Abstract:

Objective To analyze the risk factors of metabolic acidosis in children undergoing maintenance hemodialysis and construct a decision tree model. Methods The clinical data of 205 children undergoing maintenance hemodialysis admitted to a tertiary general hospital from November 2022 to November 2024 were retrospectively selected. They were divided into metabolic acidosis group and non-metabolic acidosis group according to the occurrence of metabolic acidosis. Univariate and multivariate logistic regression analyses were used to analyze the risk factors affecting metabolic acidosis in children undergoing maintenance hemodialysis. Decision tree algorithm and logistic regression algorithm were used to construct risk prediction models，and the predictive values of the two models for metabolic acidosis in children undergoing maintenance hemodialysis were compared. Results Among the 205 children undergoing maintenance hemodialysis，101 cases developed metabolic acidosis，with an incidence rate of 49.27%. Multivariate logistic regression analysis showed that diabetes mellitus，malnutrition，blood urea nitrogen，serum unsaturated iron binding capacity and blood potassium were risk factors for metabolic acidosis in children undergoing maintenance hemodialysis（P<0.05）. A decision tree model was constructed based on the risk factors，which selected 5 explanatory variables including blood potassium，blood urea nitrogen，diabetes mellitus，serum unsaturated iron binding capacity and malnutrition，with a total of 5 layers and 22 nodes. Among them，blood potassium was the most important influencing factor for metabolic acidosis in children undergoing maintenance hemodialysis. The AUC value of the decision tree model for metabolic acidosis in children undergoing maintenance hemodialysis was 0.928（95%CI：0.883~0.959），and the AUC value of the logistic regression model was 0.901（95%CI：0.852~0.938）. The Delong test results of the two models showed Z=2.453，P<0.001. Conclusion Diabetes mellitus，malnutrition，blood urea nitrogen，serum unsaturated iron binding capacity and blood potassium are risk factors for metabolic acidosis in children undergoing maintenance hemodialysis. The decision tree risk prediction model constructed based on these risk factors has high predictive efficiency，providing a scientific and effective theoretical basis for clinical medical staff to provide more accurate prediction and implement targeted preventive measures.

Key words: Maintenance Hemodialysis, Children, Metabolic Acidosis, Decision Tree Algorithm, Risk Prediction