JMIR Publications

ABSTRACT

Background:

Surgical residency training faces mounting pressure from expanding subspecialty knowledge requirements alongside compressed learning opportunities due to duty-hour restrictions and increasing clinical demands. Microlearning, defined as brief, focused educational episodes, offers a pedagogically sound approach suited to fragmented clinical schedules, while large language models (LLMs) present novel capabilities for scalable content generation. However, the feasibility, quality, and educational efficacy of LLM-generated microlearning in surgical training remain unestablished.

Objective:

This pilot study aimed to (1) assess the feasibility and acceptability of an AI-generated microlearning intervention among plastic surgery residents, (2) evaluate content quality through systematic expert faculty validation, and (3) generate preliminary efficacy estimates for knowledge and confidence improvements to inform future definitive trials.

Methods:

A single-arm pilot feasibility study with pre-post design was conducted at an academic plastic surgery residency program over 12 weeks (July–September 2025). Six microlearning modules covering core subspecialties (Hand Surgery, Reconstructive & Microsurgery, Burns & Critical Care, Aesthetic & Breast Surgery, Sarcoma & Oncologic Surgery, Core Surgical Practice) were developed using Google Gemini 2.5 Pro. From 60 AI-generated multiple-choice questions, 42 were selected following independent validation by two board-certified plastic surgeons using a standardized 5-dimension rubric (accuracy, relevance, clarity, pedagogical value, safety). All 11 program residents were enrolled; modules were released bi-weekly via mobile-accessible platform. Outcomes included retention, module completion, acceptability ratings, content quality metrics, and pre-post changes in self-reported knowledge and confidence (7-point Likert scales). Wilcoxon signed-rank tests and Cohen's d effect sizes were calculated for matched participants (n=9).

Results:

Retention was 100% (11/11), with module completion ranging from 82%–100% across subspecialties. Total time investment averaged 60–120 minutes over 12 weeks. Acceptability was high: usability (mean 5.9/7), training value (mean 5.3/7), and willingness to continue (mean 5.6/7; 82% favorable). Expert validation confirmed robust content quality (mean overall quality index 4.68/5.0), with safety ratings remaining consistently high (mean 4.80/5.0) independent of other dimensions. Educational outcomes showed medium-to-large effects: composite knowledge improved by 0.33 points (Cohen's d=0.71, p=0.094) and composite confidence by 0.52 points (d=0.65, p=0.047). Topic-specific gains were largest where baseline knowledge was lowest, with Hand Surgery demonstrating significant improvements in both knowledge (d=1.48, p=0.016) and confidence (d=1.41, p=0.016). Weekly study time, traditional resource utilization, and peer learning community measures remained stable throughout the intervention.

Conclusions:

AI-generated microlearning modules are feasible, acceptable, and potentially efficacious for surgery residency education. Expert-validated LLM content achieved consistently high quality while residents demonstrated strong engagement and meaningful learning gains with minimal time investment. The intervention integrated seamlessly into existing learning ecosystems without disrupting traditional resources or peer dynamics. These findings support proceeding to adequately powered randomized trials comparing AI-generated microlearning to standard educational approaches.