Abstract: Objective  To construct a risk prediction model for moderate to severe pruritus in hemodialysis patients based on machine learning algorithms, and to screen and verify the optimal model.  Methods  A total of 510 patients who underwent maintenance hemodialysis in the blood purification center of our hospital from January 2020 to June 2025 were studied. Logistic regression (LR), random forest (RF), lightweight gradient boosting machine (LightGBM), extreme gradient boosting (XGBoost) and support vector machine (SVM) were used to establish the prediction models. The performance of the models was evaluated by comparing the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, calibration curve and decision curve.  Results  Finally, six factors including intact parathyroid hormone (iPTH), blood phosphorus, dry skin, sleep quality score, anxiety score and depression score were included to construct a risk prediction model. The XGBoost prediction model had the highest AUC in the modeling group and the validation set, which were 0.894 (95% CI: 0.828～0.961) and 0.887 (95% CI: 0.801～0.973), respectively; the accuracies were 0.837 and 0.823, respectively, the sensitivities were 0.508 and 0.492, respectively, the specificities were 0.951 and 0.945, respectively, and the F1 scores were 0.614 and 0.591, respectively. The calibration curve was close to the ideal curve, and the decision curve analysis (DCA) showed that the model had a high net benefit in 8%～78% threshold probability range, indicating high clinical applicability.  Conclusion  The XGBoost model can effectively predict the risk of moderate to severe pruritus in hemodialysis patients, which is expected to provide a basis for early clinical warning and individualized intervention.

Key words: Hemodialysis, Moderate to severe itching, Machine learning, Prediction model