A new AI model, developed by researchers at the European Molecular Biology Laboratory (EMBL) and the German Cancer Research Center (DKFZ) in collaboration with the University of Copenhagen, makes it possible to predict individual health risks for over 1,000 diseases over a period of more than ten years. Trained with anonymized health data from the UK and Denmark, the model, published in the journal Nature, reveals new possibilities for personalized health strategies, but is not yet ready for clinical use.

The model is based on algorithms similar to those of large language models and was trained with data from 400,000 participants from the UK Biobank. It was then tested with data sets from 1.9 million people from the Danish Patient Registry. It analyzes the chronological sequence of medical events, such as diagnoses or lifestyle factors, and recognizes patterns to predict disease risks and their course over time. It is particularly reliable for diseases with clear courses such as diabetes, heart attacks or sepsis, while it loses accuracy for unpredictable or rare diseases.

The results show that the model provides probabilities for health events, similar to weather forecasts. For example, the risk of heart attack in men between the ages of 60 and 65 varies between 4 and 100 per 10,000 people per year, depending on their pre-existing conditions and lifestyle. Women show a lower risk, but a similar spread. Short-term predictions are more precise than long-term ones, and the calculated probabilities are consistent with the actual cases of illness in independent data sets.

Although the model cannot currently be used clinically, it offers valuable insights for research. It allows the study of disease progression, the analysis of lifestyle factors and the simulation of health outcomes. In the future, such models could help doctors identify high-risk patients at an early stage and support health systems in resource planning, especially in the face of an aging population and rising chronic diseases.

The development was carried out under strict ethical guidelines with anonymized data and secure data processing to protect the privacy of the participants. The work was funded by EMBL member states, the DKFZ and the Novo Nordisk Foundation. Further tests and regulatory frameworks are necessary before the model can be used in practice. The researchers see this as a significant step towards personalized and preventive health care.

Original Paper:

Learning the natural history of human disease with generative transformers | Nature

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability. ected.