Artificial intelligence developed to predict diseases years in advance

According to a study published in Nature, the AI, named Delphi-2M, was trained on the UK Biobank database, which contains biomedical data from half a million participants in the UK. The model uses a "transformer" architecture the same type underlying ChatGPT commonly applied in language-based tasks.

Moritz Gerstung from the German Cancer Research Center said, "Understanding medical diagnostic sequences is similar to learning grammar rules in text. Delphi-2M learns which combinations and sequences of diagnoses occur."

The model can assess individual risk levels for conditions such as heart attacks beyond just age and traditional risk factors. Researchers also validated it using Denmark's public health data for two million people.

Although the system shows promising results, scientists emphasize that it is not yet ready for clinical use. UK researcher Peter Bannister noted that the datasets used may be biased in terms of age, ethnicity, and existing health outcomes.

Nonetheless, experts believe that systems like Delphi-2M could transform healthcare by:

Allowing closer monitoring of patients,

Enabling earlier preventive measures,

Optimizing resource use in health systems,

Current software typically analyzes risk for a single condition, like heart attack or stroke. Delphi-2M, however, can predict long-term risk for over 1,000 diseases simultaneously.

Prof. Gustavo Sudre from King's College London described the work as "an important step toward developing a scalable, interpretable, and most importantly ethically responsible predictive model."