New AI Tool Predicts Risk of over 1,000 Diseases Years in Advance

The model, called Delphi-2M, draws on medical records and lifestyle factors to estimate the likelihood of illnesses such as cancer, skin diseases, and immune conditions up to 20 years in advance.

A new artificial intelligence (AI) tool can forecast a person’s risk of developing more than 1,000 diseases, in some cases decades before symptoms appear.

The model, called Delphi-2M, draws on medical records and lifestyle factors to estimate the likelihood of illnesses such as cancer, skin diseases, and immune conditions up to 20 years in advance.

Researchers say its multi-disease modelling capability could help clinicians identify high-risk patients early and introduce preventive care. The findings are published in Nature.

Unlike most AI-based prediction tools that focus on one illness, Delphi-2M can generate health trajectories across 1,258 conditions in a single run.

“A health-care professional would have to run dozens of them to deliver a comprehensive answer,” said study co-author Moritz Gerstung, a data scientist at the German Cancer Research Center in Heidelberg.

The tool is built on a modified large language model (LLM), similar to those underpinning AI chatbots. It incorporates details such as age, sex, body mass index, and habits like tobacco and alcohol use, in addition to past medical history.

“Delphi-2M’s generative nature also enables sampling of synthetic future health trajectories, providing meaningful estimates of potential disease burden for up to 20 years,” the researchers said.

Prof Moritz Gerstung, head of the division of AI in oncology at the German Cancer Research Centre, added: “This is the beginning of a new way to understand human health and disease progression. Generative models such as ours could one day help personalise care and anticipate healthcare needs at scale.”

Computer scientist Stefan Feuerriegel of Ludwig Maximilian University of Munich, who develops AI for medical applications, described the tool’s capabilities as “astonishing.” “It can generate entire future health trajectories,” he said.

The tool showed the strongest performance in diseases with predictable patterns, such as some cancers. Tested on 400,000 participants from the UK Biobank, Delphi-2M outperformed existing models that predict individual diseases. It also exceeded the accuracy of algorithms relying on biomarkers to assess multiple conditions. “It worked astonishingly well,” said Gerstung.

Researchers further validated the tool using health data from 1.9 million people in the Danish National Patient Registry, which has tracked hospital admissions for nearly 50 years. Predictions were only slightly less accurate than those on UK data, indicating that the model could be applied across national health systems.

Experts caution that limitations remain. Degui Zhi, a bioinformatics researcher at the University of Texas Health Science Center, noted that UK Biobank data used for training only captured participants’ first recorded encounter with a disease. “The number of times someone has had an illness is important for the modelling of personal health trajectories,” he said.

The research team plans to test Delphi-2M on additional datasets from different countries to expand its accuracy and scope. “Thinking about how this information can be combined for developing even more precise algorithms will be important,” said Gerstung.

Stay tuned for more such updates on Digital Health News