Large language models as internal consistency checkers for ophthalmic manuscripts - Takeaways - MDSpire
Coming Soon: Introducing MDSpire News. Learn more
Conexiant’s news site is now MDSpire News. Learn more

Evaluating Large Language Models for Detecting Internal Inconsistencies in Ophthalmic Manuscripts

  • By

  • Ethan La

  • Jai Paris

  • Ayushi Agarwal

  • Weng Onn Chan

  • Dinesh Selva

  • October 9, 2026

Share

  • 1

    Four large language models assessed 40 ophthalmic articles, including 30 with planted inconsistencies and 10 unchanged manuscripts.

  • 2

    The models detected 98 of 120 planted errors (81.7%), with sensitivity ranging from 63.3% for GPT-5.5 to 96.7% for GPT-5.6.

  • 3

    Detection was highest for sample-size and subgrouping inconsistencies (90.0%) and lowest for statistical reporting inconsistencies (72.5%).

  • 4

    All detected planted errors were localised correctly, and explanations and corrections were accurate for 97 of 98 errors.

  • 5

    Manual adjudication confirmed 337 of 355 additional error flags as genuine, while the authors described LLMs as a complementary manuscript quality-control layer.

Original Source(s)

Large language models as internal consistency checkers for ophthalmic manuscripts

  • by Ethan La, Jai Paris, Ayushi Agarwal, Weng Onn Chan, Dinesh Selva

  • October 9, 2026

Related Content

FDA clears targeted cholangiocarcinoma therapy

In the single-arm REFOCUS trial, 46% of patients responded to treatment, with a median response duration of 11.8 months.

Addressing the Emergency-Care Burden in Ophthalmology—Keeping an Eye on the Call

Correction: Idiopathic retinal vasculitis, aneurysms, and neuroretinitis (IRVAN) syndrome: a comprehensive literature review of a rare retinal vasculopathy