成人重症新型冠状病毒肺炎的预测模型

Study on Prediction Models of Severe Corona Virus Disease in Adults

目的建立有价值、实用的预测模型,用于早期识别重症新型冠状病毒肺炎(COVID-19)高危病例。方法纳入2020年1月17日—3月24日在宜昌市第三人民医院确诊的343例成人COVID-19患者(训练集240例,测试集103例)进行回顾性单中心研究。收集患者入院时的临床特征和实验室检查数据。采用单因素及多因素logistic回归分析探讨重症COVID-19的预测因素,构建预测模型,列线图和受试者工作特征曲线(ROC)用于评估、验证预测模型。结果训练集和测试集患者入院时的临床特征和实验室检查数据差别均无统计学意义。存在并发症、发热、咳痰、呼吸困难、血淋巴细胞数减少和D-二聚体升高是重症COVID-19的独立预测因素。构建了预测重症COVID-19的自评模型、基层模型和复杂模型:一致性指数分别为0.902、0.924和0.943;内部验证:ROC曲线下面积分别为0.898、0.920和0.937;外部验证:ROC曲线下面积分别为0.870、0.878和0.879。3种模型均具有较高的诊断敏感性(84.4%~90.0%)、特异性(67.5%~89.2%)、阴性预测值(均超过95%)和可接受的阳性预测值(内部验证:48.1%~65.0%,外部验证:40.0%~53.1%)。结论入院时存在并发症、发热、咳痰、呼吸困难、血淋巴细胞数减少和D-二聚体升高是重症COVID-19的良好预测因素。构建的自评模型、基层模型、复杂模型可在不同情况下预测重症COVID-19的发生。

Objective To develop valuable and practicable predicting models for early identification of cases at high risk of severe corona virus disease(COVID-19).Methods A retrospective,single-center study with 343(development cohort 240,validation cohort 103)confirmed adult patients with COVID-19 at Third People s Hospital of Yichang between January 17,2020 and March 24,2020 was conducted.Clinical and laboratory data at admission were collected.Exploration of predictors and construction of prediction models of severe COVID-19 were performed by logistic regression analysis.Evaluation and verification of prediction models were performed by nomograms and receiver operating characteristic curves(ROCs).Results(1)There was no significant difference in clinical and laboratory data at admission between development cohort and validation cohort.(2)Comorbidity,fever,sputum production,dyspnea,lower lymphocyte count and higher D-dimer at admission were independent predictors for severe COVID-19.(3)Self-assessment model,grass-root model and complex model were built for predicting severe COVID-19.In the three models presented by Nomograms,concordance indexes were 0.902,0.924 and 0.943.The area under ROCs were 0.898,0.920 and 0.937 in internal verification,and 0.870,0.878 and 0.879 in external verification.(4)They showed high sensitivity(84.4%-90.0%),specificity(67.5%-89.2%)and negative predictive value(exceeded 95%),and acceptable positive predictive value(48.1%-65.0%in internal verification,and 40.0%-53.1%in external verification)at each prediction model.Conclusions Comorbidity,fever,sputum production,dyspnea,lower lymphocyte count and higher D-dimer at admission are independent predictors for severe COVID-19.The self-assessment model,grass-root model and complex model constructed in this study can be used to predict occurrence of severe COVID-19 under different circumstances.