A deep learning system for predicting time to progression of diabetic retinopathy
Authors' conclusion
Altogether, the DeepDR Plus system could predict individualized risk and time to DR progression over 5 years, potentially allowing personalized screening intervals.
Study context
- Population
- Participants with diabetes; pretraining 717,308 fundus images from 179,327 participants; training/validation multiethnic dataset of 118,868 images from 29,868 participants; real-world cohortstraining/validation dataset
- Participants
- 29,868 participants
- Outcome
- Prediction of time to DR progression
- Follow-up
- Within 5 years
Methodological detailsdeep learning, multiethnic dataset, real-world validation cohorts, Pretrained on 717,308 fundus images from 179,327 participants with diabetes
- Deep learning
- Multiethnic dataset
- Real-world validation cohorts
- Pretrained on 717,308 fundus images from 179,327 participants with diabetes
Performance
As reported in the abstract
all times up to 5 years
| Test or model | Concordance index | Integrated Brier score |
|---|---|---|
| DeepDR Plus | 0.754-0.846 | 0.153-0.241 |
DeepDR Plus
Concordance index0.754-0.846
Integrated Brier score0.153-0.241
Further findings
Real-world validation: integration with clinical workflow
- Mean screening interval potentially extended from 12 months to
- 31.97 months
- Participants recommended screening at 1-5 years, respectively
- 30.62%, 20.00%, 19.63%, 11.85% and 17.89%
- Delayed detection of progression to vision-threatening diabetic retinopathy (DR)
- 0.18%
- Real-world cohorts of participants with diabetes
Safety
- Delayed detection of progression to vision-threatening DR
- 0.18%
Original publication
A deep learning system for predicting time to progression of diabetic retinopathy. Nat Med. Published 4 January 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →