Introduction

Colorectal cancer (CRC) is the second-leading cause of cancer-related mortality in Europe and the United States, but early detection boosts the 5- year relative survival rate to 91%. While colonoscopy remains the gold standard for screening, its invasive nature reduces adherence. Although less invasive stool-based and emerging blood-based biomarker tests have greatly advanced screening efforts, additional strategies are needed to enhance accuracy, particularly for detecting early-stage CRC and advanced adenomas (AA). In this study, part of the Horizon Europe-funded DIOPTRA project for Mission Cancer, a blood-based protein biomarker approach to improve early detection was investigated.

Aims & Methods

A prospective clinical study was conducted to collect blood samples from individuals undergoing colonoscopy (N=260). Participants were categorized into four groups: Healthy (no findings or hyperplastic polyps), Non-AA (NAA), AA, and CRC based on the colonoscopy findings and subsequent histopathological diagnosis. Each sample was analyzed using Olink’s next-generation proteomics platform, based on the proximity extension assay technology, measuring 3,072 proteins in blood. Bioinformatic analysis, utilizing linear machine learning algorithms and differential expression methods was performed.

Results

The analysis pipeline identified a 12-protein signature capable of accurately distinguishing the four study groups. In a 5-fold cross-validation scheme, the linear combination of these proteins achieved a sensitivity of 98.1% (95% CI:[90.1%, 100.0%]) for CRC detection and a specificity of 93.1% (95% CI:[77.2%, 99.2%]) for advanced neoplasia (calculated by combining the Healthy
and NAA groups). Notably, the model’s sensitivity for stage I/II CRC was 100.0% (95% CI: [85.8%, 100.0%]) and for AA was 50.0% (95% CI: [37.0%,63.0%]), demonstrating the potential of protein signatures in blood to detect early-stage cancer and advanced precancerous lesions.

Comparison	AUC (%)	Sensitivity (%)	Sensitivity 95% CI	Specificity (%)	Specificity 95% CI	Accuracy (%)	Accuracy 95% CI
CRC vs Normal	95.9	98.1	90.1-100.0	93.1	77.2-99.2	96.4	89.8-99.2
SI vs Normal	96.9	100.0	71.5-100.0	93.1	77.2-99.2	95.0	83.1-99.4
SII vs Normal	97.6	100.0	75.3-100.0	93.1	77.2-99.2	95.2	83.8-99.4
SI/SII vs Normal	97.3	100.0	85.8-100.0	93.1	77.2-99.2	96.2	87.0-99.5
AA vs Normal	78.0	50.0	37.0-63.0	93.1	77.2-99.2	63.7	53.0-73.6
HGD vs Normal	74.7	43.5	23.2-65.5	93.1	77.2-99.2	71.2	56.9-82.9
CIS vs Normal	66.4	50.0	06.8-93.2	93.1	77.2-99.2	87.9	71.8-96.6
AA10 vs Normal	85.8	61.1	35.7-82.7	93.1	77.2-99.2	80.9	66.7-90.9
V vs Normal	74.1	50.0	21.1-78.9	93.1	77.2-99.2	80.5	65.1-91.2

Table. The performance metrics of the 12-plex assay for different categories of CRC and AA. Normal samples are referred to Healthy + NAA. The metrics have been calculated considering disease state (CRC, AA, sub-categories) as true positives, and Normal samples as true negatives.
Results: The AI model performance of the 12-plex assay across different classifications of advanced neoplasia has outperformed the current blood-based IVD in the case of CRC (98.1% Sensitivity, 95% CI: [90.1%, 100.0%]) and AA (50.0% Sensitivity, 95% CI: [37.0%, 63.0%]), without compromising the classification of Non-neoplastic samples (93.1% Specificity, 95% CI: [77.2%, 99.2%]).

