On-Demand Learning Lab: Advanced Data Science for Healthcare Quality Improvement-Recording - Advanced Data Science for Healthcare Quality Improvement

Video Transcription

Video Summary

NACU hosted a Live Learning Lab webinar featuring Duke Institute for Health Innovation leaders Emily Starrett, Will Koneckly, and Mark Sendak on how healthcare delivery organizations are using AI to advance quality and safety. Starrett, a pediatric emergency physician and QI leader, emphasized that AI is already embedded in clinical and operational work (e.g., EHR documentation support, decision aids) and that quality professionals must help ensure AI improves systems rather than adding waste. She contrasted traditional improvement tools (Lean, PDSA, SPC) with AI’s ability to detect patterns at scale, illustrating with a medication error that shifted focus from individual noncompliance to process design—and raised the possibility of using large language models to detect safety risks beyond self-reporting. She described Duke’s collaboration to develop a pediatric sepsis detection approach using real-time EHR data, pairing clinical workflow expertise with data science, adjudication, and bedside implementation through iterative testing and human-centered design, leading to faster antibiotics.<br /><br />Koneckly framed AI as a ubiquitous tool that can amplify QI but still requires QI leaders to define problems, manage change, and govern implementation. He shared Duke examples: automating SEP-1 quality measure abstraction to reduce manual review costs; querying thousands of policies with AI; extracting social needs and goals-of-care from unstructured notes; improving surgical prior authorization notes with high accuracy; oncology intake triage support; and in-basket message triage that reduced clinician burnout. He also highlighted operational ML for ED flow and Duke’s “Sepsis Watch,” associated with improved bundle compliance and reduced mortality.<br /><br />Sendak closed by introducing the Health AI Partnership, a national community of practice providing best practices, technical assistance (including for safety-net providers), and governance resources to support safe, effective, equitable AI adoption.

Keywords

healthcare AI

quality improvement (QI)

patient safety

Duke Institute for Health Innovation

EHR data analytics

pediatric sepsis detection

large language models (LLMs)

Sepsis Watch

clinical decision support

automation of quality measure abstraction (SEP-1)

AI governance and equitable adoption