节点文献

基于临床病理特征、血液检测指标、CT影像组学的口咽癌生存预测模型研究

Study on Survival Prediction Model of Oropharyngeal Cancer Based on Clinicopathological Features,Blood Test Parameters and CT Radiomics

【作者】 李亮;

【导师】 王旭东;

【作者基本信息】 天津医科大学 , 临床医学(专业学位), 2022, 博士

【摘要】 目的:口咽癌(oropharyngeal cancer,OPC),是头颈部最常见的恶性肿瘤之一,最常见的病理类型是鳞状细胞癌,恶性程度高,易转移、预后差。目前治疗原则是手术切除,以及化疗、放疗为主的综合治疗,因发病机制、解剖部位、肿瘤异质性等原因,治疗效果存在明显的个体化差异。目前TNM分期系统仍然是制定治疗方案、评判治疗效果等的主要依据,但是对患者个体化预后的预测存在明显局限性。本研究旨在联合患者临床病理特征、血液检测指标及CT影像组学构建口咽癌生存预测模型,以个体化判断患者的预后,为口咽癌患者制定个性化治疗方案提供理论依据。方法:回顾性分析2010-2021年在天津医科大学肿瘤医院治疗的248例口咽癌患者的临床病理特征、血液检测指标、影像组学资料。对部分缺失数据进行了多重插补并进行了二次敏感性分析。采用Kaplan-Meier生存曲线及单因素Cox回归方法分析与口咽癌患者预后相关的临床病理特征及血液检测指标,将有意义的变量纳入多因素Cox回归分析,筛选出口咽癌患者预后的独立影响因素。利用影像组学的方法,对患者CT影像完成感兴趣区域(region of interest,ROI)的勾画,应用python工具包Pyradiomics(3.0.1)从ROI中提取每个患者的CT影像组学特征,采用LASSO回归的方法筛选出影响OPC患者预后的影像组学特征,并利用它们构建基于影像组学特征的Radiomics分类器,为每一个患者计算Rad-score评分,应用时间依赖性受试者工作特征曲线(Receiver operating characteristic curve,ROC)下面积(Area under receiver operating characteristic curve,AUC),评估Radiomics分类器与TNM分期系统对患者预后的判断力的差异;利用多因素COX回归分析,联合患者临床病理特征、血液检测指标及CT影像组学构建OPC患者的生存预测模型,应用AUC、C-index,校准曲线、临床决策曲线、NRI、IDI对预测模型进行评价。所有统计数据及图表均应用R统计软件(版本4.1.0)进行统计分析。所有的统计检验都以P<0.05(双侧)为差异具有统计学意义。结果:1.248例患者总体3年、5年生存率为63.1%、47.7%。单因素分析显示:T分期、N分期,TNM分期、治疗方式、γ-GGT、直接胆红素、白蛋白均与口咽癌患者的预后相关,P<0.05,差异具有统计学意义。多因素分析显示:年龄、饮酒史、治疗方式、TNM分期、γ-GGT、直接胆红素是OPC患者预后的独立影响因素。在经过多重插补以后发现NLR、PLR、PLT、白蛋白、总蛋白、纤维蛋白原是OPC患者的预后独立因素,P值均<0.05,差异具有统计学意义。2.利用LASSO回归分析采用10折交叉验证与可重复自抽样技术,对每个患者提取的875个影像组学特征进行筛选,最终选择了7个特征(特征名称详见第二部分结果),利用特征乘以相应的回归系数计算了每位患者的Rad-Score值,利用该Rad-Score值构建Radiomics分类器。X-tile软件将Rad-score的cutoff值设为7.17和7.71,并把患者划分为低危组、中危组和高危组,在训练集和验证集中,低危组、中危组和高危组患者生存率差异均存在统计学差异,P<0.0001。ROC分析结果显示:TNM分期系统和Radiomics分类器对口咽癌患者5年生存率预测的AUC为0.629和0.740,两者联合的AUC为0.772。与单独使用TNM分期系统相比,Radiomics分类器联合TNM分期能够显著提高对患者OS的预测性能。3.联合OPC患者临床病理特征、血液检测指标、影像组学进行多因素Cox回归分析,结果显示T分期、N分期、白蛋白、γ-GGT、直接胆红素、Rad-Score是口咽癌患者预后的独立影响因素。根据上述因素构建预测模型,并绘制列线图。该模型在训练组和验证组中的c-index指数为分别为0.8和0.77;校准曲线与理想的45°虚线比较贴合,该模型在预测OPC患者3年及5年生存概率方面表现出良好的符合度;决策性分析曲线显示在不同的阈值概率下均有较好的净获益,显示该模型对OPC患者3年及5年生存率方面具有良好的临床应用价值。增加Rad-Score对临床血液模型的改善程度:NRI为0.382,IDI为0.111。结论:1、口咽癌患者的年龄、治疗方式、饮酒史、TNM分期、γ-GGT、直接胆红素、NLR、血小板计数、总蛋白、PLR、白蛋白、纤维蛋白原是口咽癌患者OS的独立影响因素;血液检测指标经济、方便、易获得,可以作为口咽癌患者预后预测指标。多重插补技术,可以为我们提供更多信息,能够提升统计效能。2、基于患者CT影像组学特征的Radiomics分类器,与OPC患者的预后显著相关,是OPC患者预后的独立影响因素。它可以作为TNM分期系统的补充,来提升其预测OPC患者预后的准确性。3、基于OPC患者临床病理特征、血液检测指标及CT影像组学特征成功建立了预测患者1年、3年、5年生存率的列线图;该列线图通过了验证集的区分度、校准度、临床决策曲线的验证,具有较高的区分度,较为准确的一致性以及较好的患者净获益;该列线图可以作为评估口咽癌预后的可视化工具,具有较好的准确性和临床应用价值,为患者个性化制定治疗方案,提供了理论依据。

