Evaluating How Different AI Displays Affect Doctors Confidence and Accuracy in Predicting Preterm Birth Using Cervical Ultrasound

What this Trial Is About

This research aims to understand how different ways of presenting artificial intelligence AI decision support affect clinical judgment among medical doctors in obstetrics and gynecology. The study focuses on assessing the risk of spontaneous preterm birth using clinical case vignettes with cervical ultrasound images. It compares two AI presentation styles a simple binary classification preterm vs term birth and an individualized risk estimate of preterm birth. The main goal is to see which format improves the alignment between doctors confidence in their decisions and the accuracy of those decisions, and whether these AI presentations help or harm clinical choices. Participants are randomly assigned to one of two groups one receives AI predictions in a binary format indicating preterm or term birth, while the other group receives an AI-generated percentage risk estimate of preterm birth. These AI supports are provided alongside standard clinical information during an online questionnaire session. Each participant reviews clinical cases, makes diagnostic and management decisions, rates their confidence before and after seeing the AI output, and reports their trust in the AI system. During the approximately 20-minute online session, participants complete a questionnaire involving clinical case evaluations and decision-making tasks. Researchers measure how well clinicians confidence matches their decision accuracy immediately after AI exposure. They also assess helpful versus harmful changes in decisions, shifts in decision accuracy and confidence, trust in AI, and follow-up ultrasound planning. The study is designed to gather these outcomes in a single session without ongoing treatment or monitoring.

Does AI Make Clinicians More Appropriately Confident? A Randomized Study in Preterm Birth Prediction