目的:基于集成学习算法建立患者再入重症监护病房(intensive care unit,ICU)的风险预测模型,并比较各个模型的预测性能。方法:使用美国重症医学数据库(medical information mart for intensive care,MIMIC)-Ⅲ,根据纳入、排除标准筛选患...目的:基于集成学习算法建立患者再入重症监护病房(intensive care unit,ICU)的风险预测模型,并比较各个模型的预测性能。方法:使用美国重症医学数据库(medical information mart for intensive care,MIMIC)-Ⅲ,根据纳入、排除标准筛选患者,提取人口学特征、生命体征、实验室检查、合并症等可能对结局有预测作用的变量,基于集成学习方法随机森林、自适应提升算法(adaptive boosting,AdaBoost)和梯度提升决策树(gradient boosting decision tree,GBDT)建立再入ICU预测模型,并比较集成学习与Logistic回归的预测性能。使用五折交叉验证后的平均灵敏度、阳性预测值、阴性预测值、假阳性率、假阴性率、受试者工作特征曲线下面积(area under the receiver operating characteristic curve,AUROC)和Brier评分评价模型效果,基于最佳性能模型给出重要性排序前10位的预测变量。结果:所有模型中,GBDT(AUROC=0.858)优于随机森林(AUROC=0.827),略好于AdaBoost(AUROC=0.851)。与Logistic回归(AUROC=0.810)相比,集成学习算法在区分度上均有较大的提升。GBDT算法给出的变量重要性排序中,平均动脉压、收缩压、舒张压、心率、尿量、血肌酐等变量排序靠前,相对而言,再入ICU患者的心血管功能和肾功能更差。结论:基于集成学习算法的患者再入ICU预测模型表现出较好的性能,优于Logistic回归。使用集成学习算法建立的再入ICU风险预测模型可用于识别再入ICU风险高的患者,医务人员可针对高风险患者采取干预措施,改善患者的整体临床结局。展开更多
BACKGROUND In patients with cirrhosis,hepatic encephalopathy(HE)indicates a poor prognosis despite the use of artificial liver and liver transplantation,presenting as frequent hospitalizations and increased mortality ...BACKGROUND In patients with cirrhosis,hepatic encephalopathy(HE)indicates a poor prognosis despite the use of artificial liver and liver transplantation,presenting as frequent hospitalizations and increased mortality rate.AIM To determine predictors of early readmission and mid-term mortality in cirrhotic patients discharged after the resolution of HE.METHODS From January to February 2018,213 patients were enrolled in this observational study assessing all the successive patients with cirrhosis discharged from Department of Gastroenterology and Department of Infectious and Liver Diseases,Second Affiliated Hospital of Chongqing Medical University after the resolution of HE.The patients were followed for 6 mo.For each subject,demographic,clinical,and laboratory variables were assessed at the time of diagnosis of HE,during hospital stay,at discharge,and during follow-up.The primary endpoints were incidence of early readmission and mid-term mortality.RESULTS During follow-up,65(31%)patients experienced an early readmission.International normalized ratio(INR)[odds ratio(OR)=2.40;P=0.003)at discharge independently predicted early readmission.The incidence of early readmission was significantly higher in patients with an INR>1.62 at discharge than in those with an INR≤1.62(44%vs 19%;P<0.001).Model for End-stage Liver Disease(MELD)score(OR=1.11;P=0.048)at discharge proved to be an independent predictor of early readmission caused by HE.Hemoglobin(OR=0.97;P=0.005)at discharge proved to be an independent predictor of non-early readmission.During 6 months of follow-up,34(16%)patients died.Artificial liver use(hazard ratio=6.67;P=0.021)during the first hospitalization independently predicted mid-term mortality.CONCLUSION INR could be applied to identify fragile cirrhotic patients,MELD score could be used to predict early relapse of HE,and anemia is a potential target for preventing early readmission.展开更多
文摘目的:基于集成学习算法建立患者再入重症监护病房(intensive care unit,ICU)的风险预测模型,并比较各个模型的预测性能。方法:使用美国重症医学数据库(medical information mart for intensive care,MIMIC)-Ⅲ,根据纳入、排除标准筛选患者,提取人口学特征、生命体征、实验室检查、合并症等可能对结局有预测作用的变量,基于集成学习方法随机森林、自适应提升算法(adaptive boosting,AdaBoost)和梯度提升决策树(gradient boosting decision tree,GBDT)建立再入ICU预测模型,并比较集成学习与Logistic回归的预测性能。使用五折交叉验证后的平均灵敏度、阳性预测值、阴性预测值、假阳性率、假阴性率、受试者工作特征曲线下面积(area under the receiver operating characteristic curve,AUROC)和Brier评分评价模型效果,基于最佳性能模型给出重要性排序前10位的预测变量。结果:所有模型中,GBDT(AUROC=0.858)优于随机森林(AUROC=0.827),略好于AdaBoost(AUROC=0.851)。与Logistic回归(AUROC=0.810)相比,集成学习算法在区分度上均有较大的提升。GBDT算法给出的变量重要性排序中,平均动脉压、收缩压、舒张压、心率、尿量、血肌酐等变量排序靠前,相对而言,再入ICU患者的心血管功能和肾功能更差。结论:基于集成学习算法的患者再入ICU预测模型表现出较好的性能,优于Logistic回归。使用集成学习算法建立的再入ICU风险预测模型可用于识别再入ICU风险高的患者,医务人员可针对高风险患者采取干预措施,改善患者的整体临床结局。
文摘BACKGROUND In patients with cirrhosis,hepatic encephalopathy(HE)indicates a poor prognosis despite the use of artificial liver and liver transplantation,presenting as frequent hospitalizations and increased mortality rate.AIM To determine predictors of early readmission and mid-term mortality in cirrhotic patients discharged after the resolution of HE.METHODS From January to February 2018,213 patients were enrolled in this observational study assessing all the successive patients with cirrhosis discharged from Department of Gastroenterology and Department of Infectious and Liver Diseases,Second Affiliated Hospital of Chongqing Medical University after the resolution of HE.The patients were followed for 6 mo.For each subject,demographic,clinical,and laboratory variables were assessed at the time of diagnosis of HE,during hospital stay,at discharge,and during follow-up.The primary endpoints were incidence of early readmission and mid-term mortality.RESULTS During follow-up,65(31%)patients experienced an early readmission.International normalized ratio(INR)[odds ratio(OR)=2.40;P=0.003)at discharge independently predicted early readmission.The incidence of early readmission was significantly higher in patients with an INR>1.62 at discharge than in those with an INR≤1.62(44%vs 19%;P<0.001).Model for End-stage Liver Disease(MELD)score(OR=1.11;P=0.048)at discharge proved to be an independent predictor of early readmission caused by HE.Hemoglobin(OR=0.97;P=0.005)at discharge proved to be an independent predictor of non-early readmission.During 6 months of follow-up,34(16%)patients died.Artificial liver use(hazard ratio=6.67;P=0.021)during the first hospitalization independently predicted mid-term mortality.CONCLUSION INR could be applied to identify fragile cirrhotic patients,MELD score could be used to predict early relapse of HE,and anemia is a potential target for preventing early readmission.
