Bladder cancer model identifies risk signals in patient records before diagnosis

Researchers led by the University of Plymouth report that an AI model found patterns in medical records associated with bladder cancer before diagnosis. In records from nearly 70,000 patients, the model detected 85% of patients with the disease and correctly classified 91% of those without it. Some signals appeared up to five years before diagnosis, although performance was strongest in the final 12 months. The system combines symptoms, risk factors, prescriptions and clinical activity rather than relying on blood in the urine alone. Its evaluation used data from Wales; the researchers say it needs validation in other health systems before routine clinical use.