Improving individual prevention

Liver cancer is often detected too late

Hepatocellular carcinoma (HCC), the most common form of liver cancer, is one of the deadliest cancers worldwide. It develops when liver cells change and multiply uncontrollably, often as a result of chronic liver damage. Early diagnosis significantly improves treatment outcomes, yet many patients are only diagnosed at an advanced stage. Current screening programs focus primarily on people who have already been diagnosed with liver cirrhosis, thereby overlooking other individuals at risk.

PRE-Screen-HCC: a reliable risk score for liver cancer based on routine data

For their study, the researchers analyzed health data from more than 900,000 individuals drawn from two large-scale studies: the UK Biobank for model development and the US All of Us Research Program for external validation. In total, nearly 1,000 confirmed HCC cases were included in the analysis. The model outperformed previously established HCC risk scores. Notably, the predictive performance based on routine data was comparable to that of models relying on complex genomic or metabolomic data. PRE-Screen-HCC classifies individuals into low-, medium-, and high-risk groups and could thus help target ultrasound screening more effectively.

“Our work shows how population data that has so far been underutilized can be used to improve prevention and early detection of liver cancer,” says Jan Clusmann, lead author of the study and researcher at the EKFZ for Digital Health at TUD and Uniklinik RWTH Aachen. “The key advantage is that our model is based on routine data that is already available in everyday clinical practice. This could allow us to identify people who would benefit from ultrasound-based early detection at an earlier stage,” adds the physician and scientist. His research project was funded by the German Cancer Aid as part of the Mildred Scheel Postdoctoral Program.

Publicly available AI model enables further development and broad application

The team led by Prof. Jakob N. Kather, Professor of Clinical Artificial Intelligence at TUD, and Prof. Carolin V. Schneider, Junior Professor of Prevention and Genetics of Metabolic Liver Diseases at RWTH Aachen University, places particular emphasis on the transparency of their results. To this end, the team also analyzed which types of routine data contribute particularly to risk prediction and how robust the results are across different population groups. All models and the underlying code, as well as a web-based risk calculator, have been made publicly available. This not only enables further external validation, for example in other population groups, but also future use and integration into agentic AI systems. In the long term, PRE-Screen-HCC could help to identify people at increased risk of liver cancer earlier and enable more targeted preventive screening.

“Our results demonstrate the potential of routine data for early detection when it is systematically analyzed at scale. In the future, algorithms such as the PRE-Screen-HCC we developed could be directly linked to patient records – for example, within the framework of the European Health Data Space,” explains Prof. Jakob N. Kather, Professor of Clinical AI at TUD and a specialist in internal medicine at the NCT/UCC of Dresden University Hospital Carl Gustav Carus.

“In clinical practice, we see that patients with a relevant risk of liver cancer cannot always be identified at an early stage. Our approach offers the opportunity to identify high-risk individuals earlier in the future and thereby improve their treatment prospects,” says Prof. Carolin V. Schneider, junior professor and physician at the Uniklinik RWTH Aachen and lead of the study.