目的 分析1型糖尿病（type 1 diabetes mellitus，T1DM）患儿发生酮症酸中毒的危险因素，通过决策树算法建立T1DM患儿酮症酸中毒的风险预测模型。方法 回顾性分析2021年2月—2024年9月收治的218例T1DM患儿的临床资料，根据患儿是否发生酮症酸中毒将其分为酮症酸中毒组和非酮症酸中毒组，采用Logistic回归分析筛选T1DM患儿发生酮症酸中毒的危险因素，通过R语言构建预测T1DM患儿发生酮症酸中毒的随机森林模型，采用5折交叉验证法对模型进行内部验证，并对比模型的预测效能。结果 该研究218例T1DM患儿中，有66例患儿发生了酮症酸中毒，酮症酸中毒发生率为30.28%；单因素分析结果显示，年龄、入院随机血糖、糖化血红蛋白、前驱感染、甲状腺功能等差异有统计学意义（P<0.05）；Logistic回归结果表明，年龄增加是T1DM患儿发生酮症酸中毒的保护因素，而入院随机血糖升高、糖化血红蛋白升高、有前驱感染、甲状腺功能异常是T1DM患儿发生酮症酸中毒的危险因素（P<0.05）；模型预测总体正确性为84.2%；内部验证显示，模型预测正确率为77.6%；随机森林模型预测T1DM患儿酮症酸中毒发生的AUC值与Logistic回归模型相近，两种模型均具有较好的预测效能。结论 该研究构建的随机森林模型可预测T1DM患儿发生酮症酸中毒风险，有助于护理人员根据不同因素的重要性排序制订相应的护理干预策略。

Objective To analyze the risk factors of ketoacidosis in children with type 1 diabetes mellitus （T1DM），and establish a risk prediction model of ketoacidosis in children with T1DM by using decision tree algorithm. Methods The clinical data of 218 children with T1DM treated from February 2021 to September 2024 were retrospectively analyzed. The children were divided into ketoacidosis group and non-ketoacidosis group according to whether they had ketoacidosis. The risk factors of ketoacidosis in children with T1DM were screened by Logistic regression analysis. A random forest model for predicting ketoacidosis in children with T1DM was constructed using R language，and the model was internally verified by using 5-fold cross-validation method，and the prediction efficiency of the model was compared. Results Among 218 children with T1DM，66 cases had ketoacidosis，and the incidence of ketoacidosis was 30.28%. Univariate factor results showed that age，blood glucose at admission，glycated hemoglobin，preinfection，thyroid function had statistically differences（P<0.05）. The Logistic regression results indicated that increased age was a protective factor for ketoacidosis in children with T1DM，while random elevation of blood glucose at admission，elevated glycated hemoglobin，presence of propositional infection，and thyroid dysfunction were risk factors for ketoacidosis in children with T1DM（P<0.05）. The overall accuracy of the model prediction was 84.2%. Internal verification showed that the prediction accuracy of the model was 77.6%. The AUC value of random forest model in predicting the occurrence of ketoacidosis in children with T1DM was similar to that of Logistic regression model，and both models had better predictive efficacy. Conclusion The random forest model constructed in this study can predict the risk of developing ketoacidosis in children with T1DM，which is helpful for nursing staff to formulate corresponding nursing intervention strategies according to the importance of different factors.