AI in medicine, explained for everyoneGeneral information only, not medical advice · Our standards
How Far

Prediction and risk · Prediction and risk

Delphi-2M: AI that forecasts risk of 1,000+ diseases

How it was checked
AI-assisted source check. Not clinically reviewed. ()
Last updated
Next review
Page ID
TR-007

Our editorial standards · 3 references · Corrections

General information only, not medical advice. Talk to your doctor about your situation. Read the disclaimer.

Delphi-2M, a GPT-style model published in Nature in September 2025, learned from health records of about 400,000 UK Biobank participants to predict rates of more than 1,000 diseases. It held up on 1.9 million Danish records. It's early research and not ready for clinical use.

How far away is this?

It's a computer study on past records. No trial has tested whether using it helps patients. UK Biobank volunteers tend to be healthier and less ethnically diverse than the general population, so predictions may be less reliable for other groups.

Where is it approved?

RegionRegulatorStatusNotes
United StatesFDANot approvedResearch model; not a medical device.
European UnionCE mark / EMANot approvedResearch model; not a medical device.
United KingdomMHRANot approvedResearch model; not a medical device.
AustraliaTGANot approvedResearch model; not a medical device.

A research model. It isn't used in care anywhere that we know of. "Unknown" means we haven't verified it, not that it's unavailable. Approval in one country isn't approval in another.

What is it?

Like a chatbot predicting the next word, Delphi-2M predicts the next likely diagnosis in a person's health history, based on past diagnoses and factors like sex, body mass and smoking.

What's the evidence?

  1. Trained on data from 0.4 million UK Biobank participants and validated on 1.9 million Danish individuals without changing the model.[1]

    Evidence stage: Early research · #c1

  2. Its accuracy was comparable to existing single-disease models. The authors also note it learned biases from its training data.[1]

    Evidence stage: Early research · #c2

  3. Commentators note it isn't ready for clinical decisions and needs further validation.[3]

    Evidence stage: Early research · #c3

What does it mean for me?

There's no app or test to ask for. The ordinary things your GP already checks (blood pressure, family history, smoking) remain the basis of risk assessment.

What could I ask my doctor?

  • What are my main health risks, and what can I do about them?
  • Which risk checks are recommended for someone my age?

These are general starting points, not a recommendation to choose any particular test or treatment.

Change history

  • : First published.
  • : Updated after independent fact-check and clinical review
  • : Review label corrected: this page was not clinically reviewed. Earlier labels saying 'independent fact-check' and 'clinical review' were wrong; checks so far were AI-assisted source checks only.

General information only, not medical advice. Talk to your doctor about your situation. AI-assisted source check. Not clinically reviewed.

References

  1. Shmatko et al., Nature (2025). nature.com (2025). Peer-reviewed study. https://www.nature.com/articles/s41586-025-09529-3
  2. Open-access version (PMC12589094). pmc.ncbi.nlm.nih.gov. Peer-reviewed study. https://pmc.ncbi.nlm.nih.gov/articles/PMC12589094/
  3. VVSOR blog critique. blog.vvsor.nl. Expert commentary. https://blog.vvsor.nl/2025/10/can-we-really-learn-the-natural-history-of-human-disease/

Publisher shown as the website hosting each source. Dates are given where the source label includes them.