People with ulcerative colitis (UC), a chronic inflammatory bowel disease, face up to four times the risk of developing colorectal cancer compared to the general population. A key early warning sign is low-grade dysplasia (LGD)—abnormal, precancerous lesions—though only a subset of these cases progress to cancer. This uncertainty makes it difficult for both clinicians and patients to decide on the best course of care, ranging from routine surveillance to preventative surgery. Now, a new study led by researchers at the University of California San Diego demonstrates that artificial intelligence (AI), when combined with biostatistical risk models, can accurately predict which UC-LGD patients are most likely to develop cancer. The findings, published on February 17 in Clinical Gastroenterology and Hepatology, could transform patient counseling, improve decision-making, and ensure timely follow-up care. The research team developed a fully automated AI workflow to analyze the medical records of 55,000 patients from the U.S. Department of Veterans Affairs (VA) health care system. This dataset is the largest of its kind in the United States and includes detailed clinical notes, colonoscopy reports, and pathology records. The AI system was trained to extract critical risk factors—such as lesion size, number of lesions, and severity of colon inflammation—from unstructured narrative text in clinical notes. “Large language models were able to accurately identify colitis-associated colorectal cancer risk factors directly from the clinical notes,” said Kit Curtius, Ph.D., assistant professor of medicine in the Division of Biomedical Informatics at UC San Diego School of Medicine and a member of Moores Cancer Center. “This is a major step forward in turning qualitative clinical data into actionable, quantitative risk scores.” The AI-driven risk model provides precise, individualized risk assessments that can be seamlessly integrated into clinical workflows. Instead of relying on subjective judgment or incomplete data, clinicians can now use a clear risk score during patient visits to guide decisions about colonoscopy timing or the need for surgery. “Currently, advising patients about their risk is often based on intuition and limited data,” Curtius explained. “This AI pipeline can read the notes and deliver a specific risk score—transforming a list of risk factors into a meaningful number that supports informed decisions.” The technology also holds promise for improving patient follow-up. By identifying individuals at highest risk, it can help flag those who need earlier or more frequent screenings, reducing delays that contribute to cancer progression. Looking ahead, the team plans to validate the AI tool in diverse patient populations beyond the VA system and incorporate emerging risk factors, including genetic data. “We know genomics play a significant role in cancer development,” Curtius noted. “Future versions of the model will integrate genetic information to further refine risk prediction.” Additional co-authors include Brian Johnson, Hyrum Eddington, Samir Gupta, and Shailja C. Shah from UC San Diego and VA San Diego Healthcare System, and Misha Kabir from University College London Hospitals NHS Trust.