新生儿败血症休克预测模型的构建——1 695例患儿回顾性研究

目的 筛选新生儿发生败血症休克的危险因素，建立新生儿败血症休克临床预测模型。方法 选取2016年1月1日—2019年12月31日重庆医科大学7家附属医疗机构中患有败血症的新生儿，根据是否发生败血症休克分为研究组和对照组。采用单因素分析、LASSO和logistic回归分析筛选危险因素。采用logistic、极端梯度提升（XGBoost）、随机森林（RF）、分类回归树（CART）和人工神经网络（ANN）建立新生儿败血症休克预测模型，根据灵敏度、特异度、曲线下面积等指标评估模型性能。结果 本研究中，共有339名败血症新生儿发生败血症休克，1 356名败血症新生儿未发生败血症休克。单因素分析筛选出31项差异指标，多因素分析筛选出12项独立危险因素。测试集中，logistic、XGBoost、RF、CART、ANN模型的曲线下面积分别为0.856 (0.809～0.903),0.861 (0.819～0.904),0.880 (0.838～0.922),0.835 (0.790～0.881),0.808 (0.756～0.860)。结论 本文构建的五种预测模型相对稳定，其中，RF模型的预测性能最佳，能为新生儿败血症休克提供较好的预测。

Construction of predictive model for neonatal septic shock:a retrospective study of 1 695 patients

Objective To screen risk factors for the occurrence of septic shock in neonates and to establish a clinical prediction model for septic shock in neonates. Methods Neonates with sepsis in seven affiliated medical institutions of Chongqing Medical University from January 1, 2016 to December 31, 2019 were selected and divided into study and control groups according to whether septic shock occurred. Risk factors were screened using univariate analysis, LASSO and logistic regression analysis. Logistic, extreme gradient boosting (XGBoost), random forest (RF), categorical regression tree (CART), and artificial neural network (ANN) were used to build a neonatal septic shock prediction model, and model performance was evaluated based on sensitivity, specificity, and area under the curve. Results In this study, a total of 339 septic neonates developed septic shock, and 1 356 septic neonates did not develop septic shock. A univariate analysis screened 31 indicators of variation and a multifactorial analysis screened 12 independent risk factors. In the test set, the area under the curve of the logistic, XGBoost, RF, CART, and ANN models were 0.856 (0.809-0.903), 0.861 (0.819-0.904), 0.880(0.838-0.922), 0.835(0.790-0.881), and 0.808 (0.756-0.860). Conclusion The five prediction models constructed in our study are relatively stable. RF model has the best prediction performance and can provide better prediction for neonatal septic shock.