Natural language processing to identify and characterize spondyloarthritis in clinical practice

Area of Science:

• Rheumatology
• Medical Informatics
• Natural Language Processing

Background:

• Spondyloarthritis (SpA) is a group of inflammatory diseases affecting the spine and joints.
• Characterizing the clinical profile of SpA patients is crucial for effective management.
• Electronic Health Records (EHRs) contain vast amounts of patient data, but extracting information from unstructured text is challenging.

Purpose of the Study:

• To utilize a novel Natural Language Processing (NLP) technology to extract clinical information from EHRs.
• To characterize the clinical profile of patients diagnosed with SpA at a large hospital.
• To evaluate the performance of the NLP technology in detecting SpA clinical entities.

Main Methods:

• An observational, retrospective analysis of EHR data from SpA patients (2020-2022).
• Data extraction using Savana Manager, an NLP-based system for unstructured EHRs.
• Analysis of demographic data, SpA subtypes, comorbidities, and treatments; performance evaluation using precision, recall, and F-1 scores.

Main Results:

• 0.7% of the hospital population (4337 patients) had SpA diagnoses.
• The SpA cohort was predominantly male (55.3%) with a mean age of 50.9 years.
• Common comorbidities included hypertension (25.0%) and dyslipidaemia (22.2%). Methotrexate (25.3%) and adalimumab (10.6%) were the most used drugs. NLP demonstrated high precision and recall (F-1 scores > 0.80).

Conclusions:

• NLP technology effectively characterized the SpA patient profile, including demographics, comorbidities, and treatments.
• The study supports the utility of NLP in enhancing the understanding of SpA.
• NLP has the potential to improve SpA patient management by extracting meaningful data from unstructured EHRs.