【Abstract】 Objective:Oropharyngeal cancer(OPC)is one of the most common malignant tumors of the head and neck.The most common pathological type is squamous cell carcinoma with high degree of malignancy,easy metastasis and poor prognosis.At present,the principle of treatment is surgical resection,as well as chemotherapy,radiotherapy-based comprehensive treatment,due to the pathogenesis,anatomical site,tumor heterogeneity and other reasons,the treatment effect has obvious individual differences.At present,TNM staging system is still the main basis for making treatment plan and evaluating treatment effect,but there are obvious limitations in predicting individual prognosis of patients.The purpose of this study is to construct a survival prediction model of oropharyngeal cancer combined with clinicopathological features,blood test parameters and CT Radiomics,to individually judge the prognosis of patients with oropharyngeal cancer,and to provide a theoretical basis for individualized treatment of oropharyngeal cancer patients..Method:The clinicopathological features,blood test parameters and Radiomic data of248 patients with oropharyngeal cacer treated in Cancer Hospital of Tianjin Medical University from 2010 to 2021 were analyzed retrospectively.Multiple interpolation and secondary sensitivity analysis were carried out for some missing data.Kaplan-Meier survival curve and univariate COX regression were used to analyze the clinicopathological features and blood test parameters related to the prognosis of patients with oropharyngeal cancer.Significant variables were included in multivariate COX regression analysis to screen the independent factors affecting the prognosis of patients with oropharyngeal carcinoma.The region of interest(ROI)was delineated in the CT images of patients by using the method of Radiomics.The CT Radiomic features of each patient were extracted from ROI by python package-Pyradiomics(3.0.1).The Radiomics features affecting the prognosis of OPC patients were screened by LASSO regression,and they were used to construct a Radiomics classifier based on Radiomics features,and the Rad-score score was calculated for each patient.The area under the curve(AUC)of time-dependent receiver operating characteristics(ROC)was used to evaluate the difference of prognosis between Radiomics classifier and TNM staging system.The survival prediction model of OPC patients was constructed by multivariate COX regression analysis,combined with clinicopathological features,blood test parameters and CT Radiomic features.AUC,C-index,calibration curve,decision analysis curve,NRI and IDI were used to evaluate the prediction model.All statistical data and charts shall be statistically analyzed by R statistical software(version4.1.0).All the statistical tests were bilateral,and the difference was considered to be statistically significant when P < 0.05.Result:1.The overall 3-year and 5-year survival rates of 248 patients were 63.1% and 47.7%..Univariate analysis showed that T stage,N stage,TNM stage,treatment mode,γ-GGT,direct bilirubin and albumin were all correlated with the prognosis of patients with oropharyngeal carcinoma.Multivariate analysis showed that age,drinking history,treatment mode,TNM stage,γ-GGT and direct bilirubin were independent influencing factors of OS in OPC patients.After multiple interpolation,it was found that NLR,PLR,PLT,albumin,total protein,fibrinogen and other parameters also had independent guiding significance for the prognosis of OPC