Scientists have unveiled an artificial intelligence model capable of predicting the likelihood of developing more than 1,000 diseases years before diagnosis, in what they describe as a major step toward preventative healthcare.
The model, called Delphi-2M, is built on the same transformer technology that powers consumer chatbots like ChatGPT. Instead of processing language, it analyses patient health records to identify patterns in the progression of illnesses, researchers explained in a paper published Wednesday in Nature.
Half a million participants
The project brought together experts from institutions in the UK, Denmark, Germany, and Switzerland. They trained the system using data from the UK Biobank, a biomedical database covering half a million participants, and later validated its performance against Denmark’s public health records, which hold information on nearly two million people.
“Understanding a sequence of medical diagnoses is a bit like learning grammar in a text,” said Moritz Gerstung of the German Cancer Research Center. “Delphi-2M learns how diseases emerge in succession, enabling very meaningful and health-relevant predictions.”
Charts presented by Gerstung suggested the AI could identify individuals at much higher or lower risk of heart attacks than traditional models based solely on age and other standard factors. Unlike existing tools such as Britain’s QRISK3 – which focuses on heart attack and stroke – Delphi-2M can estimate risks across a vast range of conditions simultaneously.
System remains experimental
However, researchers cautioned that the system remains experimental. They noted that the datasets used to train it are skewed toward certain age groups and ethnicities, limiting how broadly the results can be applied.
“This is still a long way from improved healthcare,” said Peter Bannister, a fellow at the Institution of Engineering and Technology.
Even so, the team sees significant potential. In the future, Delphi-2M could guide monitoring, enable earlier interventions, and help optimise strained healthcare resources, said co-author Tom Fitzgerald of the European Molecular Biology Laboratory.
Ewan Birney, another co-author, described the tool as “a step toward preventative medicine at scale,” while King’s College London professor Gustavo Sudre called it “a significant advance towards interpretable and ethically responsible predictive modelling.”
