State of the Art Lecture: AI and Innovation in Urogynecology

State of the Art Lecture: AI and Innovation in Urogynecology

Leon N. Plowright, MD, FACOG, FACS, URPS

Assistant Professor, University of Florida College of Medicine – Jacksonville

Moderators: Jorge Milhem Haddad (Brazil) & Anna Rosamilia (Australia)

Speaker: Emanuel Trabuco (United States)

Artificial intelligence is no longer theoretical- it is rapidly becoming embedded in everyday life and increasingly in clinical medicine. At this year’s IUGA State of the Art Lecture, Dr. Emanuel Trabuco (United States), founder of the Mayo Clinic OB/GYN AI Laboratory delivered a timely and thought- provoking overview on how artificial intelligence is beginning to transform Urogynecology.

Dr. Trabuco framed the machine learning landscape with two foundational concepts.  First, Supervised learning trains on labeled data with predefined outcomes and require both discrimination and calibration- meaning, a predicted 70% risk should translate into events 70% of the time. Unsupervised learning, by contrast, identifies hidden phenotypes within heterogeneous populations, with the critical question being whether the mathematically derived clusters are biologically meaningful. He also distinguished large language models, which respond passively to prompts, from AI agents, which are LLMs that have been given access to tools and live data, enabling true workflow integration.

Three published studies illustrated the clinical relevance of these approaches. A Peking University cohort study using XGBoost demonstrated accurate prediction of post-prolapse repair stress urinary incontinence (AUC 0.71), supporting risk-tailored prophylactic sling decisions rather than universal intervention. A sacrocolpopexy cohort using unsupervised learning identified five preoperative prolapse phenotypes predictive of surgical success, with notable differences based on prior hysterectomy status. Finally, a Cedars-Sinai IC/BPS study used K-means clustering to identify distinct symptom phenotypes, demonstrating that bladder-centric pain responded preferentially to bladder instillations, while myofascial-predominant phenotypes showed greater benefit from pelvic floor physical therapy.

Importantly, Dr. Trabuco addressed the challenge of implementation—where many algorithms stall. He highlighted the practical realities of integrating AI into clinical care: building interoperable EMR data pipelines, validating models across external populations, navigating FDA pathways, and obtaining institutional risk management approval. His team’s TRIGGERS platform exemplified this translational effort, reducing 80 hours of nurse chart abstraction to just 30 minutes of compute time.

He closed with examples already active in Mayo Clinic practice, including a platform aggregating 26 petabytes of de-identified data across 64 health systems for rapid cohort discovery, and Gracie, an AI agent generating overnight patient summaries from the EMR. Notably, AI agents were not even part of his team’s working vocabulary at the end of 2024; a year later, they are clinical reality.

For urogynecology, the promise of AI is not simply automation, it is precision. The ability to move toward truly individualized care, where treatment decisions are informed not only by clinical judgment, but by predictive intelligence. If this lecture made one thing clear, it is that in our field, the pace of change is itself the message. We have only begun to scratch the surface.