Surgical Infections Curriculum Development Enhanced by Artificial Intelligence

Authors:
Kwame Wiafe, Anthony Dwyer, Neil Rubert, Laura Brown

Body of Abstract:
BACKGROUND

The explosion of medical knowledge makes it impossible to maintain current surgical infection education without new tools or approaches. Guidelines and reviews can lag behind recently published literature and require significant effort to produce. This study aims to explore whether Artificial Intelligence (AI) can be effectively used to improve the curation and development of an up-to-date, module-based curriculum, starting with a pilot topic focused on the prevention and treatment of superficial surgical site infections.

METHODS

Several Large Language Models (LLMs) were used to conduct targeted research on the treatment and prevention of superficial surgical infections. Each model was instructed to precisely identify high-quality, evidence-based articles, including randomized controlled trials, practice guidelines, systematic reviews, and primary research studies. The output from each model was compared to PubMed search results to generate articles that met quality standards (peer-review, availability, journal reputation) to reduce the likelihood of insufficient/incorrect data or AI-related hallucinations. The identified articles were uploaded into a specific LLM known for processing clinical evidence to establish a Retrieval-Augmented Generation (RAG) framework. Specific prompts and a content outline were used to minimize AI-related bias and misinterpretation. This RAG system was then used to produce the specialized content and code for a web-based application that hosts a structured, module-based curriculum.

RESULTS

LLMs effectively synthesized the curated medical literature to develop a comprehensive, module-based curriculum. The final web application features six educational modules dedicated to the pilot topic on superficial surgical site infections (Figure 1). Each curriculum includes five AI-generated clinical case scenario questions amenable to psychometric data collection for content validation. The LLMs provided a ready-to-launch file that can be easily deployed on any network.

CONCLUSION

We have successfully demonstrated the feasibility of using AI to generate structured educational didactic materials based on current medical literature curated by the LLMs. Once openly reviewed by a community of experts, the resulting curriculum will provide surgeons with an easily accessible, evidence-based resource that can be quickly deployed on any network; thus, offering a scalable solution for keeping surgical education current.