Can <scp>ChatGPT</scp> Generate Acceptable Case‐Based Multiple‐Choice Questions for Medical School Anatomy Exams? A Pilot Study on Item Difficulty and Discrimination
Yavuz Selim Kıyak, Ayşe Soylu, Özlem Çoşkun, Işıl İrem Budakoğlu, Tuncay Peker
Abstract
Developing high-quality multiple-choice questions (MCQs) for medical school exams is effortful and time-consuming. In this study, we investigated the ability of ChatGPT to generate case-based anatomy MCQs with acceptable levels of item difficulty and discrimination for medical school exams. We used ChatGPT to generate case-based anatomy MCQs for an endocrine and urogenital system exam based on a framework for artificial intelligence (AI)-assisted item generation. The questions were evaluated by experts, approved by the department, and administered to 502 second-year medical students (372 Turkish-language, 130 English-language). The items were analyzed to determine the discrimination and difficulty indices. The item discrimination indices ranged from 0.29 to 0.54, indicating acceptable differentiation between high- and low-performing students. All items in Turkish (six out of six) and five out of six in English met the higher discrimination threshold (≥ 0.30) required for large-scale standardized tests. The item difficulty indices ranged from 0.41 to 0.89, most items falling within the moderate difficulty range (0.20-0.80). Therefore, it was concluded that ChatGPT can generate case-based anatomy MCQs with acceptable psychometric properties, offering a promising tool for medical educators. However, human expertise remains crucial for reviewing and refining AI-generated assessment items. Future research should explore AI-generated MCQs across various anatomy topics and investigate different AI models for question generation.