Related Experiment Video
Updated: Jul 6, 2026
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
Undergraduate medical students and artificial intelligence: A global systematic review and meta-analysis using a
Paolo Visci1, Francesco Lupelli1, Francesco Calò1
1Section of Legal Medicine, Department of Interdisciplinary Medicine, Bari Policlinico Hospital, University of Bari, Bari, Italy.
Background:
Artificial intelligence (AI), including deep learning and large language models, is increasingly reshaping healthcare and medical education. Since the public release of generative AI tools such as ChatGPT, medical students have become a particularly exposed group, often engaging with these technologies without standardized curricular guidance. Although previous reviews have explored readiness, acceptance, or anxiety, the relationship between students' knowledge, attitudes, and practical engagement with AI has not been comprehensively synthesized. This study therefore examined undergraduate medical students' engagement with AI using a Knowledge-Attitude-Practice (KAP) framework.
Methods:
This systematic review and meta-analysis was reported in accordance with PRISMA 2020. The research question was structured using the PCC framework. PubMed, Scopus, and Web of Science were searched up to September 30, 2025. Eligible studies included cross-sectional and psychometric validation studies addressing AI-related perceptions among undergraduate medical students. Methodological quality was assessed using the AXIS tool. Because of substantial heterogeneity in instruments and reporting, quantitative synthesis was restricted to 11 studies using the MAIRS-MS scale and reporting comparable percentage-based outcomes. Pooled estimates were calculated using Freeman-Tukey double arcsine transformation and a random-effects DerSimonian-Laird model.
Results:
Forty-three studies met the inclusion criteria, including 37 cross-sectional and 6 psychometric validation studies. Attitudes were the most frequently explored dimension, followed by knowledge and practice. In the 11 studies included in the quantitative synthesis, pooled estimates showed moderate knowledge (0.55; 95% CI: 0.49-0.61), positive attitudes (0.64; 95% CI: 0.59-0.69), and comparable practice levels (0.63; 95% CI: 0.59-0.67). Heterogeneity remained high, while Egger's test showed no significant small-study effects.
Conclusions:
Medical students appear receptive to AI, but their knowledge remains moderate and their practical engagement reflects this limited preparation. These findings support the need for structured undergraduate AI curricula integrating theoretical, applied, and ethical-legal competencies.