Foundation models
Med-BERT
Transformer models pretrained on sequences of structured electronic health record codes. They transfer to new prediction tasks with far fewer labeled examples than training from scratch, and we adapt them to your coding systems, cohorts and endpoints.
- Input
- Sequences of structured diagnosis and procedure codes
- Use
- Risk prediction and other downstream clinical tasks
- Adaptation
- Fine-tuning on your cohorts and endpoints
- Research
- Rasmy et al., npj Digital Medicine, 2021
- Validation support
- Model card, training record and evaluation report