aceComparative evaluation of large language models and clinicians in real-world glaucoma clinical reasoning
Purpose Clinical decision-making in glaucoma is complex and requires integration of heterogeneous information, including patient history, examination findings, and risk stratification. While artificial intelligence (AI) has shown strong performance in image-based ophthalmic tasks, its capability in specialty-specific clinical reasoning remains insufficiently explored. Methods Performance was evaluated by glaucoma specialists using a predefined rubric across three clinically oriented domains: medical accuracy (40…
In a 34-case glaucoma reasoning test, LLM systems produced structured reasoning with weighted scores overlapping attending ophthalmologists and often included safety-critical diagnostic and management elements.
LLM reasoning did not establish clinical equivalence in this limited evaluation and requires specialist oversight and further validation before clinical use.
Findings are based on a limited 34-case dataset and described as exploratory, not establishing clinical equivalence, with substantial variability among human clinicians.
Evidence
- Peer-reviewedGraefe's Archive for Clinical and Experimental Ophthalmology2026-08-06
How should this claim be treated?
Truvace Impact Record TRV-2026-0673, v1: “Comparative evaluation of large language models and clinicians in real-world glaucoma clinical reasoning.” Truvace, 2026-08-07. /record/TRV-2026-0673 (accessed at citation time). sha256 dd4bde2565eca7c3…
Calibration history
Every change to this record since certification, in the open. None yet — the reading has held since it entered the record.
Certified into the record
How to verify without trusting this page
Fetch the canonical text of any version from /api/record/TRV-2026-0673 and hash it yourself — for example shasum -a 256 on the saved canonical field. The result must equal content_hash, and each version’s text ends with prev:followed by the prior version’s hash (version 1 chains to 64 zeros). If a single character of any version had been altered since certification, the chain would not reproduce.