Performance of Generative Large Language Models on Ophthalmology Board-Style Questions.

PURPOSE: To investigate the ability of generative artificial intelligence models to answer ophthalmology board-style questions. DESIGN: Experimental study. METHODS: This study evaluated 3 large language models (LLMs) with chat interfaces, Bing Chat (Microsoft) and ChatGPT 3.5 and 4.0 (OpenAI), using 250 questions from the Basic Science and Clinical Science Self-Assessment Program. Although ChatGPT is trained on information last updated in 2021, Bing Chat incorporates a more recently indexed internet search to generate its answers. Performance was compared with human respondents. Questions were categorized by complexity and patient care phase, and instances of information fabrication or nonlogical reasoning were documented. MAIN OUTCOME MEASURES: Primary outcome was response accuracy. Secondary outcomes were performance in question subcategories and hallucination frequency. RESULTS: Human respondents had an average accuracy of 72.2%. ChatGPT-3.5 scored the lowest (58.8%), whereas ChatGPT-4.0 (71.6%) and Bing Chat (71.2%) performed comparably. ChatGPT-4.0 excelled in workup-type questions (odds ratio [OR], 3.89, 95% CI, 1.19-14.73, P = .03) compared with diagnostic questions, but struggled with image interpretation (OR, 0.14, 95% CI, 0.05-0.33, P < .01) when compared with single-step reasoning questions. Against single-step questions, Bing Chat also faced difficulties with image interpretation (OR, 0.18, 95% CI, 0.08-0.44, P < .01) and multi-step reasoning (OR, 0.30, 95% CI, 0.11-0.84, P = .02). ChatGPT-3.5 had the highest rate of hallucinations and nonlogical reasoning (42.4%), followed by ChatGPT-4.0 (18.0%) and Bing Chat (25.6%). CONCLUSIONS: LLMs (particularly ChatGPT-4.0 and Bing Chat) can perform similarly with human respondents answering questions from the Basic Science and Clinical Science Self-Assessment Program. The frequency of hallucinations and nonlogical reasoning suggests room for improvement in the performance of conversational agents in the medical domain.

PRO score: predictive scoring system for visual outcomes after rhegmatogenous retinal detachment repair.

BACKGROUND/AIMS: To compare risk factors for poor visual outcomes in patients undergoing primary rhegmatogenous retinal detachment (RRD) repair and to develop a scoring system. METHODS: Analysis of the Primary Retinal detachment Outcomes (PRO) study, a multicentre interventional cohort of consecutive primary RRD surgeries performed in 2015. The main outcome measure was a poor visual outcome (Snellen VA ≤20/200). RESULTS: A total of 1178 cases were included. The mean preoperative and postoperative logMARs were 1.1±1.1 (20/250) and 0.5±0.7 (20/63), respectively. Multivariable logistic regression identified preoperative risk factors predictive of poor visual outcomes (≤20/200), including proliferative vitreoretinopathy (PVR) (OR 1.26; 95% CI 1.13 to 1.40), history of antivascular endothelial growth factor (VEGF) injections (1.38; 1.11 to 1.71), >1-week vision loss (1.17; 1.08 to 1.27), ocular comorbidities (1.18; 1.00 to 1.38), poor presenting VA (1.06 per initial logMAR unit; 1.02 to 1.10) and age >70 (1.13; 1.04 to 1.23). The data were split into training (75%) and validation (25%) and a scoring system was developed and validated. The risk for poor visual outcomes was 8% with a total score of 0, 17% with 1, 29% with 2, 47% with 3, and 71% with 4 or higher. CONCLUSIONS: Independent risk factors were compared for poor visual outcomes after RRD surgery, which included PVR, anti-VEGF injections, vision loss >1 week, ocular comorbidities, presenting VA and older age. The PRO score was developed to provide a scoring system that may be useful in clinical practice.

Seasonal and Environmental Variations in Endophthalmitis after Intravitreal Anti-vascular Endothelial Growth Factor Injection: A Six-year Review.

