ChatGPT (OpenAI, San Francisco, Calif., https://chat.openai.com) is a large-language model (LLM) artificial intelligence (AI) chatbot that has been suggested as a paradigm-shifting development in the way patients and physicians access health-related information.1Goodman RS Patrinely JR Osterman T Wheless L Johnson DB On the cusp: considering the impact of artificial intelligence language models in healthcare.Med. 2023; 4: 139-140Abstract Full Text Full Text PDF PubMed Scopus (22) Google Scholar, 2Golan R Ripps SJ Reddy R et al.ChatGPT's ability to assess quality and readability of online medical information: evidence from a cross-sectional study.Cureus. 2023; 15: e42214PubMed Google Scholar, 3Bernstein IA Zhang YV Govil D et al.Comparison of ophthalmologist and large language model chatbot responses to online patient eye care questions.JAMA Netw Open. 2023; 6e2330320Crossref Scopus (6) Google Scholar However, along with the excitement, there are concerns regarding the potential dissemination of misinformation (termed hallucination) and the comprehensibility of health care information it provides.1Goodman RS Patrinely JR Osterman T Wheless L Johnson DB On the cusp: considering the impact of artificial intelligence language models in healthcare.Med. 2023; 4: 139-140Abstract Full Text Full Text PDF PubMed Scopus (22) Google Scholar,2Golan R Ripps SJ Reddy R et al.ChatGPT's ability to assess quality and readability of online medical information: evidence from a cross-sectional study.Cureus. 2023; 15: e42214PubMed Google Scholar We compared the quality of ChatGPT's postoperative instructions with those from Google Search as well as 2 institutional developed handouts for common ophthalmic procedures. This study was modeled on the framework proposed by Ayoub et al.4Ayoub NF Lee Y-J Grimm D Balakrishnan K. Comparison between ChatGPT and Google Search as sources of postoperative patient instructions.JAMA Otolaryngol Neck Surg. 2023; 149: 556-558Crossref PubMed Scopus (20) Google Scholar Postoperative patient instructions were examined for 7 common ophthalmic procedures: intravitreal injections, cataract surgery, partial-thickness corneal transplants, full-thickness corneal transplants, trabeculectomy, vitrectomy, and strabismus surgery. Postoperative instructions were obtained from 4 sources: (i) ChatGPT, version 3.5, (ii) Google Search, (iii) MyHealth.Alberta.ca, a peer-reviewed database of postoperative patient instructions in Canada, and (iv) the patient information collection of the Sunderland Eye Infirmary, a National Health Services (NHS) eye hospital that part of the Sunderland and South Tyneside NHS trust, with a peer and patient review process. To obtain instructions from ChatGPT, the following phrase was used: "Please provide postoperative instructions for a patient who just underwent [procedure]." To obtain instructions from Google, the following phrase was entered into Google Search: "Postoperative patient instructions for [procedure]." The first non-sponsored result was used. Patient instructions were scored by 2 masked assessors (K.N. and C.W.Y.), with standardized fonts and identifying information removed for masking. Differences were resolved through discussion. The primary outcome was the Patient Education Materials Assessment Tool–Printable (PEMAT-P) total score.5Shoemaker SJ Wolf MS Brach C. Development of the Patient Education Materials Assessment Tool (PEMAT): a new measure of understandability and actionability for print and audiovisual patient information.Patient Educ Couns. 2014; 96: 395-403Crossref PubMed Scopus (476) Google Scholar This score consists of 19 criteria assessing understandability and 7 criteria assessing actionability. The secondary outcome was the frequency with which instructions mentioned procedure-specific items. An a priori literature search informed a list of 6 items for each procedure. This list was shortened to 4 items by a vote performed by the clinician authors of this investigation (Supplementary File 1, available online). Scores were compared with one-way analysis of variance. For significant results, post hoc comparisons of means were performed using Tukey's honestly significant difference testing to correct for multiplicity. Analyses were performed using R (version 4.3.1; R Foundation, Vienna, Austria). There was near-perfect agreement among raters (97%; Cohen's kappa = 0.92). Tables 1 and 2, as well as Supplementary Figure 1 (available online), summarize the ratings for each questionnaire. A one-way analysis of variance test revealed a significant effect for mean PEMAT-P understandability (p = 0.018; F = 4.05) and