Conclusion

Future work is focused on developing multiplex immunoassays for these biomarkers to orthogonally validate the findings and assess the signature’s analytical and clinical performance in a large participant cohort from the DIOPTRA validation studies (N=1600 participants). This protein biomarker signature could pave the way for more accurate, patient-friendly CRC screening methods that detect both cancer and its precursors at earlier, more treatable stages.

Introduction

Primary sclerosing cholangitis (PSC) is a progressive autoimmune cholestatic liver disease often associated with inflammatory bowel disease (IBD), particularly ulcerative colitis [1]. In PSC, reduced expression and function of the bile acid receptor GPBAR1 (TGR5) in cholangiocytes is associated with a dysfunctional gut–liver immune axis and impaired resolution of inflammation [2,3]. Parallelly, the transcription factor RORγt in Th17 cells sustains IL-17 perpetuating IL-23-driven intestinal [4] and liver inflammation [5]. While GPBAR1 mediates anti-inflammatory responses to secondary bile acids, RORγt governs Th17 differentiation and cytokine secretion. Recent clinical studies have demonstrated that IL-23 inhibitors, such as risankizumab [6] and mirikizumab [7], are effective in treating IBD, highlighting the significance of the IL-23/IL-17 axis. GPBAR1/RORγt dual simultaneous modulation represents a novel strategy to counteract the fibrotic and inflammatory environment characteristic of PSC.

Aims & Methods

We evaluated the therapeutic efficacy of PBT002, a dual GPBAR1 agonist and RORγt inverse agonist, in modulating immune and fibrotic pathways in PSC. Transcriptomic distribution of target genes was assessed using publicly available single-cell RNA-sequencing datasets from healthy human livers [8]. Differential gene expression was validated in two independent cohorts of patients with PSC and PBC, and further confirmed in an external validation cohort (GSE159676). Functional studies were performed in vitro using primary human cholangiocytes (NHC), immortalized hepatic stellate cells (LX-2), primary monocyte-derived macrophages and CD4+ T lymphocytes. In vivo efficacy was assessed in Abcb4^⁻/⁻ mice treated with PBT002 (10 mg/kg/day) for 12 weeks.

Results

Single-cell transcriptomic analysis revealed that IL-17 and IL-23 cytokines are predominantly expressed by immune cells, whereas their receptors (IL23R, IL17RA, IL17RC, IL17RE) are enriched in epithelial cells, including hepatocytes and cholangiocytes. Deconvolution analysis of liver bulk transcriptomes from PSC and PBC cohorts showed features of immune exhaustion, with a mild upregulation of pro-inflammatory markers and IL-17 receptor subunits, along with prominent overexpression of PD-L1, a known pro-fibrotic mediator. In vitro, exposure of hepatic stellate cells, macrophages, and CD4+ T cells to supernatants from inflamed cholangiocytes induced strong activation, which was reversed by PBT002 treatment. In vivo, PBT002 significantly reduced hepatocellular damage, as shown by decreased serum ALT, AST, ALP, and bilirubin levels. Histological analyses revealed marked reductions in periductal inflammation and liver fibrosis. Gene expression profiling confirmed significant downregulation of M1 and Th17-associated genes (Rorc, Il17f, Il22, Il23a) in colon and liver. Bile acid profiling showed restoration of key GPBAR1 agonists and RORγt antagonists (DCA, 3-oxo-DCA, allo-LCA), while microbiota analysis demonstrated increased abundance of protective taxa (e.g., Blautia, Sutterellaceae), inversely associated with disease severity.

Conclusion

PBT002 exerts potent immunomodulatory effects in experimental PSC by rebalancing M1/M2 and Th17/Treg homeostasis through dual GPBAR1 activation and RORγt inhibition. These effects correlate with restoration of bile acid and microbiota profiles, confirming the key role of the microbiota–bile acid–immune axis in PSC and supporting PBT002’s clinical development for immune-mediated cholangiopathies, particularly in PSC-IBD overlap.

