A Large Language Model Workflow for Auditable Brain Abscess Risk Stratification and Pre-residency Scholarship: A...

Area of Science:

• Medical Education
• Artificial Intelligence in Medicine
• Scholarly Workflows

Background:

• Medical trainees need transparent methods for integrating large language models (LLMs) into scholarly work.
• Existing artificial intelligence (AI) literacy curricula lack auditable frameworks for LLM use.
• Developing structured AI use skills is crucial for pre-residency readiness.

Purpose of the Study:

• To describe a mentored educational framework for teaching AI literacy and structured LLM use to medical trainees.
• To transform a clinical case report into a reproducible, LLM-supported risk-stratification model.
• To foster rigor, verification, and transparent AI engagement in scholarly activities.

Main Methods:

• A de-identified brain abscess case was reconstructed using structured clinical variables.
• LLM interactions were logged in a standardized prompt ledger, capturing rationale and decisions.
• A provisional Neurologic Deterioration in Brain Abscess Score (NDBAS v0.1) was developed based on literature-supported indicators.

Main Results:

• The LLM-assisted synthesis accelerated evidence review compared to manual methods.
• The framework produced three curriculum artifacts: a case appendix, prompt ledger, and variable dictionary.
• The index case received an NDBAS v0.1 score of 5, indicating moderate risk.

Conclusions:

• This structured, audit-ready LLM workflow enhances AI literacy and scholarly output for medical trainees.
• The 'Prompt Ledger plus ask-verify-revise' pattern facilitates transparent evidence synthesis and conceptual model-building.
• Mentored LLM workflows, with appropriate governance, can prepare learners for responsible AI engagement in clinical training.