Early Risk Prediction for Biologic Therapy in Psoriasis Using Machine Learning Models Based on Routine Health...

Area of Science:

• Dermatology
• Medical Informatics
• Machine Learning

Background:

• Psoriasis is a chronic inflammatory skin condition with unpredictable progression.
• Early identification of patients needing biologic therapy is crucial for managing complications and optimizing care.
• Predictive modeling using electronic health records (EHR) can aid in early identification.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting future biologic therapy use in psoriasis patients.
• To identify key predictors of biologic therapy initiation in psoriasis.
• To assess the utility of EHR data for predictive modeling in psoriasis management.

Main Methods:

• Retrospective study utilizing EHR data from Clalit Health Services.
• Development and comparison of KNN, SVM, Random Forest, and Logistic Regression models.
• Models trained on data from the first five years post-onset or five years preceding biologic therapy.
• Performance evaluation using AUC-ROC, precision, recall, and F1-score, prioritizing recall.

Main Results:

• The best-performing models integrated clinical, demographic, and laboratory data.
• SVM model achieved AUC=0.83, recall=0.7 using early post-onset data.
• Random Forest model achieved AUC=0.93, recall=0.95 using pre-biologic therapy data.
• Significant predictors included comorbidities, topical treatment frequency, and inflammation/metabolism markers.

Conclusions:

• EHR-based ML models effectively predict biologic therapy use in psoriasis.
• Models incorporating routine laboratory, demographic, and clinical data show high predictive performance.
• Larger datasets and more comprehensive data may further enhance model accuracy.