Two advanced predictive machine learning algorithms developed by British researchers have significantly improved the ability of general practitioners to identify currently undiagnosed patients, such as difficult-to-detect liver and oral cancer. By incorporating routine blood test results with known risk factors, these models have made significant advances in early detection of cancer within primary care.
Researchers at Queen Mary University and Oxford University in London trained and validated two new algorithms using anonymous electronic health records of more than 7.4 million adults in the UK. Compared to existing tools such as QCancer, the new model showed greater sensitivity and the potential to transform clinical decision-making and early diagnosis.
Unlike previous models, the algorithm incorporated seven standard blood tests, including whole blood count and liver function biomarkers, along with patient age, family history, existing medical conditions, symptoms and general health. This approach has enabled more accurate risk estimates for 15 cancer types, including lung, colorectal, breast, prostate, ovaries, uterus, pancreas, testes, and gastroesophageal malignant tumors.
Additionally, the researchers have identified four new medical conditions associated with increased cancer risk, as well as two new family history associations: blood and lung cancer. Seven additional symptoms, including itching, bruises, ho, back pain, and dark urine, were found to be related to multiple cancers.
In particular, the algorithm was the first of their kind that could estimate the likelihood of liver cancer that has not been diagnosed in a primary care setting.
“These algorithms are designed to be incorporated into clinical systems and used during routine consultations. They represent a significant improvement on the current model, especially when identifying cancer at a more therapeutic stage.”
“By using existing data available in medical records, this approach is both cost-effective and scalable, helping the NHS achieve its goal of improving early cancer diagnosis by 2028,” she added.
These new tools demonstrate the distinct ability to identify individuals at the highest risk for 15 cancer types, using lifestyle factors that provide the possibility of previous diagnosis, including symptoms, blood outcomes, and some rare forms of cancer. ”
For more information, please see: 10.1038/s41467-025-57990-5
