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

Background: Psoriasis is a chronic inflammatory skin disease with a variable course. Early identification of patients likely to require biologic therapy may help reduce complications and optimize care. In this study, we developed machine learning (ML) models to predict future biologic therapy use in psoriasis patients. Methods: We conducted a retrospective study using electronic health records (EHR) from Clalit Health Services in Israel, including psoriasis patients who started biologic therapy and matched psoriasis controls. Predictors included demographics, comorbidities, treatment history, and laboratory test results. KNN, SVM, Random Forest, and Logistic Regression ML models were trained on data from either the first five years post-onset or the five years preceding biologic therapy. Performance was evaluated on a held-out test set using AUC-ROC, precision, recall, and F1-score, with an emphasis on recall to maximize identification of true positive cases. Results: The best-performing models incorporated clinical, demographic, and laboratory data. Using data from the first five years after onset, the SVM model achieved the highest performance (AUC = 0.83, recall = 0.7). For data from the five years preceding biologic therapy, the Random Forest model performed best (AUC = 0.93, recall = 0.95). Key predictors included comorbid immune-mediated conditions, topical treatment frequency, and markers of inflammation and metabolism. Conclusions: EHR-based ML models, particularly those incorporating routine laboratory, demographic, and clinical data, can effectively predict future biologic therapy use in psoriasis patients. Model performance may be improved with larger cohorts and more complete clinical and laboratory data.