References

1. Fiorucci, S. et al. Bile Acids-Based Therapies for Primary Sclerosing Cholangitis: Current Landscape and Future Developments. Cells 13, (2024).
2. Di Giorgio, C. et al. Liver GPBAR1 Associates With Immune Dysfunction in Primary Sclerosing Cholangitis and Its Activation Attenuates Cholestasis in Abcb4-/- Mice. Liver Int. Off. J. Int. Assoc. Study Liver 45, e16235 (2025).
3. Reich, M. et al. Downregulation of TGR5 (GPBAR1) in biliary epithelial cells contributes to the pathogenesis of sclerosing cholangitis. J. Hepatol. 75, 634–646 (2021).
4. Biagioli, M. et al. Development of dual GPBAR1 agonist and RORγt inverse agonist for the treatment of inflammatory bowel diseases. Pharmacol. Res. 208, 107403 (2024).
5. Garcia Moreno, A. S. et al. IL-17 signaling in primary sclerosing cholangitis patient-derived organoids. Hepatol. Commun. 8, (2024).
6. Louis, E. et al. Risankizumab for Ulcerative Colitis: Two Randomized Clinical Trials. JAMA 332, 881–897 (2024).
7. D’Haens, G. et al. Mirikizumab as Induction and Maintenance Therapy for Ulcerative Colitis. N. Engl. J. Med. 388, 2444–2455 (2023).
8. Andrews, T. S. et al. Single-Cell, Single-Nucleus, and Spatial RNA Sequencing of the Human Liver Identifies Cholangiocyte and Mesenchymal Heterogeneity. Hepatol. Commun. 6, 821–840 (2022).

Introduction

Hepatocellular carcinoma (HCC) remains one of the leading causes of cancer-related mortality worldwide, with limited therapeutic options available for unresectable cases. In recent years, immune checkpoint inhibitors (ICIs) have revolutionized the treatment landscape. Among them, the combination of durvalumab (anti-PD-L1) and tremelimumab (anti-CTLA-4), collectively referred to as Dur/Tre, has demonstrated significant survival benefits in patients with advanced HCC. However, robust prognostic biomarkers specific to patients receiving Dur/Tre therapy have yet to be established. The neutrophil-to-lymphocyte ratio (NLR), a systemic marker of inflammation, has been correlated with prognosis in various malignancies, including HCC [1, 2].

Aims & Methods

This study aimed to assess the prognostic significance of baseline NLR in patients with unresectable HCC treated with Dur/Tre. This multicenter retrospective cohort study included 182 patients with unresectable HCC who received Dur/Tre therapy between May 2022 and June 2024. Eligibility criteria included the availability of baseline NLR data and complete clinical records. The treatment regimen comprised a single 300 mg dose of tremelimumab on day 1, followed by 1500 mg of durvalumab administered every four weeks. Primary endpoints were progression-free survival (PFS) and overall survival (OS), assessed using the Kaplan-Meier method and Cox proportional hazards modeling. Baseline NLR values were calculated, and the optimal cutoff for OS prediction was determined using hazard ratio (HR) spline curve analysis.

Results

The median age of patients was 74 years, and 82.9% were male. The median baseline NLR was 3.13. Dur/Tre was administered as first-line therapy in 29.7% of patients and as second-line or later therapy in 70.3%. The median PFS was 3.5 months (95% confidence interval [CI]: 2.7–4.4 months), while the median OS was not reached (95% CI: 12.1 months–not reached). Multivariate analysis identified administration of Dur/Tre after second-line treatment as an independent predictor of shorter PFS (HR: 1.819; p = 0.003). Notably, an NLR ≥2.56 independently predicted poorer OS (HR: 1.919; p = 0.039). Median OS was not reached in the low-NLR group (<2.56), whereas it was 12.1 months in the high- NLR group (≥2.56) (p = 0.016). A Sankey diagram revealed that patients with higher NLR were more likely to receive best supportive care (BSC) rather than subsequent anticancer therapies post-Dur/Tre (p = 0.046). Spline curve analysis identified an optimal NLR threshold range of 2.3–3.0 for OS prediction.

