重症监护病房急性胰腺炎患者早期院内死亡预测模型的构建与评价

Establishment and evaluation of early in-hospital death prediction model for patients with acute pancreatitis in intensive care unit

目的:探讨重症监护病房(ICU)内急性胰腺炎(AP)患者死亡风险预测因子,构建死亡预测模型并评价其效能。方法:使用美国重症监护医学信息数据库Ⅲ(MIMIC-Ⅲ)中的数据进行回顾性队列研究,收集数据库中ICU收治的285例AP患者的临床资料,包括年龄、性别、血常规和血生化指标、合并症、简化急性生理学评分Ⅲ(SAPSⅢ)及院内预后。采用单因素分析比较存活与死亡患者临床资料的差异;采用二元多因素Logistic回归分析筛选出AP患者院内死亡的独立预测因子,构建死亡预测模型,并绘制列线图;绘制受试者工作特征曲线(ROC曲线),计算ROC曲线下面积(AUC),检验预测模型的区分度,并比较该预测模型与SAPSⅢ评分对AP患者院内死亡的区分能力;通过Hosmer-Lemeshow拟合优度检验评价列线图模型的校准能力,并绘制校准图,展示预测模型的校准度。结果:285例AP患者中,院内死亡29例,存活256例。单因素分析显示,与存活组比较,死亡组患者年龄更大(岁:70±17比58±16),白细胞计数(WBC)、总胆红素(TBil)、血肌酐(SCr)、血尿素氮(BUN)、红细胞体积分布宽度(RDW)、合并充血性心力衰竭比例及SAPSⅢ评分更高〔WBC(×109/L):18.5(13.9,24.3)比13.2(9.3,17.9),TBil(μmol/L):29.1(15.4,66.7)比16.2(10.3,29.1),SCr(μmol/L):114.9(88.4,300.6)比79.6(53.0,114.9),BUN(mmol/L):13.9(9.3,17.8)比6.1(3.7,9.6),RDW:0.152(0.141,0.165)比0.141(0.134,0.150),充血性心力衰竭:34.5%比14.8%,SAPSⅢ评分(分):66(52,90)比39(30,48)〕,差异均有统计学意义(均P<0.05);二元多因素Logistic回归分析显示,年龄〔优势比(OR)=1.038,95%可信区间(95%CI)为1.005~1.073〕、WBC(OR=1.103,95%CI为1.038~1.172)、TBil(OR=1.247,95%CI为1.066~1.459)、BUN(OR=1.034,95%CI为1.014~1.055)和RDW(OR=1.344,95%CI为1.024~1.764)是AP患者院内死亡的独立危险因素,并构建Logistic回归模型:Logit(P)=0.037×年龄+0.098×WBC+0.221×TBil+0.033×BUN+0.296×RDW-12.133。ROC曲线分析显示,Logistic回归模型预测AP患者院内死亡的AUC为0.870(95%CI为0.794~0.946),敏感度为86.2%,特异度为78.5%,说明该模型具有较好的预测效能,且优于SAPSⅢ评分〔AUC为0.831(95%CI为0.754~0.907),敏感度为82.8%,特异度为75.4%〕。根据多因素分析结果构建列线图模型,校准图显示,列线图模型校准曲线与标准曲线非常接近,拟合优度检验:χ2=6.986,P=0.538,说明列线图模型预测死亡风险与实际发生风险的一致性较高。结论:AP患者年龄越大,WBC、TBil、BUN、RDW越高,院内死亡风险就越大;以上述指标构建的死亡预测Logistic回归模型及列线图模型对院内死亡高风险患者的区分能力较好,准确性较高,可以较准确地预测AP患者的死亡概率,为AP患者的预后判断及临床治疗提供依据。

Objective To investigate the death risk prediction factors of acute pancreatitis(AP)patients in intensive care unit(ICU),and to establish a death prediction model and evaluate its efficacy.Methods A retrospective cohort study was conducted using the data in the Medical Information Mart for Intensive Care-Ⅲ(MIMIC-Ⅲ).The clinical data of 285 AP patients admitted to the ICU in the database were collected,including age,gender,blood routine and blood biochemical indicators,comorbidities,simplified acute physiology scoreⅢ(SAPSⅢ)and hospital prognosis.By using univariate analysis,the differences in the clinical data of the patients were compared between the two groups.Binary multivariate Logistic regression analysis was used to screen out independent predictors of in-hospital death in AP patients.A death prediction model was established,and the nomogram was drawn.The receiver operator characteristic curve(ROC curve)was plotted,and the area under the ROC curve(AUC)was used to test the discrimination of the prediction model.In addition,the prediction model was compared with the SAPSⅢscore in predicting in-hospital death.The calibration ability of the prediction model was evaluated by the Hosmer-Lemeshow goodness of fit test,and a calibration map was drawn to show the calibration degree of the prediction model.Results Among 285 patients with AP,29 patients died in the hospital and 256 patients survived.Univariate analysis showed that the patients in the death group were older than those in the survival group(years old:70±17 vs.58±16),and had higher white blood cell count(WBC),total bilirubin(TBil),serum creatinine(SCr),blood urea nitrogen(BUN),red blood cell volume distribution width(RDW),proportion of congestive heart failure and SAPSⅢscore[WBC(×109/L):18.5(13.9,24.3)vs.13.2(9.3,17.9),TBil(μmol/L):29.1(15.4,66.7)vs.16.2(10.3,29.1),SCr(μmol/L):114.9(88.4,300.6)vs.79.6(53.0,114.9),BUN(mmol/L):13.9(9.3,17.8)vs.6.1(3.7,9.6),RDW:0.152(0.141,0.165)vs.0.141(0.134,0.150),congestive heart failure:34.5%vs.14.8%,SAPSⅢscore:66(52,90)vs.39(30,48),all P<0.05].Multivariate Logistic regression analysis showed that age[odds ratio(OR)=1.038,95%confidence interval(95%CI)was 1.005-1.073],WBC(OR=1.103,95%CI was 1.038-1.172),TBil(OR=1.247,95%CI was 1.066-1.459),BUN(OR=1.034,95%CI was 1.014-1.055)and RDW(OR=1.344,95%CI was 1.024-1.764)were independent risk predictors of in-hospital death in patients with AP.Logistic regression model was established:Logit(P)=0.037×age+0.098×WBC+0.221×TBil+0.033×BUN+0.296×RDW-12.133.ROC curve analysis showed that the AUC of the Logistic regression model for predicting the in-hospital death of patients with AP was 0.870(95%CI was 0.794-0.946),the sensitivity was 86.2%,and the specificity was 78.5%,indicating that the model had good predictive performance,and it was superior to the SAPSⅢscore[AUC was 0.831(95%CI was 0.754-0.907),the sensitivity was 82.8%,and the specificity was 75.4%].A nomogram model was established based on the result of multivariate Logistic regression analysis.The calibration map showed that the calibration curve of the nomogram model was very close to the standard curve,with the goodness of fit test:χ2=6.986,P=0.538,indicating that the consistency between the predicted death risk of the nomogram model and the actual occurrence risk was relatively high.Conclusions The older the AP patient is,the higher the WBC,TBil,BUN,and RDW,and the greater the risk of hospital death.The death prediction Logistic regression model and nomogram model constructed based on the above indicators have good discrimination ability and high accuracy for high-risk patients with hospital death,which can accurately predict the probability of death in AP patients and provide a basis for prognosis judgment and clinical treatment of AP patients.