Introduction

The number of patients with inflammatory bowel diseases continues to increase, and several biologic therapies can be used. However, there have yet been few useful biomarkers for drug selection in clinical practice, resulting in difficulty to optimize treatment with biologics. Recently, the usefulness of machine learning approach to predict treatment outcomes has been reported in the fields of several gastrointestinal diseases. Here, we applied machine learning approach to predict treatment outcomes of anti-tumor necrosis factor (TNF)-α antibody agents in patients with Crohn’s disease (CD). In addition, the clinical variables that weighed significantly for anti-TNF-α antibody agents selection for patients with CD were identified.

Aims & Methods

Consecutive CD patients who had been treated with anti-TNF-α antibody agents in Chiba University Hospital from July 2004 to March 2022 were analyzed. Medical information including age, gender, medical history, disease type, treatment history, endoscopic findings, and blood test findings at the start of biological agents administration was collected. These variables were applied to machine learning approach, and the primary outcome was to predict primary failure of anti-TNFα antibody agents. Then, factors related to the poor outcome with anti-TNFα antibody agents were identified. The used machine learning system was LightGBM, a gradient boosting method. The constructed model performance was assessed through the area under the curve (AUC) with specificity and sensitivity. The SHapley Additive exPlanations (SHAP) were implemented for each predictive factor for evaluation of the impact on primary failure of anti-TNFα antibody agents.

Results

Total 232 CD patients who were treated with anti-TNF-α antibody agents at the hospital were analyzed. Of these, 29 patients failed in remission induction with anti-TNFα antibody agents. The constructed model with LightGBM was showed high prediction performance with AUC of 0.94 (sensitivity: 1.00 and specificity: 0.88).
The SHAP value showed that several variables were correlated with primary failure; platelet count (0.045), current medical treatment with thiopurine (0.025) and elemental diet (-0.008), surgical history (0.011), and disease duration (0.019).

Conclusion

Machine learning approach using clinical parameters alone could construct a predictive model of primary failure of anti-TNF-α antibody agents in CD patients with high accuracy. Platelet count, current medical treatment with thiopurine hand elemental diet, surgical history and disease duration were correlated with the primary failure, among which platelet count appeared to affect the most to the primary failure.

Introduction

Anti-tumour necrosis factor-alpha (anti-TNF-alpha) therapy is an effective treatment for the management of Crohn’s disease (CD). However, treatment failure is common.

Aims & Methods

The aim of this study was to identify predictors of anti-TNF-alpha therapy failure.
Retrospective single-center study including anti-TNF-alpha naïve patients with CD, who started on intravenous infliximab, between January 2019 and December 2021. Biochemical parameters included erythrocyte sedimentation rate (ESR), c-reactive protein (CRP), faecal calprotectin, infliximab serum concentrations and the presence of antibodies to infliximab (ATI). Anti-TNF-alpha therapy failure was defined as the development of ATI at 6 and 12 months, absence of clinical response at 6 months, absence of clinical remission or of objective response at 12 months. Clinical response was defined as a reduction in Harvey-Bradshaw index (HBI) of ≥ 3 points comparing with initial value or HBI < 5 points if initial HBI ≥ 7 points and clinical remission as HBI ≤ 4 points. Objective response was assessed by endoscopic studies, defined by improvement of mucosal inflammation and absence of deep ulcerations, or imaging, defined as improvement in bowel wall thickness, inflammatory fat, mural blood flow and hyper-enhancement.

