Bladder cancer claims 220,000 lives globally each year, with symptoms sometimes mistaken for other conditions such as kidney stones.
Now researchers have developed an artificial intelligence system that can identify bladder cancer patients up to five years before their official diagnosis, potentially revolutionising screening for the disease.
A new study led by the University of Plymouth developed an AI tool to analyse electronic health records to detect subtle warning signs of the disease – identifying early symptoms, risk factors and clinical activities that precede diagnosis.
Uncovering hidden patterns in health data
Current referral guidelines focus primarily on visible blood in urine (haematuria), but this symptom can also indicate benign conditions like kidney stones or prostate issues. The result is a poor detection rate for bladder cancer, and it can only be fully confirmed by having a cystoscopy procedure (where a long, thin tube with a small camera inside is moved up the urethra and into the bladder).
So the research team, led by
Professor Shang-Ming Zhou at the University of Plymouth's
Centre for Health Technology , analysed records from nearly 70,000 patients collected between 1995 and 2020 to see if there was a way of more accurately predicting the disease. Using their self-built model called PRECISE-AGZ, the researchers sifted through 48,261 potential health indicators (from smoking and exercise habits to medication use) to identify 38 key features that signal bladder cancer risk.
The screening model correctly detected bladder cancer in 85% of patients who had it; showed 91% accuracy in correctly identifying cancer-free individuals; demonstrated effective detection up to 12 months before diagnosis, with some signals appearing up to five years earlier; and was more likely to detect bladder cancer than current NHS referral guidelines.
As well as confirming known risk factors like smoking and blood in urine, the AI tool identified hidden patterns invisible to conventional screening methods. For example, patients with Parkinson's disease or dementia showed lower bladder cancer risk, and patients who used tamoxifen (a breast cancer medication) long term showed a higher bladder cancer risk. While the research does not show prevention or causation of bladder cancer in either case, it does suggest there may be shared biological pathways that warrant further investigation.
The study also suggested that the significance of blood in urine might vary depending on other health indicators. For instance, when combined with benign prostate enlargement in men, it actually indicated lower cancer risk; and clarifying such associations with further research could help to reduce unnecessary referrals.
‘Impressive accuracy’ and early detection
The system was also able to classify patients into three risk categories: low-risk (below 7% probability), uncertain (grey zone: 7-55%), and high-risk (above 55%). If rolled out following further investigation and development work, this stratification could enable healthcare providers to prioritise resources and reduce the need for invasive procedures, like cystoscopy, by monitoring patients within the grey zone before doing anything clinical.
Professor Shang-Ming Zhou
