An Organized Approach to Using Large Language Models for Medical Information

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Large language models (LLMs) like ChatGPT are increasingly popular.
• Existing frameworks for medical applications of LLMs are limited.
• Standardized prompting techniques are needed for effective medical use.

Purpose of the Study:

• To establish nomenclature for LLM prompting variables and clauses in medical contexts.
• To demonstrate the utility of a structured prompting approach in healthcare.
• To outline a methodology for patient-specific LLM information generation.

Main Methods:

• Defined key terms: input prompt, variable, clause, demographic variable, independent variable, dependent variable, generative clause, and output.
• Utilized ChatGPT-4 for assessment.
• Implemented methodology with three sample patient cases from patient and physician perspectives.

Main Results:

• Precise combinations of variables and clauses yield unique, personalized LLM outputs.
• LLM outputs consider patient demographics (age, gender, weight, height, education).
• LLMs can rapidly generate comprehensive educational materials, potentially improving healthcare outcomes.

Conclusions:

• The described framework standardizes medical LLM input methodology.
• This approach enhances the potential for patient-specific information queries.
• Future research should explore specialty- and presentation-specific prompt adjustments for optimal results.