Abstract Background Surgical training suffers from a global deficit; 5 billion people lack access to safe surgery, with an estimated 143 million additional procedures needed annually. Traditional surgical education, constrained by the apprenticeship model, faces critical limitations in standardization and scalability, particularly in low- and middle-income countries where expert mentors are scarce. AI-augmented tutoring systems represent a potentially transformative solution. This systematic review and meta-analysis were conducted to address that evidence gap. Methods Following PRISMA 2020 guidelines, we systematically searched major databases for trials comparing AI tutoring with expert instruction. Primary outcomes were performance (Intelligent Continuous Expertise Monitoring System ICEMS score) and skill acquisition (Objective Structured Assessment of Technical Skills OSATS score). Cognitive load was a secondary outcome, measured using the Mental Effort Scale (MES) and Cognitive Load Index (CLI). Results Our search yielded 40 studies for narrative synthesis. Four studies (3 RCTs and 1 pilot prospective study), encompassing a total of 268 participants, were included in the meta-analysis. AI tutoring showed a small, statistically significant improvement in expert-rated OSATS scores (MD 0.20; 95% CI, 0.01 to 0.39) with no heterogeneity (I2 = 0%) but with low certainty. The AI group reported a significantly higher extraneous cognitive load (MD 0.23; p = 0.01). No significant difference was found in ICEMS scores. Conclusion AI tutoring systems demonstrated comparable effectiveness to expert instructors in simulated surgical skill acquisition. The small OSATS advantage (0.20 points) is of uncertain clinical significance, falling below commonly published competency cut-points for meaningful change on global rating scales, and is based on low-certainty evidence driven by a single high-risk-of-bias study. AI tutoring imposed a higher extraneous cognitive load. These findings do not support replacing human instructors with AI. Instead, the evidence supports a hybrid model, though this itself requires rigorous empirical validation.
Hanna et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: