Researchers across Europe have successfully developed an advanced artificial intelligence system designed to predict the onset of various diseases many years before they manifest. This significant breakthrough, which leverages an individual`s comprehensive medical history, was detailed in a study recently published in the esteemed scientific journal Nature.
The published research explains the methodology behind modifying the architecture of a Generative Pre-trained Transformer (GPT) model specifically for the purpose of simulating human disease progression. This innovative AI model, designated as Delphi-2M, is capable of forecasting the probability of over a thousand different diseases. Its predictive accuracy, based on each person`s unique health records, is on par with existing specialized models that typically focus on a single disease.
A distinctive attribute of Delphi-2M is its generative capacity, enabling it to create hypothetical future health scenarios. This provides an invaluable, informative assessment of an individual`s potential disease burden over an extended timeframe, specifically up to 20 years into the future. Such a capability holds immense promise for personalized preventative medicine and public health planning.
For its training, the scientists employed extensive data sources. These included information from 400,000 participants in the UK Biobank project—a large-scale repository containing biological samples and detailed health data from approximately half a million individuals. Additionally, the AI model integrated health data from an impressive 1.9 million Danish citizens.
The research team concluded that GPT-like models are exceptionally well-suited for both predictive and generative tasks within the healthcare domain. They highlight the potential for these models to be widely applied to national-scale health databases, suggesting a future where data-driven insights could revolutionize healthcare delivery and personalized treatment strategies.
Scientists from Germany, the United Kingdom, Denmark, and Switzerland collaboratively participated in this pioneering research effort.
Rupert Blackwood 
Clement Ashworth