基于MIMIC-Ⅲ数据库研究合并房颤的危重症患者短期死亡风险预测模型  

A risk prediction model of short‑term mortality for critically ill patients with atrial fibrillation:Based on MIMIC‑Ⅲdatabase

作　　者：杨婷 童传凤[2] YANG Ting;TONG Chuanfeng(Chengdu Second People's Hospital,Chengdu 610021,Sichuan,China;Zhongnan Hospital of Wuhan University,Wuhan 430071,Hubei,China)

机构地区：[1]成都市第二人民医院,四川成都610021 [2]武汉大学中南医院,湖北武汉430071

出　　处：《武汉大学学报（医学版）》2024年第6期674-680,共7页Medical Journal of Wuhan University

摘　　要：目的:利用MIMIC-Ⅲ数据库研究重症病区合并房颤者短期预后的独立危险因素,建立预测模型并对模型进行验证。方法:回顾性分析MIMIC-Ⅲ数据库患者出院诊断包含心房颤动,通过纳入及排除筛选标准纳入6 555名研究对象。通过SQL语言提取患者基本临床资料、实验室数据、出入院时间,死亡记录、院内死亡情况等并记录患者90 d全因死亡率。采用COX风险回归模型分析合并房颤的危重症患者90 d全因死亡的独立危险因素,依据COX分析结果建立预测模型并且绘制Nomogram列线图,同时受试者工作曲线(ROC)分析并计算曲线下面积(AUC),进行内部验证,最后进行曲线校正评估模型。结果:在90 d内1 639名患者死亡,4 916名患者存活,预测模型中基线特征提示死亡组较存活组在年龄、性别、充血性心力衰竭、电解质紊乱、慢性肺疾病、瓣膜疾病差异具有统计学意义(P<0.05),两组在阴离子间隙(AG)、血肌酐、红细胞压积(HCT)、血小板(PLT)、尿素氮、白细胞、红细胞分布宽度(RDW)、住院时长及疾病严重评分(SAPSⅡ、OASIS、SOFA)差异具有统计学意义(P<0.05);COX回归预测模型提示年龄、充血性心力衰竭、瓣膜病、电解质紊乱、慢性肺疾病是重症患者合并房颤患者90 d内发生死亡事件的风险因素,入院24 h内RDW、AG、PLT、HCT值能用于预测短期死亡。根据Cox回归模型分析结果建立预测模型为Y=1.116×X1+1.064×X2+1.283×X3+1.199×X4+1.165×X5+1.246×X6(X1=RDW;X2=AG;X3=充血性心力衰竭;X4=瓣膜疾病;X5=慢性肺疾病;X6=电解质紊乱及酸碱失衡)。模型及其他独立预测指标进行ROC分析,结果显示该模型的AUC=0.753 (P<0.001)。由Cox回归结果绘制列线图,其C-index值为0.719。最后进行Bootstrap内部验证并绘制校正曲线。结论:预测模型对合并房颤的重症患者90 d全因死亡风险预测良好。Objective:To study the independent risk factors for the short‑term prognosis of patients with atrial fibrillation in the ICU from the MIMIC‑Ⅲdatabase,established a predictive model,and verified the model.Methods:In the MIMIC‑Ⅲdatabase,6555 research subjects were included through inclusion and exclusion screening criteria.The basic clinical data,laboratory data,time of admission and discharge,death records,and in‑hospital deaths of patients were extracted from the database,and the 90‑day all‑cause mortality of patients was recorded.The independent risk factors for 90‑day all‑cause mortality in critically ill patients with atrial fibrillation were analyzed using the Cox risk regression model.The prediction model was established based on the Cox analysis results,and the Nomogram column chart was drawn.Meanwhile,the ROC was analyzed,and the AUC was calculated.After internal verification,the calibration curve was performed.Results:A total of 1639 patients died(death group)and 4916 survived(survival group)within 90 days.The baseline characteristics in the prediction model suggested that the death group was different from the survival group in age,gender,congestive heart failure,electrolyte imbalance,chronic lung diseases,and valve diseases(P<0.05).In addition,significant differences were also found in the AG,serum Cr,HCT,PLT,BUN,WBC,RDW,length of stay,and disease severity scores(SAPSⅡ,OASIS,and SOFA)(P<0.05).COX regression prediction model multivariate analysis suggested that age,congestive heart failure,valvular diseases,electrolyte disorder,and chronic lung diseases were risk factors for death in critically ill patients with atrial fibrillation within 90 days.The values within 24 hours from admission,such as RDW,AG,PLT,and HCT,could be used to predict short‑term death.The length of hospital stay and SAPSⅡscores could significantly reflect the short‑term risk of patients.According to the analysis results of the Cox regression model,the prediction model was Y=1.116×X1+1.064×X2+1.283×X3+1.

关 键 词：心房颤动 危重症患者 预测模型 MIMIC-Ⅲ数据库 

分 类 号：R541.75[医药卫生—心血管疾病]