be combined with statistical methodology to reconstruct and synthesize survival evidence from published clinical data. A central theme is the transformation of unstructured or partially structured clinical evidence into analyzable statistical objects. Examples include recovering individual patient-level survival data from published Kaplan–Meier plots, aligning extracted curves with axes and risk tables, reconstructing censoring information, validating reconstructed survival outputs, and quantifying uncertainty introduced by digitization and reconstruction.
Throughout the webinar, I will emphasize that generative AI should serve as an interface between unstructured clinical evidence and rigorous statistical inference, rather than a replacement for statistical reasoning. AI can help convert figures, tables, and text into structured statistical inputs; statistical methods remain essential for validation, uncertainty quantification, reproducibility, and valid inference.