Introduction

Barrett’s esophagus (BE) is a precursor to esophageal adenocarcinoma. Societal guidelines were developed in order to standardize surveillance, treatment and improve quality of care in patients with BE.¹ However, adherence to these guidelines varies among gastroenterologists depending on their expertise, care settings, and geographic regions. Numerous studies have highlighted suboptimal adherence rates to guideline recommendations.^2,3
This study aimed to assess the accuracy of large language models (LLMs) in delivering guideline-based recommendations for managing Barrett’s esophagus across various stages of care.

Aims & Methods

We evaluated 107 cases of patients with Barrett’s esophagus (BE). Each case included all available original and unaltered esophagogastroscopy and pathology reports for the patient. Two experts reviewed the cases and provided recommendations according to the European Society of Gastrointestinal Endoscopy (ESGE).¹ Recommendations focused on three aspects: (1) request for a pathology revision by an additional gastropathologist; (2) the next recommended action—surveillance, lesion resection, or eradication—and (3) the appropriate time interval to the next endoscopy.
OpenAIs o3-mini-high, and Gemini 2.0, were evaluated. The models were trained on 23 cases by prompt engineering incorporated ESGE recommendations and best practices, this primarily utilized chain-of-thought reasoning. The gold standard was defined as agreement with the recommendation provided by at least one of the experts. The optimized prompts were then used to evaluate the remaining 84 cases.
Cohen’s Kappa was used to assess agreement between experts. Accuracy rates with 95% confidence intervals (CIs) were calculated using bootstrapping with 1,000 resamples. McNemar’s test was employed to statistically compare the accuracies of the two models.

Results

107 patients were included, visible lesions were present in 15% of the patients, dysplasia was identified in 29%. LLMs with final prompts were tested on 84 cases. Both LLMs demonstrated high to near perfect accuracy across all three fields (Table 1) without significant differences between the two models for all parameters.
Agreement between the two LLM was higher compared experts, as LLM agreed on action 99% (CI 94-100%) of time compared to 80% (CI 72%-87%) agreement between experts, with interval agreement of 87% (CI 78-93%) between models and only 69% (CI 60-78%) between experts.

Conclusion

LLMs (o3-mini and Gemini 2.0) demonstrated high accuracy in assessing real‑world presentations of BE and providing guideline‑based recommendations for patient management. These findings suggest that LLM may have a role in supporting clinical decision making, thus promoting adherence to guidelines and aiding in the standardization of care.

References

1. Weusten BLAM, Bisschops R, Dinis-Ribeiro M, et al. Diagnosis and management of Barrett esophagus: European Society of Gastrointestinal Endoscopy (ESGE) Guideline. Endoscopy. 2023;55(12):1124-1146. doi:10.1055/a-2176-2440
2. Roumans CAM, van der Bogt RD, Steyerberg EW, et al. Adherence to recommendations of Barrett’s esophagus surveillance guidelines: a systematic review and meta-analysis. Endoscopy. 2020;52(1):17-28. doi:10.1055/a-0995-0134
3. Holmberg D, Ness-Jensen E, Mattsson F, Lagergren J. Adherence to clinical guidelines for Barrett’s esophagus. Scand J Gastroenterol. 2019;54(8):945-952. doi:10.1080/00365521.2019.1641740

Introduction

Esophageal stricture after endoscopic submucosal dissection (ESD) remains a significant clinical challenge, particularly when the post-ESD mucosal defect involves ≥75% of the esophageal circumference. Various preventive strategies, including local or systemic corticosteroids, have been explored to mitigate this risk, but their efficacy is often limited by incomplete drug coverage, suboptimal timing of administration, or considerable side effects. This study aimed to investigate a new prophylactic strategy to prevent post-ESD esophageal strictures in cases of large circumferential mucosal defects.

Aims & Methods

This was a prospective single-arm trial conducted at a large regional medical center in Southwest China. Patients were considered eligible if they met all of the following criteria: (1) early esophageal cancer confirmed by postoperative pathology; (2) an ESD-induced mucosal defect involving ≥75% of the esophageal circumference; (3) no evidence of local or distant metastasis; and (4) no contraindications to corticosteroid therapy. All eligible patients received a combined steroid regimen to prevent stricture formation. Specifically, the first triamcinolone injection was administered intraoperatively immediately after complete lesion resection. Using a 23G endoscopic needle, triamcinolone (8 mg/mL) was injected submucosally at multiple points around the resection bed, spaced approximately 1 cm apart in four quadrants (0.5-1 mL per site; total of 20-40 injections). The second injection, following the same protocol, was given on postoperative day 4. Oral prednisone was started on day 5 post-ESD at 30 mg/day for 2 weeks, then reduced to 25 mg/day for the next 2 weeks, followed by a taper of 5 mg per week until discontinuation at 8 weeks. Clinically significant stricture was defined by both an inability to pass a standard 9.9 mm endoscope and a dysphagia severity of Stooler grade ≥3. Scheduled follow-up endoscopies at 3, 6, 12, and 24 months were performed to assess stricture formation and detect any residual or recurrent disease. Dysphagia symptoms were also assessed via telephone interviews at 2, 4, 6, 8, 12, and 24 months using the Stooler grading.

Results

Between September 2024 and March 2025, ten patients meeting the inclusion criteria (6 women and 4 men; mean age 67.5 years) were enrolled and treated with the combined steroid regimen. Six patients had full circumferential mucosal defects after ESD, while the remaining four had near-circumferential defects ranging from approximately five-sixths to eight-ninths of the esophageal circumference. The median longitudinal length of the post-ESD mucosal defects was 7.5 cm (range 4-12 cm), and the median resected area was 3375 mm² (range 1500-5250 mm²). No procedure-related complications were observed within 30 days post-ESD. Over a median follow-up period of 5 months (range 2-7 months), only one patient (10%) developed an esophageal stricture. This stricture was successfully managed with a single endoscopic dilation at 2 months after ESD. All other patients exhibited good mucosal healing and remained free of significant strictures throughout the follow-up.

Conclusion

This combined approach of intraoperative and postoperative local steroid injections coupled with a short-term oral corticosteroid taper appears to be a safe and preliminarily effective strategy for preventing esophageal strictures after ESD when large circumferential mucosal defects are present. Further validation in large-scale, multicenter randomized controlled trials is warranted to confirm its efficacy and safety.

Introduction

Colorectal cancer surveillance recommendations for individuals with Lynch Syndrome (LS) differ between professional associations and the data on interval colorectal cancer (I-CRC) and interval advanced polyps (I-AP) are limited. We aimed to measure the rate of I-CRC and I-AP within a surveillance program.

Aims & Methods

Retrospective study involved individuals with LS who participated in CRC surveillance program at Clalit Health hospitals, Israel. The main outcomes were the rates of I-CRC and I-AP ≤36 months, categorized by 12-month intervals. Included only colonoscopies with adequate preparation. I-CRC and I-AP (≥10 mm or containing high-grade dysplasia or villous histology) were defined as interval advanced neoplasia (I-AN).

Results

The cohort included 342 individuals, 52.3% females, with a median age at data analysis of 49.0 (IQR 37.7-64). At interval ≤ 12 months, there were 7 cases of I-CRC [6/142 (4.2%) of MSH2; 1/46 (2.2%) of MLH1] and 13 cases of I-AP (10/142 (7.0%) of MSH2, 2/46 (4.3%) of MLH1, 1/136 (0.7%) of MSH6]. At 12-24 months, there were 3 cases of I-CRC [2/142 (1.4%) of MSH2, 1/136 (0.7%) of MSH6] and 10 cases of I-AP [8 /142 (5.6%) of MSH2, 1/46 (2.2%) of MLH1, 1/136 (0.7%) of MSH6]. At 24-36 months, there were 3 cases of I-CRC [2/142 (1.4%) of MSH2, 1/18 (5.6%) of PMS2] and 4 cases of I-AP [1 /142 (0.7%) of MSH2, 3/136 (2.2%) of MSH6]. Of all I-CRC, four were diagnosed at stage 3 [4/342 (1.2%)], all at interval ≤24months [3-MSH2, 1-MSH6]. Increasing age and path_MSH2 were significantly associated with I-AN.

Conclusion

Extending surveillance colonoscopy intervals could result in significant rates of I-CRC and I-AP. Diagnosis of I-CRC at advance stage can occur at interval less than 24 months.

