Named Delphi-2M, the model marks a major advance over earlier AI-based risk predictors, which typically focus on a single illness, said study co-author Moritz Gerstung, a data scientist at the German Cancer Research Center in Heidelberg. He noted that Delphi-2M could help doctors identify high-risk patients early, opening opportunities for preventive measures long before symptoms appear.
In the study, researchers adapted a large language model to forecast the likelihood of 1,258 diseases based on a person's medical history, combined with factors such as age, sex, body mass index, and lifestyle habits like tobacco and alcohol use. The tool was trained on health data from 400,000 participants in the UK Biobank, which tracks long-term biomedical information.
Results showed that Delphi-2M matched or exceeded the accuracy of conventional single-disease models and outperformed algorithms relying on biomarker data to estimate risk across multiple conditions. The AI excelled in forecasting illnesses with predictable progression, including certain cancers, and could calculate disease probabilities decades in advance depending on medical records.
To test its broader applicability, the team validated Delphi-2M using health data from 1.9 million people in the Danish National Patient Registry, a database of nearly 50 years of hospital records. Predictions for the Danish cohort proved only marginally less accurate than those for UK Biobank participants.
Despite its promise, the researchers acknowledged limitations. The UK Biobank, for example, records only a participant's first incidence of disease, which could affect long-term prediction accuracy. The team plans to expand testing with datasets from multiple countries to improve the model's precision and global relevance.
Xinhua
Source : https://en.igihe.com/science-technology/article/new-ai-tool-predicts-disease-risks-decades-ahead
