AI Predicts Bladder Cancer Years Before Diagnosis via Health Records

Artificial intelligence can spot warning signs of bladder cancer up to five years before a formal diagnosis by scanning routine electronic health records, ScienceDaily reported.

Researchers at the University of Plymouth developed the model to parse through complex medical histories and catch subtle patterns that standard screening methods miss.

Bladder cancer kills about 220,000 people globally each year, making early detection vital for survival. Current referral guidelines rely heavily on visible blood in the urine, a symptom that frequently overlaps with benign conditions like kidney stones and prostate enlargement. Confirming the disease typically requires cystoscopy, an invasive procedure using a lighted tube inserted through the urethra.

A team led by Professor Shang-Ming Zhou at the University’s Centre for Health Technology studied records from nearly 70,000 patients spanning a 25-year period between 1995 and 2020. The researchers created an AI model named PRECISE-AGZ, narrowing 48,261 health indicators down to 38 key features for predicting risk. The system performed better at detecting the disease than existing NHS referral guidelines.

Model identifies 85% of bladder cancer cases

The model correctly identified 85% of patients who developed bladder cancer and accurately classified 91% of healthy individuals, ScienceDaily reported. While it flagged established risks like smoking and hematuria, the technology also uncovered unexpected associations. Patients diagnosed with Parkinson’s disease or dementia appeared to have a lower risk of bladder cancer, while long-term use of tamoxifen linked to higher risk.

“This work represents a paradigm shift towards precision screening for bladder cancer,” said Shang-Ming Zhou, professor at the University of Plymouth. “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.”

Researchers emphasized that these unexpected links do not establish direct causation. Instead, the findings suggest shared biological mechanisms that warrant closer scientific examination. The AI also showed that the significance of blood in the urine depends heavily on a patient’s concurrent conditions, such as benign prostate enlargement in men.

Stratifying Patient Risk Groups

Beyond flagging early indicators, the AI system categorizes patients into three distinct risk groups based on probability. The tool designates individuals as low-risk when probability falls below 7%, uncertain in a grey zone between 7% and 55%, and high-risk when probability exceeds 55%. This stratification helps doctors prioritize urgent investigations while monitoring borderline cases.

“Bladder cancer ranks as the ninth most common cancer worldwide, with 614,000 new cases diagnosed in 2022,” said Xu Wang, a PhD student who led the data analysis. “Our system goes far beyond traditional symptoms, and it’s great to have uncovered the results that we have. The research shows that the combination of multiple factors – from medication histories to seemingly unrelated conditions – can reveal patterns invisible to conventional screening methods.”

Clinical leaders note that a reliable screening tool addresses a major unmet need in oncology. Catching the disease earlier expands treatment options and reduces the physical toll of invasive procedures.

“It could identify those most at risk more accurately, reduce harm from invasive procedures, and catch cancer earlier, when survival outcomes and quality of life can be dramatically improved,” said Dr. Helen Winter, Clinical Director for the Somerset, Wiltshire, Avon and Gloucestershire Cancer Alliance.

Validation Across Healthcare Systems

Despite promising performance metrics, the technology requires further testing before deployment in routine clinical settings. All patient data used in the initial study came from the SAIL database in Wales. Researchers must now determine whether the algorithm maintains its accuracy across different patient populations and varied electronic health record systems.

“It’s important to emphasize that further validation across different healthcare 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 datasets too,” Zhou said. Additional studies are also required to confirm the causal relationships suggested by the model’s predictions.

Frequently Asked Questions

What data sources did the AI model analyze?

The researchers used electronic medical records from nearly 70,000 patients covering the period from 1995 through 2020 via the SAIL database in Wales.

How early can the system detect bladder cancer risks?

The AI demonstrated effective detection as early as 12 months before an official diagnosis, with some subtle warning signals traceable up to five years prior.

What specific risk categories does the model use?

The system divides patients into three groups: low-risk for probabilities below 7%, an uncertain grey zone for 7% to 55%, and high-risk for probabilities above 55%.

Founded in 1862 as the Plymouth School of Navigation, the University of Plymouth traces its roots to maritime education before gaining university status in 1992. The institution hosts the Royal Statistical Society’s Centre for Statistical Education, selected in 2008, and runs specialised courses in marine engineering, marine biology, and maritime business alongside its health technology research.

Where was the study published?

The findings appear in the research paper titled “Early Detection of Bladder Cancer Using Advanced Feature Engineering and Swarm Intelligence Optimization on EHRs,” published in IEEE Transactions on Biomedical Engineering.