New AI Model Predicts Risk for Over 1,000 Diseases, Possibly 20 Years in Advance
LONDON – A new artificial intelligence tool, dubbed Delphi-2M, is capable of predicting an individual’s risk of developing more than a thousand diseases, offering a significant leap forward in preventative healthcare. Developed by researchers, the system functions similarly to generative language models like ChatGPT, identifying patterns in medical data to forecast health outcomes.
Delphi-2M analyzes a patient’s medical history, factoring in age, gender, smoking habits, and obesity, to assess the likelihood of conditions such as cancer, diabetes, heart disease, and airway disorders. Unlike existing AI risk assessment tools focused on single conditions, Delphi-2M offers a broad, multi-disease prediction capability.
The AI was trained using anonymized data from 400,000 individuals in the United Kingdom and its reliability was subsequently tested against the medical data of nearly 2 million citizens in Denmark, with comparable results in both countries.
Researchers report Delphi-2M achieves an average accuracy score of 0.76, where 1.0 represents perfect prediction accuracy.Predictions extending beyond 10 years achieved an average score of 0.7.In some instances, the model can predict the onset of diseases up to 20 years in advance, especially for conditions with predictable progression like heart attacks and various cancers.
The potential applications extend beyond individual risk assessment. Scientists envision the technology informing lifestyle recommendations – such as weight loss or smoking cessation – and assisting hospitals in proactively allocating resources based on predicted patient needs.
Further testing is planned in diverse populations and countries before the system is expected to be available for use in a few years.