Performance of Large Language Models for Structured Recognition and Refractive Prediction
For patients and families
In plain language
An automatic summary of structured registry data. It is an orientation aid, not a substitute for the official protocol or a physician assessment.
- What is being studied
- This is an observational study: the protocol does not assign a study treatment.
- Who it may be relevant to
- Registry conditions: Cataract. Basic parameters: from 18 years · All.
- What needs checking
- Age, condition and sex are only basic indicators. Prior treatment, laboratory values and other mandatory requirements appear in the eligibility criteria below.
- Where it takes place
- China
- Next step
- Save the trial, show it to the treating physician, and confirm current recruitment with the study center. Costs, documents and travel →
Unsure about the terms? Read our patient guide →
Official title
Head-to-Head Evaluation of ChatGPT 4o, GPT-5, and DeepSeek for Structured Extraction, Toric IOL Recommendation, and Refractive Prediction
Overview
We conducted a single-center, retrospective observational study to evaluate large language models (ChatGPT 4o, GPT-5, DeepSeek) for automated interpretation of de-identified IOLMaster 700 reports provided as raster images. Models produced structured biometric extraction, toric IOL recommendation, and refractive predictions (sphere, cylinder, axis). Primary outcomes included parameter-level agreement and refractive error metrics; secondary outcomes included decision-support performance for toric IOL selection and agreement on ordered T-codes. No clinical intervention was performed.
Detailed description
This study compares three large language models accessed in their native configurations, without fine-tuning or external tools. For each examination, the original IOLMaster 700 report image was supplied without manual annotation or pre-processing. A standardized instruction required: (i) structured extraction of AL, ACD, LT, WTW, K1/K2 and axes, ΔK, TK1/TK2 and axes, and ΔTK; (ii) binary toric candidacy and T-code according to institutional ALCON mapping; and (iii) refractive recommendations (sphere, cylinder, implantation axis). Each model generated three independent outputs per case. De-identification and IRB oversight (waiver of consent) were implemented according to institutional policy. The unit of enrollment is participants (n=54), with outcomes analyzed per eye (162 eyes) and per model generation where applicable.
Primary outcome measures
- Refractive prediction error for sphere [Time frame: At index examination]
- Cohen's kappa with 95% CIs between model [Time frame: At index examination (single time point)]
Secondary outcome measures (2)
- Cylinder prediction error [Time frame: At index examination]
- Axis prediction error [Time frame: At index examination]
Eligibility criteria
Inclusion criteria
-postoperative corrected distance visual acuity (CDVA) of 0.10 logMAR or better -an absolute IOL rotational stability of less than 10∘ at the 1-month follow-up examination
Exclusion criteria
- incomplete biometric data on the examination report;
- a history of previous ocular surgery or ocular trauma
- the occurrence of intraoperative complications, such as an anterior capsular tear or posterior capsular rupture
- the development of significant postoperative complications, including but not limited to severe intraocular infection or inadequate pupillary dilation.
Criteria are shown verbatim from the registry (in English). Final eligibility is always assessed by the study center.
Healthy volunteers: No
Study design
- Observational model
- Case-only
Study locations
China · 1 center
- Eye and ENT hospital of Fudan University — Shanghai
Identifiers
NCT: NCT07183891 · Totic-2025-001