Tejal Somvanshi
Karmactive Team
A new artificial intelligence system can now forecast diseases up to 20 years before they occur, according to researchers from the University of Copenhagen and international partners. The system, named Delphi-2M, works like a weather forecast for human health.
The AI model can predict more than 1,000 common diseases by analyzing patterns in medical records and lifestyle factors. Published in the journal Nature, this research marks a significant step toward early disease detection and prevention.
“Today, we survive many diseases that would previously have been fatal. As we grow older, we face a future in which many people suffer from multiple conditions simultaneously. That’s why we need to understand how diseases interact,” says Søren Brunak, Professor at the University of Copenhagen’s Department of Public Health.
The model was trained using health data from 400,000 participants in the UK Biobank who agreed to share their information for research. Researchers then tested its accuracy using data from 1.9 million Danish patients, working within a secure computing environment under the Danish Health Data Authority.
Delphi-2M shows varying levels of accuracy depending on the condition. It performs better at predicting diseases with consistent patterns, such as certain cancers, heart attacks, and sepsis. Conditions like pregnancy complications and some mental health issues are more difficult to forecast due to their unpredictable nature, according to the researchers.
Similar Posts
The system functions by recognizing patterns in how health evolves over time. Similar to how ChatGPT predicts the next word in a sentence, this AI calculates the next most likely diagnosis among numerous diseases.
Professor Brunak explained that patients with multiple conditions are difficult to manage, raising questions about treatment priorities and where they should go in the healthcare system. He noted that multimorbidity is both costly and complex, which is why mapping the common pathways of disease progression is necessary.
While currently more useful for population-level planning than individual diagnoses, the researchers believe the model could eventually help doctors make more informed treatment decisions.
Professor Brunak indicated that the model aims to project disease trajectories so physicians know how aggressively to treat patients from the beginning. He gave the example of diabetes patients, where some might only need lifestyle changes initially, while others require immediate medication.
“The more we understand about disease progression, the better we can reduce unnecessary overtreatment,” Brunak adds.
Despite its promise, researchers emphasize that Delphi-2M remains a prototype. For the model to predict not just the next disease but subsequent ones, it would need training on larger datasets.
Professor Brunak stated that they wanted to explore whether developing a method that handles more than 1,000 diseases simultaneously was possible, and their study confirmed it is.
Professor Laust Mortensen from the University of Copenhagen and the Rockwool Foundation observed that their method has great potential, noting that despite being trained on British data, it demonstrated high accuracy when applied to predict disease in Denmark.
The project received partial funding from the Novo Nordisk Foundation and involved collaboration with researchers from several European institutions.
