You have full access to this article via your institution. Download PDF Over the last 5 years, artificial intelligence (AI) has seen widespread adoption in various fields, including

healthcare, finance, and transportation. This growth can be attributed to significant advancements in AI sub-areas [1]. Large language models (LLMs), which are advanced AI models trained on

extensive textual data, have gained immense popularity among the general population and have also impacted patients who increasingly rely on large language models (LLMs) to seek clinical

information. In particular, in vitreoretinal conditions, there is a lack of patient awareness of “alarm” symptoms that may delay presentation for medical attention [2]. In response, patients

tested were ChatGPT 3.5 (OpenAI), Bing AI (powered by GPT-4 and Microsoft), and Docs-GPT Beta (optimized for healthcare and medical contexts by OpenAI). We categorized the questions into two

groups: medical advice (Table 1) and medical conditions/post-operative advice (Table 2). The answers provided by the LLMs were reviewed by three vitreoretinal surgeons and classified as

follows: Accurate and sufficient: Correct content with all important information present. Partially accurate and sufficient: Some incorrect information, but overall answer content

understandable and informative. Inaccurate: Completely wrong answer or fundamental errors in the response. No human subjects were involved in our study, and the questions used did not

include any personal information about patients. Regarding medical advice questions, ChatGPT and DocsGPT had 80% of their answers classified as accurate and sufficient and 20% being

partially accurate and sufficient. In contrast, Bing AI achieved a 100% accuracy rate (Table 1). For pre- and post-operative advice questions, ChatGPT provided accurate and sufficient

responses in 88% of cases, partially accurate and sufficient, and inaccurately in 6%, respectively. Bing AI and DocsGPT scored 81% for accurate and sufficient answers, and 19% for partially

accurate and sufficient responses. ChatGPT offered the most detailed information among the LLMs, while Bing AI was the only one providing verifiable references. However, these differences

did not significantly impact the accuracy of the responses. All LLMs demonstrated acceptable performance. This research did not aim to determine the best LLMs LLMs could play a crucial role

in vitreoretinal care, assisting with various tasks like summarizing topics for patients, addressing their questions and emails, and facilitating communication with non-English speakers

through translation services [4]. These capabilities are especially valuable in virtual clinics, where patients can seek clarifications after receiving virtual medical evaluations [5].

Accessibility is a key advantage, enabling patients to obtain quick answers anytime, which is particularly beneficial for those in remote areas. Moreover, LLMs generate responses easier to

understand than medical terminology. LLMs offer potential benefits, but we must address inherent limitations before incorporating this technology into medical practice. While our evaluation

showed generally acceptable performance, further in-depth and extensive testing is necessary. Developing specific training models for vitreoretinal surgery can enhance accuracy and coverage

of complex topics, ensuring LLMs become reliable tools. Proper patient education on LLMs usage is vital, understanding they complement, not replace, medical professionals. Ethical and legal

concerns about data collection and dissemination require attention, as sensitive patient information may be at risk [6]. Strict data protection measures, guidelines, and informed consent are

essential. LLMs have great potential to be a valuable tool in vitreoretinal surgery. The key is now how we, as ophthalmologists engage with, and integrate LLMs into our daily practice.

REFERENCES * Michael LL, Ifeoma A, Guy B, Craig B, Morgan C, Finale D-V. et al. Gathering strength, gathering storms: The one hundred year study on artificial intelligence (AI100) 2021 study

panel report. Stanford University, Stanford, CA, 2021. http://ai100.stanford.edu/2021-report. Accessed 16 June 2023. * Anguita R, Ting MYL, Makuloluwa A, Charteris DG. Causal factors for

late presentation of retinal detachment. Eye (Lond). 2023;37:185–6. https://doi.org/10.1038/s41433-022-02109-z. Article  PubMed  Google Scholar  * Ruran HB, Petty CR, Eliott D, Rao RC,

Phipatanakul W, Young BK. Patient perceptions of retinal detachment management and recovery through social media. Semin Ophthalmol. 2023;38:498–502.

https://doi.org/10.1080/08820538.2023.2168492. Article  PubMed  Google Scholar  * Large language models in medicine: The potential to reduce workloads, leverage the EMR for Better

Communication & More (2023a) The Rheumatologist. Available at:

https://www.the-rheumatologist.org/article/large-language-models-in-medicine-the-potential-to-reduce-workloads-leverage-the-emr-for-better-communication-more/ * Hanumunthadu D, Adan K,

Tinkler K, Balaskas K, Hamilton R, Nicholson L. et al. Outcomes following implementation of a high-volume medical retina virtual clinic utilising a diagnostic hub during COVID-19. Eye

(Lond). 2022;36:627–33. https://doi.org/10.1038/s41433-021-01510-4. Article  CAS  PubMed  Google Scholar  * Li H, Moon JT, Purkayastha S, Celi LA, Trivedi H, Gichoya JW. Ethics of large

language models in medicine and medical research. Lancet. Digit Health. 2023;5:e333–35. https://doi.org/10.1016/S2589-7500(23)00083-3. Article  CAS  PubMed  Google Scholar  Download

references AUTHOR INFORMATION AUTHORS AND AFFILIATIONS * Vitreoretinal Unit, Moorfields Eye Hospital NHS Foundation Trust, City Road London, London, EC1V 2PD, UK Rodrigo Anguita, Achini

Makuloluwa, Jennifer Hind & Louisa Wickham * Department of Ophthalmology, Inselspital, University Hospital of Bern, Bern, Switzerland Rodrigo Anguita Authors * Rodrigo Anguita View

author publications You can also search for this author inPubMed Google Scholar * Achini Makuloluwa View author publications You can also search for this author inPubMed Google Scholar *

Jennifer Hind View author publications You can also search for this author inPubMed Google Scholar * Louisa Wickham View author publications You can also search for this author inPubMed 

Google Scholar CONTRIBUTIONS RA and AM conceived and designed the research. RA, AM, and JH analyzed the data. RA, AM, JH, and LW analyzed and interpreted the literature. RA, AM, JH, and LW

drafted the manuscript and made critical revisions of the manuscript. CORRESPONDING AUTHOR Correspondence to Rodrigo Anguita. ETHICS DECLARATIONS COMPETING INTERESTS The authors declare no

competing interests. ADDITIONAL INFORMATION PUBLISHER’S NOTE Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. RIGHTS AND

PERMISSIONS Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other

rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law. Reprints and

permissions ABOUT THIS ARTICLE CITE THIS ARTICLE Anguita, R., Makuloluwa, A., Hind, J. _et al._ Large language models in vitreoretinal surgery. _Eye_ 38, 809–810 (2024).

https://doi.org/10.1038/s41433-023-02751-1 Download citation * Received: 11 July 2023 * Revised: 12 August 2023 * Accepted: 12 September 2023 * Published: 19 September 2023 * Issue Date:

March 2024 * DOI: https://doi.org/10.1038/s41433-023-02751-1 SHARE THIS ARTICLE Anyone you share the following link with will be able to read this content: Get shareable link Sorry, a

shareable link is not currently available for this article. Copy to clipboard Provided by the Springer Nature SharedIt content-sharing initiative