In a groundbreaking medical advancement, researchers at the University of California San Diego have harnessed the power of artificial intelligence (AI) alongside sophisticated biostatistical modeling to revolutionize risk prediction for patients with ulcerative colitis (UC) presenting low-grade dysplasia (LGD). This chronic inflammatory condition already places patients at an elevated risk for developing colorectal cancer, yet discerning which lesions might progress malignantly has long eluded clinicians. By leveraging AI to analyze massive datasets drawn from the U.S. Department of Veterans Affairs (VA) health system, the new technology produces highly accurate, individualized cancer risk assessments, forging a path toward more nuanced, personalized patient care.
Ulcerative colitis is an inflammatory bowel disease characterized by persistent inflammation and ulceration in the colon, increasing the risk for colorectal cancer roughly fourfold compared to the general population. One of the earliest precursors signaling this progression is low-grade dysplasia—microscopic abnormalities in the colon’s cellular architecture considered precancerous. However, the clinical challenge lies in the fact that most low-grade dysplasia cases do not evolve into cancer, leaving physicians in a quandary of uncertain prognosis when determining surveillance intervals or recommending prophylactic surgery.
Addressing this clinical ambiguity, the UC San Diego research team developed a fully automated AI pipeline to comb through a vast repository of over 55,000 patients’ medical records within the VA system—the largest such dataset in the United States. This deep dive into colonoscopy narratives, pathology reports, and clinical notes applied large language models—advanced AI systems capable of natural language comprehension—to extract discrete clinical variables relevant to cancer risk. These included lesion size, multiplicity of dysplastic sites, inflammation severity, and whether dysplastic lesions were visibly resected completely during procedures.
The predictive capacity of this AI workflow is striking. It successfully stratified patients into five risk categories based on four established, empirically validated clinical factors, aligning predictions closely with actual patient outcomes tracked over a decade following LGD diagnosis. Nearly half the patients fell into the lowest risk group, with predictions indicating that around 99% of these individuals would remain cancer-free two years post-assessment. Such precision offers profound reassurance to low-risk patients and could usher in longer surveillance intervals, lessening the frequency of invasive colonoscopies—a significant quality-of-life improvement.
Crucially, the AI model illuminates risk nuances previously underestimated through subjective clinical judgment. For example, patients harboring unresectable visible lesions—those too extensive or inaccessible for complete surgical removal—exhibited substantially elevated cancer risk profiles. This insight calls for reconsideration of surgical versus surveillance strategies for certain cohorts, enhancing decision-making paradigms with robust data rather than intuition alone.
This novel AI-driven approach promises more than just risk stratification; it integrates seamlessly into routine clinical workflows, automating what has traditionally been a manual, subjective process. By translating complex clinical narratives into quantifiable risk scores in real time, it empowers doctors and patients to make data-backed decisions about surveillance timing, intervention thresholds, and proactive care management—ultimately aiming to preempt cancer development and improve survival outcomes.
Moreover, the system could play a critical role in healthcare administration by promptly flagging individuals who require follow-up colonoscopies, a measure that addresses a major contributor to preventable colorectal cancers due to delayed surveillance. Early and precise identification of patients at heightened risk equips clinicians to prioritize resources efficiently within often overburdened clinical settings, elevating both care quality and system performance.
The implications of this work extend into future research horizons, with plans underway to validate the AI model across more diverse patient populations beyond the VA system, ensuring generalizability and equity in risk prediction. Furthermore, the integration of emerging biomarkers, such as genetic and genomic data, holds promise to refine risk calculations even further. The interplay between genomics and cancer progression remains a vibrant avenue for exploration, one that this AI framework is well poised to incorporate.
This study, published in the prestigious journal Clinical Gastroenterology and Hepatology, exemplifies the transformative potential of marrying artificial intelligence with biostatistics for disease prevention and personalized medicine in gastroenterology. By demystifying risk for clinicians and empowering patients with precise prognostic information, it marks a new era in managing ulcerative colitis and colorectal cancer prevention.
The project was spearheaded by Dr. Kit Curtius, an assistant professor of medicine in UC San Diego’s Division of Biomedical Informatics and a member of the Moores Cancer Center, who also serves as a research health scientist at the VA San Diego Healthcare System. The endeavor involved a collaborative team, including co-authors from UC San Diego and University College London Hospitals NHS Trust, and received support through funding from the U.S. Department of Veterans Affairs Biomedical Laboratory Research and Development Service as well as the National Institutes of Health.
Dr. Curtius emphasized the novel capability of large language models to autonomously extract nuanced risk factors such as dysplasia size, inflammation severity, and lesion visibility directly from unstructured clinical notes—a formidable challenge in traditional risk assessment that often suffers from incomplete or fragmented data. This precision augments clinicians’ toolkit, transforming how surveillance schedules and treatment plans are formulated, potentially minimizing unnecessary procedures and focusing interventions where they are most urgently needed.
In summation, this AI-driven risk prediction paradigm embodies a crucial leap toward precision oncology in colorectal cancer prevention for ulcerative colitis patients. By providing actionable insights grounded in comprehensive data analysis, it offers hope for more tailored, effective, and patient-centered care pathways, ultimately aspiring to reduce the burden of colorectal cancer in vulnerable populations worldwide.
Subject of Research: Risk Prediction in Ulcerative Colitis Patients With Low-Grade Dysplasia Using Artificial Intelligence and Biostatistical Models
Article Title: AI-Driven Automated Risk Stratification in Ulcerative Colitis Low-Grade Dysplasia: A Decade-Long Predictive Model
News Publication Date: February 17, 2026
Web References: Clinical Gastroenterology and Hepatology Article
References: DOI: 10.1016/j.cgh.2026.01.037
Image Credits: Not provided
Keywords
Artificial intelligence, Colorectal cancer, Disease prevention, Ulcerative colitis, Low-grade dysplasia, Biostatistical risk modeling, Large language models, Precision medicine, Gastroenterology, Cancer risk stratification
