血压指数与脓毒症患者预后的关联--基于MIMIC-Ⅲ数据库的队列研究

Correlation between blood pressure indexes and prognosis in sepsis patients:a cohort study based on MIMIC-Ⅲdatabase

目的探讨脓毒症患者早期血压指数与预后的关联。方法将美国重症监护医学信息数据库Ⅲ(MIMIC-Ⅲ)中2001至2012年诊断为脓毒症患者的病历资料进行回顾性队列研究,根据患者28 d预后情况分为存活组和死亡组。统计患者的一般资料及入ICU即刻和24 h内的心率(HR)、血压数据,计算各项血压指数〔包括收缩指数、舒张指数及平均动脉压(MAP)指数的最高值、中位数和均值〕。将数据随机分为训练集和验证集(4:1)。采用单因素Logistic回归分析筛选协变量,进一步构建多因素Logistic逐步回归模型,分别建立模型1(包含P<0.1的HR、血压、血压指数相关变量及P<0.05的其他变量)和模型2(仅包含P<0.1的HR、血压、血压指数相关变量)。采用受试者工作特征曲线(ROC曲线)、精度召回曲线(PRC)和决策曲线分析(DCA)曲线评价2个模型的优劣,分析脓毒症患者预后的影响因素。以最佳Logistic回归模型建立列线图模型并进行效果评价。结果共有11559例脓毒症患者纳入研究,其中存活组10012例,死亡组1547例。两组间的年龄、存活时间、Elixhauser合并症评分等46项指标比较差异均有统计学意义(均P<0.05)。单因素Logistic回归分析初筛出37项指标。经过多因素Logistic逐步回归模型筛选,在HR、血压、血压指数相关指标中,筛选出入ICU即刻HR〔优势比(OR)=0.992,95%可信区间(95%CI)为0.988~0.997〕、最高HR(OR=1.006,95%CI为1.001~1.011)、最高MAP指数(OR=1.620,95%CI为1.244~2.126)、平均舒张指数(OR=0.283,95%CI为0.091~0.856)、中位收缩指数(OR=2.149,95%CI为0.805~4.461)、中位舒张指数(OR=3.986,95%CI为1.376~11.758)6项指标(均P<0.1);其他P<0.05的变量有14个,包括年龄、Elixhauser合并症评分、连续性肾脏替代治疗(CRRT)、使用呼吸机、镇静镇痛、去甲肾上腺素、去氧肾上腺素、最高血肌酐(SCr)、最高血尿素氮(BUN)、最高凝血酶原时间(PT)、最高活化部分凝血活酶时间(APTT)、最低血小板计数(PLT)、最高白细胞计数(WBC)、最低血红蛋白(Hb)。ROC曲线显示,模型1和模型2的曲线下面积(AUC)分别为0.769、0.637,提示模型1的预测准确性更高;PRC曲线显示,模型1和模型2的AUC分别为0.381、0.240,提示模型1的效果更好;DCA曲线显示,阈值在0~0.8(即死亡概率为0~80%)时,模型1的净获益率高于模型2。校准曲线提示,模型1的列线图模型预测效果与实际结局符合较好;Bootstrap验证结果证明,该列线图模型与前述结果一致,具有较好的预测效果。结论列线图模型对脓毒症患者28 d预后具有良好的预测效果,其中血压指数是重要的预测因素。

Objective To investigate the correlation between early-stage blood pressure indexes and prognosis in sepsis patients.Methods A retrospective cohort study was conducted on the medical records of patients diagnosed with sepsis from 2001 to 2012 in the Medical Information Mart for Intensive Care-Ⅲ(MIMIC-Ⅲ)database.Patients were divided into survival group and death group according to the 28-day prognosis.General data of patients and heart rate(HR)and blood pressure at admission to ICU and within 24 hours after admission were collected.The blood pressure indexes including the maximum,median and mean value of systolic index,diastolic index and mean arterial pressure(MAP)index were calculated.The data were randomly divided into training set and validation set(4:1).Univariate Logistic regression analysis was used to screen covariates,and multivariate Logistic stepwise regression models were further developed.Model 1(including HR,blood pressure,and blood pressure index related variables with P<0.1 and other variables with P<0.05)and Model 2(including HR,blood pressure,and blood pressure index related variables with P<0.1)were developed respectively.The receiver operator characteristic curve(ROC curve),precision recall curve(PRC)and decision curve analysis(DCA)curve were used to evaluate the quality of the two models,and the influencing factors of the prognosis of sepsis patients were analyzed.Finally,nomogram model was developed according to the better model and effectiveness of it was evaluated.Results A total of 11559 sepsis patients were included in the study,with 10012 patients in the survival group and 1547 patients in the death group.There were significant differences in age,survival time,Elixhauser comorbidity score and other 46 variables between the two groups(all P<0.05).Thirty-seven variables were preliminarily screened by univariate Logistic regression analysis.After multivariate Logistic stepwise regression model screening,among the indicators related to HR,blood pressure and blood pressure index,the HR at admission to ICU[odds ratio(OR)=0.992,95%confidence interval(95%CI)was 0.988-0.997]and the maximum HR(OR=1.006,95%CI was 1.001-1.011),maximum MAP index(OR=1.620,95%CI was 1.244-2.126),mean diastolic index(OR=0.283,95%CI was 0.091-0.856),median systolic index(OR=2.149,95%CI was 0.805-4.461),median diastolic index(OR=3.986,95%CI was 1.376-11.758)were selected(all P<0.1).There were 14 other variables with P<0.05,including age,Elixhauser comorbidity score,continuous renal replacement therapy(CRRT),use of ventilator,sedation and analgesia,norepinephrine,norepinephrine,highest serum creatinine(SCr),maximum blood urea nitrogen(BUN),highest prothrombin time(PT),highest activated partial thromboplastin time(APTT),lowest platelet count(PLT),highest white blood cell count(WBC),minimum hemoglobin(Hb).The ROC curve showed that the area under the curve(AUC)of Model 1 and Model 2 were 0.769 and 0.637,respectively,indicating that model 1 had higher prediction accuracy.The PRC curve showed that the AUC of Model 1 and Model 2 were 0.381 and 0.240,respectively,indicating that Model 1 had a better effect.The DCA curve showed that when the threshold was 0-0.8(the probability of death was 0-80%),the net benefit rate of Model 1 was higher than that of Model 2.The calibration curve showed that the prediction effect of the nomogram model developed according to Model 1 was in good agreement with the actual outcome.The Bootstrap verification results showed that the nomogram model was consistent with the above results and had good prediction effects.Conclusion The nomogram model constructed has good prediction effects on the 28-day prognosis in sepsis patients,and the blood pressure indexes are important predictors in the model.