Results

A total of 53 CD patients were included, 30 were female (56.6%), with a mean age at the beginning of treatment of 39 ± 2 years. Considering the Montreal classification, 31 patients had ileal disease (58.5%), 7 colonic (13.2%) and 15 ileocolonic disease (28.3%); 22 patients had nonstricturing, nonpenetrating disease (41.5%), 22 stricturing (41.5%) and 9 penetrating disease (17.0%). Eleven patients had perianal disease (20.8%). A total of 26 patients (49.1%) were treated with combination of anti-TNF-alpha and immunomodulatory therapy (thiopurine or methotrexate). Anti-TNF-alpha therapy failure occurred in 21 patients (39.6%). At 6 months, the development of ATI occurred in 6 patients (11.3%) and absence of clinical response in 9 patients (17.0%). At 12 months, absence of clinical remission was seen in 13.6% of patients and absence of objective response in 27.0% of patients.
At 6 months, the development of ATI was significantly higher in patients with lower infliximab serum concentrations at week 14 (with antibodies 5.9±3.2 µg/mL vs without 14.3±7.7 µg/mL, p<0.001). Additionally, the development of ATI was significantly higher in patients with a higher initial value of ESR (with antibodies median 35 vs without median 13, p=0.045). The infliximab serum concentrations at week 14 (AUC 0.828; p=0.009; sensitivity 0.833 and specificity 0.404 for values ≤11.6) and the initial value of ESR (AUC 0.754; p=0.045; sensitivity 0.833 and specificity 0.468 for values ≥15) had very good and good discriminative capacity, respectively, in predicting the development of ATI. No statistically significant differences were found in initial values of CRP and faecal calprotectin, between both groups. Moreover, the other definers of anti-TNF-alpha therapy failure were not associated with the combination with immunomodulatory therapy, infliximab serum concentrations at week 14 or initial values of ESR, CRP and faecal calprotectin.

Conclusion

Infliximab serum concentration after induction therapy is the most important factor in the development of ATI, influencing treatment response, regardless of the combination of anti-TNF-alpha therapy with immunomodulatory therapy. A higher initial value of ESR can also predict the development of ATI.

Clinical Case Summary

Infectious aortitis is a rare but life-threatening condition in immunosuppressed patients. Early recognition and intervention are critical to prevent fatal outcomes. We report the case of a 65-year-old former smoker with Crohn’s disease diagnosed in February 2023. After a severe relapse in June 2024 treated with corticosteroids and infliximab, she achieved remission and infliximab was maintained. In March 2025, she developed retrosternal chest pain, dyspnea, dysphagia, and chills. Physical examination was unremarkable. Laboratory results revealed leukocytosis (12,700/µL) and elevated C-reactive protein (209 mg/dL). Contrast-enhanced Thoracic CT revealed a mycotic aneurysm of the descending thoracic aorta arising from an ulcerated atherosclerotic plaque, complicated by periaortic phlegmon and esophageal compression. Blood cultures were positive for oxacillin-sensitive Staphylococcus aureus. Empiric vancomycin and piperacillin–tazobactam were narrowed to flucloxacillin, then switched to oral trimethoprim–sulfamethoxazole and doxycycline. Autoimmune and neoplastic causes were excluded. She underwent endovascular stent-graft placement with prompt improvement. Infectious aortitis requires a high index of suspicion, especially in immunocompromised patients. A comprehensive diagnostic approach relies on laboratory tests, immunologic screening, targeted infectious studies, and imaging assessment (1,2). While autoimmune etiologies predominate, pathogens like Salmonella and Staphylococcus aureus may infect atherosclerotic or aneurysmal segments, requiring six to twelve weeks of antibiotics, often extended due to delayed clearance and risk of recurrence or rupture, followed by surgical or endovascular repair (1). This case highlights the importance of considering infectious aortitis in Crohn’s disease patient on infliximab with atypical thoracic symptoms. Prompt diagnosis, extensive evaluation, antibiotics, and vascular intervention are essential to improve patient outcomes.

References

1) Gornik, H. L., & Creager, M. A. (2008). Aortitis. Circulation, 117(23), 3039–3051.
2) Lötscher, F., Pop, R., Seitz, P. et al. Spectrum of Large- and Medium-Vessel Vasculitis in Adults: Neoplastic, Infectious, Drug-Induced, Autoinflammatory, and Primary Immunodeficiency Diseases. Curr Rheumatol Rep 24, 293–309 (2022).

