IDDF2022-ABS-0157 CT-based radiomics signature of visceral adipose tissue early predicts adverse outcomes in patients with Crohn's disease: a multicenter cohort study. (2nd September 2022)

• Record Type: Journal Article
• Authors: Li, Xuehua; Zhang, Naiwen; Hu, Cicong; Mao, Ren; Huang, Bingsheng
• Is Part Of: Gut. Volume 71(2022)Supplement 2
• Page Start: A147
• Page End: A148
• Publication Date: 2022-09-02
• Subjects: Gastroenterology -- Periodicals; 616.33
• DOI: 10.1136/gutjnl-2022-IDDF.203
• Languages: English
• ISSNs: 0017-5749

Abstract

Abstract : Background: Increasing data have shown that visceral adipose tissue (VAT) is involved in the pathogenesis of Crohn's disease (CD). As lack of a satisfactory approach to noninvasively characterize visceral adipose tissue (VAT), its effect on the clinical outcome of patients with CD remains unclear. We developed and validated a VAT-radiomics model (RM) using baseline computed-tomography (CT) images to predict CD adverse outcomes, and compare its predictive efficacy with subcutaneous adipose tissue (SAT)-RM and other conventional metrics of adipose tissue. Methods: This retrospective-prognostic study consecutively included 271 CD patients (training, n=167; test, n=104) who underwent baseline CT examination at one of the eight tertiary referral centres from March 2010 through April 2021. Adverse outcomes during follow-up referred to the presence of penetrating diseases, bowel obstruction, or CD-related surgery. 1130 radiomics features were extracted from VAT on plain CT in training cohort; subsequently, a machine learning-based VAT-RM was developed using the selected reproducible features and validated in the external test cohort. Using a similar modeling process, an SAT-RM was developed. The predictive performance of VAT-RM was compared with SAT-RM and the other six metrics of adipose tissue (e.g., VAT volume, etc). Results: In the test cohort, the area under the ROC curve (AUC) of VAT-RM for predicting adverse outcomes was higher than SAT-RM (AUC=0.873 vs. AUC=0.805; IDDF2022-ABS-0157 Figure 1). According to multivariate cox regression analysis, VAT-RM (hazard ratio [HR]=8.482, P =0.006) was the most important independent predictor, followed by SAT-RM (HR=4.792, P =0.007). However, SAT-RM failed to significantly improve predictive efficacy after adding it to VAT-RM (integrated discrimination improvement=0.183, P =0.005). Decision curve analysis further confirmed a better net benefit of VAT-RM than SAT-RM (IDDF2022-ABS-0157 Figure 2). None of the six metrics of adipose tissue correlated well with adverse outcomes (all P >0.05). Conclusions: Our VAT-RM allows for early and accurately identifying high-risk patients who were prone to suffer adverse outcomes and outperforms SAT-RM and other conventional fat metrics.