Studies on AI Use for Residents and Medical Learners in Clinical Settings
Educational Integration and Training
The evidence demonstrates that AI education for medical learners remains underdeveloped, with systematic reviews revealing minimal formal AI training in current medical curricula despite growing recognition of its necessity. 1
Current State of AI Education
Medical and health informatics students require systematic AI training that has not yet been standardized, with recent publications recommending integration of AI competencies into curricula but lacking defined frameworks for specialized training. 1
A cross-sectional study of 265 graduating medical students revealed that only 20.4% used ChatGPT to complete written assessments and merely 9.4% utilized it in clinical work during medical school, indicating minimal practical exposure during training. 2
Students planned to use AI more during residency (63.4% for exploring medical topics and 57% for exam preparation), suggesting anticipated increased utilization despite limited formal preparation. 2
Male students demonstrated significantly more positive perceptions of AI's potential, with 51.7% believing AI will improve diagnostic accuracy compared to 39.7% of female students (P=0.001). 2
Competency Requirements
Physicians require both critical human skills and growing technical/digital competencies for AI-assisted clinical settings, though concrete guidance remains ambiguous and requires further specification. 3
Dissensus exists regarding whether physicians are adequately equipped to use and monitor AI in clinical settings in terms of competencies, skills, expertise, and normative guidance ownership. 3
Clinicians and health informaticians need strong backgrounds in data analytics, data visualization, machine learning, and deep learning to effectively use, evaluate, and develop AI applications in clinical practice. 1
Implementation Challenges in Clinical Learning Environments
Structural Barriers
AI development often focuses on innovative technologies (tech-push) rather than clinical needs (demand-pull), resulting in limited success in large-scale clinical implementation that affects learner exposure. 4
Current AI initiatives generate knowledge without clear implementation paths once proof-of-concept is achieved, limiting opportunities for learners to engage with functional AI systems. 4
Lack of multidisciplinary collaboration hinders AI integration, with stakeholders including learners often involved too late and facing challenging communication barriers. 4
Knowledge and Skills Gaps
Stakeholders including IT departments and healthcare professionals often lack required skills and knowledge for effective AI integration, creating an inadequate learning environment for residents and students. 4
Hospitals need clear vision for integrating AI, including meeting preconditions in infrastructure and expertise, which directly impacts the quality of AI-related clinical education. 4
Structured curricula and formal policies are needed to adequately prepare medical learners for the forthcoming integration of AI in medicine. 2
Reporting and Evaluation Guidelines for AI Studies
DECIDE-AI Framework
The DECIDE-AI reporting guideline comprises 17 AI-specific reporting items (28 subitems) and 10 generic items developed through multi-stakeholder consensus for early-stage clinical evaluation of AI decision support systems. 5
Early-stage clinical evaluation is critical to assess AI system's actual clinical performance at small scale, ensure safety, evaluate human factors, and pave the way for larger trials - directly relevant to resident learning environments. 5
Few AI tools have demonstrated real benefit to patient care despite promising preclinical performance, highlighting the "AI chasm" that affects educational integration. 5
Key Evaluation Considerations
AI systems must account for complex intervention nature, user variability, human-computer collaboration, and changing system versions - all factors that impact learner interaction and education. 5
Learning curves should be analyzed by graphically plotting user performance against experience, providing specific metrics for assessing resident and learner competency development with AI tools. 5
Both "live evaluation" (affecting patient care) and "shadow mode" (not affecting care) should be distinguished, with implications for appropriate learner involvement at different training stages. 5
Clinical Applications Relevant to Learner Training
Cardiovascular Medicine
AI applications in cardiovascular imaging include automated segmentation, volumetric analysis, ejection fraction calculation, and automated disease detection - specific skills residents must learn. 5
The American Heart Association emphasizes that AI should enable precision medicine approaches while addressing bias, facilitating education, and ensuring implementation across training programs. 5, 6
Cancer Care
AI methods in oncology demonstrate robustness leading to improved clinical decision-making across cancer types, providing concrete learning opportunities for oncology residents. 7
Deep learning models provide clinician-level interpretation of medical imaging across CT, MRI, mammography, and digital pathology - competencies that must be integrated into radiology and pathology training. 7
Critical Gaps and Future Directions
Previous AI experience was significantly associated with positive AI perception (P=0.001 for improving patient care, P<0.001 for decreasing errors), emphasizing the importance of early exposure during training. 2
The lack of appropriate reporting guidelines obstructs reproducibility of published studies, creating challenges for evidence-based AI education. 5
AI development should incorporate patient-centered outcomes research (PCOR) principles to ensure tools address meaningful clinical questions - a framework that should guide learner education. 6
Multidisciplinary teams including bioinformatics experts, medical specialists, and patient representatives should develop AI tools, providing diverse learning opportunities for trainees. 6