A groundbreaking artificial intelligence model, dubbed “Delphi-2M” has been developed by European researchers with the ability to forecast an individual’s susceptibility to over a thousand diseases years before they manifest. The technology, which functions much like a weather forecast for human health, analyzes anonymous medical records to anticipate long-term health trajectories and provide a probability of future diagnoses.
The Technology Behind the Forecast
Developed by a team from the European Molecular Biology Laboratory, the German Cancer Research Center, and the University of Copenhagen, Delphi-2M uses a similar architecture to large language models like ChatGPT. However, instead of learning the patterns of language, it is trained to understand the “grammar” of healthcare data, including hospital admissions, GP records, and lifestyle factors. By processing this sequence of events, the model can generate a probabilistic risk assessment for 1,231 diseases, rather than a definitive diagnosis or a specific date.
According to Prof. Ewan Birney, the interim executive director of the European Molecular Biology Laboratory, this is a significant leap forward. “Just like weather, where we could have a 70% chance of rain, we can do that for healthcare,” he stated. “And we can do that not just for one disease, but all diseases at the same time—we’ve never been able to do that before.”
Data and Validation
The model was initially trained on a large dataset of over 400,000 people from the UK Biobank research project, which tracks long-term biomedical information. To test its accuracy and broader applicability, the researchers then validated it using the medical records of 1.9 million people in Denmark. The results showed that Delphi-2M performed exceptionally well in predicting diseases with clear progression patterns, such as Type 2 diabetes, heart attacks, and sepsis. The model’s accuracy was comparable to or even better than existing single-disease prediction tools.
A Tool for Prevention and Planning
The researchers envision several key applications for this technology. Clinicians could use Delphi-2M to identify high-risk patients years in advance, allowing for early, targeted interventions that could prevent the onset of a disease. For instance, a patient identified with a high risk of liver disease could be given proactive lifestyle advice.
Beyond individual care, the model holds immense potential for large-scale healthcare planning. Hospitals and health systems could use the tool to forecast future demand for specific services, such as the number of heart attack cases in a particular region in 2030, to better allocate resources and prepare for future needs.
While still an experimental tool that requires further testing before clinical use, the successful validation in two separate national datasets marks a major step toward a new era of preventative and personalized medicine.
Source: Nature
