Evaluation of the diagnostic concordance of artificial intelligence (ChatGPT) in the identification of indications for emergency cesarean section from obstetric records of the gynecology-obstetrics department of the Regional University Hospital Center of Fada N’Gourma (CHRU-FG).

Abstract

Emergency cesarean sections require rapid and accurate identification of their indications to reduce maternal and fetal morbidity and mortality. The emergence of artificial intelligence, particularly language models such as ChatGPT, offers new possibilities for clinical decision support. However, their performance in identifying obstetric indications in real-world settings remains poorly documented, particularly in resource-limited settings. This was a retrospective cross-sectional study of diagnostic accuracy conducted at the Fada N’Gourma Regional University Hospital. The medical records of patients who underwent an emergency cesarean section were analyzed. Clinical case scenarios submitted to ChatGPT were used to identify the primary indication, which was then compared to clinicians’ decisions. Agreement was assessed using the Kappa coefficient, and diagnostic performance was measured by sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Logistic regression was used to identify factors associated with this agreement. A total of 62 cases were analyzed. The overall agreement was 75.8%, with a Kappa coefficient of 0.69, indicating good agreement. The best performance was observed for preeclampsia (sensitivity 91.67%; specificity 98%). Complete records (OR=3.2; p=0.01) and frequent indications (OR=2.3; p=0.001) were significantly associated with better agreement. ChatGPT demonstrates good diagnostic performance in identifying indications for emergency cesarean section, although this depends on the quality of the clinical data and the type of indication.