Introduction

Biologic treatment plays a crucial role in managing Crohn’s disease (CD) prior to considering surgical options. However, treatment failure and withdrawal are common and jeopardize the maintenance of remission in CD.

Aims & Methods

This study aims to analyse and compare drug survival – used as a surrogate marker for treatment efficacy – and reasons for withdrawal of the commonly used biologics in CD. This is a retrospective, monocentric study of CD patients treated at the University Hospital Heidelberg and registered in the IBD database from 2017 to April 2024. The primary endpoint was drug survival; the secondary endpoint was reason for treatment withdrawal. Drug survival was analysed using Kaplan-Meier estimators in SPSS. First-line treatments with Adalimumab (ADA) and Infliximab (IFX) were compared using log-rank tests following 1:2 Propensity Score Matching in R Studio.

Results

Of 1020 IBD patients, 396 CD patients were eligible. The included cohort had a median age of 48 years (IQR: 35-60) with 44.9 % male and a median disease duration of 9 years (IQR: 2-20) until therapy initiation. 25.5 % reported to have perianal disease. Among their 1035 immunosuppressive treatments, 771 involved the biologics ADA, IFX, Vedolizumab (VDZ) and Ustekinumab (UST). ADA showed the highest median drug survival (27 months; 95 % CI: 19.8-34.2) which was significantly longer than that of IFX (14 months; 95 % CI: 10.6-17.4). Median drug survival of UST was 22 months (95 % CI: 13.4-30.6) and 15 months for VDZ (95 % CI: 7.2-22.8).
The most common reason for withdrawal for ADA (41.3 %), VDZ (76.8 %) and UST (61.8 %) was insufficient response which included persistent clinical symptoms, elevated inflammatory markers, or endoscopic activity. For IFX side effects (24.4 %), insufficient response (21.1 %) and immunogenicity (20.3 %) were almost equally present.
In first-line treatments (47 IFX vs. 94 ADA-cases), the log-rank test showed significantly longer drug survival for ADA compared to IFX (49.7 vs. 33.6 months; p=0.034).

Conclusion

Adalimumab showed greater overall and first-line drug survival than infliximab, hinting at a preference of adalimumab in clinical practice. Most biologic treatments were withdrawn due to insufficient response. More prospective, multicentric studies are encouraged to confirm these findings and to support optimised treatment decisions.

Disclosure

Annika Gauss received lecture fees and consulting fees from AbbVie, Johnson & Johnson, Takeda, Tillotts Pharma, Hexal, Pfizer and Sanofi.

Introduction

Artificial intelligence tools for endoscopic assessment of Ulcerative Colitis remain limited due to suboptimal agreement with expert readers and reliance on small, inconsistently labeled datasets. We developed a novel computer vision model (DovaVision) trained on a large-scale, expert-annotated video dataset to assess UC activity using detailed frame-level features.

Aims & Methods

DovaVision was pre-trained using DINOv2 self-supervised learning on 845 colonoscopies and millions of GI images, then significantly refined using annotations collected via a proprietary platform. 15 IBD central readers (CRs) scored video frames for five UC features (vascular pattern, bleeding, ulcers/erosions, friability, erythema) across multiple severity levels. Annotators underwent standardized onboarding to ensure high inter-rater reliability. The model comprises two vision transformer architectures: a Quality Control (QC) module to exclude low-quality/non-diagnostic frames, and a Scoring module that outputs feature-specific severity grades. Model outputs include burned-in heatmaps (for mapping of disease activity to specific timepoints) and structured JavaScript Object Notation (JSON) files.

Results

A total of 83.8 million frames from 200 colonoscopy videos were labeled by 3 expert CRs across 7 categories per video, yielding 16.9 million merged labels by majority vote. Inter-rater agreement on modified Mayo Endoscopic Score (MES) showed strong reliability (ICC = 0.86; QWK = 0.88). On a 31-video validation set from 3 separate data sources, DovaVision showed equivalency with the gold standard of CR consensus (2+1 majority vote), achieving a QWK of 0.904 and ICC of 0.907. In comparison, the pool of 3x CRs assigned/video on this same set showed ICC of 0.905. On friability assessment—a challenging feature-- DovaVision showed 90% agreement and 0.811 ICC when compared to gold standard (3x CR majority vote), while the agreement for 3 CRs per video was only 0.447 ICC on the same set. The average model inference time was short at 3.5 minutes per video. The model also outputs continuous “frame-by-frame” level information on the sub-score characteristics. The model is an entire data processing pipeline deployed on a cloud architecture with an API for secure authentication, bulk data upload, and secure results storage/retrieval.

Conclusion

DovaVision is a validated AI-based model for automated grading of UC severity on endoscopy. Its high concordance with expert readers, detailed feature-level (sub-score characteristics) analysis, and rapid inference support real-world integration into clinical and research workflows. Although further validation with additional datasets is required.

