预测肺腺癌EGFR突变的列线图模型的建立及验证

Establishment and validation of a nomogram model for predicting EGFR mutations in lung adenocarcinoma

目的基于临床因素及^(18)F-FDG PET/CT代谢参数建立预测肺腺癌表皮生长因子受体(EGFR)突变的列线图模型并验证。方法回顾性分析2014年1月至2019年1月间哈尔滨医科大学附属第一医院的114例肺腺癌[男59例、女55例,年龄(60.0±10.8)岁]患者的临床资料[吸烟状态、肿瘤位置、临床分期及癌胚抗原(CEA)水平]、^(18)F-FDG PET/CT代谢参数[SUVmax、肿瘤代谢体积(MTV)及病灶糖酵解总量(TLG)]及EGFR突变检测结果。将患者分为训练组(80例)及验证组(34例)。在训练组中,采用单因素分析(两独立样本t检验、Wilcoxon秩和检验、χ2检验或Fisher确切概率法)选取EGFR突变组与野生组间差异有统计学意义的变量。计算方差膨胀系数(VIF)删除存在共线性的变量后,基于赤池信息准则(AIC)构建最优logistic模型的列线图模型。在训练组及验证组中采用一致性指数(C-index)、灵敏度、特异性、准确性、校准度及决策曲线分析(DCA)等评估模型效果。结果114例患者中,EGFR突变型56例、EGFR野生型58例。在训练队列中,EGFR突变组与野生组间性别(男/女:14/26与25/15;χ2=6.05,P=0.014)、吸烟状态(有/无吸烟史:4/36与22/18;χ2=18.46,P<0.001)及SUVmax[5.72(3.90,8.32)与8.09(4.56,12.55);W=1045.50,P=0.018]的差异有统计学意义;余指标差异均无统计学意义(t=-0.54,χ2值:0.20和0.20,W值:921.50和983.00,均P>0.05)。性别、吸烟状态和SUVmax的VIF均小于10,同时由3种因素构成的列线图模型具有最小AIC(90.06)。模型在训练组中C-index值为0.798(95%CI:0.699~0.897)、灵敏度为85.0%(34/40)、特异性为70.0%(28/40)、准确性为77.5%(62/80)。在验证组中C-index值为0.854(95%CI:0.725~0.984)、灵敏度为13/16、特异性为14/18、准确性为79.4%(27/34)。模型具有良好的校准度,DCA示模型在较大的阈值范围内(训练组:0~0.59,验证组:0~0.65)能使患者临床获益。结论基于性别、吸烟状态及SUVmax的列线图模型能够协助临床便捷预测肺腺癌EGFR突变状态。

Objective To construct and validate a nomogram model based on clinical factors and PET/CT metabolic parameters of ^(18)F-FDG for predicting epidermal growth factor receptor(EGFR)mutations in lung adenocarcinoma.Methods From January 2014 to January 2019,114 patients(59 males,55 females,age(60.0±10.8)years)with lung adenocarcinoma in the First Affiliated Hospital of Harbin Medical University were retrospectively enrolled.Clinical data(smoking status,tumor location,clinical stage and carcinoembryonic antigen(CEA)level),^(18)F-FDG PET/CT metabolic parameters(SUVmax,metabolic tumor volume(MTV)and total lesion glycolysis(TLG))and EGFR mutation status were analyzed.Patients were divided into training group(80 cases)and validation group(34 cases).In the training group,univariate analyses(independent-sample t test,Wilcoxon rank sum test,χ2 test or Fisher′s exact probability method)were used for categorical variables.Variables that showed significant differences between EGFR mutation group and wild type group were selected.Variance inflation factors(VIF)were calculated and the collinearity variables were deleted,and a nomogram model of optimal logistic model was constructed based on Akaike information criterion(AIC).The effect of the model was evaluated by the concordance index(C-index),sensitivity,specificity,accuracy,calibration and decision curve analysis(DCA)in the training group and the validation group.Results Among 114 patients,56 were with EGFR mutations and 58 were with EGFR wild type.In the training group,there were significant differences in gender(male/female:14/26 vs 25/15;χ2=6.05,P=0.014),smoking status(with/without smoking history:4/36 vs 22/18;χ2=18.46,P<0.001)and SUVmax(5.72(3.90,8.32)vs 8.09(4.56,12.55);W=1045.50,P=0.018)between EGFR mutation group and wild type group.However,there were no significant differences in other factors(t=-0.54,χ2 values:0.20 and 0.20,W values:921.50 and 983.00,all P>0.05).The VIF of gender,smoking status and SUVmax were all less than 10,and the nomogram model with three factors showed the minimum AIC(90.06).In the training group,C-index value of the model was 0.798(95%CI:0.699-0.897),with the sensitivity of 85.0%(34/40),the specificity of 70.0%(28/40)and the accuracy of 77.5%(62/80).In the validation group,C-index value was 0.854(95%CI:0.725-0.984),with the sensitivity of 13/16,the specificity of 14/18,and the accuracy of 79.4%(27/34).The calibration curve and the goodness of fit test showed good calibration,and DCA showed that the model could benefit patients clinically within a large risk threshold range(training group:0-0.59,validation group:0-0.65).Conclusion The nomogram model based on gender,smoking status and SUVmax can be used to easily predict EGFR mutation status in lung adenocarcinoma.