Working Group Webinar Library
Webinar Library
From Clinical Notes and Patient Narratives to Real-World Evidence: Case Studies and Evaluation Approaches for LLM-Based Clinical NLP Pipelines
Large language models are transforming clinical NLP, but real-world healthcare pipelines still combine LLMs with traditional methods to turn clinical notes and patient narratives into structured evidence. This talk covers two active projects using this approach — one studying medication adherence across five diseases, another improving drug safety prediction — and discusses best practices for evaluating LLM systems in clinical settings.
FACMI and FAMIA: What They Are and How to Prepare.
In this webinar, we will learn what FACMI and FAMIA are, why they matter, and how to prepare for a successful nomination. Presenters
From Models to Mission: Deploying National-Scale Clinical AI through VA–DOE Collaboration
Deploying NLP and machine learning in real-world health systems requires far more than model development—it demands scalable infrastructure, rigorous evaluation, and integration within complex operational environments.
Data-Driven Scientific Hypothesis Generation in Clinical Research: The Roles of Human and AI
Hypothesis generation is an early and critical step in any hypothesis-driven clinical research project. However, the process of hypothesis generation is less understood. In this talk, Dr. Jing will briefly introduce the VIADS (a Visual Interactive Analytic tool for filtering and summarizing large health Data Sets coded with hierarchical terminologies) project and its results* in the last decade. The focus will be on a study that recruited clinical researchers who used (or did not use) VIADS for data-driven hypothesis generation in the clinical research context. The talk will summarize the experiments and principal findings and share the lessons learned. The role of humans and/or AI in the process and challenges in measuring the processes and results will also be discussed. * Selected publications— PMID: 39819516, 40768504, 40417518, 39211055, 37011112, 38384898, 35736798, 30764811, 24727931, 22195119 Presenter
Graph-based Prediction of Spatio-Temporal Vaccine Hesitancy from Insurance Claims Data
The VaxHesSTL framework combines Graph and Recurrent Neural Networks to predict vaccine hesitancy at the ZIP code level by capturing both spatial relationships and historical trends, outperforming existing models when trained on a six-year, five-million-person insurance claims dataset from Virginia. To address the high cost of such data, the research also explores an active learning approach to optimize which ZIP codes are selected for training.