AI Model Delphi-2M Predicts Individual Disease Risks Over Time
A European research team at the European Molecular Biology Laboratory (EMBL) has developed an AI model called Delphi-2M, which predicts individual disease risks over time. Published in Nature in 2025, the study highlights the model's potential and limitations.
Delphi-2M, a generative transformer model, learns from health records to estimate risks for over 1,000 diseases. It processes events along a timeline, predicting not just the 'what' (next diagnosis), but also the 'when' (time until then).
The model, trained on nearly 400,000 UK Biobank participants and tested on 1.93 million Danes, showed robust predictions for diseases with clear patterns like cardiovascular, diabetes, and sepsis. However, it's less suitable for rare congenital diseases or diagnoses heavily influenced by external factors.
Delphi-2M uses neutral placeholders to bridge gaps in medical history and updates predictions with new information. Internal tests showed it recognized patterns for almost all diseases significantly better than random, with predictions remaining useful but less accurate after ten years. External tests on Danish data were only slightly less accurate than internal tests, indicating potential for wider use.
While Delphi-2M offers promising disease prediction capabilities, researchers acknowledge its limitations, including demographic biases in the UK Biobank data and the need to avoid causal interpretations. They see potential in care planning by using aggregated predictions to estimate expected cases in regions and age groups.
Read also:
- Capella Nursing Students Gear Up for Crucial FPX 4050 Assessments
- Comprehensive Overview of Addressing Traumatic Brain Injuries (TBIs)
- Enhanced Health Services Provisioned by San Diego Academic Health Partnership Continues During COVID-19 and Beyond
- Vaccination drive targeting infants under 6 months old against bronchiolitis in the region of Andalucia