PURPOSE: To evaluate seasonal and environmental variations on the incidence and outcomes of postinjection endophthalmitis. METHODS: A single-center, retrospective cohort study was conducted including all patients diagnosed with post-injection endophthalmitis between 2013-2018. Associations between climate variables and endophthalmitis incidence were evaluated. RESULTS: Of 423,297 injections administered, seasonal distribution in spring, summer, autumn, and winter was 26%, 27%, 25%, and 22%, respectively. Of 171 cases of endophthalmitis identified, seasonal distribution over the spring, summer, autumn, and fall was 25%, 23%, 26%, and 26%, respectively. Endophthalmitis incidence was not correlated with monthly precipitation (p = 0.45), monthly snowfall (p = 0.49), or monthly temperature (p = 0.65). Worse visual outcomes at initial endophthalmitis presentation were correlated with increased precipitation level (p = 0.025) but were not correlated with snowfall level (p = 0.228) or mean monthly temperature (p = 0.132). Although there were no seasonal variations of visual acuity at endophthalmitis presentation (p = 0.894), odds of final visual acuity returning to within two lines of pre-endophthalmitis visual acuity were worse among patients with endophthalmitis diagnosed in the spring (OR, 0.041; p = 0.016). CONCLUSION: In contrast to previous work on postcataract endophthalmitis, seasonal and weather factors were not associated with post-injection endophthalmitis risk or bacterial species isolated. Visual outcomes at initial endophthalmitis presentation were correlated with precipitation, and worse visual outcomes were seen in patients who developed endophthalmitis in the spring.

Endophthalmitis After Intravitreal Anti-Vascular Endothelial Factor Injections: Outcomes of Eyes Managed Without Microbiologic Cultures.

PURPOSE: To evaluate outcomes of eyes that developed endophthalmitis after intravitreal anti-vascular endothelial growth factor injections that were managed without microbiologic cultures. DESIGN: Retrospective, single-center, comparative cohort study. METHODS: We included all eyes with postinjection endophthalmitis from July 1, 2013, to September 1, 2019. Endophthalmitis cases were divided into the culture group if treated with intravitreal antibiotics and a vitreous or aqueous tap sent for microbiologic sampling or into the no culture group if treated with immediate injection of intravitreal antibiotics with an anterior chamber paracentesis that was not sent for microbiologic sampling. The main outcome measures were visual acuity, the incidence of retinal detachment, and the need for additional procedures. RESULTS: Of 165 endophthalmitis cases identified, 119 (72%) were in the culture group and 46 (28%) were in the no culture group. At endophthalmitis presentation, eyes in the culture group had a mean logMAR VA of 1.98 (∼20/1900) compared with 1.90 (∼20/1600) for eyes in the no culture group (P = .589). At the 6-month follow-up, the mean vision loss was 5.5 lines lost from baseline for the culture group compared with 2.5 lines lost for the no culture group (P = .017). Eyes in the culture group required a subsequent pars plana vitrectomy in 29 of 119 cases (24%) compared with 7 of 46 cases (15%) in the no culture group (P = .29). Six of 119 eyes (5%) in the culture group developed secondary retinal detachments compared with none in the no culture group (P = .143). CONCLUSIONS: When access to microbiologic facility is not available, the management of postinjection endophthalmitis using intravitreal antibiotics without microbiologic cultures may be an acceptable treatment strategy.

Barriers to Follow-Up Retinal Care During the COVID-19 Pandemic: A Survey Study.

BACKGROUND AND OBJECTIVE: To characterize patient-identified barriers to care in those non-compliant with retina appointments during the coronavirus disease 2019 (COVID-19) pandemic. PATIENTS AND METHODS: Inclusion criteria included non-compliant patients from March 1, 2020 to May 1, 2020. Ultimately, 1,345 patients were invited to complete a 14-question survey. A retrospective chart review correlated clinical and demographic information. Univariate logistic regression, independent-samples t-test, and Pearson correlation coefficient identified differences among subgroups. RESULTS: Of the 1,345 patients, 181 (13.5%) completed the survey. The most significant barriers to care included fear of COVID (76/181; 42.0%), wait times (21/181; 11.6%), and costs (11/181; 6.1%). Patients who got their COVID information from the Centers for Disease Control and Prevention (7.8 ± 2.4) and televised news (8.0 ± 2.0) had higher levels of fear. Finally, patients with diabetic retinopathy and higher Charlson Comorbidity Index scores had greater concerns of COVID (P = .034 and P = .047, respectively). CONCLUSION: This survey study suggests fear of COVID-19 is a prominent new barrier to retinal care. Identifying those at risk for loss to follow-up can guide practices as the pandemic continues. [Ophthalmic Surg Lasers Imaging Retina. 2021;52:526-533.].

Systemic considerations with pigmented fundus lesions and retinal pigment epithelium hamartomas in Turcot syndrome.

PURPOSE OF REVIEW: To illustrate retinal pigment epithelium hamartomas (RPEHs) in a patient with known Turcot syndrome and to demonstrate the significance of these lesions in other colonic polyposis syndromes. RECENT FINDINGS: The presence of multiple, bilateral RPEHs is a specific marker for familial adenomatous polyposis (FAP). Recent studies have noted RPEHs in associated syndromes including Gardner and Turcot syndromes. Herein, we illustrate an 18-year-old male patient with known brain medulloblastoma who was documented 7 years later to have asymptomatic RPEHs in both eyes, demonstrating hypo-autofluorescence. SUMMARY: This patient with previous medulloblastoma and known Turcot syndrome was later found to have typical RPEHs, a feature that is important in family screening. Confirmation of RPEHs related to Turcot syndrome was made, and genetic evaluation confirmed germline mutation in the adenomatous polyposis coli (APC) gene. Observation was advised. Due to the fairly rapid progression to colorectal cancer if untreated, ophthalmologists should be aware that RPEHs can be the first extracolonic manifestation of FAP, Gardner syndrome, and Turcot syndrome.

