背景：置入外周静脉留置针是患儿住院期间最常见的侵入性操作之一，静脉炎、外渗、渗出、堵管、意外脱管等并发症的发生率达18.9%~71.0%，从而导致外周静脉留置针的过早拔除，增加住院期间医疗费用，给患儿带来再次置针的疼痛和恐惧。作为重要的患儿安全问题，识别风险因素并采取积极有效的前馈管理可降低外周静脉留置针并发症对住院患儿造成的不良影响，而目前国内研究对风险因素及其交互效应挖掘不够充分，故有必要对数据进行深入分析和挖掘，为提高住院患儿外周静脉留置针管理质量提供科学依据。

目的：（1）通过横断面研究调查住院患儿外周静脉留置针并发症现状；（2）分析住院患儿外周静脉留置针并发症严重度的风险因素，基于SPSS 24.0 CART决策树算法原理构建住院患儿外周静脉留置针并发症严重度风险预测模型。

方法：本研究为描述性横断面研究，便利选取2020年11月至2021年9月北京市2家三级甲等医院儿科病房符合入选标准的303名住院患儿。采用自行设计的外周静脉留置针并发症风险因素收集表记录患儿风险因素及并发症信息。应用SPSS 24.0 对数据进行统计和分析，基于SPSS 24.0 CART决策树算法原理，构建住院患儿外周静脉留置针并发症严重度风险因素预测模型。

结果：（1）303名患儿中并发症发生率最高的为渗出，130例（42.9%）；其次为渗血或渗液，67例（22.1%）；堵管51例（16.8%）；静脉炎47例（15.5%）；意外脱管8例（2.7%）。并发症轻度组121例（39.9%），中度及以上组182例（60.1%）。（2）两组患儿中，单因素分析有统计学意义（P＜0.05）的变量有17个：年龄分期、科室、疾病、输液接头、穿刺部位、血管状态、置针过程、敷料是否无张力粘贴、固定是否有效、穿刺过程中患儿配合、输注方式、留置期间患儿配合、留置期间评估及时全面、冲封管时机、是否及时更换敷料、每日输液时长、输液速度。（3）基于SPSS 24.0 CART决策树算法原理，构建住院患儿外周静脉留置针并发症严重度风险预测模型，输出决策树风险预测模型深度5层、叶节点18个、决策路径18条，筛选出10个决策因素：固定是否有效、穿刺部位、置针过程、疾病、血管状态、每日输液时长、穿刺过程中患儿配合、输液速度、年龄分期、留置期间患儿配合，其中固定是否有效是最重要的决策因素。风险预测模型的准确率为78.55%，灵敏度为78.57%，特异度为78.51%。

结论：基于决策树算法构建住院患儿外周静脉留置针并发症严重度风险预测模型具有良好的预测效能，为今后我国住院患儿外周静脉留置针并发症风险因素预防与管理提供了理论和实践参考。

Background: Inserting Peripheral Intravenous Catheters (PIVCs) is one of the most common invasive procedures on hospitalized children. The complications of PIVCs include phlebitis, infiltration, extravasation, occlusion and accidental removal. The incidence rate of them fluctuates between 18.9% and 71.0%, which result in the need of replacing PIVCs before their normal life cycles; increasing medical expenses and aggravating the pain and fear for children. As an important patient safety problem for hospitalized children, identifying risk factors of PIVCs complications, and taking effective management in advance can reduce the negative influence. Currently, the interaction effect of risk factors is explored insufficient in domestic studies. It is necessary to conduct in-depth data analysis and mining to provide references to improve the quality of PIVCs for hospitalized children.

Objectives: (1) To investigate the current situation of PIVCs complications for hospitalized children. (2) To analyze the risk factors and construct a risk prediction model for the severity of PIVCs complications for hospitalized children based on the principle of SPSS 24.0 CART decision tree.

Methods: This study is a descriptive cross-sectional study. The study cases include 303 hospitalized children who met the selection criteria in three pediatric wards of two tertiary hospitals in Beijing during the period from November 2020 to September 2021. In this study, a self-designed table, which is called Risk Factors Collection Table, was used for recording the collected risk factors and complication information. Also, this study used SPSS 24.0 tool to analyze data, then it built a prediction model for the severity of PIVCs complications for hospitalized children based on the principle of SPSS 24.0 CART decision tree.

Results: (1) The incidence of rate PIVCs complications among 330 hospitalized children: 130 participants (42.9%) experienced a complication with infiltration, followed by leakage (22.1%), occlusion (16.8%), phlebitis (15.5%), and accidental removal (2.7%). 121 cases (39.9%) of them were in the mild complication group and 182 cases (60.1%) were in the moderate or above groups. (2)There are 17 variables showing statistical significance (P＜0.05) in the univariate analysis of hospitalized children in these two groups. They are age stage, inpatient ward, disease, transfusion joint, puncture sites, vascular status, insertion process, dressing tension-free, the valid fixation of PIVCs, cooperation of children during the puncture process, the method of transfusion, cooperation of children during the indwelling period, evaluation was timely, comprehensive during indwelling period, the condition of sealing, dressing is changed in time, the daily length of infusion and the infusion speed. (3) Based on the principle of SPSS 24.0 CART decision tree, a risk prediction model of PIVCs complications for hospitalized children was constructed. The exported decision tree in the model had 5 layers, 18 leaf nodes, and 18 combination rules. A total of 10 risk factors were selected: the valid fixation of PIVCs, puncture sites, insertion process, disease, vascular status, the daily length of infusion, cooperation of children during the puncture process, the infusion speed, age stage and the cooperation of children during the indwelling period. The result showed that the valid fixation of PIVCs was the most important factor. Regarding the performance of this risk prediction model: the accuracy is 78.55%; the sensitivity is 78.57%; and the specificity is 78.51%.

Conclusion: Based on the CART decision tree algorithm, the risk prediction model of PIVC complications’ severity for hospitalized children has good predictive efficacy, which provides a theoretical and practical reference for preventing and managing PIVCs complications for hospitalized children in China in the future.