Objective To investigate the influencing factors of cerebral edema in children with diabetic ketoacidosis (DKA) and construct a decision tree model.

Methods A retrospective selection of medical records from 158 children with DKA was conducted. The patients were divided into cerebral edema group (n=36) and non-cerebral edema group (n=122) based on the presence or absence of concurrent cerebral edema. General clinical data of all children were collected and analyzed, and differences in relevant indicators were compared between the two groups. Multivariate logistic regression analysis was employed toscreen for factors influencing the development of cerebral edema in children with DKA. The SPSS Modeler software was used to construct a decision tree model for predicting cerebral edema in children with DKA, and the predictive efficacy of the model was analyzed.

Results A total of 158 children with DKA were included in this study, of whom 36 developed cerebral edema, resulting in an incidence rate of 22.78%. The cerebral edema group had significantly higher proportions of patients aged ≤6 years, with blood glucose levels >20 mmol/L, pH values≤7.35, bicarbonate concentrations >10 mmol/L, slow increases in serum sodium levels, persistent hyponatremia and elevated blood urea nitrogen levels compared to the non-cerebral edema group (P < 0.05). Age≤ 6 years, blood glucose levels >20 mmol/L, pH values ≤7.35, bicarbonate concentrations>10 mmol/L, slow increases in serum sodium levels and elevated blood urea nitrogen levels were identified as independent risk factors for cerebral edema in children with DKA (P < 0.05). Seven items showing statistical differences in univariate analysis were incorporated into the decision tree model, from which six explanatory variables were selected: bicarbonate concentration >10 mmol/L, blood glucose level >20 mmol/L, persistent hyponatremia, slow increase in serum sodium levels, pH value≤7.35 and elevated blood urea nitrogen level. The decision tree model consisted of 5 layers and 13 nodes, with a slow increase in serum sodium levels being the most important predictor. The area under the curve (AUC) of the decision tree model for predicting cerebral edema in children with DKA was 0.880 (95%CI, 0.819 to 0.927), while that of the logistic regression model was 0.735 (95%CI, 0.659 to 0.802). The Delong test results for the two models were Z=2.790, P=0.005, indicating that the decision tree model had a superior AUC (P < 0.05).

Conclusion The decision tree prediction model for cerebral edema in children with DKA constructed in this study significantly outperforms the logistic regression model. This study confirms that a slow increase in serum sodium levels is the strongest predictor and establishes a three-step early warning pathway involving bicarbonate: >10 mmol/L, blood glucose >20 mmol/L and serum sodium dynamics. The model has direct clinical guiding value but requires multicenter validation. The mechanistic association between serum sodium dynamics and cerebral edema warrants further investigation.