Clinical Case Summary

We present a case, which demonstrates endoluminal vacuum (EndoVac) therapy as an efficient technique when used for postoperative esophago-jejunal anastomotic leakage. A 59-year-old male, who underwent a total gastrectomy with esophago-jejunal Roux-en-Y anastomosis for stomach adenocarcinoma, presented with abdominal pain, tachycardia, fever and dyspnea a few days after the operation. A CT demonstrated a large anastomotic leakage in the esophago-jejunoanastomosis. An upper endoscopy was performed, revealing an anastomotic defect with a diameter up to 20mm. A 120mm fully covered esophageal stent was inserted. The stent was fixated with three 16mm clips. Due to clinical deterioration a follow up CT was performed. The images showed proximal migration of the stent with a perianastomotic inflammatory cavity. The stent was retrieved and 2.5L of pus-like fluid was evacuated, followed by placement of EndoVac System. During the procedure, a 15fr PEG-J feeding tube was placed, ensuring enteral nutrition. Additionally VATS was performed immediately after the endoscopy and two redon drains were placed.
Four EndoVac sponges were exchanged in a period of 12 days, during which a significant clinical improvement was observed and the patient was discharged shortly after. Eight weeks later gastroscopy and CT were performed with no evidence of a fistula and complete recovery of the patient, the PEG-J was removed. We demonstrate with our case that EndoVac therapy is an effective method in managing postoperative esophageal leaks and it should be considered as an initial technique when dealing with such complications.

References

Endoscopy photo of the ansatomotic leak
Endoscopy photo EndoVac
Endoscopy photo of the PEG-J
Endoscopy photo of the final result

Introduction

Lynch syndrome (LS) is a hereditary condition associated with an increased risk of early colorectal cancer (CRC), due to mutations in DNA mismatch repair (MMR) genes, accounting for 3-5% of all CRC. High-definition screening colonoscopies in specialized centers reduce CRC mortality in this setting.

Aims & Methods

We conducted a clinical audit in a large cohort of these patients, ultimately to enhance the quality of care provided. We performed a retrospective evaluation of the clinical characteristics and risk factors of LS patients under surveillance at our reference center. CRC surveillance program includes colonoscopy every 2 years, starting at ages 20–25 for MLH1/MSH2 carriers and 30–35 for MSH6/PMS2 carriers. We further analyzed compliance to surveillance protocol and its clinical outcomes – our primary endpoint was mortality rate due to CRC and secondary endpoints were time to first advanced colorectal lesion and CRC staging at diagnosis.

Results

A total of 508 patients were included (44% MSH2, 23% MLH1, 20% MSH6, and 13% PMS2). The mean age at surveillance initiation was 44.4 years, with 35.6% starting after the diagnosis of a Lynch syndrome-associated neoplasia. The mean surveillance duration was 5.6 years. Each patient underwent an average of four colonoscopies, with 50% performed within the recommended surveillance intervals.
During the follow-up period, 35 patients were diagnosed with CRC, with a mean time to first advanced colorectal polyp of 3.7 years and for the first CRC diagnosis of 4.8 years. Having any colonoscopy outside the established interval was not associated with higher CRC incidence (p = 0,647). CRC cases diagnosed prior to surveillance initiation (n=143) were more advanced (58.1% stage ≥ II) and were treated with segmental colectomy in 80% of cases. In contrast, CRC cases diagnosed during surveillance (n=48) were detected at earlier stages (64.6% Tis/T1) and managed with endoscopic resection (29%) or total colectomy (56%) (p < 0.001).
The incidence of CRC was higher in MLH1/MSH2 carriers compared to MSH6/PMS2 carriers (36% vs. 22%, p = 0.0020). The mean age at first CRC diagnosis was 45.6 years for MLH1/MSH2 carriers and 49.9 years for MSH6/PMS2 carriers (p = 0.053). The CRC-related mortality rate was 0.6% (n=3), all cases diagnosed before surveillance initiation.

Conclusion

Although adherence to the CRC surveillance program in LS patients can be optimized, our findings underscore its importance in specialized reference centers. Surveillance was associated with earlier CRC detection, less invasive therapeutic approaches, and reduced CRC-related mortality. Our results also show a lower incidence of CRC and an higher age at first CCR diagnosis in MSH6/PMS2 carriers, strengthening the personalized surveillance recommendations dependent on gene specificities.

