AI Model Predicts Bladder Cancer Up to 5 Years Ahead with 85% Accuracy

An artificial intelligence system developed by researchers at the University of Plymouth has flagged patterns in electronic health records that precede bladder cancer diagnosis by as many as five years. The model, named PRECISE-AGZ, achieved an 85% detection rate among confirmed cases and correctly classified 91% of patients without the disease.

How the Model Analyzed Decades of Health Data

The research team, led by Professor Shang-Ming Zhou of the University’s Centre for Health Technology, examined nearly 70,000 patient records spanning 1995 to 2020. From an initial pool of 48,261 possible indicators—including smoking status, exercise habits, prescription histories, and clinical activity—the model distilled 38 features most strongly associated with subsequent bladder cancer diagnosis.

These features included both well-established risk markers—such as hematuria and tobacco use—and less obvious associations uncovered only through large-scale pattern recognition across longitudinal records.

Why Hematuria Alone Falls Short

Current NHS referral guidelines rely heavily on visible blood in the urine (hematuria) as the primary trigger for further investigation. Yet because hematuria occurs in numerous noncancerous conditions—including kidney stones and benign prostate enlargement—this approach leads to both unnecessary invasive testing and missed diagnoses.

A definitive diagnosis typically requires cystoscopy: insertion of a thin, camera-equipped tube through the urethra into the bladder. More accurate pre-referral risk stratification could help clinicians prioritize who needs this procedure most urgently—and who may be safely monitored.

Unexpected Associations in the Data

Precise-AGZ confirmed known risk factors but also revealed novel statistical associations not captured by standard protocols. For instance, patients diagnosed with Parkinson’s disease or dementia showed a lower likelihood of later developing bladder cancer. Conversely, long-term use of tamoxifen—a medication prescribed for breast cancer—correlated with elevated risk.

The study explicitly notes these associations do not imply causation. Rather, they point to potential biological links meriting future investigation. Similarly, hematuria carried different implications depending on context: among men with benign prostate enlargement, its presence was linked to reduced cancer probability.

Three-Tier Risk Stratification Framework

PRECISE-AGZ classifies patients into three risk categories: low (below 7% probability), uncertain (7–55%), and high (above 55%). This tiered output is designed to support clinical decision-making—directing urgent diagnostic resources toward those most likely to have bladder cancer while enabling closer observation rather than immediate intervention for individuals in the uncertain range.

Such stratification could reduce the number of avoidable cystoscopies, though the researchers stress that further validation is required before integration into routine care.

Clinical and Technical Limitations

“Bladder cancer ranks as the ninth most common cancer worldwide, with 614,000 new cases diagnosed in 2022. Despite its prevalence, no routine screening program exists for the general population, and—following symptoms—detection relies heavily on invasive cystoscopy procedures,” said PhD student Xu Wang, who led the data analysis.

Dr. Helen Winter, clinical director for the Somerset, Wiltshire, Avon, and Gloucestershire (SWAG) Cancer Alliance, described the work as a “paradigm shift toward precision screening.” She emphasized the potential to improve survival outcomes and quality of life by catching the disease earlier.

Professor Zhou added that all training and validation data came exclusively from the Secure Anonymised Information Linkage (SAIL) Databank in Wales. He underscored the necessity of replicating results across additional health systems before broader deployment—and cautioned that confirming causal relationships suggested by the model will require dedicated follow-up studies.

Reference: “Early Detection of Bladder Cancer Using Advanced Feature Engineering and Swarm Intelligence Optimization on EHRs” by Xu Wang, Andrea Preston, Jonathan Aning, Michael Loizou and Shang-Ming Zhou, 26 January 2026, IEEE Transactions on Biomedical Engineering. DOI: 10.1109/TBME.2026.3658230