接受PD-1抑制剂治疗的肿瘤患者发生肾损伤的风险预测模型构建与评价

Construction and evaluation of risk prediction model for renal injury in tumor patients receiving PD-1 inhibitor treatment

目的探讨接受程序性死亡受体-1(PD-1)抑制剂治疗的肿瘤患者发生肾损伤的危险因素,并进一步构建列线图模型预测患者发生肾损伤的可能性。方法本研究为单中心回顾性分析。纳入苏州大学附属第三医院2018年1月-2021年1月使用PD-1抑制剂治疗的肿瘤患者447例,随访至2022年1月。肾损伤定义为急性肾脏疾病(AKD)。根据随访结束时是否发生PD-1抑制剂相关AKD将患者分为AKD组(n=71)与非AKD组(n=376)。比较两组患者的基本信息、疾病信息、用药情况、实验室指标,以及随访期间肾外免疫相关不良事件(irAEs)的发生情况。采用单因素及多因素logistic回归模型确定PD-1抑制剂相关AKD的独立危险因素。将所有患者按照7∶3的比例随机分成训练集(n=313)和验证集(n=134),根据筛选出的独立危险因素在训练集中建立列线图预测模型,通过受试者工作特征(ROC)曲线对模型的区分度进行评价,绘制校准曲线对模型的校准度进行评价,绘制临床决策曲线分析(DCA)探讨模型的临床有效性与获益率。结果AKD组患者联用抗生素、合并糖尿病、高血压、肾外irAEs的比例及胱抑素C(CysC)水平明显高于非AKD组(P<0.05),而血红蛋白(Hb)水平低于非AKD组(P<0.05)。单因素logistic回归分析显示,联用抗生素,合并糖尿病、高血压、肾外irAEs,较低的Hb、估算肾小球滤过率(eGFR)、较高的血尿素氮(BUN)、血清肌酐(SCr)、CysC、空腹血糖(FBG)、谷丙转氨酶(ALT)是PD-1抑制剂相关AKD的危险因素(P<0.05)。多因素logistic回归分析显示,合并肾外irAEs、较低的Hb、较高的SCr及直接胆红素(DBIL)是PD-1抑制剂相关AKD的独立危险因素(P<0.05)。基于以上独立危险因素进一步建立列线图预测模型,并对该模型进行验证,结果显示该模型的训练集和验证集ROC曲线下面积(AUC)分别为0.703(95%CI0.628~0.777)、0.791(95%CI0.671~0.911),提示其具有良好的区分度。训练集和验证集的校准曲线均徘徊在45°的理想线附近,提示该模型具有良好的校准度。DCA显示构建的模型曲线远离两条极端线(净获益为0的曲线和所有样本都是阳性的曲线),提示该模型具有良好的临床效益。结论合并肾外irAEs、较低的Hb、较高的SCr和较高的DBIL是PD-1抑制剂相关AKD的独立危险因素,据此建立的列线图模型具有较好的区分度和校准度,可为临床提供指导。

Objective To explore the risk factors for renal injury in tumors patients treated with programmed death receptor-1(PD-1)inhibitor,and further construct a column chart model to predict the likelihood of renal injury in patients.Methods The present study is a single center retrospective analysis.447 patients with tumors treated with PD-1 inhibitors in the Third Affiliated Hospital of Soochow University between January 2018 and January 2021 were included and followed up until January 2022.Kidney injury was defined as acute kidney disease(AKD).All patients were divided into AKD group(n=71)and non-AKD group(n=376)according to whether PD-1 inhibitor associated with AKD development at the end of follow-up.Basic information,disease and medication situation,laboratory indicators,and the incidence of extrarenal immune related adverse events(irAEs)during follow-up period were compared between the two groups.Univariate and multivariate logistic regression models were used to identify independent risk factors for PD-1 inhibitor associated AKD.The present study randomly divided all samples(n=447)into training set(n=313)and validation set(n=134)in a 7:3 ratio,built nomogram prediction models in the training set according to the screened independent risk factors,drawn the receiver operating characteristic(ROC)curves to evaluate the discrimination of the models,drawn calibration curves to evaluate the calibration of the models,and drawn clinical decision curve analysis(DCA)to explore the clinical validity and benefit rate of the models.Results The combination of antibiotics,diabetes,hypertension,extrarenal irAEs and cystatin C(Cys C)in AKD group were significantly higher than those in non-AKD group(P<0.05),but hemoglobin(Hb)was significantly lower than that in non-AKD group(P<0.05).Single factor logistic regression analysis showed that combination of antibiotics,diabetes,hypertension,extrarenal irAEs,lower Hb,estimated glomerular filtration rate(eGFR),higher blood urea nitrogen(BUN),serum creatinine(SCr),Cys C,fasting blood glucose(FBG),and alanine transaminase(ALT)were risk factors for PD-1 inhibitor related AKD(P<0.05).Multivariate logistic regression analysis showed that concomitant extrarenal irAEs,lower Hb,higher SCr,and direct bilirubin(DBIL)were independent risk factors for PD-1 inhibitor associated AKD(P<0.05).Based on the independent risk factors mentioned above,a column chart prediction model was further established and validated.The results showed that the area under the ROC curve(AUC)of the training and validation sets of the model were 0.703(95%CI 0.628-0.777)and 0.791(95%CI 0.671-0.911),respectively,indicating good discrimination.The calibration curves of both the training and validation sets hover around the ideal line of 45°,indicating that the model has good calibration.DCA shows that the constructed model curve is far away from the two polar lines(the curve with a net benefit of 0 and the curve with all samples being positive),indicating that the model has good clinical benefits.Conclusion The combination of extrarenal irAEs,lower Hb,higher SCr,and higher DBIL are independent risk factors for the occurrence of PD-1 inhibitor related AKD;The established column chart model has good discrimination and calibration,which can provide guidance for clinical practice.