中华护理杂志 ›› 2024, Vol. 59 ›› Issue (17): 2100-2107.DOI: 10.3761/j.issn.0254-1769.2024.17.008

• 专科护理实践与研究 • 上一篇    下一篇

急性中毒患者洗胃期间误吸风险预测模型的构建及验证

张烁妮(), 王俊杰(), 张波, 刘雪兰   

  1. 315000 宁波市医疗中心李惠利医院急诊科(张烁妮,张波,刘雪兰);浙江中医药大学护理学院(王俊杰)
  • 收稿日期:2023-12-13 出版日期:2024-09-10 发布日期:2024-09-02
  • 通讯作者: 王俊杰,E-mail:wjjie2000@163.com
  • 作者简介:张烁妮:女,硕士,主管护师,E-mail:1176856452@qq.com
  • 基金资助:
    浙江省医药卫生科技计划项目(2023KY249)

Construction and validation of a prediction model of aspiration risk of acute poisoning patients during gastric lavage

ZHANG Shuoni(), WANG Junjie(), ZHANG Bo, LIU Xuelan   

  • Received:2023-12-13 Online:2024-09-10 Published:2024-09-02

摘要:

目的 分析急性中毒患者洗胃期间误吸风险影响因素,构建并验证急性中毒患者洗胃期间误吸风险预测模型。方法 通过文献检索与分析,归纳急性中毒患者洗胃期间误吸风险影响因素。以2020年1月—2023年6月在宁波市某三级甲等综合医院急诊科急性中毒洗胃患者为研究对象进行回顾性研究。采用R 4.2.1与Python 3.11统计软件分析机器学习中的随机森林、逻辑回归、极致梯度提升树和梯度提升决策树算法,建立急性中毒患者洗胃期间误吸风险预测模型并进行内部验证。通过混淆矩阵、校准曲线、受试者操作特征曲线、受试者操作特征曲线下面积、区分度指标Kolmogorov-Smirnov值、准确度、精确度、召回率和F1值对4种预测模型的预测效果进行评估,选择最佳模型。结果 4种机器学习算法建模结果显示,随机森林、逻辑回归、极致梯度提升树、梯度提升决策树算法曲线下面积分别为0.954(0.934~0.974)、0.878(0.843~0.913)、0.910(0.880~0.939)、0.917(0.889~0.945)。内部验证结果显示,随机森林、逻辑回归、极致梯度提升树、梯度提升决策树算法曲线下面积分别为0.910(0.864~0.955)、0.877(0.824~0.931)、0.849(0.790~0.908)、0.873(0.819~0.928)。年龄、意识状况、D-二聚体与吸收毒物时间是急性中毒患者洗胃期间误吸影响因素中重要性排序较为突出的4个特征。结论 4种预测模型中随机森林模型预测效果最好,对急性中毒患者洗胃期间误吸发生风险有良好的辨别能力,且临床应用便捷,可为医护人员采取预防性治疗措施和护理提供参考。

关键词: 急性中毒, 洗胃, 误吸风险, 预测模型, 护理

Abstract:

Objective To analyze the influencing factors of aspiration risk in patients with acute poisoning during gastric lavage,and to build and validation a prediction model of aspiration risk in patients with acute poisoning during gastric lavage. Methods Through literature search and analysis,the risk factors of aspiration during gastric lavage was summarized in patients with acute poisoning. A retrospective study was conducted on patients with acute poisoning in the emergency department of a tertiary A general hospital in Ningbo from January 2020 to June 2023. Through R 4.2.1 and Python 3.11 programming language,the random forest,logistic regression,extreme gradient boosting tree and gradient boosting decision tree algorithms in machine learning were used to establish a prediction model of aspiration risk during gastric lavage in patients with acute poisoning and carry out internal verification. The prediction effects of the 4 prediction models were evaluated by confusion matrix,calibration curve,receiver operating characteristic curve,area under curve,Kolmogorov-Smirnov value,accuracy,precision,recall rate and F1 score,and the best model was selected. Results The modeling results of the 4 machine learning algorithms show that the area under the curve of the Random Forest,Logistic Regression,Extreme Gradient Boosting Tree,and Gradient Boosting Decision Tree algorithms are 0.954(0.934~0.974),0.878(0.843~0.913),0.910(0.880~0.939),and 0.917(0.889~0.945),respectively. The internal validation results show that the area under the curve of the random forest,logistic regression,extreme gradient boosting tree,and gradient boosting decision tree algorithms are 0.910(0.864~0.955),0.877(0.824~0.931),0.849(0.790~0.908),and 0.873(0.819~0.928),respectively. Age,state of consciousness,D-dimer and the time of absorption of poison are the 3 characteristics that are particularly prominent in the order of importance of the influencing factors of aspiration during gastric lavage in patients with acute poisoning. Conclusion Among the 4 prediction models,random forest model has better prediction effect,with good discrimination ability for the risk of aspiration during gastric lavage in patients with acute poisoning,and it is convenient for clinical use,which can provide references for medical staff to take preventive treatment and care.

Key words: Acute Poisoning, Gastric Lavage, Aspiration Risk, Predictive Model, Nursing Care