Predicting Cancer in Inflammatory Bowel Disease Starts With Good Notes

Models for colorectal cancer (CRC) risk prediction in patients with inflammatory bowel disease are freely available in validated online calculators to use during shared decision-making.1, 2, 3 The statistical models use standard variables typically included in colonoscopy and pathology reports that can then be captured as free text in the electronic health record and interpreted by increasingly accurate large language models.⁴ By incorporating artificial intelligence, we will soon have tools that automatically extract variables, combine different risk weighting for each variable, and project predicted CRC risk in the electronic health record without additional human effort. This will give precious time back to patients and health-care providers at the bedside. However, the accuracy of these models critically hinges on appropriate documentation in the procedure and pathology notes. When variables such as biopsy location and nature (eg, random vs targeted), lesion characteristics or lack thereof, severity of inflammation, among others, are uncertain from the notes, artificial intelligence methods cannot reliably assign a risk category or predict the patient’s future CRC risk. Taking thorough notes, and ultimately using standardized note templates, is worthwhile both for creating data fit for research purposes (such as training/testing new models) and for real-time model implementation with patients, which is imminent for medicine.