Objective:

We propose the use of GPT-4 to facilitate initial history-taking in neurology and to demonstrate their utility as augmentable patient digital twins for personalized medicine.

Background:

An artificial intelligence (AI) large language model (LLM) could be utilized as a digital twin which would be enhanced from the electronic medical record (EMR), with additional patient data from wearables. This digital twin would then be coupled with an LLM diagnostician to provide healthcare conversational agents (HCAs) to automate smart waiting-room questionnaires.

Design/Methods:

We presented verbatim history of present illness (HPI) narratives from published case reports of headache, stroke, and neurodegenerative diseases. Three standard GPT-4 models were designated Models P: Patient digital twin; N: Neurologist to query Model P; and S: Supervisor to synthesize the N-P dialogue to summarize HPI and formulate differential diagnosis. Given the variability of GPT-4 output, each case was presented five separate times to check consistency and reliability.

Results:

Using the case report diagnosis as the standard, this method achieved 81% overall HPI content retrieval accuracy: 84% for headache subtypes, 82% for stroke subtypes, and 77% for neurodegenerative disease identification. Retrieval accuracies for individual HPI components were as follows: 93% for chief complaints, 47% for associated symptoms and review of systems, 76% for relevant symptom details, and 94% for histories of past medical, surgical, allergies, social, and family factors. The ranking of case diagnoses in the differential diagnosis list averaged in the 89th percentile.

Conclusions:

Our tripartite model — LLM digital twin patient P dialogues with LLM neurologist N with output read by LLM supervisor S — successfully extracted essential information from case report HPIs. We expect our further studies with EMR HPIs will improve due to their completeness compared to published case reports. Subsequent study with direct patient care would allow addition of real-time data from health-monitoring devices and self-monitoring assessments.