Identification of clinical phenotypes and disease trajectories in SLE using AI through a natural language processing...

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Informatics
• Rheumatology Research

Background:

• Electronic Health Records (EHRs) contain valuable unstructured data for patient analysis.
• Leveraging Artificial Intelligence (AI) and Natural Language Processing (NLP) can unlock insights from complex medical data.
• Systemic Lupus Erythematosus (SLE) is a heterogeneous autoimmune disease requiring detailed patient phenotyping.

Purpose of the Study:

• To develop an NLP pipeline for identifying clinical phenotypes and disease trajectories in SLE patients using EHR data.
• To create a framework combining AI and human intelligence (HI) for robust data extraction and analysis.
• To characterize SLE patient phenotypes at first contact and analyze their longitudinal disease progression.

Main Methods:

• A standardized, stepwise framework integrating AI and HI was employed.
• Ontology-based definitions were created for clinical domains, flares, and complexity phenotypes.
• An NLP pipeline was utilized to extract relevant data from EHRs of SLE patients meeting specific inclusion criteria.

Main Results:

• The study analyzed 262 SLE patients with a median follow-up of 6 years.
• Hematological, articular, cutaneous, and renal domains were frequently involved at first contact.
• A significant portion of patients (43%) presented with high-complexity phenotypes, correlating with more flares and increased medication use over time.

Conclusions:

• The developed AI-driven framework enables longitudinal phenotyping of SLE patients from real-world EHR data.
• This approach offers a feasible method for studying SLE heterogeneity and disease progression.
• The findings suggest potential applications in clinical research and improved patient management strategies for SLE.