预测肝细胞癌术后结局的列线图分析:一项国内两中心的回顾性研究

Nomogram analysis for predicting postoperative outcomes in hepatocellular carcinoma:a retrospective two-center national study

目的建立肝细胞癌术后结局影响的预测模型,帮助临床医生评估患者预后,指导临床治疗实施。方法筛选2015年1月1日至2018年10月31日评估患者苏州大学附属第一医院(n=366)和华中科技大学同济医学院附属同济医院(n=99)接受过手术切除并随访近4年的患者。每个临床数据结果的列线图分析均通过随机森林-决策树分类的机器学习法得出。从临床特征的80个变量中筛选出20个自变量,使用Calibration曲线来分析临床预测模型拟合度,并用受试者工作特征曲线下面积评估模型准确性。结果预测模型预测HCC术后第1、2、3年无进展生存的可靠性较好,其特异性分别为0.75、0.65、0.85。预测模型经过体内验证拟合度好,其预测结果与观察结果相似。预测模型对HCC术后3年无进展生存预测准确性最高。结论利用随机森林-决策树分类的机器学习法建立的肝细胞癌术后结局预测模型可帮助临床医生评估患者的预后情况,对其临床治疗实施具有一定指导价值。

Objective To establish a prediction model for the impact of postoperative outcome of hepatocellular carcinoma(HCC),to help clinicians judge patient prognosis and guide the implementation of clinical treatment.Methods The patients with HCC who underwent surgical resection in First Affiliated Hospital of Soochow University(n=366)and Tongji Hospital(n=99)between January 1,2015 and October 31,2018,and were followed up for almost 4 years.Nomogram for each outcome was obtained using a decision tree classification model and variable selection by the machine learning method of Random Forest-Decision Tree Classification method.Twenty independent variables were selected from 80 variables of clinical characteristics.Calibration curves were used to assess the degree of clinical pre-model fit and ultimately the area under the receiver operating characteristic curve to assess the accuracy of the model.Results The prediction model had good reliability in predicting progression-free survival at 1,2 and 3 years after HCC surgery,with specificity of 0.75,0.65 and 0.85,respectively.The predictive model was fully evaluated and calibrated,and its predictive results were similar to the observed results.The prediction model had the highest accuracy for 3-year PFS for postoperative HCC.Conclusion The prediction model of HCC outcome was established by using the machine learning method of Random Forest-Decision Tree Classification,which can help clinicians to assess the severity and prognosis of HCC patients and has a guiding value for their clinical treatment implementation.