Machine Learning-Based Predictive Model With Routine Blood Work Identifies Moderate-Severe Alopecia Areata

This study explored the use of routine blood work to predict moderate-to-severe alopecia areata (AA) using machine learning. The analysis included 53 AA patients and 111 healthy controls, revealing that AA patients had significantly higher BMI and total serum protein levels. The machine learning model, utilizing the XGboost library, achieved an AUC of 0.76, indicating a moderate ability to distinguish AA patients from controls. The top predictors of AA were HbA1c, hematocrit, and red blood cell distribution width. These findings suggest that moderate-to-severe AA has a systemic component with significant metabolic differences, which may not be evident through univariate analysis alone.