基于机器学习的儿童营养风险筛查预测研究

Prediction study of nutritional risk screening in children based on machine learning method

目的:基于机器学习方法预测住院儿童的疾病风险和营养摄入风险。方法:将某医院住院患儿的现病史、医嘱、诊断等信息加工后作为输入特征,分别运用随机森林、逻辑回归、决策树、支持向量机和极端梯度提升树(XGBoost)算法,训练模型,并调整参数,以人工评估结果为依据,比较各机器学习模型预测住院患儿的疾病风险和营养摄入风险的性能。结果:在疾病风险预测方面,极端梯度提升树模型性能最好,敏感度调整到90.3%时,特异度为44.1%;在营养摄入风险预测方面,支持向量机模型表现最好,泛化能力强,在敏感度达到90.5%时,特异度为65.6%。结论:机器学习在住院患儿STAMP评分的疾病风险预测和营养风险预测方面具有良好的效果,可用于协助开展住院患儿的初步营养筛查,在儿童营养风险筛查方面具有较好的应用前景。

Objective To predict disease risk and nutritional intake risk of hospitalized children based on machine learning methods.Methods After processing the medical history,medical orders,diagnosis and other information of hospitalized children in a hospital as input features,the Random Forest,Logistic Regression,Decision Tree,Support Vector Machine and Extreme Gradient Boosting Tree(XGBoost)algorithms were respectively used to train the models and adjust the parameters.The manual evaluation results were used as the evaluation basis to compare the performance of various machine learning models in predicting the disease risk and nutritional intake risk of hospitalized children.Results In terms of predicting disease risk,the XGBoost model performs the best,and the specificity was 44.1%when the sensitivity was adjusted to 90.3%.In terms of predicting nutritional intake risk,the Support Vector Machine model performs the best with strong generalization ability,and the specificity reached 65.6%when the sensitivity reached 90.5%.Conclusion Machine learning has shown good performance in disease risk prediction and nutritional risk prediction based on the STAMP score of hospitalized children,which can be used to assist in their preliminary nutritional screening and has great application prospect in pediatric nutritional risk screening.