Kyle Wang
Research Mentor: Lam Tsoi
Mentor Department: Dermatology / Computational Medicine and Bioinformatics, Medicine
Author(s): Kyle Wang, Lam Tsoi, Matthew Patrick
Session: Session 6 (3:00 PM – 3:50 PM)
Presentation Type: Poster 127
Abstract
The risk of developing psoriatic arthritis (PsA) involves complex family and clinical factors. However, tracking exactly how patients progress from psoriasis to PsA over time remains difficult using population-level data. This study uses large-scale electronic health record (EHR) data from Michigan Medicine to map PsA risk. We used a large language model (LLM) to extract family history from clinical notes and built timelines to track how related conditions develop. An LLM processed clinical notes for 67,988 patients to identify a family history of PsA. We then linked these patients to their ICD-10 billing codes to see what other inflammatory conditions they had. To evaluate longitudinal risk, we built a timeline-based computational pipeline tracking how newly diagnosed psoriasis patients developed other conditions (like PsA and type 2 diabetes) after taking biologic or topical medications. To ensure accurate timelines and remove data bias, the pipeline required patients to have at least one year of medical history before their psoriasis diagnosis. We also excluded early joint pain, restricted drug exposure to the first year, and tracked their true last medical visit across 4- to 10-year periods. We used propensity score matching (PSM) to ensure we were comparing patients with similar baseline clinical characteristics. Among the 365 patients with a family history of PsA, 94.2% did not have a personal PsA diagnosis. However, 12.8% had psoriasis and 22.7% had rheumatoid arthritis, showing that a family history of PsA is linked to a broader risk for inflammatory diseases. Over time, when we balanced the biologic and topical treatment groups using PSM, the difference in how many patients developed PsA became very small (12.9% for biologics vs. 10.8% for topicals). When looking at type 2 diabetes as a separate long-term condition over 10 years, the rates of developing the disease were practically the same between biologic and topical users. In conclusion, while a family history indicates an early risk for inflammatory diseases, the long-term development of conditions like PsA and type 2 diabetes appears to converge over time, regardless of whether a patient initially received biologic or topical therapies. This computational approach provides a strong foundation for future research to better predict and personalize care for patients with autoimmune diseases.