Conclusion

This large, real-world, multicenter study demonstrated that baseline NLR serves as a significant and independent prognostic biomarker for overall survival in patients with unresectable HCC undergoing Dur/Tre therapy. Higher baseline NLR values (≥2.56) were associated with reduced survival and a lower likelihood of receiving subsequent anticancer treatment, whereas patients with lower NLR values were more likely to continue active therapeutic interventions. These findings suggest that NLR may function not only as a prognostic indicator but also as a practical tool for therapeutic decision-making and patient stratification in clinical settings. Further prospective studies with extended follow-up are warranted to validate these findings and refine NLR-based stratification strategies to optimize immunotherapy outcomes in HCC.

References

1. Tada T, Kumada T, Hiraoka A, et al. Neutrophil-to-lymphocyte ratio is associated with survival in patients with unresectable hepatocellular carcinoma treated with Lenvatinib. Liver Int. 2020; 40: 968-976.
2. Tada T, Kumada T, Hiraoka A, et al. Neutrophil-lymphocyte ratio predicts early outcomes in patients with unresectable hepatocellular carcinoma treated with atezolizumab plus bevacizumab: a multicenter analysis. Eur J Gastroenterol Hepatol. 2022; 34: 698-706.

Disclosure

Toshifumi Tada: lecture fees from AbbVie, AstraZeneca, Eisai, and Chugai
Atsushi Hiraoka: lecture fees from Eisai, Bayer, Eli Lilly, and Otsuka
Hidenori Toyoda: lecture fees from AbbVie, Eisai, Gilead, Terumo, and Bayer

Introduction

Since 2029 national curricula for sedation and emergency management in endoscopy has established for endoscopy staff. NAPS are trained in three-day courses. After a baseline assessment in 2013, the effect of the training was evaluated in a survey in 2018.

Results & Findings

The questionnaire was completed by 356 endoscopy departments (95% in hospitals, 6% in private practice).

• Most departments operate in a multidisciplinary manner (85% gastroenterology, 59% TEE, 13% cardiology, 1.12% urology).
• In 52% of departments, 100% of staff have been trained, while 30% have trained 75-100% of staff. Refresher courses are completed every 2-5 years by 77% of units, while annual resuscitation courses are established in 93% od units.
• 98% have examination rooms equipped with BB, pulse oximeter, ECG, and oxygen supply. The second suction is available in 80%, and capnography in 9%. 83% have their own recovery room, 16% use other recovery rooms.
• ASA classification is used in 80% of unit for risk assessment, while Mallampati score is used in 7% only. 38% developed department specific scores. 63% perform the sign-in for all patients, 16% for 75-100% of patients. Only 10% of departments never perform a sign-in.
• 35% of units perform team-time-out (TTO) for all patients, 11% for 75-100% of patients. Only 26% of departments never use TTO.
• In 47% of departments, a dedicated person for patient monitoring and sedation is available. 30% have this person in 75-100% of patients, and 14% have this person in 25-50% of cases.
• Propofol mono sedation by bolus application is used in 88% of units, TCI in 17% of units, combination Propofol and Midazolam is used in 27%.
• In 77% of units, inpatients and outpatient are equally monitored in the recovery area while 21% of units only use the recovery area for outpatients.
• In 73% of units, physicians discharge the patient, in 12% nurses discharge patients.
• In 80% of departments, patients are only allowed to leave the endoscopy with an accompanying person.
• 70% of departments document all complications, but only 42% evaluate the complication rate statistically.
• 42% of units reported to have no problems to establish all requirements for sedation and patient monitoring. 37% reported about staff shortage. 12% could not find new staff, 25% reported about financial problems. 44% reported about resistance in their own department or with hospital management to establish improvements.

Summary & Discussion

• Comparing data from 2013, 2018 and 2024, improvements could be identified in the establishment of riss assessment by using ASA Score and individual checklists, Malampati score is still not well-known. The establishment of TTO increased from 20 to 35% in all patients, but still needs to be improved.
• The availability of a sedating person remains unchanged with 40 to 47% in all patients and 30% in 75-100% of patients. The recovery situation remains unchanged.