Disclosure

Dr. Michael Byrne is the Chairman, Founder and CMO at Dova Health Intelligence

Introduction

Epithelial injury is a hallmark of chronic mucosal damage in Crohn’s disease (CD), which can lead to fibrostenotic strictures. Type VII collagen, a critical component of the epithelial basement membrane, forms U-shaped anchoring fibrils that link the basement membrane to the interstitial matrix by binding fibrillar collagen. Genetic mutations in the COL7A1 gene, which encodes type VII collagen, are known to impair fibril formation and reduce tissue tensile strength, resulting in epithelial dysfunction, as seen in disorders such as dystrophic epidermolysis bullosa. This highlights the essential structural role of type VII collagen in maintaining matrix stability. In CD, the importance of this protein is underscored by the presence of autoantibodies targeting type VII collagen in up to 50% of patients with CD. Loss of tissue integrity, because of loss of the interaction between the basement and interstitial matrix, leads to tissue dysfunction and fibrosis.

Aims & Methods

We investigated C7M, a biomarker of type VII collagen degradation, to assess its specificity for intestinal mucosal damage. We analyzed serum samples from CD patients enrolled in a prospective cohort study. The cohort included 62 CD patients undergoing surgical resection for fibrostenotic strictures and 49 patients with luminal disease receiving treatment adjustments without surgery. Serum was collected at baseline, post-operative day 1 (POD1), and one-month post-resection. C7M levels were quantified using a competitive ELISA.

Results

At baseline, C7M levels were significantly elevated in fibrostenotic CD patients (median: 9.79ng/mL) compared to those with luminal CD patients (median: 6.03ng/mL) (p<0.01, AUC: 0.69). In the resected group, C7M levels rose sharply on POD1 (p<0.0001, AUC: 0.81), suggesting acute release in response to surgical injury. By one-month post-surgery, C7M levels had returned to baseline, closely resembling those observed in luminal CD patients (p<0.0001, AUC: 0.80). C7M also demonstrated mild-moderate correlations to biochemical markers of inflammation, including CRP (rho=0.62, p<0.0001), hemoglobin (rho=-0.28, p=0.0024), thrombocytes (rho=0.24, p=0.0105), albumin (rho=-0.52, p<0.0001), and fecal calprotectin (rho=0.25, p=0.019).

Conclusion

C7M was elevated in fibrostenotic CD patients relative to luminal CD, indicating its potential as a biomarker of more severe or complex mucosal pathology. The pronounced but transient increase following surgery, with normalization within one month, further supports C7M as a dynamic marker of intestinal epithelial damage. These findings suggest C7M may serve as a novel, disease-specific biomarker for assessing mucosal integrity and inflammatory burden in CD.

Disclosure

Full time employee at Nordic Bioscience

Introduction

The intestinal mucosa undergoes constant renewal, with epithelial and immune cells routinely shed into the gut lumen. Recent findings suggest that these shed cells remain viable and retain transcriptomic signatures relevant to bowel pathology. Analysis of luminal content obtained during colonoscopy indicates that Fecal Host Transcriptomes (FHT) mirror histologic inflammation in IBD. This study aimed to evaluate whether sequencing of stool samples can noninvasively reflect the presence of intestinal inflammation in patients with IBD, predict the location of inflammation within the bowel and distinguish between IBD type; Crohn’s Disease and Ulcerative Colitis.

Aims & Methods

Our aim was to investigate whether transcriptomic analysis of stool samples that are taken ambulatory or at home could reveal the inflammation signature of inflamed IBD patients.
Bulk RNA sequencing of stool samples from 71 IBD patients and healthy controls was used to generate FHTs. Each sample was paired with an endoscopic assessment and clinical data to capture a time-matched view of intestinal inflammation. Three random forest classifiers were trained to: (1) distinguish active inflammation from remission; (2) predict inflammation location in Crohn’s disease (colon vs. small intestine). and (3) classifier based on gene modules expression was used to differ between CD and UC in patients with active inflammation.

Results

Fecal RNA quality was high, with a median of 2,140 human genes detected per sample. The FHT test demonstrated robust performance across all tested applications. The first model, designed to distinguish active inflammation from remission (n = 71), achieved a sensitivity of 85% and a specificity of 92%. The second model, which predicted the location of inflammation in Crohn’s disease patients (n = 21), achieved 100% sensitivity and 100% specificity. The third model, aimed at differentiating Crohn’s Disease from Ulcerative Colitis (n = 35), reached a sensitivity of 83% and a specificity of 100%. These findings underscore the potential of FHTs as a non-invasive tool for comprehensive inflammation profiling in IBD.

