AI Models Use Routine Blood Tests and Health Records to Predict Cancer Risk

Routine health data, including blood tests and clinical history, may provide additional signals for personalized risk stratification. New findings show that machine learning models can use these data to identify people at higher risk of several common cancers.

Researchers from the Else Kröner Fresenius Center (EKFZ) for Digital Health at TU Dresden and collaborating institutions developed machine learning models to predict risk across 20 common cancers. The work also included PRE-Screen, a toolbox designed to allow artificial intelligence agents based on large language models to run the risk models consistently. The models incorporated a broad range of routine and advanced health information, including medical history, lifestyle factors, blood test results, genetic data, and metabolomic data.

The international study analyzed data from 463,620 UK Biobank participants who were followed for approximately 18 years, during which 57,055 developed cancer. The models were also evaluated in participants from Scotland and Wales, 42,036 people enrolled in the U.S. All of Us Research Program, and more than 3 million people from Japan’s JMDC database. To reduce the potential influence of cancers that were already present but undiagnosed, cases diagnosed within the first year were excluded from the relevant prediction models.

Overall, the machine learning models outperformed traditional statistical approaches, although predictive performance varied by cancer type. For a combined outcome covering cancers that could be predicted from routine clinical data, the models achieved an area under the receiver operating characteristic curve of 0.83 at 15 years.

Performance was particularly strong for lung, esophageal, and liver cancers but more limited for ovarian, breast, and cervical cancers. For acute myeloid leukemia and multiple myeloma, predictive performance declined over longer time periods, suggesting that some stronger short-term predictions may reflect early signs of cancer that was already present but had not yet been diagnosed.

Much of the predictive information came from routine blood tests and clinical history. Blood tests continued to add predictive value for cancers diagnosed more than five years later, while genetic and metabolomic data provided only a small further improvement. The findings were presented at ESMO Congress 2026 under the abstract “Development and validation of a machine-learning-toolbox for autonomous multicancer risk stratification from routine clinical data.” Because the study analyzed existing data retrospectively, further research is needed to establish whether the approach improves earlier diagnosis or outcomes.

“For selected cancers, these models could serve as a low-cost pre-screening step using data already available in routine care. Existing screening programmes cover only certain cancers and often use broad eligibility criteria, rather than assessing each person’s individual risk,” said lead author Dr. Paul-Henry Koop, Else Kröner Fresenius Center for Digital Health, TU Dresden, Germany. 

“Cancer types differ substantially in their carcinogenic mechanisms and in how strongly their risk factors and precursor conditions are captured in routine data. A positive flag should therefore trigger minimally invasive testing first,” added Dr. Koop.

Related Links
EKFZ for Digital Health
ESMO