Predictive Modeling of the Risk of Hyperglycemia in Psoriasis Patients Using Machine Learning: A Multicenter Retrospective Study.

Purpose: This study aims to develop and validate machine learning models for predicting hyperglycemia risk in psoriasis patients. Methods: Clinical data from 575 psoriasis patients admitted to the Department of Dermatology were collected and randomly split into a training set and an internal test set in a 7:3 ratio. An external test set was derived from 135 psoriasis patients enrolled in the National Health and Nutrition Examination Survey (NHANES) during 2003-2004 and 2011-2012. Eleven machine learning algorithms, including decision trees, random forests, extreme gradient boosting (XGboost), light gradient boosting machine, support vector machines, multilayer perceptron, K-nearest neighbors, logistic regression, lasso regression, ridge regression, and elastic net, were systematically compared. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), calibration curves and clinical decision curve (DCA). Results: The extreme gradient boosting (XGboost) was selected as the final predictive model due to its robust performance across multiple evaluation metrics. The area under the curve values for the training, internal, and external test sets were 0.821 (95% confidence Interval (CI): 0.775-0.866), 0.820 (95% CI: 0.751-0.888), and 0.788 (95% CI: 0.695-0.881), respectively. Calibration and clinical decision curve analysis confirmed the model's accuracy and clinical utility. Additionally, a web-based calculator was developed to improve the model's accessibility and application. Conclusion: The XGBoost-based model effectively predicts hyperglycemia risk in psoriasis patients, emphasizing personalized treatment plans for high-risk individuals to manage hyperglycemia progression and psoriasis-related inflammation.