Conclusion

Staff shortage and resistance within the own team are the main challenges when establishing safe sedation and patient monitoring in Endoscopy.

Disclosure

no conflict of interest

Introduction

Lymphocytes inhibit the progression of hepatocellular carcinoma (HCC) through immune surveillance within an immunosuppressive microenvironment. Reduced lymphocyte counts, often associated with systemic inflammation, have been implicated in facilitating tumor progression. In contrast, elevated monocyte counts have been reported to be associated with microvascular invasion and poor prognosis in patients with HCC. Therefore, a high lymphocyte-to-monocyte ratio (LMR), calculated by dividing the lymphocyte count by the monocyte count, is considered a favorable prognostic indicator and is associated with improved clinical outcomes in patients with HCC.

Aims & Methods

We assessed the utility of the LMR in predicting the prognosis of patients with unresectable hepatocellular carcinoma who were treated with atezolizumab plus bevacizumab (Atezo/Bev). The study involved 941 patients who received Atezo/Bev treatment between September 2020 and February 2025. The median age was 74.0 years, and 754 were male. Performance status (PS) was distributed as follows: 741 patients with PS 0, 169 with PS 1, and 31 with PS ≥2. The liver background conditions were as follows: 149 patients with hepatitis B, 282 with hepatitis C, and 449 with non-hepatitis B or C (NBNC). Child-Pugh scores were distributed as follows: 522 patients with a score of 5, 254 with a score of 6, 67 with a score of 7, and 37 with a score of ≥8. The Barcelona Clinic Liver Cancer (BCLC) stages were as follows: 77 patients with stage ≤A, 317 with stage B, and 547 with stage ≥C. The median values of key biomarkers were as follows: LMR: 3.33, NLR: 2.63, AFP: 34.1 ng/mL. The median observation period was 14.3 months. NLR was included in Model 1, and LMR was incorporated in Model 2 of the multivariate analysis.

Results

The progression-free survival (PFS) was 6.8 months, and the overall survival (OS) was 23.6 months. The optimal cutoff for LMR, determined via time-dependent receiver operating characteristic (ROC) analysis based on median OS, was 3.6. The objective response rate (ORR) and disease control rate (DCR) were 27.3% and 76.5%, respectively. In the multivariate analysis of PFS, Model 1 identified the following factors as significant: BMI (HR: 0.98), Child-Pugh score ≥7 (HR: 1.34), BCLC stage C or higher (HR: 1.23), Log AFP (HR: 1.16), second or later treatment line (HR: 1.19). In Model 2, LMR was included as a significant factor (HR: 0.94). For OS, the following factors were identified as significant: age (HR: 1.01), Child-Pugh score ≥7 (HR: 2.09), BCLC stage C or higher (HR: 1.24), Log AFP (HR: 1.28), second or later treatment line (HR: 1.24), and NLR (HR: 1.03). The integrated discrimination improvement (IDI) for NLR and LMR at PFS (at 3 and 6 months) was 0.012 (95% CI: 0.012). The net reclassification improvement (NRI) was 0.016 (95% CI: 0.006-0.030, p<0.001) at 3 months, 0.023 (95% CI: 0.006-0.030, p<0.001) at 6 months, and 0.023 (95% CI: 0.004-0.024, p<0.001) at 12 months. The IDI for NLR and LMR at OS (12 and 24 months) was 0.027 (95% CI: 0.011-0.049, p=0.002) and 0.034 (95% CI: 0.012-0.064, p<0.001), with LMR demonstrating superior performance. The NRI for OS at 12 and 24 months was 0.036 (95% CI: 0.016-0.062, p<0.001) and 0.053 (95% CI: 0.020-0.091, p<0.001), respectively, both favoring LMR. In the time-dependent ROC analysis, the area under the curve (AUROC) was higher for LMR compared to NLR throughout the study period, with the exception of PFS at 3 months and OS at 18 months.

