AI Tool Predicts Bladder Cancer from Electronic Health Records

Reliance on visible hematuria and confirmatory cystoscopy may delay diagnosis while exposing patients to invasive procedures. Health systems therefore need noninvasive risk tools that can identify earlier warning signs within routine care. Researchers have now developed an artificial intelligence system that predicts bladder cancer risk years before formal diagnosis.

Developed at the University of Plymouth, the PRECISE-AGZ model analyzes electronic health records to identify subtle patterns that emerge before a bladder cancer diagnosis. The system examines routine clinical data, including symptoms, risk factors, and health care activity, allowing it to move beyond referral pathways based primarily on individual symptoms. Current pathways often emphasize visible hematuria, which can also arise from benign conditions and still requires cystoscopy for diagnostic confirmation.

The study analyzed nearly 70,000 patient records collected between 1995 and 2020. Using a purpose-built model, investigators evaluated 48,261 potential indicators spanning patient behaviors, comorbidities, medication histories, and other routinely recorded information. From this large set of variables, the algorithm identified 38 key features that were most strongly associated with bladder cancer risk.

PRECISE-AGZ detected bladder cancer in 85% of patients who had the disease and correctly identified cancer-free individuals with 91% accuracy. The model demonstrated effective detection up to 12 months before diagnosis, with some predictive signals appearing as early as five years beforehand, and it outperformed current National Health Service referral guidelines. 

The tool classified patients into low-risk, gray-zone, and high-risk groups, corresponding to probabilities below 7%, between 7% and 55%, and above 55%, respectively. This stratification could help prioritize patients for further evaluation, while those in the gray zone could potentially undergo continued monitoring before invasive testing, reducing unnecessary cystoscopies.

The model confirmed established associations such as smoking and hematuria while also identifying patterns that are less apparent through conventional screening approaches. Patients with Parkinson’s disease or dementia showed a lower observed risk, while long-term tamoxifen use was associated with higher risk, although these findings do not establish prevention or causation. 

The predictive significance of hematuria also changed when considered alongside other clinical indicators; for example, hematuria combined with benign prostate enlargement in men was associated with lower risk. These findings illustrate how the model integrates multiple factors rather than relying on individual warning signs in isolation.

The authors emphasize that further validation across different health care systems will be needed because all study data were drawn from the SAIL database in Wales. The findings, published in IEEE Transactions on Biomedical Engineering, suggest that incorporating AI-based risk assessment into routine clinical data could help identify bladder cancer earlier, improve referral prioritization, and reduce unnecessary invasive procedures.

“This work represents a paradigm shift toward precision screening for bladder cancer. By harnessing the power of interpretable machine learning and comprehensive health records, we’re moving closer to detecting this disease at its earliest, most treatable stages,” said Professor Shang-Ming Zhou at the University of Plymouth’s Center for Health Technology.

“It’s important to emphasize that further validation across different health care systems is essential before anything is rolled out more widely—for example, all of the data analyzed came from the SAIL database in Wales, so we’d want to investigate other data sets, too. But it’s a very exciting early study. Additional studies would also be needed to confirm the causal relationships suggested by the model’s predictions,” added Prof. Zhou.

Related Links
University of Plymouth’s Center for Health Technology