基于随机森林的血糖变异预测ICU监护时长研究

ICU Stay Time Prediction with Glucose Variability Based on Random Forest Model

为预测ICU患者重症监护时长并研究血糖变异对时长的影响,分别构建K最近邻、决策树、支持向量机以及随机森林4种模型,使用重症患者的血糖变异情况和基本病例信息构成实验数据集进行训练,预测患者能否在72h内转出ICU病房。训练后的4种模型在判断患者重症监护时长的准确率分别为63.94%、78.77%、81.07%和84.14%。实验结果表明,血糖变异情况对患者重症监护时长有重要影响,且随机森林模型相比其它机器学习算法能较好地预测ICU患者的重症监护时长,能帮助医生合理安排治疗计划,在提高医疗资源利用效率上具有参考意义。

To establish an ICU stay time predictive model by using random forest and study the effect of glucose variability to the stay time,the k-nearest neighbor model,decision tree model,support vector machine model and random forest model were constructed and trained by ICU glucose data set to predict whether the patients can leave from ICU in 72 hours.The results of accuracy in predicting ICU stay time were 63.94%,78.77%,81.07%and 81.14%.The results show that glucose variability has an important influence on ICU stay time,and random forest model can predict the time better than other models.This research is valuable to help doctors arrange treatment plans and improve the efficiency of medical resource utilization.