AI Detects Hidden Bladder Cancer Warning Signs up to 5 Years Before Diagnosis

Researchers at the University of Plymouth built PRECISE-AGZ, an AI system that scans electronic health records for combinations of symptoms, prescriptions and risk factors linked to bladder cancer. Across nearly 70,000 patients (1995–2020), it detected 85% of cancer cases and 91% of cancer-free patients, outperforming current NHS referral guidelines.

• Model examined 48,261 indicators and narrowed them to 38 key features
• Sensitivity 85%, specificity 91% on the SAIL database in Wales
• Some risk signals appeared up to 5 years before diagnosis, strongest in the final 12 months
• Patients split into low (below 7%), uncertain (7–55%) and high (above 55%) risk groups