Citations

Abstract Objectives Accurate and timely prediction of physiological deterioration in hospitalized patients is essential, but existing deep learning models, which rely solely on structured electronic health record (EHR) data, generate high false-positive rates by missing crucial contextual information in clinical notes. We present DETERIO-LLM, a hybrid prediction model that selectively integrates large language models (LLMs) with a deep learning model (DETERIO) to reclassify borderline-risk alerts using narrative clinical notes. Materials and Methods This retrospective study used a cohort of 1000 inpatients (4.6% deterioration prevalence). DETERIO-LLM selectively analyzed narrative notes for alerts in the uncertainty range (scores 2.5-3.5). Performance was evaluated using sensitivity, positive predictive value (PPV), and F1 score, benchmarked against comparators including the best-established ML model, eCART. Results DETERIO-LLM significantly improved predictive precision over all comparators. At its optimal threshold, the model achieved a PPV of 30.6% and an F1 score of 37.3%, outperforming eCART. This selective integration reduced false-positive alerts by 46.5% within the uncertainty range while maintaining comparable sensitivity. Discussion This study demonstrates a practical, interpretable method for integrating LLM-derived contextual knowledge into existing structured EHR prediction workflows. By focusing LLM analysis on high-uncertainty alerts, DETERIO-LLM substantially improves clinical utility through a marked reduction in false-positive cases, which is critical for mitigating alert fatigue. Conclusion Selective LLM integration with structured EHR models offers a viable, interpretable path to improving early warning systems in clinical care by leveraging narrative context to reduce alert fatigue while enhancing predictive performance.

The Ajmera Transplant Center and Mayo Clinic hosted the third annual Transplant Artificial Intelligence (AI) Symposium in Toronto, Canada, bringing together expert clinicians, researchers, scientists, and trainees to discuss the current role of AI in transplant medicine. This paper summarizes the third annual Transplant AI Symposium proceedings and talks. Presentations covered a wide range of topics across the transplant continuum, highlighting numerous benefits of AI in transplantation such as organ matching, human-AI collaboration, and survival/risk prediction. Artificial intelligence is most useful when linked to specific clinical problems, especially those involving multimodal or longitudinal data. However, speakers also emphasized ongoing limitations in data quality, generalizability, workflow integration, and fairness. Multiple presentations highlighted the importance of clinician oversight. Overall, the symposium highlighted that the future of transplant AI will depend on careful validation, clinically meaningful implementation, and attention to patient outcomes ( Figure 1 ).

Background: Hepatocellular carcinoma (HCC) has a poor long-term prognosis due to high recurrence and cirrhosis-related mortality, even after potentially curative treatments such as liver transplantation (LT), surgical resection, or ablation. This study aimed to identify factors associated with ≥10-year survival in HCC patients. Methods: A retrospective cohort study was conducted among HCC patients diagnosed between 2004 and 2022 using the National Cancer Database. Multivariable Cox regression was used to identify predictors of overall survival, and logistic regression was used to identify predictors of ≥10-year survival. Results: Among 249,600 HCC patients, 177,585 (71.2%) died within 5 years, 8613 (3.5%) died at 5–10 years, 54,988 (22.0%) were alive with <10 years of follow-up, and 8219 (3.3%) survived ≥10 years. LT, resection, and ablation were performed in 6.6%, 9.3%, and 11% of patients, respectively. Compared with ablation as the reference group, LT [adjusted odds ratio (aOR) 11.96, 95% confidence interval (CI): 11.27–13.29] and resection (aOR: 2.83, 95% CI: 2.57–3.08) increased the odds of ≥10-year survival, while non-curative treatments reduced the odds compared with ablation (aOR: 0.50, 95% CI: 0.47–0.55). Cox regression results were consistent with the logistic model, confirming the association. Decision tree analysis confirmed LT as the dominant determinant of long-term survival. Black individuals were associated with lower odds of ≥10-year survival (aOR: 0.88, 95% CI: 0.820–0.96) and decreased likelihood of receiving LT (aOR: 0.73, 95% CI: 0.55–0.96). Conclusions: LT offers the best chance of ≥10-year survival in HCC. Ensuring equitable access is essential, especially for Black patients who have lower transplant rates and worse outcomes.