Conclusion

The LMR serves as a valuable biomarker for predicting prognosis in patients receiving Atezo/Bev.

Disclosure

Toshifumi Tada: lecture fees from AbbVie, AstraZeneca, Eisai, and Chugai
Atsushi Hiraoka: lecture fees from Eisai, Bayer, Eli Lilly, and Otsuka
Hidenori Toyoda: lecture fees from AbbVie, Eisai, Gilead, Terumo, and Bayer

Introduction

Colon capsule endoscopy (CCE) or panenteric capsule endoscopy (PCE) offers a promising, non-invasive diagnostic approach for patients with iron deficiency anaemia (IDA). However, high rates of conversion to conventional colonoscopy (CCC) following capsule procedures reduce cost-effectiveness and patient satisfaction. Optimising the faecal immunochemical test (FIT) threshold may improve patient stratification and reduce unnecessary conversions in future applications within the IDA diagnostic pathway.

Aims & Methods

This study aimed to evaluate the accuracy of faecal immunochemical testing (FIT) in predicting colorectal cancer (CRC) and polyp burden in patients with iron deficiency anaemia (IDA), and to extrapolate these findings to assess FIT’s utility in predicting the likelihood of conversion from CCE to conventional colonoscopy (CCC) based on underlying pathology in IDA.
The CLEAR IDA study was a multicentre, retrospective observational study conducted across four UK hospitals. Data were collected over a six-month study period and included patients referred via the two-week-wait (2WW) cancer pathway for iron deficiency, with or without anaemia, over a 12-month timeframe. Colonoscopy findings were analysed and extrapolated using NHS England’s CCE-to-colonoscopy referral criteria to assess the predictive value of FIT for colorectal cancer (CRC), polyp burden, and CCC using ROC curve analysis. The optimal FIT threshold was identified through three complementary approaches: threshold-based analysis, decision curve analysis, and cost-benefit modelling.

Results

A total of 1,531 patients were analysed; only 1.6% underwent small bowel capsule endoscopy. The diagnostic accuracy (AUC) of FIT for predicting CRC, polypoidal lesions, and CCC was 0.78, 0.58, and 0.69, respectively. Threshold-based analysis identified FIT = 15 µg/g as the lowest level at which CCC rates significantly increased (p = 0.02; OR = 1.87; 95% CI: 1.07–3.14). Decision curve analysis showed maximum net benefit at FIT = 17.6 µg/g, while cost-benefit modelling identified 9 µg/g as the most cost-effective. An optimal cost-effective FIT threshold range was identified between 10 and 17.6 µg/g. Threshold selection should be tailored to local service capacity and resource availability.

Conclusion

Significant pathologies in the IDA pathway were uncommon. A FIT threshold range between 10–17 µg/g may enhance cost-effectiveness and guide appropriate PCE use in IDA.

References

1) Snook J, Bhala N, Beales ILP, et al. British Society of Gastroenterology guidelines for the management of iron deficiency anaemia in adults. Gut. 2021 Nov;70(11):2030-2051.
2) (NICE) NIFHACE. PillCam COLON 2 for investigation of the colon through direct visualisation: NATIONAL INSTITUTE FOR HEALTH AND CARE EXCELLENCE; August 2024 [cited 2025 23/03/2025]; Diagnostics Assessment Programme]. Available from: https://www.nice.org.uk/guidance/gid-dg10083/documents/final-scope-2
3) Turvill J, Haritakis M, Pygall S, et al. Multicentre Study of 10,369 Symptomatic Patients Comparing the Diagnostic Accuracy of Colon Capsule Endoscopy, Colonoscopy and CT Colonography. Aliment Pharmacol Ther. 2025 Feb 26;0:1–13.
4) Cilona A, Zullo A, Hassan C, et al. Is faecal-immunochemical test useful in patients with iron deficiency anaemia and without overt bleeding? Dig Liver Dis. 2011 Dec;43(12):1022-4.
5) Turvill JL, Turnock D, Cottingham D, et al. The Fast Track FIT study: diagnostic accuracy of faecal immunochemical test for haemoglobin in patients with suspected colorectal cancer. Br J Gen Pract. 2021 Aug;71(709):e643-e651.

