Prediction of Lupus Classification Criteria via Generative AI Medical Record Profiling.

Area of Science:

• Rheumatology
• Artificial Intelligence
• Medical Informatics

Background:

• Systemic Lupus Erythematosus (SLE) presents diagnostic challenges due to heterogeneous patient symptoms and severity.
• Generative artificial intelligence (genAI) and large language models (LLMs) offer potential solutions for complex medical record analysis.
• Automated patient profiling can assist in assessing key medical criteria for SLE classification.

Purpose of the Study:

• To evaluate the efficacy of genAI in profiling patient medical records for Systemic Lupus Erythematosus (SLE) classification.
• To assess genAI's accuracy in determining SLE classification criteria based on the ACR 1997 guidelines.
• To determine the overall predictive success rate of genAI in classifying SLE patients.

Main Methods:

• Utilized generative artificial intelligence (genAI) with large language models (LLMs) to process patient medical records.
• Developed LLM prompts based on the American College of Rheumatology (ACR) 1997 SLE classification criteria.
• Computationally profiled records from 78 patients across five genAI replicate runs.

Main Results:

• GenAI achieved perfect concordance with clinical classification for "Discoid Rash" and "Pleuritis or Pericarditis" criteria.
• Accuracy for "Immunologic Disorder" criterion was 56%, indicating statistical unreliability for some factors.
• The genAI approach demonstrated a 72% overall predictive success rate compared to clinical classification.

Conclusions:

• GenAI-based patient profiling shows potential as a tool to assist clinicians in evaluating SLE patients.
• The accuracy of genAI criteria assessment correlates inversely with the complexity of clinical determination.
• Advancements in clinical classification efficacy may drive improvements in AI-driven patient profiling tools for SLE.