INTRODUCTION: Chronic gastrointestinal (GI) conditions, including irritable bowel syndrome and other disorders of gut-brain interaction, are prevalent and costly, and fragmented care contributes to suboptimal outcomes. Virtual and multidisciplinary care models may improve outcomes. Objectives: We aimed to (i) characterize patients receiving virtual GI care, (ii) report patient engagement with and patient-reported outcomes after virtual multidisciplinary care, and (iii) report healthcare utilization and costs from a payer perspective among patients vs propensity score-matched controls identified from claims data. METHODS: We collected descriptive data on patient demographics, clinic provider visits, symptoms, and patient-reported outcomes from 3/2021 to 9/2022. To evaluate healthcare utilization and costs, we conducted difference-in-difference analyses using claims data compared with a propensity score-matched control group. RESULTS: Among 234 patients (71% women, mean age 45.4 ± 13.2 years), 51% received a new GI diagnosis, predominately disorders of gut-brain interaction (63%). Engagement criteria were met by 80% (n = 187), who completed a mean of 10 visits and median time-to-first appointment of 6 days. Dietitians (93%), psychologists (76%), and health coaches (64%) were frequently used. Engaged patients exhibited significant improvements from preclinic to postclinic in irritable bowel syndrome severity (226.7–145.6), symptom control (20.2%–86.6%), quality-of-life dysphoria (27.5–32.8), missed workdays (1.0–0.7), and satisfaction (37.4%–96.3%) (all P < 0.05). Compared with matched controls, a significantly smaller percent of patients had imaging (15% vs 44%; P < 0.0001) and GI-related emergency department visits (4% vs 10%; P = 0.0028), which contributed to reductions in GI-related costs ($443 per member per month; P = 0.047) and all-cause costs ($676 per member per month; P = 0.043), equating to annualized savings of $5,316 and $8,112, respectively. DISCUSSION: Virtual multidisciplinary GI care was associated with high engagement, improved clinical outcomes, and reduced healthcare utilization and costs.

INTRODUCTION: The American College of Gastroenterology assembled a multidisciplinary task force to evaluate the current state and future direction of artificial intelligence (AI) in gastroenterology, hepatology, and endoscopy leading to the development of consensus-based recommendations for responsible AI integration in clinical practice. METHODS: A total of 32 subject-matter experts and 12 industry partners, representing diverse practice settings and expertise, conducted subgroup literature reviews across 5 key areas (endoscopy, practice management clinical applications, training and education, inflammatory bowel disease and liver disease, ethics and equity). Draft statements were developed and rated on a 5-point Likert scale using a modified Delphi process. A consensus was set at ≥70% combined agreement. Nonconsensus items were revised and revoted electronically. RESULTS: A total of 43 statements, 40 (93%) reached consensus in round 1 and the remaining 3 achieved consensus after round 2. Evidence supports computer-aided detection improving adenoma detection rate and miss rate in controlled studies, with mixed real-world impact and insufficient long-term outcomes (e.g., interval colon cancer rate). Recommendations emphasize thorough validation and reduction of bias by heterogeneous Data sets. Outside endoscopy, ambient AI scribes, natural language processing (NLP)-enabled coding, workflow optimization, and previous authorization support show potential. Training recommendations endorse a structured AI curriculum while preserving independent procedural competence to avoid deskilling. In inflammatory bowel disease and hepatology, AI could help improve diagnostic accuracy, help predict risk of disease progression, and help guide therapy. Equity, governance, and reimbursement statements call for chain-of-custody data protections, specialty-society oversight, and payment models that reward quality and cost reduction. DISCUSSION: This consensus outlines how AI can augment rather than replace clinical expertise while promoting safety, transparency, interoperability, and equity. Priorities include pragmatic and prospective trials, multi-institutional data-sharing consortia, bias mitigation, and workforce training to enable trustworthy and clinically impactful AI adoption in gastroenterology, liver, and endoscopy care.

