Scientists announced on Wednesday, September 17, that they have developed an artificial intelligence model capable of predicting medical diagnoses years before they occur, using the same underlying technology that powers consumer chatbots like ChatGPT.
The new system, called Delphi-2M, can forecast “the rates of more than 1,000 diseases” well into the future by analysing a patient’s medical history, according to a paper published in the journal Nature by researchers from institutions in Britain, Denmark, Germany, and Switzerland.
The model was trained using data from the UK Biobank, a large-scale biomedical research project that contains detailed health and genetic information on about half a million participants.
Delphi-2M is based on neural networks using transformer architecture, the “T” in “ChatGPT”, which has been most prominently used in language-based tasks, including generative chatbots. Researchers said that deciphering medical records is not unlike learning the grammar of language.
“Understanding a sequence of medical diagnoses is a bit like learning the grammar in a text,” explained Moritz Gerstung, an AI expert at the German Cancer Research Center. Delphi-2M, he said, “learns the patterns in healthcare data, preceding diagnoses, in which combinations they occur and in which succession, enabling very meaningful and health-relevant predictions.”
Charts presented by Gerstung showed that the AI could identify individuals with a significantly higher or lower risk of experiencing a heart attack than would typically be predicted based solely on age or other conventional factors.
To verify the model’s accuracy, the team tested Delphi-2M against health data from nearly two million people contained in Denmark’s public health database. The results reinforced the system’s predictive capabilities.
However, the researchers cautioned that Delphi-2M is not yet ready for clinical use. “This is still a long way from improved healthcare,” said Gerstung, emphasising that the datasets used so far from Britain and Denmark are biased in terms of age, ethnicity, and health outcomes.
Peter Bannister, a health technology researcher and fellow at Britain’s Institution of Engineering and Technology, also warned that the limitations of the data need to be addressed. Still, he said the work represents progress in harnessing AI for preventative medicine.
In the future, Gerstung suggested, systems like Delphi-2M could guide patient monitoring and allow earlier interventions, effectively advancing preventative care. On a broader scale, co-author Tom Fitzgerald of the European Molecular Biology Laboratory said such tools could aid in “optimisation of resources across a stretched healthcare system.”
Doctors in many countries already rely on computer tools to predict disease risks, such as QRISK3, which helps UK general practitioners estimate the likelihood of heart attack or stroke. But co-author Ewan Birney said Delphi-2M is a significant leap forward because it “can do all diseases at once and over a long time period.”
Gustavo Sudre, a professor at King’s College London specialising in medical AI, described the work as “a significant step towards scalable, interpretable and, most importantly, ethically responsible predictive modelling.”
He noted that one of the major challenges in AI research is explainability, as the internal decision-making processes of large AI models often remain opaque even to their creators. The Delphi-2M project, he said, shows promise in addressing that concern while opening new possibilities for long-term healthcare innovation.
