Analysis of Risk Factors for Colorectal Cancer Associated with Ulcerative Colitis Using Machine Learning: A Retrospective Longitudinal Study Using a National Database in Japan.

Background/

Objectives

Ulcerative colitis (UC) is a chronic inflammatory bowel disease that significantly increases the colorectal cancer (CRC) risk. This study used nationwide data on intractable diseases to clarify the clinical epidemiology of UC-related CRC in Japan.

Methods

Patients diagnosed with UC between FY 2003 and 2011 were included. The relative incidence ratio (RR) was calculated using the standardized incidence ratio from the National Cancer Registry. To compare prognostic factors, outcomes were evaluated using the Cox proportional hazards model analysis for cancer occurrence, and a prognostic prediction model was developed using machine learning.

Results

Among 78,556 patients with UC, CRC was identified in 141 patients. The RR of CRC peaked in both males and females in the 25-39 age group. Univariate analysis revealed several risk factors, including pseudo-polyps observed during endoscopy (hazard ratio 2.92, p = 0.001), abnormal crypt architecture (hazard ratio 3.14, p < 0.001), and dysplasia (hazard ratio 11.31, p < 0.001) in biopsy. Conversely, 5-ASA was associated with reduced CRC risk (hazard ratio 0.36, p = 0.003). The machine learning model categorized patients into three groups, demonstrating that the group with the highest number of patients with pancolitis had a significantly higher risk of CRC than did the other groups.

Conclusions

Pseudo-polyps and dysplasia represent CRC risk factors in patients with UC. Additionally, machine learning analysis indicates that pancolitis in individuals in their 50s increases the risk of colon cancer, while proctitis in those in their 30s raises rectal cancer risk. These findings aim to enhance early detection and improve prevention efforts for UC-related CRC.