References

Castillo-Iturra, J., Sánchez, A. and Balaguer, F. (2024) ‘Colonoscopic surveillance in Lynch Syndrome: Guidelines in perspective’, Familial Cancer [Preprint]
del Carmen, G. et al. (2023) ‘Colorectal surveillance outcomes from an institutional longitudinalcohort of Lynch syndrome carriers’, Frontiers in Oncology, 13
Kastrinos, F. et al. (2021) ‘Gene-specific variation in colorectal cancer surveillance strategies for Lynch syndrome’, Gastroenterology, 161(2)

Introduction

Patients with gastric intestinal metaplasia (GIM) are at an increased risk of developing gastric cancer (GC). The histological Operative Link on Gastric Intestinal Metaplasia (OLGIM) classification aids in risk stratification. The recently developed endoscopic grading of GIM (EGGIM) may accurately predict the OLGIM stage. This study aimed to evaluate the diagnostic accuracy of EGGIM in assessing histological GIM using data from two large, prospective European consortia.

Aims & Methods

Interim data were derived from the Artificially Intelligent Diagnostic Assistant (AIDA) and Towards Gastric Cancer Screening Implementation in the European Union (TOGAS) consortia, both part of a European Horizon programme focused on gastric cancer prevention. Patients in AIDA underwent oesophagogastroduodenoscopy (OGD) for GIM surveillance or treatment of neoplastic lesions, while participants in TOGAS (aged 50–74) were offered an OGD during screening colonoscopy or polyp surveillance. Procedures were performed at eight academic centres by both expert and non-expert endoscopists, using high-resolution white-light endoscopy and virtual chromoendoscopy. The EGGIM score (range 0–10) was based on the extent of GIM (0: 0%, 1: ≤30%, 2: >30% of mucosa) across five anatomical sites: the lesser and greater curvatures of the antrum and corpus, and the incisura. Targeted biopsies were taken if GIM was suspected; otherwise, random biopsies were collected according to the Sydney system to determine the OLGIM stage. Healthy controls and GIM patients with an EGGIM score ≥1 but OLGIM stage 0 were also included. ROC curve and Spearman analyses assessed the diagnostic and ordinal correlation between EGGIM and OLGIM stages III–IV.

Results

A total of 252 patients were included (median age 60 years, IQR 9; 57.4% female): 91 from the TOGAS study (75 cancer screening, 16 polyp surveillance) and 161 from AIDA (117 GIM surveillance, 21 neoplasia treatment, 23 healthy controls). OLGIM staging identified 89 (35.3%) patients in stage 0, 74 (29.4%) in stage I, 49 (19.4%) in stage II, 35 (13.9%) in stage III, and 5 (2.0%) in stage IV. The area under the ROC curve for EGGIM in detecting OLGIM III/IV was 0.90 (95% CI 0.79–0.93, p < 0.001), with an optimal cut-off at EGGIM ≥5 (sensitivity 85%, specificity 89%).

Conclusion

Our findings prospectively validate the EGGIM score across two large European cohorts. This scoring system may serve as a practical tool to streamline gastric cancer surveillance by reducing the need for biopsies and enhancing endoscopic risk stratification.

References

Esposito G, Pimentel-Nunes P, Angeletti S, et al. Endoscopic grading of gastric intestinal metaplasia (EGGIM): a multicenter validation study. Endoscopy. 2019;51(6):515-521. doi:10.1055/a-0808-3186

Introduction

In recent years, artificial intelligence (AI) has been widely explored to enhance capsule endoscopy (CE), with the goal of improving the efficiency of the reading process. While most AI models have been developed for small bowel and colon analysis, the development of esophagogastric (E-G) models has been limited due to the scarcity of frames captured during the procedure, making it challenging to develop a robust model. This study aims to develop an interoperable ubiquitous model capable detecting pleomorphic lesions in the E-G tract.

Aims & Methods

We included 59,482 E-G frames, from 774 CE procedures of 5 centers, to develop a Convolutional Neural Network (CNN). The dataset was divided following an exam-based split, with 90% allocated for training – including a 5-fold cross validation – while the remaining was used for testing. The primary outcomes were: sensitivity, specificity, accuracy and area under the curve (AUC).

Results

During training, the CNN achieved mean sensitivity of 85.0% (IC95% 76.5–93.5), specificity of 96.3% (IC95% 93.9–98.7), accuracy of 93.6% (IC95% 91.7–95.4), and AUC-ROC of 0.98 (IC95% 0.97–0.98). During testing, the sensitivity, specificity and accuracy of CNN were 92.2%, 95.1% and 94.6%, respectively.

Conclusion

CE-AI models capable of assessing E-G are crucial step toward developing a truly robust panendoscopic model. This ubiquitous and interoperable model capable of lesion detection in both esophagus and stomach, achieved good overall accuracy. However, prospective real-world studies comparing its performance with standard upper endoscopy are still needed to validate its clinical applicability.