乳腺癌化学治疗致肝损伤预测模型的构建及其评价

Construction and evaluation of a prediction model for liver injury induced by chemotherapy for breast cancer

目的·分析乳腺癌化学治疗(化疗)致肝损伤的危险因素,构建乳腺癌化疗致肝损伤的预测模型并对其进行评价。方法·以2019年4月至2021年9月在南昌大学第二附属医院接受蒽环类联合环磷酰胺序贯紫杉醇类±曲妥珠单抗化疗方案的乳腺癌患者为研究对象,分为肝损伤组与未损伤组。采用χ^(2)检验及二元Logistic回归分析化疗致肝损伤的危险因素并建立预测模型。采用受试者工作特征(receiver operating characteristic,ROC)曲线评价回归模型的预测效能并对其进行外部验证。结果·入选符合纳入标准的乳腺癌患者207例。其中,肝损伤组患者69例(33.3%),未损伤组138例。2组间的化疗方案(χ^(2)=44.851,P=0.000)、感染乙肝病毒(χ^(2)=16.682,P=0.000)、既往高血压病史(χ^(2)=13.211,P=0.004)、肿瘤TNM分期(χ^(2)=14.422,P=0.001)、既往糖尿病史(χ^(2)=4.839,P=0.028)、化疗前白蛋白水平低下(χ^(2)=10.073,P=0.002)以及三酰甘油升高(χ^(2)=39.367,P=0.000)的差异存在统计学意义。二元Logistic回归分析显示,化疗方案(OR=4.734,95%CI为1.687~13.283,P=0.003)、感染乙肝病毒(OR=4.530,95%CI为1.806~11.366,P=0.001)、肿瘤TNM分期Ⅲ期(OR=5.304,95%CI为1.802~15.608,P=0.002)、既往糖尿病史(OR=3.041,95%CI为1.196~7.729,P=0.019)以及化疗前白蛋白水平低下(OR=3.744,95%CI为1.413~9.920,P=0.008)是化疗致肝损伤的独立危险因素。Logistic回归模型:logit(P)=-3.471+1.511·X1+1.112·X2+1.320·X3+0.755·X4+0.691·X5+0.973·X6+1.258·X7+0.741·X8+1.668·X9+1.555·X10。ROC曲线下面积为0.874,标准误0.024,95%CI为0.827~0.922,P=0.000。外部验证显示,预测模型特异度为86.9%,灵敏度76.9%;阳性预测值为71.4%,阴性预测值为89.8%;准确度为83.9%;Kappa值为0.624,标准误为0.091,P=0.000。结论·此Logistic回归模型具有较高的预测效能,对预测患者是否出现肝损伤有一定的参考价值。

Objective·To analyze the risk factors of liver injury induced by chemotherapy for breast cancer,and to construct and evaluate a prediction model of liver injury induced by chemotherapy.Methods·Breast cancer patients hospitalized at the Second Affiliated Hospital of Nanchang University from April 2019 to September 2021 who received anthracycline combined with cyclophosphamide sequential paclitaxel±trastuzumab chemotherapy were enrolled in the study,and were divided into liver injury group and non-injury group.The risk factors of liver injury induced by chemotherapy were analyzed byχ^(2)test and binary Logistic regression,and the prediction model was established.The receiver operating characteristic(ROC)curve was used to evaluate the predictive performance of the regression model,and then the predictive model was externally validated.Results·Two hundred and seven patients with breast cancer met the inclusion criteria in this study,and sixty-nine patients with liver injury(33.3%).Univariate analysis showed that the difference of chemotherapy regimen(χ^(2)=44.851,P=0.000),hepatitis B virus infection(χ^(2)=16.682,P=0.000),previous history of hypertension(χ^(2)=13.211,P=0.004),tumor TNM staging(χ^(2)=14.422,P=0.001),previous history of diabetes(χ^(2)=4.839,P=0.028),low albumin level before chemotherapy(χ^(2)=10.073,P=0.002)and elevated triacylglycerol(χ^(2)=39.367,P=0.000)were statistically significant between the two groups.Binary Logistic regression showed that chemotherapy regimen(OR=4.734,95%CI 1.687-13.283,P=0.003),hepatitis B virus infection(OR=4.530,95%CI 1.806-11.366,P=0.001),tumor TNM stageⅢ(OR=5.304,95%CI 1.802-15.608,P=0.002),previous history of diabetes(OR=3.041,95%CI 1.196-7.729,P=0.019),and low albumin level before chemotherapy(OR=3.744,95%CI 1.413-9.920,P=0.008)were independent risk factors for liver injury caused by chemotherapy.logit(P)=-3.471+1.511·X1+1.112·X2+1.320·X3+0.755·X4+0.691·X5+0.973·X6+1.258·X7+0.741·X8+1.668·X9+1.555·X10.The area under the ROC curve was 0.874,the standard error was 0.024,95%CI 0.827-0.922,P=0.000.External validation showed that the specificity of the prediction model was 86.9%,the sensitivity was 76.9%;the positive predictive value was 71.4%,the negative predictive value was 89.8%;the accuracy was 83.9%;the Kappa value was 0.624,the standard error was 0.091,P=0.000.Conclusion·This Logistic regression model has high predictive performance and has certain reference value for breast doctors to predict whether patients have liver damage.