Evaluating Knowledge Gaps in Cardio-Obstetrics: A Comparative Analysis of
Cardiologists, Obstetricians, and Generative Artificial Intelligence

Background: Recent advancements in artificial intelligence (AI) have shown
potential in various medical disciplines, often achieving comparable performance to
human physicians. This study aimed to evaluate the knowledge gaps between
cardiologists and obstetricians in the cardio-obstetrics domain and to compare their
performance with a generative AI chatbot.
Methods: We developed 12 guideline-based standardized questions addressing key
cardio-obstetrics topics, including anticoagulation management, cardiac risk
stratification, high-risk pregnancy monitoring, and gestational hypertensive disorders.
Human participants, comprising cardiologists and obstetricians (Maternal-Fetal
Medicine specialists), were permitted to select "I don’t know" alongside multiple-
choice answers. Responses were assessed against predefined correct, well-
referenced answers, and accuracy rates were calculated for each group, GPT-4,
Claude 3.0, Gemini 2.0, and Microsoft Copilot in Bing.
Results: The accuracy rates for 20 cardiologists and 26 obstetricians who answered
the survey were 47.5% and 67%, respectively. GPT-4 and Gemini 2.0 achieved the
highest accuracy rate of 85%, while Copilot scored 75 % and Perplexity AI scored
67%. Notable knowledge gaps among physicians were identified in areas such as
anticoagulation management during pregnancy, high-risk pregnancy monitoring, and
hypertensive disorders. Obstetricians performed better than cardiologists in
hypertensive disorder management, while cardiologists demonstrated greater
accuracy in maternal cardiovascular risk assessment during pregnancy.
Conclusion: Our findings highlight substantial knowledge gaps in cardio-obstetrics,
underscoring the need for interdisciplinary collaboration to optimize patient care.
Remarkably, GPT-4 and Gemini 2.0 outperformed cardiologists and obstetricians,
demonstrating their potential as an adjunctive tool for decision-making and education
in this complex field. Integrating advanced AI systems into clinical workflows could
enhance diagnostic accuracy and provide continuous learning opportunities for
healthcare professionals.