临床因素与影像组学联合模型列线图预测克罗恩病患者英夫利西单抗治疗后继发性无反应的价值

The value of combined model nomogram based on clinical characteristics and radiomics in predicting secondary loss of response after infliximab treatment in patients with Crohn′s disease

目的探讨基于CT小肠造影(CTE)的影像组学特征联合临床因素构建列线图预测克罗恩病(CD)患者英夫利西单抗(IFX)治疗后继发性无反应(SLOR)的价值。方法该研究为病例对照研究。回顾性收集2015年3月至2022年7月在安徽医科大学附属第一医院确诊的155例CD患者的临床和影像资料, 采用分层抽样法按7∶3的比例分成训练集(n=108)和测试集(n=47)。所有患者均按规范方案治疗, 根据治疗反应分为SLOR(训练集43例、测试集18例)和非SLOR(训练集65例、测试集29例)。基于训练集数据, 使用单因素和多因素logistic回归分析在临床资料中筛选出IFX治疗后SLOR的独立预测因素, 建立临床模型。选择肠期CTE图像沿病灶轮廓逐层勾画以获得感兴趣区体积并提取影像组学特征, 使用单因素分析及最小绝对收缩和选择算子筛选影像组学特征建立影像组学模型。采用多因素logistic回归分析基于筛选出的临床独立预测因素和影像组学特征建立临床-影像组学联合模型, 并绘制列线图。使用受试者操作特征曲线评估3个模型对IFX治疗后SLOR的预测效能, 并计算曲线下面积(AUC), 采用决策曲线来评价模型的临床实用价值。结果临床资料中病程(OR=1.983, 95%CI 1.966~2.000, P=0.046)、肠腔狭窄(OR=1.246, 95%CI 1.079~1.764, P=0.015)是SLOR的独立预测因素, 建立临床模型。最终9个影像组学特征被纳入影像组学模型。临床、影像组学和联合模型预测CD患者IFX治疗后SLOR的AUC在训练集中分别为0.691(95%CI 0.591~0.792)、0.896(95%CI 0.836~0.955)、0.910(95%CI 0.855~0.965), 在测试集中分别为0.722(95%CI 0.574~0.871)、0.866(95%CI 0.764~0.968)、0.889(95%CI 0.796~0.982)。测试集中决策曲线分析显示影像组学模型和联合模型的临床净效益均高于临床模型, 在大部分阈值概率区间内, 联合模型的临床净效益高于影像组学模型。结论基于CTE的影像组学模型可有效预测CD患者IFX治疗后SLOR, 加入临床因素病程和肠腔狭窄的联合模型可进一步提高预测效能。

Objective To investigate the value of nomogram based on radiomics features of CT enterography(CTE)combined with clinical characteristics to predict secondary loss of response(SLOR)after infliximab(IFX)treatment in patients with Crohn′s disease(CD).Methods This study was a case-control study.Clinical and imaging data of 155 patients with CD diagnosed at the First Affiliated Hospital of Anhui Medical University from March 2015 to July 2022 were retrospectively collected.The patients were divided into a training set(n=108)and a testing set(n=47)in the ratio of 7∶3 by stratified sampling method.All patients were treated according to the standardized protocol and were classified as SLOR(43 in the training set and 18 in the testing set)and non-SLOR(65 in the training set and 29 in the testing set)according to treatment outcome.Based on the data from the training group,independent clinical predictors of SLOR after IFX treatment were screened in the clinical data using univariate and multivariate logistic regression analysis to establish a clinical model.Intestinal phase images were selected to be outlined layer by layer along the margin of the lesion to obtain the volume of the region of interest to extract the radiomics features.The radiomics features were screened using univariate analysis and the minimum absolute shrinkage and selection operator to establish the radiomics model.Multivariate logistic regression analysis was used to build a combined clinical-radiomics model based on the screened clinical independent predictors and radiomics characters,then a nomogram was drawn.The predictive efficacy of the 3 models for SLOR after IFX treatment was assessed by receiver operating characteristic curves,and the area under the curve(AUC)was calculated.The decision curve analysis was applied to evaluate the clinical utility of the models.Results Disease duration(OR=1.983,95%CI 1.966-2.000,P=0.046)and intestinal stenosis(OR=1.246,95%CI 1.079-1.764,P=0.015)were identified as the independent predictors of SLOR in the clinical data,and a clinical model was established.Totally 9 radiomics features were included in the radiomics model.The AUCs of clinical,radiomics,and combined models for predicting SLOR after IFX treatment in CD patients were 0.691(95%CI 0.591-0.792),0.896(95%CI 0.836-0.955),and 0.910(95%CI 0.855-0.965)in the training set,and 0.722(95%CI 0.574-0.871),0.866(95%CI 0.764-0.968),and 0.889(95%CI 0.796-0.982)in the testing set.Decision curve analysis in the testing set showed higher net clinical benefits for both the radiomics model and combined model than the clinical model,and combined model had higher net clinical benefits than the radiomics model over most threshold probability intervals.Conclusions CTE-based radiomics model can effectively predict SLOR after IFX treatment in patients with CD,and a combined model by incorporating clinical characteristics of disease duration and intestinal stenosis can further improve the predictive efficacy.