Imagine if your doctor could look at your medical records and say, “You’ve got a 70% chance of a heart attack in the next 5 years—unless we do something about it.”
Thanks to a new artificial intelligence model, that future may not be far off. Scientists have developed a groundbreaking AI system that works like a weather forecast—but instead of predicting rain, it forecasts your future health risks, years in advance.
Meet Delphi-2M, an AI model trained on the medical records of millions, capable of calculating your likelihood of developing more than 1,200 diseases, from diabetes and heart failure to sepsis.
What Is Delphi-2M?
Developed by researchers at the European Molecular Biology Laboratory (EMBL), the AI model—called Delphi-2M—uses technology similar to popular AI chatbots like ChatGPT. However, instead of predicting text, this model forecasts disease risk by identifying patterns in anonymized medical data.
“It’s like saying there’s a 70% chance of rain—but for your health,” explained Prof. Ewan Birney, interim executive director at EMBL. “And we can do that not just for one disease, but for over a thousand at the same time. That’s never been possible before.”
How Does It Work?
Delphi-2M was trained using the anonymized medical records of more than 400,000 participants from the UK Biobank, which includes data such as:
- Hospital admissions
- GP visits
- Lifestyle factors (e.g., smoking, alcohol use)
The model was then tested on medical records from 1.9 million people in Denmark, where it showed strong accuracy. For instance, if it predicted a one-in-10 risk of disease within a year, that estimate matched real-world outcomes.
“It’s good—really good—in Denmark,” said Prof. Birney. “The predictions are holding up.”
What Can It Predict?
The model is most effective at forecasting diseases with clear progression, such as:
- Type 2 diabetes
- Heart attacks
- Sepsis
It’s less accurate for unpredictable conditions like infections. Instead of pinpointing exact dates, it gives probability-based forecasts, helping clinicians and policymakers make more informed decisions.
How Could It Be Used?
While not yet ready for clinical deployment, Delphi-2M could be used in the near future to:
- Identify high-risk patients early for preventive care
- Offer tailored lifestyle advice or treatments
- Guide disease screening programmes
- Forecast healthcare demand years in advance, such as estimating heart attack rates in a given city by 2030
For example, someone with a high predicted risk of liver disease might be advised to cut back on alcohol more aggressively than the general population.
“This is the beginning of a new way to understand human health and disease progression,” said Prof. Moritz Gerstung, head of AI in oncology at the German Cancer Research Centre (DKFZ). “Generative models like ours could one day personalize care and predict healthcare needs at scale.”
Key Takeaways:
- AI health forecasting works like a weather forecast, offering probabilities for disease risk.
- The Delphi-2M model can estimate the likelihood of over 1,200 conditions, based on medical and lifestyle data.
- It aims to prevent disease, guide resource planning, and eventually personalize patient care.
- Not yet in clinical use, but has shown strong performance in UK and Danish datasets.