Introduction

Vedolizumab (VDZ), a gut-selective anti-lymphocyte trafficking agent, was approved for the treatment of moderate-to-severe Crohn’s disease (CD) in Europe and the USA in 2014. We conducted a systematic literature review and meta-analysis of studies reporting real-world effectiveness of VDZ in patients with CD. Here we describe treatment persistence in biologic-naïve patients with CD treated with VDZ compared with anti-tumor necrosis factor-alpha (TNFα) treatment at 12 and 24 months to address the gap of long-term data available on patients with CD.

Aims & Methods

Systematic literature reviews were performed using MEDLINE, MEDLINE In-Process, Embase, Cochrane Library, clinicaltrials.gov, and the World Health Organization International Clinical Trials Registry Platform to identify relevant studies published from May 2014 to July 2022. In addition, if the congress abstracts were not indexed in Embase, key congresses from 2015 to July 2022 were hand searched. Data from adults (aged ≥ 18 years) with CD treated with VDZ in observational/real-world studies were extracted for meta-analysis. The analysis was conducted using the DerSimonian-Laird random-effects method to obtain weighted mean rates (percentage of patients) and the corresponding 95% confidence interval (CI) for each treatment persistence or serious adverse event (SAE) as performed for a meta-analysis published in 2018.¹

Results

Overall, data from 78 studies of 30 158 patients with CD had at least one outcome of interest for analysis. Four and three studies of 7082 and 5040 patients with CD, respectively, reported the rate of treatment persistence in biologic-naïve patients treated with VDZ versus anti-TNFα treatment at 12 and 24 months. The rate of treatment persistence was higher with VDZ (83.8% [95% CI 71.4–96.1]) compared with anti-TNFα treatment (75.3% [69.9–80.7]) at 12 months. Similarly, the rate of treatment persistence was higher with VDZ (70.4% [95% CI 61.3–79.6]) compared with anti-TNFα treatment (64.5% [56.9–72.2]) at 24 months. Two studies of 491 and 86 biologic-naïve patients with CD treated with VDZ versus anti-TNFα treatment reported SAEs. In a retrospective study, safety outcomes were followed from treatment initiation up to five half-lives post-treatment discontinuation; 8.3% treated with VDZ experienced SAEs compared with 19.0% with anti-TNFα treatment.² In an additional retrospective study that reported SAEs within the first 12 months of treatment; the rate of SAEs was comparable between patients treated with VDZ (18.2%) and anti-TNFα treatment (17.0%).³ Among these patients, only those receiving anti-TNFα treatment experienced drug-induced SAEs (3.8%; pancreatitis and systemic lupus erythematosus).³

Conclusion

This meta-analysis of real-world studies demonstrated higher rates of long-term treatment persistence and suggests that VDZ may have a favorable safety profile compared with anti-TNFα treatment in biologic-naïve patients with CD.

References

1. Schreiber S et al. J Gastroenterol 2018;53:1048–64.
2. Bressler B et al. J Crohns Colitis 2021;15:1694–706.
3. Moens A et al. Inflamm Bowel Dis 2021;28:1135–42.

