基于重症监护医学信息数据库-Ⅲ建立预测心脏术后再次转入重症监护室风险的模型

Development of model predicting the risk of readmission to the intensive care unit after cardiac surgery based on MIMIC-Ⅲ database

目的确定接受心脏手术后的患者再次转入重症监护室(intensive care unit, ICU)的独立预测因素。方法从重症监护医学信息数据库-Ⅲ(Medical Information Mark for Intensive Care Ⅲ,MIMIC-Ⅲ)中提取研究所需的患者数据;根据"再次转入ICU"的定义将其分为对照组与病例组。主要结果采用最小绝对收缩与选择算子(The least absolute shrinkage and selection operator, LASSO)回归、单因素以及多因素Logistic回归分析确定术后再次转入ICU的独立预测因素,并在此基础上构建列线图模型。结果最终共纳入4 420例接受心脏手术后的患者进行回顾性分析,其分为对照组(n=4 183,占94.6%)与病例组(n=237,占5.4%)。多因素logistic回归分析显示术前并发心律失常(OR=0.723,95%CI:0.546-0.958,P<0.05),外周血管疾病(OR=0.658,95%CI:0.465-0.930,P<0.05),肾功能不全(OR=0.649,95%CI:0.428-0.983,P<0.05),电解质紊乱(OR=0.549,95%CI:0.380-0.792,P<0.01),充血性心力衰竭(OR=0.476,95%CI:0.358-0.633,P<0.01),药物滥用(OR=0.453,95%CI:0.206-0.992,P<0.05),酗酒(OR=0.402,95%CI:0.206-0.786,P<0.01),失血性贫血(OR=0.260,95%CI:0.085-0.796,P<0.05),由急诊手术入院(OR=2.906,95%CI:2.118-3.986,P<0.01),术后并发重症脓毒症(OR=0.304,95%CI:0.095-0.974,P<0.05)为心脏术后患者再次转入ICU的独立预测因素。并且,该模型具有良好的区分能力。结论本研究建立的预测心脏术后再次转入ICU风险的列线图模型可能有助于ICU医生识别高风险患者。然而,在模型推荐用于临床实践之前,该模型需要进一步的外部验证。

AIM To determine the independent predictors of readmission to intensive care unit(ICU)after cardiac surgery. METHODS Patient data needed for research is extracted from Medical Information Mark for Intensive Care Ⅲ(MIMIC-Ⅲ);They were assigned into control group and case group according to the definition of "readmission to ICU". The least absolute shrinkage and selection operator(LASSO) regression model was used to select the feature for multivariable logistic regression analysis that applied to build a predicting model incorporating the feature selected in the LASSO regression model;and based on the selected patients, a nomogram model predicting the risk of readmission to ICU was developed. RESULTS A total of 4 420 patients after cardiac surgery were included for retrospective analysis, divided into control group(n=4 183, accounting for 94.6%) and case group(n=237, accounting for 5.4%);Multivariate logistic regression analysis showed that various perioperative parameters were implicated as independent predictors for ICU readmission after cardiac surgery, such as history of cardiac arrhythmias(OR=0.723,95%CI:0.546-0.958,P<0.05), peripheral vascular disease(OR=0.658, 95%CI: 0.465-0.930, P<0.05), renal failure(OR=0.649,95%CI:0.428-0.983, P<0.05), fluid electrolyte disorder(OR= 0.549, 95%CI: 0.380-0.792, P<0.01), congestive heart failure(OR = 0.476, 95%CI: 0.358-0.633, P<0.01), drug abuse(OR = 0.453, 95%CI:0.206-0.992,P<0.05), alcohol abuse(OR=0.402, 95%CI:0.206-0.786, P<0.01) and blood loss anemia(OR=0.260,95%CI:0.085-0.796,P<0.05);and underwent non-elective surgery(OR= 2.906, 95%CI: 2.118-3.986,P<0.01) and complicated with severe sepsis after surgery(OR = 0.304, 95%CI: 0.095-0.974,P<0.05).Moreover, the model display good discrimination, with a AUC of 0.737. CONCLUSION The nomogram model developed in our study to predict the risk of readmission to ICU after cardiac surgery may assist ICU physicians to identify high-risk patients. However, the model requires external validation before being recommended for clinical practice.