IDDF2021-ABS-0039 Identifying patients with crohn’s disease at high risk of primary nonresponse to infliximab therapy using a radiomic-clinical nomogram

Background13%–40% of patients with Crohn’s disease (CD) show a primary loss of response to infliximab therapy. This study aimed to develop and validate a novel radiomic-clinical nomogram to identify CD patients at high risk of primary nonresponse to infliximab therapy.MethodsThis retrospective study included 224 CD patients (training, n=118; test 1, n=52; test 2, n=54) recruited from one of the two-three tertiary referral centers. The computed tomography enterography (CTE) and clinical data before infliximab treatment of CD patients were collected and the outcome of infliximab induction therapy at week 14 was classified as response or primary nonresponse. In the training cohort, 1130 radiomics features were extracted from the whole inflamed gut of CD patients on arterial-phase CTE; Least absolute shrinkage and selection operator and support vector machine methods were performed to select key features and construct a radiomics signature; moreover, a clinical factors model was developed based on the selected clinical risk factors. Subsequently, a radiomic-clinical nomogram which integrated the radiomics signature and the selected clinical predictors was developed based on a multivariable logistic regression analysis. The predicting performance of this nomogram was validated in two external independent test cohorts.ResultsIn the training and two test cohorts, the radiomics signature showed an acceptable discrimination for predicting primary nonresponse to infliximab therapy (an area under the curve [AUC], 0.861, 0.827, and 0.769, respectively; all P<0.05); after adding the clinical predictors (albumin and body mass index) to radiomic signature for developing a radiomic-clinical nomogram (IDDF2021-ABS-0039 Figure 1. Radiomic-clinical nomogram), its prediction performance (AUC, 0.891 ,0.841, and 0.804, respectively; all P<0.05; IDDF2021-ABS-0039 Figure 2. ROC analysis of the prediction performance of the radiomic-clinical nomogram) was significantly improved comparing with radiomics signature alone. Decision curve analysis demonstrated that the radiomic-clinical nomogram provided a better net benefit to predicting primary nonresponse to infliximab than radiomics signature and the clinical factors model (IDDF2021-ABS-0039 Figure 3. Decision curve analysis of radiomic-clinical nomogram, radiomic signature and clinical factors model).ConclusionsThe radiomic-clinical nomogram may be a promising tool to allow accurately identify CD patients at high risk of primary nonresponse to infliximab therapy.