Introduction

Gastroenterologists face dual demands from clinical practice and academic research, necessitating efficient literature processing. While large language models (LLMs) show promise as healthcare tools, their variable performance in literature interpretation remains understudied, particularly regarding content completeness and reliability. This study systematically evaluates four prominent LLMs (ChatGPT-4o, Claude-3.5sonnet, Doubao-pro, Deepseek-V3) in explaining five gastroenterology literature types (original articles, case reports, reviews, editorials, guidelines), providing evidence-based recommendations for practical applications.

Aims & Methods

We tasked four LLMs with interpreting 125 gastroenterology publications (25 per literature type, 5 interpretations per article) spanning artificial intelligence, endoscopy, gastrointestinal cancers, and inflammatory bowel disease. Five gastroenterologists assessed interpretations using five criteria:
1. Accuracy: Correct identification of study design.
2. Format compliance: Adherence to predefined structure.
3. Content completeness: Key concept coverage (quantified via standardized rating scales).
4. Conciseness: Absence of extraneous content.
5. Reliability: Absence of hallucinated content.
Domain experts developed literature-specific completeness scales. Statistical analysis employed Pearson's chi-square tests for categorical variables (accuracy, format compliance, conciseness, reliability) and Kruskal-Wallis tests with Dunn's post-hoc comparisons for completeness scores. We applied Bonferroni correction for multiple comparisons (R version 4.4.2), with α=0.05 defining statistical significance.

Results

The 625 evaluations revealed significant inter-model differences across all metrics (p<0.001). ChatGPT-4o and Deepseek-V3 demonstrated superior accuracy (99.2% vs 100.0%) and format compliance (97.6% vs 99.2%) compared to Claude-3.5sonnet and Doubao-pro. Claude-3.5sonnet produced the most concise outputs (99.4%), while Deepseek showed the lowest concision (92.8%). Doubao exhibited perfect reliability (100.0%) versus 93.6%-97.0% in other models. Median completeness scores ranked: ChatGPT-4o (84.5%) > Claude-3.5sonnet (80.0%) = Doubao-pro (80.0%) > Deepseek-V3 (79.3%), with ChatGPT-4o significantly worse than others (Dunn's test, p<0.05).

Table 1. Large Language Model Performance by Evaluation Metric
LLM	ChatGPT-4o		Claude-3.5sonnet		Doubao-pro		Deepseek-V3
Evaluation metric	Yes, n(%)	No, n(%)	Yes, n(%)	No, n(%)	Yes, n(%)	No, n(%)	Yes, n(%)	No, n(%)
Accuracy	620(99.2)b	5(0.8)	580(92.8)a	45(7.2)	560(89.6)a	65(10.4)	625(100.0)b	0(0)
Format compliance	610(97.6)b	15(2.4)	520(83.2)a	105(16.8)	530(84.8)a	95(15.2)	620(99.2)b	5(0.8)
Conciseness	613(98.1)a,c	12(1.9)	621(99.4)a	4(0.6)	600(96.0)b,c	25(4.0)	580(92.8)b	45(7.2)
Reliability	606(97.0)a	19(3.0)	603(96.5)a,b	22(3.5)	625(100.0)c	0(0)	585(93.6)b	40(6.4)
Superscripts denote significant differences between models (same letter = no significant difference).

Conclusion

Gastroenterologists can select LLMs based on the application priority of literature processing: ChatGPT-4o and Deepseek-V3 excel in accuracy and format compliance; Claude-3.5sonnet offers superior conciseness; while Doubao-pro demonstrates optimal reliability.