论文部分内容阅读
目的:建立一种能预测肝细胞癌(HCC)患者腹腔镜肝切除术(LH)术后复发的极端梯度上升法(XGBoost)模型。方法:回顾性选取2013年1月—2016年9月在承德医学院附属医院首次接受LH治疗的原发性HCC患者440例为研究对象，确诊方式为病理诊断。使用随机数表法，以2∶8的比例将研究对象分为训练组(n n＝88)和验证组(n n＝352)。采用Kaplan-Meier法绘制无复发生存曲线，并采用Log-rank检验比较两组的生存情况；采用训练组建立COX回归模型和XGBoost模型，筛选预测LH术后复发的独立预测因素；采用受试者工作特征曲线(ROC)分析两种模型的预测能力，并在验证组中进行内部验证；采用Hosmer and Lemeshow Test来评价两种模型的校准度，以n P>0.05为模型与实际情况拟合度良好。n 结果:多因素COX回归模型和XGBoost模型均筛选出了癌栓、分化程度低、肿瘤微血管浸润、肿瘤个数、肿瘤较大、乙肝表面抗原阳性是肿瘤复发的独立预测因素(n HR＝2.477、0.769、1.786、1.905、1.544、1.805；95%n CI: 1.465～4.251、0.619～0.819、1.263～2.546、1.354～2.704、1.272～1.816、1.055～2.555)。XGboost模型评分依次为32、29、24、18、16、11分。训练组中COX回归模型和XGBoost模型预测复发的曲线下面积(AUC)分别为0.746(0.730～0.762)和0.802(0.785～0.818)，XGBoost模型预测能力较强，且在验证队列中也得到了证实。n 结论:本研究建立和验证了能够预测接受LH的HCC患者术后复发的XGBoost模型。该模型可应用于临床工作中，辅助医师为患者制订个性化的术后监测方案。HCC患者复发率高，预后差，早发现、早诊断、早治疗、加强术后复诊是改善患者预后的重要措施。“,”Objective:This study aimed to establish an eXtreme Gradient Boosting(XGBoost) model that can predict the recurrence of hepatocellular carcinoma(HCC)patients after laparoscopic hepatectomy (LH) surgery.Methods:A total of 440 patients with primary HCC who received LH treatment for the first time from January 2013 to September 2016 in Affiliated Hospital of Chengde Medical University were selected as the research objects. The diagnosis method was pathological diagnosis. Research objects were divided into training group (n n=88) and verification group (n n=352) at a ratio of 2∶8 by random number table method. The Kaplan-Meier method was used to draw the recurrence-free survival curve, and the Log-rank test was used to compare the survival of the two groups; the training group was used to establish the COX regression model and the XGBoost model to screen independent predictors of recurrence after LH; receiver operating characteristic(ROC) curve was used to analyze the predictive abilities of the two models, and conducted internal verification in the verification group; Hosmer and Lemeshow Test was used to evaluate the calibration of the two models, and n P>0.05 was used as a good fit between the model and the actual situation.n Results:Both the COX regression model and the XGBoost model screened out tumor thrombus, low degree of differentiation, tumor microvascular infiltration (MVI), number of tumors, large tumors, and positive hepatitis B surface antigen were independent predictors of tumor recurrence(n HR=2.477, 0.769, 1.786, 1.905, 1.544, 1.805; 95%n CI: 1.465-4.251, 0.619-0.819, 1.263-2.546, 1.354-2.704, 1.272-1.816, 1.055-2.555). The XGboost model scores were 32 points, 29 points, 24 points, 18 points, 16 points, 11 points, respectively. In the training group, the area under the curve (AUC) of the COX regression model and XGBoost model for predicting recurrence were 0.746 (0.730-0.762) and 0.802 (0.785-0.818), respectively. The XGBoost model had strong predictive ability and was confirmed in the validation cohort.n Conclusions:This study had established and verified the XGBoost model that can predict the recurrence of HCC patients after receiving LH for the first time. It can be used in clinics to assist doctors in formulating personalized postoperative monitoring programs for patients. Early detection, early diagnosis and early treatment of tumors and strengthening of postoperative follow-up are important measures to improve the prognosis of patients.