1. Delphi-2M predicts over 1,000 diseases using medical records.
2. Achieved average AUC ~0.76, 0.97 for predicting death.
3. Trained on UK Biobank data from 403,000 participants.
4. Longitudinal validation maintained predictive accuracy.
5. Performance varies significantly across disease categories.
6. Generates synthetic health trajectories valid for 20 years.
7. Limitations include potential biases and demographic coverage.

Source: Nature