The deployment of text-based AI models at the point of care introduces new opportunities—and risks—for improving clinical documentation, communication, and decision support. In this talk, I will share lessons learned from the real-world bedside implementation of AI systems that use the EHR notes for screening opioid misuse to Ambient AI for automating clinical documentation, as well as methods for monitoring, evaluation, and continuous improvement. Drawing on our Learning Health System at University of Wisconsin and operational frameworks embedded within clinical practice, I will discuss pragmatic strategies for safety monitoring, workload reduction, provider experience, and performance drift. I will also highlight novel evaluation tools, including the development and validation of the Provider Documentation Summarization Quality Instrument (PDSQI-9), to assess AI-generated clinical summaries. Together, these approaches offer a roadmap for translating generative AI into sustainable clinical workflows with a focus on rigor, transparency, and meeting the quintuple aims of healthcare.
Presenter
