Uncertainty estimation in diagnosis generation from large language models: next-word probability is not pre-test probability
Published in JAMIA Open, 2025
Recommended citation: Yanjun Gao, Skatje Myers, Shan Chen, Dmitriy Dligach, Timothy Miller, Danielle S Bitterman, Guanhua Chen, Anoop Mayampurath, Matthew M Churpek, Majid Afshar, Uncertainty estimation in diagnosis generation from large language models: next-word probability is not pre-test probability, JAMIA Open, Volume 8, Issue 1, February 2025, ooae154, https://doi.org/10.1093/jamiaopen/ooae154 https://doi.org/10.1093/jamiaopen/ooae154
Abstract: Cardiac valve abnormalities are one of the most common congenital cardiac malformations in children; however, relatively few valve implant devices have been approved by the US Food and Drug Administration (FDA) for pediatric use. Therefore, off-label use of adult valves to treat valve disease in children is a standard practice.1,2 Challenges in developing pediatric-specific devices are largely related to barriers in performing clinical trials in small patient populations and perceived low market potential by manufacturers.3 To define gaps in treatment options and support efforts to advance pediatric device research and development, we analyzed data from a large database of cardiac operations performed in the US and evaluated the FDA approval status of cardiac valve devices implanted in pediatric patients.