Updated
What if an AI could predict the likelihood of developing diabetes, Alzheimer’s, or cardiovascular disease up to two decades in advance? This is the promise of Delphi-2M , an artificial intelligence model capable of forecasting the probability of over a thousand diseases based on an individual’s health history.
The work, published in Nature, takes a significant step beyond current tools that typically focus on a handful of specific diseases. Delphi-2M not only calculates immediate risks but also simulates health trajectories over a span of 20 years, paving the way for a new paradigm in preventive medicine. Specifically, this tool can predict the probability of developing over 1,000 diseases .
The system was trained using clinical records from 400,000 individuals in the UK (sourced from the UK Biobank ) and validated with data from nearly two million citizens in Denmark. According to the authors, this is the first time a predictive model at this scale has been developed, capable of simultaneously managing hundreds of pathologies and dynamically projecting them over time .
The utility of such a tool is easy to envision: identifying a patient at high risk of heart attack before symptoms manifest, tailoring cancer screenings, or proactively planning the follow-up for neurodegenerative diseases.
To learn more
A Brain-Computer System Quadruples Its Capacity Thanks to AI
“This research appears to be a significant step toward a scalable, interpretable, and, more importantly, ethically responsible form of predictive modeling in medicine,” says Gustavo Sudre , a professor of Genomic Neuroimaging and Artificial Intelligence at King’s College London, as reported by Science Media Center (SMC).
“The clear demonstration of how explainable AI can be utilized for predictive modeling is crucial if this technology is to be applied in clinical practice and suggests that it might be possible to identify high-risk individuals requiring intervention ,” the expert continues.
Moreover, Sudre emphasizes that Delphi-2M is not limited to current data alone. “It is encouraging to see that the model’s architecture has been deliberately designed to accommodate richer data types, such as biomarkers , medical imaging , or even genomics . With these future integrations, the platform is well positioned to evolve into a truly multimodal precision medicine tool,” he adds.
How far can “AI Doctor” go?
The potential of Delphi-2M is enormous, but it is not without limitations. Like any AI system trained on population data, predictions may be influenced by biases related to age, ethnicity, or the socioeconomic background of the patients included in the records.
“The authors have developed an AI model capable of predicting diseases with accuracy and demonstrated that it works with data from the UK Biobank, as well as with data from nearly two million people in Denmark,” acknowledges Peter Bannister , a member of the Institution of Engineering and Technology. However, he warns: “There is still much work to be done to improve healthcare, as the authors themselves acknowledge that both datasets exhibit age, ethnic, and current health outcome biases.”
For Bannister, the immediate challenge is not just refining algorithms but ensuring that there is the necessary digital infrastructure and training for these technologies to reach everyone. “The challenge is to ensure that they are provided equitably , regardless of socioeconomic background , and that they do not exacerbate inequalities in treatment access,” he details.
The research presents a profound shift in understanding medicine: moving from a reactive practice—centered on treating diseases once diagnosed—to a predictive and preventive model, where the information of each patient allows for timely intervention and action.
Does this mean one can self-diagnose by consulting a chat interface? Absolutely not. The authors themselves acknowledge that Delphi-2M does not replace doctors , but it can serve as an invaluable complement to health planning and clinical decision-making. It remains to be seen how it will be implemented in practice and under what ethical and legal conditions, but the path toward more anticipatory medicine appears increasingly tangible.