Conclusion

Fecal shed cell transcriptomics offers a novel, non-invasive approach for assessing mucosal inflammation at the endoscopic level in IBD patients. This approach enables differentiation between disease activity states, anatomical localization of inflammation, and IBD subtype classification. Ongoing studies aim to evaluate its utility in grading inflammation severity and predicting therapeutic response.

References

Dan et al., CMGH 2023
Ungar et al., GUT 2022

Introduction

Hepatobiliary disorders are recognised extraintestinal manifestations in inflammatory bowel disease (IBD). Up to 30% of patients exhibit abnormal liver function tests and 5% will develop chronic liver disease (1). IBD-associated liver diseases (IBDALD), comprising primary sclerosing cholangitis (PSC), autoimmune hepatitis (AIH) and AIH/PSC overlap syndrome carry significant morbidity through progressive hepatobiliary damage and increased cancer risk. Early recognition is challenging due to non-specific biochemical derangement and heterogeneous presentation. No validated stratification tools exist for prospective risk identification. This study investigated whether routinely collected biomarkers convey predictive value for IBDALD, leveraging large-language model (LLM) refined phenotyping.

Aims & Methods

We analysed 1,571 paediatric and adult-onset IBD patients recruited to the “Genetics of IBD” study (Southampton, UK; 2010–). IBDALD cases were identified through LLM-assisted (Llama 3.1-70B-Instruct) structuring of 32,041 free-text imaging and histopathology reports, followed by clinician verification (2). Participants diagnosed with IBDALD prior to IBD were excluded from further analysis. Examined biomarkers included ALT, ALT and ESR (n=335,605 datapoints for the examined cohort). For each biomarker, patient-level medians were calculated from all available results within ±6 months of IBD diagnosis. A 1:4 individual matched case-control design was applied (matched on age at IBD diagnosis, gender, and IBD subtype) in R. Conditional logistic regression models assessed associations between IBDALD and biomarker values, treated as both continuous (patient-level medians) and binary – normal versus abnormal - (age- and sex-adjusted lab range flags) variables. Longitudinal biomarker trajectories were explored using LOESS and linear mixed-effects models.

Results

Thirty-five cases of IBDALD were identified (27 post-IBD diagnosis). Median follow-up duration was 11.5 (0.5–59.5) years and median 4.5 years (0.4–39.1) from IBD diagnosis to liver disease onset. In whole cohort analyses, ALT, ALP, and ESR were significantly elevated in IBDALD at diagnosis of IBD (p < 0.01). In matched case-control models, higher median ALT (OR = 1.04 per U/L increase, 95% CI: 1.01–1.07, p = 0.012), ALP (OR = 1.01, 95% CI: 1.00–1.02, p = 0.014), and ESR (OR = 1.05, 95% CI: 1.00–1.09, p = 0.034) were each associated with IBDALD. In a combined model including all three biomarkers, none retained statistical significance. When analysed as binary (normal/abnormal) variables, ALT and ALP remained significantly associated with IBDALD (ALT: OR = 5.10, 95% CI: 1.57–16.59, p<0.01; ALP: OR = 15.33, 95% CI: 1.87–125.77, p<0.05). ESR did not reach significance (OR = 3.71, 95% CI: 0.75-18.32 p=0.107). Longitudinal modelling supported distinct ALT and ALP trajectories in the IBDALD group and distinct trends over time were visualised for ALT, ALP and ESR in LOESS scatterplots.

Conclusion

These data suggest that routinely measured biomarkers may carry early predictive signals for IBD-associated liver disease, identifiable at the time of IBD diagnosis. The use of LLM-assisted phenotyping enabled high-fidelity case definition within real-world data. These findings should be interpreted cautiously given the rarity of IBDALD and potential overfitting in multivariable models. Future prospective studies are warranted to validate predictive models and support integration into IBD care pathways.

References

1. Scott JA, Mysko C, Purssell H, et al. Investigation of abnormal liver blood tests in patients with inflammatory bowel disease. Frontline Gastroenterology 2024;15:516-522.
2. Kadhim A, Green Z, Nazari I, et al. Application of Generative Artificial Intelligence to Utilise Unstructured Clinical Data for Acceleration of Inflammatory Bowel Disease Research. Article in PrePrint medRxiv 2025.03.07.25323569; doi: https://doi.org/10.1101/2025.03.07.25323569

Disclosure

JJA is a SAB member for Orchard Therapeutics.

Introduction

Early identification of factors that predict disease course in ulcerative colitis (UC) is essential to guide therapeutic decisions and improve patient outcomes. Recognizing patients at higher risk of unfavorable evolution allows for more intensive monitoring and timely escalation of therapy. Clinical severity, disease extent, histological activity, and concomitant infections may all influence prognosis. Among these, Clostridioides difficile infection (CDI) is increasingly recognized in UC patients, but its independent impact on disease evolution remains uncertain. The objective of our study was to assess various baseline severity markers, including histological activity, and concomitant CDI at the time of UC diagnosis as potential predictors of disease-related morbidity during the first year following UC diagnosis.

