New AI Tool Spots Bladder Cancer Warning Signs Up to Five Years Early
A UK research team built an AI system that mines health records for subtle bladder cancer warning signs, though it's still years from clinical use.
What happened: Researchers at the University of Plymouth built an AI tool called PRECISE-AGZ that scans electronic health records for subtle signs of bladder cancer. Trained on records from nearly 70,000 Welsh patients spanning 1995 to 2020, it sifted through 48,261 potential health indicators "from smoking habits to medication use" and narrowed them to 38 features that reliably flagged risk. In testing, it correctly identified 85% of patients who actually had bladder cancer and correctly cleared 91% of those who didn't, with some warning signals showing up as much as five years before official diagnosis. The findings were published in IEEE Transactions on Biomedical Engineering.
Why it matters: Bladder cancer kills roughly 220,000 people worldwide every year and ranks as the ninth most common cancer globally, yet there's no routine screening program for it. Diagnosis usually hinges on someone noticing blood in their urine, a symptom that can just as easily mean kidney stones or prostate trouble, and confirming cancer requires an invasive cystoscopy. A tool that flags risk earlier and more precisely could mean people get treated while the disease is still highly curable, and fewer people undergo procedures they don't need.
How it works, plainly: Instead of waiting for one classic symptom, the AI weighs dozens of everyday record details together "medications, coexisting conditions, exercise habits" and sorts patients into low-risk, high-risk, or an uncertain "gray zone" worth monitoring. It surfaced patterns doctors don't currently act on: long-term users of the breast cancer drug tamoxifen showed higher bladder cancer risk, while patients with Parkinson's or dementia showed lower risk. The researchers are careful to note these are statistical links, not proven causes, and that blood in urine meant different things depending on a patient's other conditions.
What's still missing: Every patient in this study came from a single database covering Wales, and the research team says validation on other countries' and systems' records is essential before anything moves toward real use. There's no announced timeline for a pilot or clinical trial. The unusual risk patterns the AI found, like the tamoxifen and dementia links, still need dedicated follow-up research to confirm they're clinically meaningful rather than coincidental. For now this is an early, promising study "not a screening test anyone can access."