Disclosure

This study was funded by Takeda Pharmaceuticals International AG. Medical writing services were provided by Elena Sugrue, PhD, of Oxford PharmaGenesis, and funded by Takeda Pharmaceuticals International AG.
AA Advisory board/lecture fees from AbbVie, Amgen, Arena, Biogen, Boehringer Ingelheim, Bristol-Myers Squibb, Celgene, Celltrion, Eli Lilly, Ferring, Galapagos, Gilead, Janssen, MSD, Novartis, Pfizer, Protagonist Therapeutics, Roche, Samsung Bioepis, Sandoz, Takeda.
RJB is an employee of PHMR Ltd., who received funding for this study from Takeda Pharmaceuticals International AG.
PB is an employee and stockholder of Takeda Pharmaceuticals International AG, Zurich, Switzerland.
MN is an employee and stockholder of Takeda Pharmaceuticals International AG, Zurich, Switzerland.
SW is an employee and stockholder of Takeda Pharmaceutical Company Limited, Cambridge, MA, USA.
SS has received personal fees from AbbVie, Amgen, Arena, Biogen, Bristol-Myers Squibb, Celgene, Celltrion Healthcare, Dr. Falk Pharma, Fresenius, Galapagos/Gilead, Hikma, I-Mab, Janssen, MSD, Mylan, Pfizer, Protagonist Therapeutics, ProventionBio, Takeda and Theravance Biopharma.

Introduction

Inflammatory bowel disease (IBD), including both Crohn’s disease (CD) and ulcerative colitis (UC), are difficult to diagnose due to a broad symptoms spectrum.1,2 If symptoms remain uncontrolled with conventional therapy (CT), patients are transitioned to biologics. Little is understood about which CD and UC patients will respond to CT and which patients may have a more aggressive disease progression, and therefore should be treated with biologic therapies sooner.

Aims & Methods

We aimed to develop and evaluate machine learning (ML) models to identify non-responders to CT based on patients’ demographic characteristics and on data available before their treatment with CT. The study cohort included all Finnish patients ≥18 years of age, with a diagnosis for CD or UC between 1.1.2000–31.12.2019. Data were collected from electronic health records and national healthcare and prescription registers. CT drugs included corticosteroids (CS), immunomodulators and 5-aminosalicylic acids. CT failure outcomes are defined in Table 1. A set of 6,782 predictive features including demographic and clinical characteristics, comorbidities, healthcare visits, procedures, laboratory measurements, and medication were created for three different baseline periods. A Lasso model was run both separately and in combination for each period, splitting the data into training and testing subcorpora ten different times to estimate the error bounds.

Results

After applying specific attrition criteria, the final analysis cohort included 12,818 CD and 30,178 UC patients. In total, 64.2% of CD and 61.1% of UC patients experienced CT failure (Table 1). For CD, CS at baseline and young age (0–18 and 18–30 years) were the highest predictors of CT failure. For UC, the model selected 78 predictive features, the strongest being: baseline CS use, age (0–18, 18–30, and 65+ years), gastrointestinal and schizoaffective disorders, and secondary care visits to rheumatology, general medicine, and emergency. The key findings of the Lasso model were: 1) The performance of CD-models was slightly better than the UC-models (median test accuracy was 0.65 [CD] and 0.62 [UC]); 2) a slight discrepancy between test and training statistics suggested overfitting, and 3) most features appeared in all ten splits indicating their relevance.

Table 1. CT failure outcomes (within one year from CT initiation) for CD, UC, and combined cohorts.

Conclusion

Data quality is critical to build high-performance models. Even though Finnish registries provide access to comprehensive longitudinal data allowing the generation of relatively large analysis cohorts, our ML model demonstrated a weak predictive potential, with low AUCs and accuracies (0.6 range), in line with previous findings.3 Due to scarcity of information on additional clinical variables available in the Finnish IBD registry, their use did not improve prediction of CT failure. Additional prediction difficulty is introduced by non-standardized data recording practices. Confounding factors and random chance affect complicated outcomes, such as CT failure, making them especially difficult to predict.

References

1. Wang, R., Li, Z., Liu, S. & Zhang, D. Global, regional and national burden of inflammatory bowel disease in 204 countries and territories from 1990 to 2019: a systematic analysis based on the Global Burden of Disease Study 2019. BMJ Open 13, e065186 (2023).
2. Yantiss, R. K. & Odze, R. D. Diagnostic difficulties in inflammatory bowel disease pathology. Histopathology 48, 116–132 (2006).
3. Chen, I. et al. PGI27 Applying Machine Learning to Large Databases to Predict Nonresponse to Conventional Treatment in Patients with Ulcerative Colitis. Value Health 24, S99 (2021).