patients.2.Using LASSO regression analysis,10-fold cross-validation and repeatable self-sampling technology were used to screen 875 radiomics features extracted from each patient,and finally 7 features were selected(see the second part of the results for the names of features),The Rad-Score value of each patient was calculated by multiplying the feature by the corresponding regression coefficient,and a Radiomics classifier was constructed using the Rad-Score value.X-tile software sets the cutoff value of Rad-score to 7.17 and 7.71,and divides patients into low-risk group,intermediate-risk group and high-risk group.In training group and validation group,low-risk group,intermediate-risk group and high-risk group There were statistical differences in the survival rate between the two groups,P<0.0001.ROC analysis showed that the AUC predicted by TNM staging system and radiomics classifier for5-year survival rate of patients with oropharyngeal cancer was 0.629 and 0.740,and the combined AUC was 0.772.Compared with TNM staging system alone,radiomics classifier combined with TNM staging can significantly improve the prediction performance of patient OS.3.Multivariate COX regression analysis combined with clinicopathological features,blood test parameters and CT Radiomic features of OPC patients showed that T stage,N stage,albumin,γ-GGT,direct bilirubin and Rad-Score were independent factors affecting the prognosis of patients with oropharyngeal cancer.According to the above factors,the prediction model is constructed,and the Nomogram is drawn.The c-index index of the model in the training group and the verification group was 0.8 and 0.77,the calibration curve was consistent with the ideal 45 °dotted line,the model showed a good coincidence in predicting the 3-year and 5-year survival probability of OPC patients,and the decision analysis curve showed a good net benefit under different threshold probabilities,indicating that the model has good clinical application value in 3-year and 5-year survival rates of OPC patients.Increase the improvement degree of rad-score on clinical blood model: NRI was 0.382 and IDI was 0.111.Conclusion:1.Age,treatment,drinking history,TNM stage,γ-GGT,direct bilirubin,NLR,platelet count,total protein,PLR,albumin and fibrinogen are independent influencing factors of OS in patients with oropharyngeal cancer.Blood test parameters are very economical,convenient and easy to obtain prognostic indexs.Multiple interpolation technology can provide us with more information and improve statistical efficiency.2.The Radiomics classifier based on CT Radiomic features is significantly related to the prognosis of OPC patients.It is an independent factor affecting the prognosis of patients with OPC.It can be used as a supplement to the TNM staging system to improve its accuracy in predicting the prognosis of patients with OPC.3.Based on the clinicopathological features,blood test parameters and CT Radiomics of OPC patients,the nomograms for predicting the 1-year,3-year and 5-year survival rates were successfully established;The nomogram has passed the verification of the discrimination,calibration and clinical decision curve of the verification set,and has high discrimination,more accurate consistency and better net benefit of patients;The nomogram can be used as a visual tool to evaluate the prognosis of oropharyngeal cancer.It has good accuracy and clinical application value,and provides a theoretical basis for personalized treatment.

  • 【分类号】R739.63
节点文献中: