Integrated image-based deep learning and language models for primary diabetes care
Authors' conclusion
Given its multifaceted performance, DeepDR-LLM holds promise as a digital solution for enhancing primary diabetes care and DR screening.
Study context
- Population
- Patients in primary diabetes care and diabetic retinopathy (DR) screening; single-centre real-world prospective deployment
- Participants
- 372 vs 397
- Exposure
- PCP+DeepDR-LLM vs unassisted PCP
- PCP+DeepDR-LLMn = 372
- Unassisted PCPn = 397reference
- Outcome
- Diabetes management adherence
Methodological detailssingle-center
- Single-center
Main finding
As reported in the abstract
throughout follow-up
- PCP+DeepDR-LLM vs Unassisted PCP: P < 0.05
- PCP+DeepDR-LLM vs Unassisted PCP: P < 0.01
Observational data: associations, not effects.
Further findings
Referable DR identification accuracy, average PCP (retrospective)
- Unassisted
- 81.0%
- Assisted by DeepDR-Transformer
- 92.3%
LLM module performance (retrospective)
- English: comparable to primary care physicians (PCPs) and endocrinology residents
- Chinese: outperformed PCPs; comparable to endocrinology residents
Quality and empathy of management recommendations
- DeepDR-LLM deployment improved quality and empathy level of recommendations
Original publication
Integrated image-based deep learning and language models for primary diabetes care. Nat Med. Published 19 July 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →