10 months ago
AI Model Predicts Future Diseases Based on Patient History
Scientists have created an AI model that can predict which diseases a person might get in the future by looking at their past health information.
It's like having a crystal ball for your health!
The model is really good at predicting what might happen in the next few years, but it's not perfect.
It's still being tested and improved, so it won't be used in hospitals yet.
But even now, it's helping scientists learn more about how diseases are connected.
AI model Delphi-2M predicts future diseases based on patient history
Model can forecast over 1,000 conditions, including Alzheimer's, cancer, and heart attacks
Trained on UK Biobank data and validated on Danish health records
Performs at an average AUC of 0.76 for predictions within five years
Real-world applications still far off, pending clinical trials and data updates
- Who
- Researchers at EMBL in Cambridge and the German Cancer Research Centre in Heidelberg
- What
- Developed an AI model named Delphi-2M to predict future diseases
- Where
- Trained on UK Biobank data and validated on Danish health records
- When
- Details published in Nature on September 17th
- Why
- To help doctors predict high-risk patients and allocate healthcare budgets
Optimists
Pessimists
Potential Impact
Optimists
Could revolutionize healthcare by predicting high-risk patients
Pessimists
May not be accurate enough for real-world use, causing false alarms or missed diagnoses
Key facts
- Model Name
- Delphi-2M
- Conditions Predicted
- Over 1,000, including Alzheimer's, cancer, and heart attacks
- Training Data
- UK Biobank (400,000 people) and Danish health records (1.9 million people)
- Average AUC Score (5-year predictions)
- 0.76 (UK) and 0.67 (Denmark)
- Next Steps
- Clinical trials and data updates for real-world applications
Quotes
Ewan Birney, a geneticist at EMBL
A geneticist at the European Molecular Biology Laboratory
“I’m like a kid in a candy shop.”
livemint.com