PEMAT-P overall scores (p = 0.043; F = 3.15). There was no significant effect when comparing actionability scores (F = 0.18; p = 0.910) or number of prespecified items (F = 1.6; p = 0.220; Supplementary Table 1, available online).Table 1Rating summary for each postoperative instruction sheetPostoperative instructionPEMAT-P scoreProcedure-specific itemsUnderstandability*Understandability refers to the ease at which patients of different backgrounds and levels of health literacy can process and explain the messages from the education material provided.5Actionability†Actionability refers to whether patients of different backgrounds and levels of health literacy can identify steps they can take based on the information provided.5Total scoreCataract surgeryChatGPT77%60%72%75%Google Search85%60%78%75%Canadian institution88%60%81%100%U.K. institution77%60%72%25%Intravitreal injectionChatGPT77%60%72%50%Google Search69%60%78%75%Canadian institution73%60%81%75%U.K. institution85%60%72%50%Strabismus surgeryChatGPT77%60%72%75%Google Search69%60%67%75%Canadian institution75%60%71%75%U.K. Institution85%60%78%75%Partial-thickness corneal transplantChatGPT77%60%72%75%Google Search85%60%78%100%Canadian institution81%60%76%100%U.K. institution92%40%78%100%Full-thickness corneal transplantChatGPT77%60%72%75%Google Search85%60%78%75%Canadian institution81%60%76%75%U.K. institution92%60%83%100%Pars plana vitrectomyChatGPT77%60%72%75%Google Search85%60%78%100%Canadian institution88%60%81%100%U.K. institution94%100%96%100%TrabeculectomyChatGPT77%60%72%50%Google Search85%60%78%100%Canadian institution81%60%76%75%U.K. institution85%60%78%75% Understandability refers to the ease at which patients of different backgrounds and levels of health literacy can process and explain the messages from the education material provided.5Shoemaker SJ Wolf MS Brach C. Development of the Patient Education Materials Assessment Tool (PEMAT): a new measure of understandability and actionability for print and audiovisual patient information.Patient Educ Couns. 2014; 96: 395-403Crossref PubMed Scopus (476) Google Scholar† Actionability refers to whether patients of different backgrounds and levels of health literacy can identify steps they can take based on the information provided.5Shoemaker SJ Wolf MS Brach C. Development of the Patient Education Materials Assessment Tool (PEMAT): a new measure of understandability and actionability for print and audiovisual patient information.Patient Educ Couns. 2014; 96: 395-403Crossref PubMed Scopus (476) Google Scholar Open table in a new tab Table 2Summary of PEMAT-P scores and procedure-specific items for each instruction setCategoryMean (SD) scores (%)ChatGPTGoogle SearchCanadian institutionU.K. institutionPEMAT-P total72% (0%)75% (5%)76% (4%)80% (7%)PEMAT-P understandability77% (0%)80% (8%)81% (5%)87% (6%)PEMAT-P actionability60% (0%)60% (0%)60% (0%)63% (18%)Procedure-specific items68% (12%)86% (13%)86% (13%)75% (29%)PEMAT-P, Patient Education Materials Assessment Tool–Printable. Open table in a new tab PEMAT-P, Patient Education Materials Assessment Tool–Printable. Post hoc comparisons of means demonstrated significantly higher scores from the U.K. institution when compared with ChatGPT scores for both PEMAT-P understandability (mean difference [MD] = 0.101; 95% CI, 0.019–0.184; p = 0.011) and PEMAT-P total scores (MD = 0.081; 95% CI, 0.006–0.156; p = 0.030; Supplementary Table 2, available online). ChatGPT represents a ground-breaking technology and the first conversational LLM AI to potentially affect health care.1Goodman RS Patrinely JR Osterman T Wheless L Johnson DB On the cusp: considering the impact of artificial intelligence language models in healthcare.Med. 2023; 4: 139-140Abstract Full Text Full Text PDF PubMed Scopus (22) Google Scholar,3Bernstein IA Zhang YV Govil D et al.Comparison of ophthalmologist and large language model chatbot responses to online patient eye care questions.JAMA Netw Open. 2023; 6e2330320Crossref Scopus (6) Google Scholar,6Lim ZW Pushpanathan K Yew SME et al.Benchmarking large language models' performances for myopia care: a comparative analysis of ChatGPT-3.5, ChatGPT-4.0, and Google Bard.EbioMedicine. 2023; 95104770Abstract Full Text Full Text PDF Scopus (11) Google Scholar Our investigation suggests that the postoperative instructions from ChatGPT, Google Search, and 2 institutions examined provide procedure-specific information at a comparable rate. However, the ChatGPT instructions were less understandable than the instructions from the U.K. institution. Therefore, while the results suggest that ChatGPT postoperative instructions can aid in postoperative patient management, patients and clinicians should be aware of the limitations associated with these instructions, including potentially reduced understandability, a lack of references or peer review, and a knowledge cutoff date of September 2021. Future research is important to better define the role, benefits, risks, and limitations of ChatGPT and other LLM AI for aiding patient management.6Lim ZW Pushpanathan K Yew SME et al.Benchmarking large language models' performances for myopia care: a comparative analysis of ChatGPT-3.5, ChatGPT-4.0, and Google Bard.EbioMedicine. 2023; 95104770Abstract Full Text Full Text PDF Scopus (11) Google Scholar Sobha Sivaprasad reports receiving research grants from Novartis, Bayer, Allergan, Roche, Boehringer, Ingelheim, and Optos Plc; travel grants from Novartis and Bayer; speaker fees from Novartis, Bayer, and Optos Plc; and reimbursements for attending advisory board meetings for Novartis, Bayer, Allergan, Roche, Boehringer, Ingelheim, Optos Plc, Apellis, Oculis, and Eyebiotech; David H. Steel reports receiving consultant fees from Gyroscope, Roche, Alcon, and BVI and research funds from Alcon, Bayer, DORC, Gyrsocope, and Boehringer-Ingelheim; Charles C. Wykoff reports receiving consulting fees from Acuela, Adverum Biotechnologies, Inc, Aerpio, Alimera Sciences, Allegro Ophthalmics LLC, Allergan, Apellis Pharmaceuticals, Bayer AG, Chengdu Kanghong Pharmaceuticals Group Company, Ltd., Clearside Biomedical, DORC, EyePoint Pharmaceuticals, Gentech/Roche, GyroscopeTx, IVERIC Bio, Kodiak Sciences Inc., Novartis AG, ONL Therapeutics, Oxurion NV, PolyPhotonix, Recens Medical, Regeron Pharmaceuticals, Inc., REGENXBIO Inc., Santen Pharmaceutical Company, Ltd., and Takeda Pharmaceutical Company Ltd. And receiving research funding from Adverum Biotechnologies, Aerie Pharmaceuticals, Aerpio, Alimera Sciences, Allergan, Apellis Pharmaceuticals, Chengdu Kanghong Pharmaceutical Group Company, Ltd., Clearside Biomedical, Gemini Therapeutics, Genentech/Roche, Graybug Vision, GyroscopeTx, Ionis Pharmaceuticals, IVERIC Bio, Kodiak Sciences, Neurotech LLC, Novartis AG, Opthea, Outlook Therapeutics, Recens Medical, Regeneron Pharmaceuticals, REGENXBIO Inc., Samsung Pharm Company, Ltd., Santen Pharmaceutical Company, Ltd., and Xbrane Biopharma AB; Varun Chaudhary reports acting as an advisory board member and receiving grants and other remuneration from Novartis, acting as an advisory board member and receiving grants and other remuneration from Bayer, receiving grants from Allergan, and acting as an advisory board member for Roche outside the submitted work. The remaining authors have no conflicts of interest to report. Download .docx (.06 MB) Help with docx files Download .docx (.01 MB) Help with docx files Download .docx (.01 MB) Help with docx files Download .docx (.01 MB) Help with docx files
Purpose Patients with non-proliferative macular telangiectasia type 2 (MacTel) have ganglion cell layer (GCL) and nerve fibre layer (NFL) loss, but it is unclear whether the thinning is progressive. We quantified the change in retinal layer thickness over time in MacTel with and without diabetes. Methods In this retrospective, multicentre, comparative case series, subjects with MacTel with at least two optical coherence tomographic (OCT) scans separated by >9 months OCTs were segmented using the Iowa Reference Algorithms. Mean NFL and GCL thickness was computed across the total area of the early treatment diabetic retinopathy study grid and for the inner temporal region to determine the rate of thinning over time. Mixed effects models were fit to each layer and region to determine retinal thinning for each sublayer over time. Results 115 patients with MacTel were included; 57 patients (50%) had diabetes and 21 (18%) had a history of carbonic anhydrase inhibitor (CAI) treatment. MacTel patients with and without diabetes had similar rates of thinning. In patients without diabetes and untreated with CAIs, the temporal parafoveal NFL thinned at a rate of -0.25 +/- 0.09 mu m/year (95% CI [-0.42 to -0.09]; p=0.003). The GCL in subfield 4 thinned faster in the eyes treated with CAI (-1.23 +/- 0.21 mu m/year; 95% CI [-1.64 to -0.82]) than in untreated eyes (-0.19 +/- 0.16; 95% CI [-0.50, 0.11]; p<0.001), an effect also seen for the inner nuclear layer. Progressive outer retinal thinning was observed. Conclusions Patients with MacTel sustain progressive inner retinal neurodegeneration similar to those with diabetes without diabetic retinopathy. Further research is needed to understand the consequences of retinal thinning in MacTel.
