摘要
目的 基于临床病理特征和影像学及血清生物指标分析构建肺结节(pulmonary nodules,PN)性质预测模型并验证,为肺癌早诊断和早治疗提供科学决策.方法 对 2019 年 1 月～2023 年 2 月深圳市中医医院胸外科和肿瘤科816例行手术切除或肺活检病理诊断明确的PN患者进行回顾性分析.其中,剔除不符合纳入标准者113 例,余下 703 例纳入研究.该研究基于PN患者临床病理特征(年龄、性别、吸烟史、戒烟史、癌症家族史)、胸部影像学(结节最大直径、病变位置、边界清晰、分叶、毛刺、空泡、血管集束征、钙化、空气支气管征、肺气肿、结节性质及胸膜凹陷、结节数量)和血清癌胚抗原(carcinoembryonic antigen,CEA)、细胞角蛋白19片段(cytokeratin 19 fragment,CYFRA21-1)、鳞状细胞癌抗原(squamous cell carcinoma antigen,SCCA),将上述病例随机分为建模组(n=552,良性237例,恶性315例)和验证组(n=151,良性85例,恶性66例).首先,对研究对象进行单变量分析以筛选有统计学意义的PN性质预测因子.然后,进行多变量回归分析以筛选PN性质的独立预测因子.最后,采用logistic回归分析构建PN性质的预测模型.再将验证组数据分别代入该模型与梅奥诊所(Mayo clinic,Mayo)模型、退伍军人事务(veterans affairs,VA)模型、Brock大学(Brock University,Brock)模型、北京大学(Peking University,PKU)模型和广州医科大学(Guangzhou Medical University,GZMU)模型计算PN恶性概率,绘制受试者工作特征(receiver operating characteristic,ROC)曲线.根据曲线下面积(area under the curve,AUC)比较各模型的诊断效能.结果 单变量分析筛选的具有统计学意义的变量包括年龄、癌症家族史、结节最大直径、结节性质、肺上叶、钙化、血管集束征、分叶、边界清晰、毛刺以及血清CEA,SCCA,CYFRA21-1等.多变量回归分析显示年龄、CEA,边界、CYFRA21-1,SCCA,肺上叶、结节最大直径、癌症家族史、毛刺、结节性质等为PN恶性的独立预测因子.该研究构建的PN性质预测模型方程如下:f(x)=ex/(1+ex),X=(-6.318 8+0.020 8年龄+0.527 4×CEA-0.928 4×边界+0.294 6×Cyfra21-1+0.294×结节最大直径+1.220 1×癌症家族史+0.573 2×肺上叶+0.064 8×SCCA +1.461 5×毛刺 +1.497 6×结节性质).该模型与Mayo 模型和VA 模型比较,AUC(0.799 vs 0.659,0.650)差异具有统计学意义(Z=3.029,2.638,P=0.003,0.008).然而,该模型与Brock 模型、PKU 模型、GZMU 模型比较,AUC(0.799 vs 0.762,0.773,0.769)差异无统计学意义(Z=1.063,0.686,0.757,P=0.288,0.493,0.449).结论 该研究构建的PN性质预测模型较为准确可靠,可帮助临床实现早诊断和早干预,值得推广应用.
Abstract
Objective The study aimed to construct and validate a predictive model for pulmonary nodules(PN)nature based on clinicopa-thological features,imaging,and serum biomarkers,so as to provide scientificdecision-making for early diagnosis and treatment of lung cancer.Methods A retrospective was performed on 816 PN patients with definited pathological diagnosis who received surgical resection analysisor lung biopsy in the Department of Thoracic Surgery and Oncology of Shenzhen Traditional Chinese Medicine Hospital from January 2019 to February 2023.Among them,113 cases that did not meet the inclusion criteria were excluded,and the remaining 703 cases were included in the study.The study based on the clinicopathologic features(age,gender,smoking history,smoking cessation history and family history of cancer),chest imaging(maximum diameter of nodule,location of lesion,clear border,Lobulation,spiculation,vascular convergence sign,vacuole,calcification,air bronchial sign,emphysema,nodule type and pleural indentation,nodule number)and serum carcinoembryonic antigen(CEA),cytokeratin 19 fragment(CYFRA21-1),squamous cell carcinoma antigen(SCCA)in patients with PN.These cases were randomly divided into a modeling group(n=552,237 benign,315 malignant)and a validation group(n=151,85 benign,66 malignant).First,univariate analysis was performed to screen for statistically significant predictors of nodules nature.Then,multivariate regression analysis was performed to screen for independent predictors of nodules nature.Finally,the prediction model of PN nature was constructed by logistic regression analysis.Subsequently,the validation group data were entered into the proposed model and Mayo clinic(Mayo)model,veterans affairs(VA)model,Brock University(Brock)model,Peking University(PKU)model and Guangzhou Medical University(GZMU)model,respectively.PN malignancy probability was calculated.The receiver operating characteristic(ROC)curves were plotted.The diagnostic efficiency of each model was compared according to the area under the curve(AUC).Results There were statistically significant variables including age,family history of cancer,maximum nodule diameter,nodule type,upper lobe of lung,calcification,vascular convergence sign,lobulation,clear border,spiculation,and serum CEA,SCCA,CYFRA21-1 using univariate analysis.Multiple regression analysis showed that age,CEA,clear border,CYFRA21-1,SCCA,upper lobe of lung,maximum nodule diameter,family history of cancer,spiculation and nodule type were independent predictors of PN nature.The prediction model equation constructed in this study is as follows:f(x)= ex/(1+ex),X=(-6.318 8+0.020 8×Age+0.527 4×CEA-0.928 4×clear border+0.294 6×Cyfra21-1+0.294×maximum nodule diameter+1.220 1×family history of cancer +0.573 2×upper lobe of lung +0.064 8×SCCA +1.461 5×Spiculation +1.497 6×nodule type).The AUC(0.799 vs 0.659,0.650)of the proposed model was significantly higher compared with Mayo model and VA model,and there were statistically significant differences(Z=3.029,2.638,P=0.003,0.008).However,compared with Brock model,PKU model and GZMU model,the differences of AUC(0.799 vs 0.762,0.773,0.769)were not statistically significant(Z=1.063,0.686,0.757,P=0.288,0.493,0.449).Conclusion The prediction model for PN nature established in this study is accurate and reliable,which can help clinics with early diagnosis and early intervention,and this prediction model deserves to be popularized.
基金项目
深圳市科技计划(JCYJ20190812180001770)
国家科技期刊平台
NSTL
维普
万方数据