Harnessing artificial intelligence to improve diagnostic precision of CTCL

Summary: Abstract Body: Cutaneous T-cell lymphoma (CTCL) may mimic eczema or psoriasis, contributing to diagnostic delay and improper treatments. This retrospective study utilizes natural language processing (NLP) and machine learning (ML) to extract and analyze patient histories from notes prior to diagnosis and develop predictive models for earlier and more accurate CTCL diagnosis. Using electronic health records, unstructured notes were analyzed via NLP to identify terms associated with CTCL versus controls with benign inflammatory skin conditions. Pre-diagnosis notes were reviewed for presentations, treatments, and laboratory results. ML models were then developed using patient histories and laboratory data. Key findings included significantly (p<0.05) higher frequencies of terms associated with failed topical steroids, scaling, patches, plaques, tumors, and history of multiple biopsies in CTCL patients compared to controls. Conversely, xerosis, allergies, and systemic biologic use were significantly (p<0.05) more frequent in controls than CTCL patients. NLP achieved an accuracy of 82.8% at identifying terms in patient notes. ML models distinguishing CTCL patients from controls achieved 73% accuracy (ROC 0.805) with sensitivity (87%) prioritized to minimize false negatives. An interactive tool based on these models was generated for real-time CTCL risk prediction. This study underscores the potential of AI-driven approaches to enhance CTCL diagnosis, emphasizing its utility in cases of treatment failure, inconclusive biopsies, and atypical presentations. Future research should refine these tools, validate their applicability across diverse populations, and integrate them into clinical workflows to improve diagnostic precision and patient outcomes. Emily R. Gordon¹, Sophia T. Luyten¹, Megan H. Trager², Casey Ta², Cong Liu², Thomas Litman³, Herbert Chase², Chunhua Weng², Larisa J. Geskin² 1. Columbia University Vagelos College of Physicians and Surgeons, New York, NY, United States. 2. New York-Presbyterian/Columbia University Irving Medical Center, New York, NY, United States. 3. Kobenhavns Universitet Biologisk Institut, Copenhagen, Capital Region of Denmark, Denmark. Clinical Research: Epidemiology and Observational Research