基于胸部CT平扫和临床参数的主动脉夹层预测模型的建立与初步验证  

Establishment and preliminary validation of a prediction model for aortic dissection based on non-contrast chest CT and clinical features

作　　者：张亚楠 崔磊[1] 薛颖 刘佳 明佳蕾 ZHANG Yanan;CUI Lei;XUE Ying;LIU Jia;MING Jialei(Department of Radiology,Nantong First People’s Hospital,Nantong 226001,China)

机构地区：[1]南通市第一人民医院影像科,南通226001

出　　处：《国际医学放射学杂志》2023年第5期516-522,共7页International Journal of Medical Radiology

基　　金：南通市卫健委青年课题A(QA2020001)。

摘　　要：目的建立并验证一种基于胸部CT平扫征象结合临床参数的主动脉夹层(AD)预测模型。方法回顾性收集行胸部CT平扫并短期内行胸主动脉CTA的胸痛病人322例,其中男222例、女100例,平均年龄(60.6±13.2)岁。按7∶3的比例采用完全随机法将病人分为训练集(225例,其中AD 105例)与验证集(97例,其中AD 45例)。采用单因素分析和多因素Logistic回归分析筛选CT平扫征象及临床参数中AD的危险因素,并建立临床模型、影像模型及联合的Logistic回归预测模型,采用受试者操作特征(ROC)曲线、特异度、敏感度评估模型的预测效能,以筛选最佳预测模型。采用Delong检验比较各模型的曲线下面积(AUC),通过一致性指数(C指数)评估预测模型的区分度。采用Hosmer-Lemeshow检验评价模型的校准度。结果Logistic回归分析显示男性、D-二聚体升高、管腔内线样高密度和钙化斑块内移为AD的独立危险因素(均P<0.05)。联合的Logistic回归预测模型的AUC高于临床模型、影像模型(均P<0.05),联合模型的诊断效能最佳;在训练集中模型的AUC为0.937,敏感度91.4%,特异度82.5%;在验证集中模型的AUC为0.933,敏感度95.6%,特异度82.7%。联合模型具有较好的区分度(训练集:C指数=0.937,验证集:C指数=0.933)和较好的拟合效果(均P>0.05)。校准曲线显示联合模型预测的AD发生风险与胸主动脉CTA得到的结果有较好的一致性。结论建立基于胸部CT平扫征象结合临床参数的AD个体化Logistic回归预测模型,有助于快速筛查和早期识别AD病人。Objective To establish and validate a prediction model for aortic dissection(AD)based on non-contrast chest CT signs and clinical parameters.Methods We retrospectively collected data from 322 patients with chest pain,including 222 males and 100 females,with an average age of(60.6±13.2)years,who underwent chest CT scan and CTA of the thoracic aorta in a short period.Patients were randomly divided into the training set(225 cases,including 105 cases with AD)and the verification set(97 cases,including 45 cases with AD)using a 7∶3 ratio.We conducted univariate and multivariate Logistic regression analyses to identify risk factors for AD based on non-contrast CT signs and clinical parameters.We established clinical,imaging,and combined Logistic regression prediction models.Model performance was evaluated using ROC curves,specificity,and sensitivity to select the best predictive model.The Delong test compared the AUC of each model,and the consistency index(C index)assessed the differentiation of the prediction models.We used the Hosmer-Lemeshow test to evaluate model calibration.Results Logistic regression analysis identified male gender,elevated D-dimer,intraluminal line-like hyperdensity,and calcified plaque migration as independent risk factors for AD(all P<0.05).The combined Logistic regression model had a higher AUC than the clinical and imaging models(both P<0.05).The combined Logistic regression prediction model showed the best diagnostic efficacy,with an AUC of 0.937,a sensitivity of 91.4%,and a specificity of 82.5%for the training set,and an AUC of 0.933 and a sensitivity of 95.6%for the validation set,with a specificity of 82.7%.The combined prediction model demonstrated superior discrimination(training set:C index=0.937,validation set:C index=0.933)and a better fit(both P>0.05 in the Hosmer-Lemeshow test).The calibration curve indicated that the combined model’s predictions of AD risk closely matched the results obtained by CTA of the thoracic aorta.Conclusion Establishing an individualized Logistic regr

关 键 词：主动脉夹层 体层摄影术 X线计算机 预测模型 

分 类 号：R814.42[医药卫生—影像医学与核医学] R816.2[医药卫生—放射医学]