Generative Artificial Intelligence in Medical Education: A Systematic Review an d Meta-Analysis of Randomized Controlled Trials

Cheng-bei LU^1*Yimei Hu^2*Rui KANG¹Zheng LIU¹

• ¹School of Management, Chengdu University of Traditional Chinese Medicine, Chengdu, China
• ²Office of Academic Affairs, Chengdu University of Traditional Chinese Medicine, Chengdu, China

Objective: This study aimed to evaluate the effectiveness of generative artificial intelligence (GAI) in medical education and provid e evidence-based support for its broader implementation. Methods: Randomized controlled trials (RCTs) were systematically searched in databases including PubMed, Web of Science, The Cochrane Library, Embase, CNKI, WanFang Data, and VIP, covering publications from January 2023 to March 2025. Eligible studi es involved medical students, residents, or trainees, with interventions utilizing Generative AI-based teaching tools or methods comp ared to traditional teaching approaches. Outcome measures included theoretical scores, operational skills, and teaching satisfaction. D ata analysis was performed using RevMan5.4 and Stata17 software to calculate effect sizes, 95% confidence intervals (CIs), heterog eneity, and publication bias. Subgroup analyses were conducted based on AI types and learning stages. Results: Eleven studies (total n = 770; participants: medical students) were included. These studies were conducted across four cou ntries (the United States, China, Turkey, and Germany). Meta-analysis demonstrated that Generative AI significantly improved theor etical exam scores [SMD = 1.03; 95% CI (0.64, 1.42); P < 0.001], operational skills [SMD = 0.64; 95% CI (0.32, 0.96); P < 0.0 01], and teaching satisfaction [RR = 1.28; 95% CI (1.10, 1.49); P = 0.002; SMD = 0.97; 95% CI (0.53, 1.40); P < 0.001]. Subgr oup analyses indicated that AI modality and participants' learning stages were potential moderators of effect sizes and heterogeneit y. Conclusion:Generative AI demonstrates substantial potential in enhancing the efficiency and effectiveness of medical education. Ho wever, the overall findings are tempered by considerable heterogeneity across the included studies—a factor that may undermine th e generalizability of the pooled estimates. Accordingly, additional high-quality randomized controlled trials, particularly those with st andardized protocols and diverse educational contexts, are essential to confirm the robustness and broader applicability of these find ings.