ABSTRACT Background Systemic infections are a leading cause of hospitalization and death among patients with cirrhosis. Timely and accurate infection identification is essential for both clinical care and the development of predictive models. However, existing methods such as ICD-10 coding are unreliable, and manual chart review is resource-intensive and difficult to scale. This study aimed to develop and validate an automated large language model (LLM)-based approach for infection classification and subtyping in patients with cirrhosis presenting to the emergency department (ED). Method We developed INFEHR (INfection identification and subtyping using Free-text EHR analysis), an LLM-powered pipeline utilizing Claude 3.5 Sonnet to analyze clinical notes from the first 72 hours of admission. Model outputs were compared against a physician-adjudicated gold standard in a cohort of 1,000 encounters from patients with cirrhosis who presented to the ED. Performance was benchmarked against ICD-10 code–based labeling and CDC Adult Sepsis Event criteria. Results INFEHR achieved 94.7% overall accuracy, with 99.5% sensitivity and 92.8% positive predictive value for identifying infection presence, outperforming ICD-10–based classification across all metrics ( p < 0.0001). The model also demonstrated strong performance in classifying pathogen type and infection site. This pipeline processed notes within seconds, offering improvements in efficiency and scalability over manual review. Conclusion INFEHR offers a scalable, reproducible, and accurate method for infection phenotyping in cirrhosis. By overcoming limitations of traditional coding and manual review, it supports high-throughput infection surveillance, improves cohort construction for clinical research, and enables future integration into real-time decision-support tools in hepatology.

Background Antibiotic use may alter gut microbiota, which could be associated with various diseases, including cardiovascular disease (CVD). Accordingly, further epidemiological evidence is needed to clarify the potential risk of CVD associated with antibiotic exposure. We evaluated the association between long‐term antibiotic exposure and CVD risk in South Korea. Methods and Results This nationwide retrospective cohort study used data from the Korean National Health Insurance Service database. A total of 2 159 965 individuals aged 40 to 79 years who had no history of antibiotic prescriptions from 2004 to 2005 were included and followed through December 31, 2021. Antibiotic exposure was defined as the cumulative days of antibiotic prescriptions from 2006 to 2010. The primary end point was incident CVD with 2 or more days of hospitalization from 2011 to 2021. Adjusted hazard ratios (aHR) and 95% CIs for CVD according to antibiotic exposure were estimated using multivariable Cox proportional hazards models, adjusting for a range of covariates. Longer cumulative antibiotic prescription days were associated with an increased risk of CVD. The aHRs for CVD associated with 1 to 14, 15 to 90, 91 to 180, 181 to 364, and ≥365 days of antibiotic exposure were 0.98 (95% CI, 0.96–1.01), 1.01 (95% CI, 0.99–1.03), 1.05 (95% CI, 1.03–1.08), 1.06 (95% CI, 1.04–1.09), and 1.10 (95% CI, 1.07–1.13) compared with those with no antibiotic prescriptions. Conclusions In this large population‐based cohort study of Korean adults, long‐term antibiotic exposure may be associated with a higher risk of CVD. Further research is warranted to establish causality and elucidate the underlying mechanisms for this association.

Background and Aims: Immunotherapy has emerged as an effective treatment for advanced HCC. We aimed to investigate the real-world effectiveness of immunotherapy compared to lenvatinib in HCC. Approach and Results: From the TriNetX database, we used a target trial emulation framework and identified patients with HCC who received first-line treatment with immunotherapy (atezolizumab/bevacizumab or tremelimumab/durvalumab) or lenvatinib between or between August 2018 and December 2023. OS was compared using Kaplan-Meier analysis and Cox proportional hazards regression. After propensity score matching, 1203 patients were included in each group. Immunotherapy was associated with improved OS versus lenvatinib (median survival: 545 vs. 425 d; HR: 0.86, 95% CI: 0.76–0.97). Regarding treatment type, atezolizumab plus bevacizumab showed improved survival compared to lenvatinib (n=1070 in each group; HR: 0.87, 95% CI: 0.77–0.99), while the point estimate favored durvalumab plus tremelimumab versus lenvatinib (HR: 0.81, 95% CI: 0.59–1.12), though this difference was not statistically significant, likely due to small sample size. Regarding etiology, immunotherapy had improved OS compared to lenvatinib in viral hepatitis (n=510 in each group; HR: 0.74, 95% CI: 0.61–0.89) and alcohol-associated liver disease (n=190 in each group; HR: 0.65, 95% CI: 0.49–0.87), but not in metabolic dysfunction–associated steatotic liver diseases (n=156 in each group; HR: 0.96, 95% CI: 0.70–1.31). Conclusions: In this real-world analysis, immunotherapy was associated with improved OS compared to lenvatinib in advanced HCC, with consistent benefit across most subgroups. These findings support the use of immunotherapy as a first-line treatment for advanced HCC.