直肠癌淋巴结转移风险预测模型构建研究  

Construction of a risk prediction model for lymphatic metastasis in rectal cancer

作　　者：蒋怡[1] 郭杨 胡鑫晔[1] 蒋美娜[1] 贾艳清[1] 郑伯安[1] JIANG Yi;GUO Yang;HU Xinye;JIANG Meina;JIA Yanqing;ZHENG Bo'an(Department of Colorectal Surgery,Zhejiang Provincial People's Hospital(Hangzhou Medical College Affiliated to People's Hospital),Hangzhou 310014,China)

机构地区：[1]浙江省人民医院(杭州医学院附属人民医院)结直肠外科,杭州310014

出　　处：《浙江医学》2024年第20期2161-2166,2172,共7页Zhejiang Medical Journal

基　　金：浙江省自然科学基金项目(LQ21H160042)。

摘　　要：目的构建直肠癌(RC)淋巴结转移(LNM)风险预测模型。方法回顾性选取浙江省人民医院2015年6月至2023年12月经根治性手术后病理检查证实为RC的患者745例,按7∶3比例分为训练集(用于模型构建)522例和验证集(用于模型验证)223例。比较训练集LNM阳性与阴性患者临床资料,采用多因素logistic回归模型分析训练集RC患者LNM影响因素,基于差异有统计学意义的影响因素绘制RC LNM风险预测模型的列线图,并采用ROC曲线、校准曲线、决策曲线分析和Bootstrap法进行模型验证。结果RC根治术后根据TNM分期将训练集患者分为LNM阳性224例和LNM阴性298例,LNM阳性与阴性患者在是否印戒细胞癌、组织学分级、T分期、癌胚抗原(CEA)、糖类抗原19-9(CA19-9)、血小板与淋巴细胞比值(PLR)等方面比较,差异均有统计学意义(均P<0.01)。组织学分级低或极低分化(OR=7.524)、T3~4(OR=3.421)、CEA>5 ng/mL(OR=1.652)、CA19-9>39 U/mL(OR=2.240)、PLR>132.5(OR=1.625)均是训练集RC患者LNM的独立危险因素(均P<0.05)。基于组织学分级、T分期、CEA、CA19-9、PLR构建的RC LNM风险预测模型在训练集、验证集中的AUC及95%CI分别为0.755(0.714~0.797)、0.705(0.636~0.774),提示具有较强的预测能力;校准曲线显示对LNM风险的预测概率和实际概率之间一致性高,提示具有良好的准确性和稳定性;决策曲线显示训练集、验证集的阈值概率分别为0~0.75、0~0.60,净收益率均远高于2条无效线,提示具有较强的临床适用性;Bootstrap法显示准确率为0.73,Kappa值为0.430,提示泛化能力较好。结论本研究基于组织学分级、T分期、CEA、CA19-9、PLR构建的RC LNM风险预测模型具有潜在的临床应用价值。Objective To construct a risk prediction model for lymphatic metastasis(LNM)of rectal cancer(RC).Methods Clinical data of 745 patients with pathologically confirmed RC underwent radical surgery in Zhejiang Provincial People's Hospital from June 2015 to December 2023 were collected.Patients were randomly assigned in a training set(n=522)and a validation set(n=223)with a ratio of 7∶3.The clinical data of LNM positive and negative patients in the training set were compared,and the influencing factors of LNM in RC patients in the training set were analyzed using multi-factor logistic regression model.The column graph of the RC LNM risk prediction model was drawn based on the influencing factors with statistically significant differences.The model was verified by ROC curve,calibration curve,decision curve analysis and Bootstrap method.Results After RC radical surgery,the patients in the training set were divided into 224 patients with LNM positive and 298 patients with LNM negative according to TNM stage.There were statistically significant differences in signet ring carcinoma,histological grade,T stage,carcinoembryonic antigen(CEA),carbohydrate antigen 19-9(CA19-9),platelet to lymphocyte ratio(PLR)between the patients with LNM positive and negative(all P<0.01).The RC with low or very low differentiation(OR=7.524),T3~4(OR=3.421),CEA>5 ng/mL(OR=1.652),CA19-9>39 U/mL(OR=2.240),and PLR>132.5(OR=1.625)were independent risk factors for LNM in patients of training set(all P<0.05).A prediction nomogram was constructed based on above risk factors.The AUC of the nomogram for predicting LNM risk in RC patients was 0.755(95%CI:0.714-0.797)for the training set and 0.705(95%CI:0.636-0.774)for the validation set,respectively.The calibration curves showed high consistency between predicted probability of LNM risk and the actual probability.The decision curve showed that the threshold probabilities of the training set and validation set were 0-0.75 and 0-0.60,respectively,and the net yields were much higher than the 2 invalid lines

关 键 词：直肠癌 淋巴结转移 临床病理特征 预测模型 列线图 

分 类 号：R735.37[医药卫生—肿瘤]