Hannah Gordon, Silvia Minozzi, Uri Kopylov, Bram Verstockt, Maria Chaparro, Christianne Buskens, Janindra Warusavitarne, Manasi Agrawal, Mariangela Allocca

Machine learning models and over-fitting considerations

This article is the second in a series of two publications on the European Crohn’s and Colitis Organisation [ECCO] evidence-based consensus on the management of Crohn’s disease. The first article covers medical management; the present article addresses surgical management, including preoperative aspects and drug management before surgery. It also provides technical advice for a variety of common clinical situations. Both articles together represent the evidence-based recommendations of the ECCO for Crohn’s disease and an update of prior ECCO Guidelines.

Summary: Background: Biologic therapies in patients with Crohn's disease often yield low clinical and endoscopic remission rates. After multiple failed therapies, combining two biologic therapies is possibly the sole medical alternative to recurrent surgery. However, data on this approach are limited.Aims: To assess the efficacy and safety of concomitant use of two biologic therapies in the largest cohort to date of refractory Crohn's disease patients.Methods: Data were extracted from Crohn's disease patients started on dual biologic therapy at two referral centres. Biologics utilised include infliximab, adalimumab, vedolizumab, ustekinumab, certolizumab and golimumab. The primary outcome was endoscopic improvement (>50% reduction in Simplified Endoscopic Score‐Crohn's disease [SES‐CD] or explicitly stated). Endoscopic remission (SES‐CD < 3 or stated), clinical response (Crohn's disease–patient‐reported outcome‐2 score [PRO2] reduced by 8), clinical remission (PRO2 < 8), and C‐reactive protein (CRP) were also assessed.Results: A total of 22 patients with 24 therapeutic trials of dual biologic therapy were identified. The majority of patients had prior surgical resections (91%), stricturing (59%) or penetrating (36%) phenotype, and perianal fistulas (50%). Median number of prior failed biologics was 4. Endoscopic improvement occurred in 43% of trials and 26% achieved endoscopic remission. Fifty per cent had clinical response and 41% achieved clinical remission. There were significant post‐treatment reductions in median SES‐CD (14.0 [12.0‐17.5] to 6.0 [2.5‐8.0], P = 0.0005], PRO‐2 (24.1 [20.3‐27.0] to 13.4 [4.6‐21.8], P = 0.002] and CRP (17.0 [11.0‐24.0] to 9.0 [4.0‐14.0], P = 0.02). Presence of perianal fistulas decreased from 50% to 33%. Adverse events occurred in 13% of trials.Conclusion: Dual biologic therapy was associated with clinical, biomarker and endoscopic improvements in selected patients with refractory Crohn's disease who failed multiple biologics. Further studies are needed to validate this approach.

Noninvasive tests to measure endoscopic activity in patients with Crohn's disease (CD) have limitations. We aimed to develop a test to identify patients in remission, based on endoscopic analysis, and monitor CD activity based on serum levels of proteins. We developed a test to measure 13 proteins in blood (ANG1, ANG2, CRP, SAA1, IL7, EMMPRIN, MMP1, MMP2, MMP3, MMP9, TGFA, CEACAM1, and VCAM1), called the endoscopic healing index [EHI], using samples from 278 patients with CD from a multinational training cohort. We validated the test using 2 independent cohorts of patients with CD: 116 biologic-naive patients with early-stage CD (validation cohort 1) and 195 biologic-exposed patients with chronic CD (validation cohort 2). The ability of the test to identify patients with active disease vs patients in remission (defined as a simple endoscopic score for CD of ≤2 and ≤1 in each segment, or a total CD endoscopic index of severity score <3) was assessed by using area under receiver operating characteristic curve (AUROC) analysis. The diagnostic accuracy of the test was compared with that of measurement of serum C-reactive protein (CRP) and fecal calprotectin. The EHI scores range from 0 to 100 units; higher scores indicate more severe CD activity, based on endoscopy findings. The EHI identified patients in remission with an AUROC of 0.962 in validation cohort 1 (95% confidence interval, 0.942-0.982) and an AUROC of 0.693 in validation cohort 2 (95% confidence interval, 0.619-0.767), regardless of CD location or phenotype. A cutoff value of 20 points identified patients in remission with the highest level of sensitivity (97.1% in validation cohort 1 and 83.2% in validation cohort 2), with specificity values of 69.0% and 36.6%, respectively. A cutoff value of 50 points identified patients in remission with the highest level of specificity (100% in validation cohort 1 and 87.8% in validation cohort 2), with sensitivity values of 37.3% and 30.0%, respectively. The EHI identified patients in remission with a significantly higher AUROC value than the test for CRP (0.876, P < .001 in validation cohort 1 and 0.624, P = .109 in validation cohort 2). In analysis of patients with available FC measurements, the AUROC value for the EHI did not differ significantly from that of measurement of FC (AUROC, 0.950 for EHI vs AUROC, 0.923 for FC; P = .147 in validation cohort 1 and AUROC, 0.803 for EHI vs AUROC, 0.854 for FC; P = .298 in validation cohort 2). We developed an index called the EHI to identify patients with CD in endoscopic remission based on blood levels of 13 proteins. The EHI identified patients with resolution of endoscopic disease activity, with good overall accuracy, although with variation between the 2 cohorts assessed. The EHI AUROC values were comparable to measurement of FC and higher than measurement of serum CRP. The test might be used in practice to assess endoscopic activity in patients with CD.