Generative AI ‘Delphi-2M’ Forecasts 1,200 Diseases a Decade Ahead

A team of researchers has unveiled Delphi-2M, a groundbreaking generative transformer model capable of forecasting the risk of more than 1,200 diseases up to 10 years in advance. The system marks a major leap in predictive healthcare, combining large-scale biomedical data with generative AI to model the entire ICD-10 disease spectrum — something no single model has achieved before.

Delphi-2M was trained on millions of health records from the UK Biobank and further validated on Denmark’s national patient registry, ensuring robustness across diverse populations and healthcare systems. Using multi-modal datasets that include demographics, lab results, medical history, and prescription data, the model generates individualized forecasts of disease onset and progression years before symptoms appear.

Unlike traditional statistical tools, Delphi-2M leverages transformer-based attention mechanisms to identify which clinical factors most strongly influence each prediction. The model’s attention maps highlight the specific medical features — such as lab values, prior diagnoses, or family history — that contribute most to each forecast, offering doctors an interpretable view into the “reasoning” behind its outputs.

The implications are vast. By anticipating disease risk years in advance, healthcare providers could intervene earlier, personalize treatments, and prioritize high-risk patients for screenings or preventive therapies. On a larger scale, Delphi-2M could enable population-level forecasting for health systems, helping policymakers allocate resources more efficiently and plan for long-term public-health challenges.

Researchers emphasize that Delphi-2M is not intended to replace clinicians, but to serve as a decision-support tool that enhances preventive care. The model’s developers plan to expand its training data to include additional regions and imaging modalities, further refining its accuracy and global applicability.

With the launch of Delphi-2M, generative AI is moving beyond language and image generation — into the future of predictive medicine, where algorithms might detect tomorrow’s illnesses long before they arrive.