New AI Can Predict Risk of Over 1,000 Diseases, Years Before Symptoms Appear

In a landmark achievement for preventative medicine, European scientists have developed a powerful new artificial intelligence system that can analyze a person's health records and predict their individual risk for more than 1,000 different diseases, up to two decades before they might occur.

The new generative AI tool, named "Delphi-2M," represents a monumental leap beyond current risk-assessment models, which typically focus on a single disease like heart disease or cancer. This new system can simultaneously forecast the long-term risk for a vast spectrum of illnesses, from diabetes to blood poisoning, effectively creating a personalized "health forecast."


The breakthrough, detailed in the prestigious journal Nature, is the result of a collaboration between researchers at the European Molecular Biology Laboratory (EMBL), the German Cancer Research Centre (DKFZ), and the University of Copenhagen.

Learning the "Grammar" of Human Disease
Delphi-2M was built using a similar architecture to the large language models (LLMs) that power chatbots like ChatGPT. But instead of learning the patterns of language, it was trained to learn the "grammar" of human health.


Researchers fed the AI a massive, anonymized dataset of health records from over 2.3 million people from the UK and Denmark. The AI learned to identify the complex sequences and patterns of medical events—diagnoses, lifestyle factors like smoking, and the time between them—that often precede a specific disease.


"Medical events often follow predictable patterns," said Tom Fitzgerald, a scientist at EMBL's European Bioinformatics Institute. "Our AI model learns those patterns and can forecast future health outcomes."

A 'Health Forecast,' Not a Certainty
The researchers are careful to emphasize that the system provides probabilities, not certainties, much like a weather forecast. It might predict a person has a 70% chance of developing a certain condition in the next 10 years, allowing for early and personalized intervention.


In testing, Delphi-2M's accuracy was comparable to or even better than existing single-disease prediction models. Its ability to work across two completely different national healthcare systems (the UK and Denmark) without significant changes suggests it has captured fundamental patterns of human disease progression.


While the creators stress that Delphi-2M is not yet ready for clinical use, they believe it is a huge step towards a new era of proactive, preventative medicine.

"This is the beginning of a new way to understand human health and disease progression," said Professor Moritz Gerstung of the German Cancer Research Centre. "Generative models such as ours could one day help personalize care and anticipate healthcare needs at scale."

The ultimate vision is a future where a doctor can use such a tool to give a patient a clear, personalized roadmap of their future health risks and provide highly targeted advice—whether it's lifestyle changes or early screening—to change that future for the better