Clinical AI Competence in Obstetrics and Gynecology

This Clinical Opinion argues that competent, accountable use of generative artificial intelligence is becoming a component of safe professional practice in obstetrics and gynecology. The authors define clinical AI competence as the ability to use AI tools for appropriate clinical tasks while critically appraising output, recognizing hallucinations and outdated guidance, verifying sources, preserving confidentiality, communicating uncertainty, and maintaining independent clinical judgment. Their central claim is not that clinicians will use AI, but that the risk lies in using it poorly.

The evidentiary tension is a competence gap. Survey data cited in the paper place physician awareness or use of AI at 81%, while fewer than 15% of students and faculty report formal expertise in its clinical application; roughly one in three US adults now report using large language models for health information. On the benefit side, a 70-clinician randomized trial found structured clinician-LLM workflows reached 82% to 85% diagnostic accuracy versus 75% with conventional resources, and documentation, readability, and evidence-retrieval tasks show the clearest efficiency evidence. Autonomous clinical decision-making without physician review is described as inconsistent with professional responsibility and unsupported by current evidence.

The argument is framed historically. Obstetric ultrasound, electronic fetal monitoring, cell-free DNA screening, and cervical-length screening each provoked early resistance before becoming standard, a pattern the authors term primum non mutare (first do no change), which they propose replacing with primum benefacere (first do good), under structured use, source verification, and explicit physician oversight. They are explicit that direct evidence showing improved maternal, fetal, or neonatal outcomes from clinical AI use has not been established and that outcome-based evaluation should be a research priority.