Abstract:

Objective To explore the risk factors of enteral nutrition-associated diarrhea（ENAD）,and to construct a risk prediction model based on random forest. Methods This study adopted a retrospective data collection method. A total of 537 ICU patients who received enteral nutrition in a tertiary hospital in Tianjin from August 2017 to May 2021 were enrolled. Based on univariate analysis and multivariate logistic regression analysis and VIF,the variables of the risk prediction model for enteral nutrition-associated diarrhea were determined. All data are randomly divided into a training set and a validation set by 6：4. Based on the training set data,the random forest model of ENAD is established,and the importance of ENAD risk factors was ranked by variable importance score using a random forest model. Confusion matrix analysis,accuracy,sensitivity,specificity,and area under the ROC curve were used to evaluate the predictive effect of the model on the validation dataset. Results The incidence of ENAD in this study was 38.73%. After univariate analysis,multivariate Logistic regression analysis and VIF,9 variables were determined to be included in the prediction model. The top 5 variable importance scores of the random forest prediction model include days of taking potassium preparations,APACHE Ⅱ score,daily doses of enteral nutrition,fasting days before enteral nutrition and hypoproteinemia. The accuracy of the random forest model on the validation set is 76.27%;the sensitivity is 70.73%;the specificity is 79.70%;the area under the ROC curve is 0.810（95%CI：0.638~0.827）. Conclusion The predictive model has good predictive performance,which provides a reference for the precise prevention of clinical nurses with certain practical significance.

Key words: Intensive Care Units, Enteral Nutrition, Diarrhea, Random Forest Algorithm, Nursing Care

摘要：

目的 探讨ICU患者肠内营养相关性腹泻（enteral nutrition-associated diarrhea,ENAD）的危险因素,基于随机森林算法,构建并验证风险预测模型。 方法 回顾性收集2017年8月—2021年5月在天津市某三级甲等医院ICU住院且行肠内营养的537例患者的资料,采用单因素分析、多重共线性分析、Logistic回归分析筛选ENAD的危险因素。再将数据按6：4的比例随机分为训练组和验证组。基于训练组数据,利用随机森林算法,建立ENAD风险预测模型,对危险因素的重要性进行排序,并绘制部分依赖图,分析危险因素的变化对ENAD发生风险的影响。对验证组数据进行混淆矩阵分析,采用准确率、灵敏度、特异度和受试者操作特征曲线下面积评价模型的预测效果。 结果 ENAD发生率为38.73%,经单因素分析、多重共线性分析、Logistic回归分析后初步确定9个危险因素,其中重要性排序前5位的危险因素为口服钾制剂天数、急性生理与慢性健康状况Ⅱ评分、平均每日肠内营养量、禁食天数、白蛋白浓度≤35 g/L。验证组中,风险预测模型的准确率76.27%,灵敏度70.73%,特异度79.70%,受试者操作特征曲线下面积为0.810（95%CI：0.638~0.827）。 结论 该风险预测模型的预测效能良好,可为护理人员制订个性化预防措施提供参考。

关键词: 重症监护病房, 肠内营养, 腹泻, 随机森林算法, 护理