Background/Purpose: To determine and compare the efficacy of a surgical internal limiting membrane (ILM) flap technique with the traditional ILM peel on long-term visual and anatomical outcomes for large (>400 mu m) full-thickness macular holes. Methods: From October 2016 to July 2022, patients undergoing initial full-thickness macular hole repair with the ILM flap or ILM peel technique were reviewed. Final outcomes were recorded and based on size in microns: 401 to 800, 801 to 1,200, and >1,200. Results: Patients treated with ILM flap (n = 52, 94.2% closure rate) or ILM peel (n = 407, 93.6% closure rate) were followed with a mean follow-up time of 15.0 +/- 10.2 and 20.0 +/- 13.4 months, respectively. Success rates for ILM flaps and ILM peels were compared for full-thickness macular holes of 401 to 800 (100%, 95.8%, P = 0.39), 801 to 1,200 (95%, 93%, P = 0.74), and >1,200 (86.7%, 86.7%, P = 1.0) mu m. Mean best-recorded logarithm of the minimal angle of resolution visual acuity for ILM flaps and ILM peels, respectively, was 1.02 +/- 0.46 and 0.87 +/- 0.47 preoperatively, with follow-up acuity of 0.48 +/- 0.32 (P < 0.03) and 0.39 +/- 0.42 (P < 0.01) at Year 3. Conclusion: Both techniques provide a similar anatomical closure rate and functional improvement in vision. Comparisons should be cautiously made based on difference in preoperative hole size.
Background/Purpose: To determine and compare the efficacy of a surgical internal limiting membrane (ILM) flap technique with the traditional ILM peel on long-term visual and anatomical outcomes for large (>400 µ m) full-thickness macular holes. Methods: From October 2016 to July 2022, patients undergoing initial full-thickness macular hole repair with the ILM flap or ILM peel technique were reviewed. Final outcomes were recorded and based on size in microns: 401 to 800, 801 to 1,200, and >1,200. Results: Patients treated with ILM flap (n = 52, 94.2% closure rate) or ILM peel (n = 407, 93.6% closure rate) were followed with a mean follow-up time of 15.0 ± 10.2 and 20.0 ± 13.4 months, respectively. Success rates for ILM flaps and ILM peels were compared for full-thickness macular holes of 401 to 800 (100%, 95.8%, P = 0.39), 801 to 1,200 (95%, 93%, P = 0.74), and >1,200 (86.7%, 86.7%, P = 1.0) µ m. Mean best-recorded logarithm of the minimal angle of resolution visual acuity for ILM flaps and ILM peels, respectively, was 1.02 ± 0.46 and 0.87 ± 0.47 preoperatively, with follow-up acuity of 0.48 ± 0.32 ( P < 0.03) and 0.39 ± 0.42 ( P < 0.01) at Year 3. Conclusion: Both techniques provide a similar anatomical closure rate and functional improvement in vision. Comparisons should be cautiously made based on difference in preoperative hole size.
PurposeTo analyse the morphological characteristics of eyes with neovascular age-related macular degeneration (AMD) with good long-term visual acuity after anti-VEGF (vascular endothelial growth factor) therapy.MethodsRetrospective, observational study of 175 patients with neovascular AMD with >5 years of follow-up after initiating anti-VEGF therapy. Spectral-domain optical coherence tomography images were assessed for thickness of pigment epithelial detachment (PED), subretinal hyper-reflective material (SHRM), subretinal fluid and subfoveal choroidal, as well as the integrity of the outer retinal bands.ResultsThe final analysis cohort included 203 eyes (175 patients) followed for a mean of 7.84±1.70 years (range: 5–11). The maximum PED thickness in the foveal central subfield (FCS) was significantly lower (p<0.001) in the poor vision group (13.11 μm) compared with the intermediate (86.25 μm) or good (97.92 μm) vision groups, respectively. In contrast, the maximum thickness of SHRM in the FCS was significantly thicker (p<0.001) in eyes with poor vision (149.46 μm) compared with eyes with intermediate vision (64.37 μm) which in turn were significantly thicker (p<0.001) than eyes with good vision (9.35 μm). The good vision group also had better continuity of all outer retinal bands (external limiting membrane, ellipsoid zone, and retinal pigment epithelium) compared with the other two groups (all p<0.001).ConclusionA thicker PED and thinner SHRM were correlated with better vision in eyes with neovascular AMD following long-term anti-VEGF therapy. If replicated in future prospective studies, these findings may have implications for design of optimal anatomic endpoints for neovascular AMD treatment.