Nature Podcast This AI tool predicts your risk of 1,000 diseases — by looking at your medical records
Sep 17, 2025
Moritz Gerstung, a computational biologist at the German Cancer Research Centre, discusses the groundbreaking Delphi2M AI tool that predicts an individual's risk of over 1,000 diseases using their health records. The system, trained on extensive data from 400,000 individuals, aims to streamline healthcare by forecasting disease progression. Gerstung also highlights concerns about biases in the AI’s training data and its potential transformative role in population health management. Additionally, the podcast explores intriguing research on AI's influence on unethical behavior in task delegation.
AI Snips
Chapters
Transcript
Episode notes
Biases From Training Cohort
- Training data biases limit model generalisability; UK Biobank skews toward affluent, white British participants.
- The cohort's entry age also hides early-life mortality, creating unrealistic assumptions about deaths before 40.
Broaden Data Before Clinical Use
- Validate Delphi2M on broader, more diverse datasets before clinical use.
- Add richer inputs like surgeries and prescriptions to improve prediction accuracy.
Tool For Population-Level Planning
- Delphi2M could forecast regional disease burden to inform healthcare planning.
- Predictions can guide decisions on hospital capacity and specialist services needed.