Disclosure

This study was sponsored by Takeda Oy. TK, JM, and HL are employees by MedEngine Oy. TY is the CEO and founder of MedEngine Oy. KE and PK are employed by Takeda and hold Takeda stocks/stock options. Mónica Ferreira (PhD) from MedEngine Oy is acknowledged for medical writing support.

Introduction

There is an increasing diversification in the treatment landscape for Crohn’s disease (CD) driven by heterogeneity of individual phenotypes, disease courses and previous treatments. This complexity makes universal evidence-based algorithms challenging. A “treat-to-target approach” with tight inflammation monitoring after starting biologics or small molecules therapy is recommended. Intestinal ultrasound (IUS) has proven effective in monitoring treatment response in large CD cohorts. Transmural healing (TH) is a robust but puzzling endpoint that can be achieved in about one-fourth of patients treated with biological therapy and few prognostic indicators to predict it were analyzed.

Aims & Methods

We aimed to describe treatment patterns and effective therapeutic sequences to achieve TH. Secondary aim was to identify prognostic factors to predict TH. This retrospective analysis included CD patients treated with biological and small molecules therapies according to the ECCO recommendations. Patients received biological and small molecules therapies across up to four treatment lines. Biological and small molecules treatment pathways prior to achieve TH were visualized using a Sankey diagram. IUS outcomes were evaluated one year after each biological and small molecules therapy. Logistic regression analysis was performed to identify clinical predictive factors of TH, that was defined as bowel wall thickness <3 mm and no vascularization at IUS.

Results

Among 139 patients treated with biological and small molecules therapies, TH was achieved in 25.9% (n = 36). Infliximab was used as the first-line agent in 55 patients (39.6 %), adalimumab in 69 (49.6%), vedolizumab in 6 (4.4%), certolizumab pegol in 1 (0.7%), Ustekinumab in 8 (5.7%). Twenty-five patients (20%) achieved TH after the first-line therapy (n=10 with infliximab; n=9 with adalimumab; n=3 with vedolizumab; n=3 with ustekinumab). TH rates declined with each additional line of therapy. The median time to achieve TH was 25.6 months (range 12–60). Treatment sequencing analysis revealed a progressive shift toward ustekinumab in later lines, with anti-TNFs remaining the most frequent first-line agents. Multivariate analysis showed that colonic lesion and early treatment (within 2 years from diagnosis) were associated with a higher likelihood of achieving TH [OR = 7.76, p<0.001; OR 6.705, p<0.001, respectively], while perianal disease was associated with a lower likelihood of TH [OR 0.151, p=0.02].

Conclusion

This study highlights a progressive decline in TH rates with successive lines of biological and small molecule therapies in CD patients and identifies several clinical factors that influence this outcome. Long-term follow-up and prospective studies specifically designed to evaluate sequencing in patients exposed to anti-TNFs are needed to guide challenging decisions.

Introduction

Advances in therapeutic drug monitoring (TDM) have significantly improved the understanding of the relationship between anti-TNF-alpha drug concentrations and disease activity in patients with Crohn’s disease (CD). Early dose optimization of adalimumab (ADAL) and higher serum trough concentrations are associated with increased clinical, biochemical, and endoscopic remission in CD^1,2,3,4. This study aimed to evaluate whether early dose optimization of ADAL to a trough level >15 µg/mL improves clinical and biochemical remission rates at 12 weeks compared to standard dosing.

