肝癌TACE抵抗的危险因素分析及其预测模型构建与验证

Risk factors for TACE resistance in hepatocellular carcinoma and the construction and validation of prediction model

目的分析肝癌经导管肝动脉化疗栓塞术(TACE)抵抗的危险因素,构建肝癌TACE抵抗的预测模型并验证其预测效能。方法选择接受至少3次TACE治疗的肝癌患者164例,按1∶1随机分配至训练集和测试集。根据TACE抵抗的CCI定义及专家共识判定TACE抵抗,训练集TACE有效49例、TACE抵抗33例;验证集TACE有效58例、TACE抵抗24例。收集所有研究对象的一般资料、实验室检查资料及影像学检查资料,采用多因素Logistic回归模型分析训练集TACE抵抗的危险因素。通过机器学习随机森林分类筛选其预测变量的特征重要性并绘制预测模型列线图。通过受试者工作特征(ROC)曲线下面积验证该模型的预测效能。采用Hosmer-Lemeshow拟合优度检验评价该模型的校准能力。结果多因素Logistic回归分析发现,合并乙肝、高ALBI评分、高水平甲胎蛋白、无肿瘤包膜以及强化模式3、4型是导致肝癌TACE抵抗的独立危险因素(P均<0.05)。通过机器学习随机森林分类评估训练集TACE抵抗5个独立危险因素的特征重要性,结果发现甲胎蛋白(36.1%)、强化模式(30.4%)、ALBI评分(26.5%)的特征重要性较大,而无肿瘤包膜(2.8%)和合并乙肝(4.1%)的特征重要性较小,故将其排除。利用甲胎蛋白、强化模式、ALBI评分构建肝癌TACE抵抗的预测模型并绘制列线图。ROC曲线分析发现,在训练集中该模型预测肝癌TACE抵抗的曲线下面积为0.899,在测试集中为0.753。Hosmer-Lemeshow拟合优度检验发现,该模型在训练集和测试集中的一致性较好(训练集:χ^(2)=11.829,P>0.05;测试集:χ^(2)=7.927,P>0.05)。结论合并乙肝、高ALBI评分、高水平甲胎蛋白、无肿瘤包膜以及强化模式3、4型是导致肝癌TACE抵抗的独立危险因素;以ALBI评分、甲胎蛋白和强化模式构建了肝癌TACE抵抗的预测模型,经验证,该模型的预测效能较好。

Objective To analyze the risk factors of resistance to transcatheter arterial chemoembolization(TACE)in patients with hepatocellular carcinoma(HCC),to construct a prediction model,and to verify its prediction efficiency.Methods A total of 164 HCC patients who received at least 3 times of TACE were randomly divided into the training set(82 cases)and the test set(82 cases)at a ratio of 1∶1.According to the CCI definition of TACE resistance and expert con-sensus,TACE resistance was determined,and 49 cases were TACE effective and 33 cases were TACE resistant in the training set.In the validation set,58 cases were TACE effective and 24 cases were TACE resistant.The general data,lab-oratory examination data and imaging examination data of all subjects were collected.Multivariate Logistic regression model was used to analyze the risk factors of TACE resistance in the training set.The feature importance of predictor variables was screened by machine learning random forest classification and the prediction model nomogram was drawn.The area under the receiver operating characteristic(ROC)curve was used to verify the predictive efficacy of the prediction model.The Hosmer-Lemeshow goodness-of-fit test was used to evaluate the calibration ability of the prediction model.Results Mul-tivariate Logistic regression analysis showed that hepatitis B,high ALBI score,high level of alpha-fetoprotein(AFP),no tumor capsule,and enhancement pattern 3 and 4 were independent risk factors for TACE resistance in HCC(all P<0.05).The feature importance of 5 independent risk factors of TACE resistance in the training set was evaluated by machine learning random forest,and we found that AFP(36.1%),enhancement pattern(30.4%),and ALBI score(26.5%)had higher feature importance,while non-tumor capsule(2.8%)and hepatitis B(4.1%)had lower feature importance,so they were excluded.AFP,enhancement pattern and ALBI score were used to construct a prediction model for TACE resistance of liver cancer and we drew a nomogram.ROC curve analysis showed that the area under the curve of the model in predicting TACE resistance of liver cancer was 0.899 in the training set and 0.753 in the test set.Hosmer-Lemeshow goodness-of-fit test showed that the model had good consistency in the training set and the test set(training set:χ^(2)=11.829,P>0.05;test set:χ^(2)=7.927,P>0.05).Conclusions Hepatitis B,high ALBI score,high level of AFP,absence of tumor capsule,enhancement patterns 3 and 4 were independent risk factors for TACE resistance in HCC.Based on ALBI score,AFP,and enhancement mode,a prediction model for TACE resistance of hepatocellular carcinoma was established,and the prediction efficiency of the model was good.