Abstract: PUB082
Clinical Questions and Uncertainty in the ASN Open Forum: An Artificial Intelligence-Assisted Analysis
Session Information
Category: Educational Research
- 1000 Educational Research
Authors
- Bergling, Karin, Renal Research Institute, New York, New York, United States
- Fitzgerald, Mark, American Society of Nephrology, Washington, District of Columbia, United States
- Rodby, Roger A., Rush University Medical Center, Chicago, Illinois, United States
- Glassock, Richard J., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States
- Woo, Karen, University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States
Background
Connecting ASN members from numerous countries and cities around the world, the ASN Communities Open Forum provides a space for the kidney community to ask questions, share experiences, and seek peer input. With approximately 21,000 members, these discussions offer a real-world view of clinical uncertainty, evidence gaps, and educational needs. Our objective was to evaluate whether artificial intelligence (AI) could characterize the topics, question types, and sources of uncertainty raised by members in thread-starting posts.
Methods
Discussion threads from 2025 were extracted and de-identified using code-based screening and manual review. Thread-start posts were classified with GPT-5.5 (OpenAI) using predefined labels describing patient population, disease area, question type, and cause of uncertainty/complexity. AI-assigned label distributions were summarized using counts and percentages. AI classification accuracy was assessed by two independent reviewers in 50 randomly selected posts, with discrepancies resolved through discussion.
Results
532 thread-start posts were de-identified and labeled. Table 1 summarizes the five most common AI-assigned labels across disease area, clinical question type, and uncertainty theme. The leading disease areas were glomerular disease or vasculitis, excluding IgA nephropathy, which had its own label, followed by acute kidney injury and hypoalbuminemia or nephrotic syndrome. Questions most concerned treatment selection, with risk-benefit tradeoffs as the leading source of uncertainty. Human review found AI-generated labels fully correct in 44 of 50 posts (88%); 6 posts (12%) had at least one partially correct category. Information was missed in 2 posts (4%; 2 categories), and labels unsupported in 4 posts (8%; 6 categories).
Conclusion
AI-assisted analysis identified recurring clinical questions and uncertainty in real-world nephrology discussions, demonstrating that Open Forum data can help ASN understand member needs and guide targeted education and research.
Acknowledgment
Artificial intelligence tools were used as part of the study methods to classify information in thread-start posts and to assist with abstract preparation. The authors take full responsibility for the integrity and accuracy of the final work.
Top five AI-assigned disease areas, clinical questions, and uncertainty themes. Percentages reflect the proportion of posts assigned each label; each post could receive multiple labels.