Aims & Methods

We conducted a retrospective study on a cohort of 125 patients newly diagnosed with UC and followed for at least 12 months. Data collected included demographic characteristics, clinical presentation, biochemical markers, endoscopic findings, histologic disease activity (assessed using the Nancy Histological Index – NHI), and the presence of CDI at diagnosis. Disease-related morbidity was defined as any UC-related hospitalization occurring in the first year after diagnosis, excluding the initial admission for diagnostic purposes. Statistical analysis involved bivariate comparisons and multivariate logistic regression to identify independent predictors of hospitalization.

Results

Independent predictors of UC-related hospitalization were Mayo clinical score (OR = 1.32; p = 0.016), pancolitis (OR = 2.82; p = 0.024), and histological activity (NHI; OR = 1.97; p = 0.014). CDI was identified in 20% of patients and was associated with significantly higher histological activity (median NHI = 3.5 vs. 3.0; p = 0.036), but did not independently predict hospitalization (OR = 1.84; p = 0.265).

Conclusion

Clinical activity, disease extent, and histological severity, at the time of diagnosis are strong, independent predictors of UC-related morbidity within the first year, underscoring their prognostic value beyond the initial presentation. These parameters retain their significance over the medium-term disease course, supporting early risk stratification and individualized treatment planning. Although CDI is associated with greater histologic inflammation at baseline, it does not independently influence outcomes during the first year. Notably, the identification of UC in 20% of patients during a CDI episode suggests that acute infection may unmask previously subclinical disease. These findings highlight the need for comprehensive and careful diagnostic evaluation - including clinical, endoscopic, and histologic assessment, in all patients with newly diagnosed UC.

Introduction

In corticosteroid-resistant acute severe ulcerative colitis (ASUC), second-line treatment involves infliximab or cyclosporine (CYCLO). With increasing antiTNFα exposed patients, CYCLO is often chosen to prevent emergency colectomy but is limited to the induction phase due to tolerability, serving as a bridge until effective maintenance therapy. In this setting, limited data on vedolizumab (VDZ) and ustekinumab (UST) as maintenance make the choice of an optimal bridge strategy challenging. This study aimed to compare the efficacy of VDZ and UST after CYCLO induction in corticosteroid-resistant ASUC.

Aims & Methods

This single-center retrospective cohort included all patients hospitalized for ASUC, responding to CYCLO induction, and starting VDZ or UST maintenance. ASUC was defined by modified Truelove and Witts criteria, with response to CYCLO defined by a Lichtiger score reduction to <10 points for 2 consecutive days, with a minimum 3-point decrease by day 7. The primary outcome was colectomy-free survival; secondary outcomes were treatment persistence and safety.

Results

From 2014-2024, 62 ASUC patients received CYCLO. Eighteen were excluded (non-response, tofacitinib [n=7], infliximab [n=3], and/or emergency colectomy [n=14]), leaving 44 CYCLO-responsive ASUC patients (median age 39, IQR 30-58, 60% female), with 35 on VDZ (79%) and 9 on UST (21%). ASUC was related to ulcerative colitis in 84% (73% pancolitis) and undetermined or Crohn's colitis in 16%. Disease duration was 7 years (IQR 2-15), with 91% prior biologic exposure, primarily antiTNF. Presentation included Lichtiger score 10 (IQR 10-12), CRP 46 mg/L (22-79), albumin 29 g/L (27-30), fecal calprotectin 778 µg/g (450-1650), UCEIS 7 (5-8), and Mayo endoscopic score 3 (3-3). Colectomy-free survival at 2 years was 79% with VDZ and 47% with UST (p=0.03). Treatment persistence at 2 years was 43% with VDZ, and 47% with UST (p=0.47). After adjusting for maintenance therapy, disease type, and severity, colectomy risk factors included hypoalbuminemia (HR=1.2; 95%CI:1.02-1.4, p=0.018) and prior tofacitinib exposure (HR=9.1; 95%CI:1.5-55.6, p=0.017). Over a median 37-month follow-up (95%CI:35-40), 34% of patients experienced adverse events, with 19% severe (mainly CYCLO-related), including 2 deep vein thromboses in the VDZ group and 1 infection in the UST group. No deaths occurred.

Conclusion

Our study suggests that VDZ may provide higher colectomy-free survival at 2 years than UST in CYCLO-responsive ASUC, with comparable treatment persistence. Hypoalbuminemia and prior tofacitinib exposure were identified as independent risk factors for colectomy, emphasizing the role of patient profiling in treatment selection. Further prospective studies are needed to refine bridge strategies in ASUC.