Aims & Methods

This study aimed to evaluate whether early dose optimization of ADAL to a trough level >15 µg/mL improves clinical and biochemical remission rates at 12 weeks compared to standard dosing. This was a randomized, multicenter, open-label trial evaluating TDM-guided ADAL dose optimization versus standard dosing in biologic-naïve and experienced patients with active CD. In the dose optimization arm , after standard induction doses at Weeks 0 (160 mg) and 2 (80 mg), patients with trough levels ≤15 µg/mL at week 4 received an additional 80 mg at Week 6 followed by 40 mg every week. Patients in the optimization arm who had trough levels >15 µg/mL at week 4 and all patients in the standard care arm remained on 40 mg every 2 weeks. The primary endpoint was a composite of clinical and biochemical remission at Week 12, defined as Harvey-Bradshaw Index (HBI) <5 and C-reactive protein (CRP) <5 mg/L or fecal calprotectin (FCal) <250 µg/g.

Results

A total of 77 patients were enrolled in the trial (39 in optimization arm vs 38 in standard care arm). At week 4, 15/39 patients in the intervention arm were eligible for dose optimization, with twelve going on to receive the accelerated regimen. At week 12, 6/12 (50%) of the optimized patients achieved an ADAL trough level >15 µg/mL. A numerically higher proportion of patients in the dose optimization arm achieved the primary endpoint compared to the standard care arm, however this difference was not statistically significant (17 (43.6%) vs 12 (31.6%), p = 0.394). In total, eleven patients in the optimization arm (28.2%) and seven patients in the standard care arm (18.4%) achieved an ADAL trough >15 µg/mL, with mean trough levels being numerically higher in the optimization group (18.1 vs 12.8 µg/mL, p = 0.160). ROC analysis of the predictive value of Week 12 trough levels on the primary outcome determined an optimal threshold level of 18.2 µg/mL (AUC 0.68, sensitivity 88.9%, specificity 39.3%). When analyzing the entire cohort, patients who achieved an ADAL trough >15 µg/mL (whether they were dose-optimized or not) at Weeks 8 or 12 were significantly more likely to achieve combined clinical and biochemical remission (19/27 patients (70.4%) vs 10/50 patients (20.0%), p < 0.001). Taken separately, both clinical and biochemical remission occurred more frequently in patients who achieved an ADAL trough >15 µg/mL (clinical: 22/27 (81.5%) vs 16/50 (32.0%), p <0.001; biochemical: 22/27 (81.5%) vs 14/50 (28.0%), p < 0.001). No difference in adverse effects was observed between trial arms. The major limitation of the trial was the small sample size and patient withdrawal due to the COVID 19 pandemic.

Conclusion

Although proactive dose optimization did not significantly increase remission rates compared to standard dosing, achieving an ADAL trough level >15 µg/mL at Week 8 or 12 was strongly associated with higher remission rates. These findings support early TDM-driven dose escalation and suggest that maintaining ADAL >15 µg/mL is a key target for achieving biochemical remission.

References

1. Little RD, Swaine A, Reynolds R, et al. Adalimumab Drug Levels at Secondary Loss of Response Do Not Predict Response to Dose-intensification in Crohn’s Disease: A Retrospective, International Multicenter Study. Inflamm Bowel Dis. 2024;30(10):1714-1723. doi:10.1093/ibd/izad248
2. Assa A, Matar M, Turner D, et al. Proactive Monitoring of Adalimumab Trough Concentration Associated With Increased Clinical Remission in Children With Crohn’s Disease Compared With Reactive Monitoring. Gastroenterology. 2019;157(4):985-996.e2. doi:10.1053/j.gastro.2019.06.003
3. Chaparro M, Guerra I, Iborra M, et al. 538 Correlation Between Adalimumab Serum Levels and Remission After the Induction Phase in Crohn’s Disease Patients. Gastroenterology. 2015;148(4):S-107-S-108. doi:10.1016/S0016-5085(15)30370-X
4. Kim MJ, Kim E, Kang B, Choe YH. Therapeutic Drug Monitoring of Adalimumab During Long‐term Follow‐up in Paediatric Patients With Crohn Disease. J Pediatr Gastroenterol Nutr. 2021;72(6):870-876. doi:10.1097/MPG.0000000000003070