血清NGAL与Fetuin A对脓毒症患者28天死亡的预测价值

Predictive value of serum NGAL and Fetuin A in 28-day mortality in the patients with sepsis

目的探讨中性粒细胞明胶酶相关脂质运载蛋白(NGAL)与胎球蛋白A(Fetuin A)对脓毒症患者28 d死亡的预测,并构建死亡风险预测模型。方法收集2018年2月至2020年12月我院收治的163例脓毒症患者临床资料,根据28 d生存情况将患者分为生存组(n=120)和死亡组(n=43),治疗后进行为期28 d的随访。将受试者资料按7∶3的比例随机分为训练集和验证集,比较患者临床特征,将有差异的因素纳入Cox多因素回归分析和XGBoost学习模型,使用验证集数据对预测模型效能进行评价。结果生存组白细胞计数、红细胞比容、血小板计数和血小板与淋巴细胞比值高于死亡组,序贯器官衰竭评分(SOFA)、急性生理与慢性健康状况评估系统Ⅱ(APACHEⅡ)评分、降钙素原(PCT)、D-二聚体、NGAL、肌酐(SCr)、乳酸(Lac)、Fetuin A水平和中性粒细胞与血小板比值低于死亡组,差异有统计学意义(P<0.05);两组性别比例、年龄、BMI、血小板、C-反应蛋白(CRP)、血尿素氮(BUN)差异无统计学意义(P>0.05)。多因素回归分析显示,SOFA评分≥2分、APACHEⅡ评分≥18分、PCT≥30 ng/mL、D-二聚体≥2.35 mg/L、NGAL≥350 mg/L、SCr≥410μmol/L、Lac≥2 mmol/L、Fetuin A≥0.3 g/L是脓毒症患者28 d死亡的危险因素,白细胞计数≥12×10^(9)/L、红细胞比容≥30%、血小板计数≥175×10^(3)/μL是保护性因素(P<0.05)。XGBoost学习模型结果显示,上述因素预测能力由大到小依次为NGAL、Fetuin A、PCT、APACHEⅡ、SOFA、SCr、Lac、D-二聚体、血小板计数、白细胞计数、红细胞比容。ROC曲线分析显示,XGBoost模型AUC(0.964,95%CI:0.885~0.983,P<0.001)高于Cox回归分析模型(0.891,95%CI:0.863~0.952,P<0.001)、NGAL(0.853,95%CI:0.782~0.895,P<0.001)和Fetuin A(0.726,95%CI:0.613~0.798,P<0.001);校准曲线显示,XGBoost模型的预测值相较于Cox回归分析模型与实际观测值更为一致。结论联合NGAL和Fetuin A构建的脓毒症患者28 d死亡风险预测模型中XGBoost学习模型有较好的预测效能,有助于医生结合预测结果采取干预治疗,改善患者预后。

Objective To explore the prediction of the 28-day mortality by neutrophil gelatinase-associated lipocalin(NGAL)and Fetuin A for the patients with sepsis,and to construct a death risk prediction model.Methods The 163 sepsis patients who were treated in our hospital from February 2018 to December 2020 were selected as the research objects.The clinical data of the patients were collected,and the patients were divided into survival group(n=120)and death group(n=43)according to the 28-day mortality rate.After treatment,a 28-day follow-up was carried out.The subjects’data was randomly divided into training set and validation set according to the ratio of 7∶3,the clinical characteristics of patients were compared,the different factors were incorporated into Cox multivariate regression analysis and XGBoost learning model,and the validation set data was used to evaluate the model efficacy.Results The results showed that WBC count,erythrocyte volume,platelet count and platelet to lymphocyte ratio in survival group were higher than those in death group.Sequential organ failure assessment(SOFA)score,APACHEⅡscore,PCT,D-dimer,NGAL,serum creatinine(SCr),lactic acid(Lac),Fetuin A level,neutrophil to platelet ratio in survival group were lower than those in death group,the differences were statistically significant(P<0.05).There were no significant differences in sex ratio,age,BMI,platelet,CRP and BUN between the two groups(P>0.05).Multivariate regression analysis showed SOFA score≥2 points,APACHEⅡscore≥18 points,PCT≥30 ng/mL,D-dimer≥2.35 mg/L,NGAL≥350 mg/L,SCr≥410μmol/L,Lac≥2 mmol/L,Fetuin A≥0.3 g/L were risk factors for death within 28 days in the patients with sepsis.WBC count≥12×10^(9)/L,erythrocyte volume≥30%and platelet count≥175×10^(3)/μL were protective factors(P<0.05).The results of XGBoost learning model showed that the prediction ability of the above factors were NGAL,Fetuin A,PCT,APACHEⅡ,SOFA,SCr,Lac,D-dimer,platelet count,WBC count and erythrocyte volume in order.ROC curve analysis showed that AUC of XGBoost model(0.964,95%CI 0.885-0.983,P<0.001)was higher than that of Cox regression model(0.891,95%CI 0.863-0.952,P<0.001)and NGAL(0.853,95%CI 0.782-0.895,P<0.001)and Fetuin A(0.726,95%CI 0.613-0.798,P<0.001).The calibration curve showed that the prediction value of the XGBoost model was more consistent with the actual observations than the Cox regression analysis model.Conclusions The XGBoost learning model has good predictive performance for the 28-day mortality of the sepsis patients,which helps the doctors to intervene and treat the patients based on the prediction results and improves the prognosis of patients.