GEOGRAPHIC ACCESS DISPARITIES TO CLINICAL TRIALS IN RETINOPATHY OF PREMATURITY IN THE UNITED STATES.

BACKGROUND/PURPOSE: To identify geographic and socioeconomic variables predictive of residential proximity to retinopathy of prematurity (ROP) clinical trial locations. METHODS: This cross-sectional epidemiological study used census tract-level data from three national public data sets and trial-level data from ClinicalTrials.gov. Socioeconomic predictors of driving distance and time to the nearest ROP clinical trial location were identified. Primary outcomes were time >60 minutes and distance >60 miles traveled to the nearest ROP clinical trial site. RESULTS: Multivariate analysis showed that residents were more likely to travel >60 minutes to the nearest ROP clinical trial site if they lived in census tracts that were rural (adjusted odds ratio 1.20, P = 0.0002), had higher percentages of the population living ≤ federal poverty level (fourth quartile vs. first quartile, adjusted odds ratio 1.19, P < 0.0001), or had less education (associate vs. bachelor's degree, adjusted odds ratio 1.01, P <0.007). By contrast, counties with higher percentages of births with birth weight <1500 g (adjusted odds ratio 0.88, P = 0.0062) were less likely to travel >60 minutes. Similar variables predicted travel distance. CONCLUSION: Although counties with higher incidences of very low-birth-weight infants were closer to ROP clinical trial sites, residents living in rural and low-income census tracts had significantly greater travel burdens.

Current Applicant Perceptions of the Ophthalmology Residency Match.

Importance: The ophthalmology residency application process is critical for applicants and residency programs, and knowledge about the preferences of applicants would assist both groups in improving the process. Objective: To evaluate the experiences and preferences of ophthalmology residency applicants. Design, Setting, and Participants: This cross-sectional, nonvalidated survey was conducted online. All applicants to the Bascom Palmer Eye Institute ophthalmology residency program during the 2018-2019 application cycle were invited to complete the survey. Data collection occurred from April 1, 2019, to April 30, 2019. Main Outcomes and Measures: Applicant demographics, application submissions, interview experiences, financial considerations, match results, and suggestions for improvement of the application process. Results: Responses were received from 185 applicants (36.4%), including 77 women (41.6%). A successful match into an ophthalmology residency was achieved by 172 respondents (93.0%). There was a mean (SD) US Medical Licensing Examination Step 1 score of 245.8 (13.3) points. Respondents applied to a mean (SD) of 76.4 (23.5) ophthalmology residency programs, received 14.0 (9.0) invitations to interview, and attended 10.3 (4.4) interviews. Choices regarding applications and interviews were based mostly on program reputation, location, and advisor recommendation. A usual lead time of at least 3 weeks between the invitation and interview was reported by 126 respondents (69.2%), which was reduced to 14 respondents (15.1%) when a wait-list was involved. The ophthalmology residency application process cost a mean (SD) of $5704 ($2831) per applicant. Respondents reported that they were most able to reduce costs through housing choices (hotel stays or similar arrangements) and least able to reduce costs by limiting the number of programs to which they applied or at which they interviewed. Conclusions and Relevance: The ophthalmology residency application process is complex and poses substantial challenges to applicants and residency programs. These findings suggest that many current applicants have difficulty selecting programs to apply to, and most respondents desired changes to the current system of interview invitations and scheduling.

Punctuated evolution of canonical genomic aberrations in uveal melanoma.

Cancer is thought to arise through the accumulation of genomic aberrations evolving under Darwinian selection. However, it remains unclear when the aberrations associated with metastasis emerge during tumor evolution. Uveal melanoma (UM) is the most common primary eye cancer and frequently leads to metastatic death, which is strongly linked to BAP1 mutations. Accordingly, UM is ideally suited for studying the clonal evolution of metastatic competence. Here we analyze sequencing data from 151 primary UM samples using a customized bioinformatic pipeline, to improve detection of BAP1 mutations and infer the clonal relationships among genomic aberrations. Strikingly, we find BAP1 mutations and other canonical genomic aberrations usually arise in an early punctuated burst, followed by neutral evolution extending to the time of clinical detection. This implies that the metastatic proclivity of UM is "set in stone" early in tumor evolution and may explain why advances in primary treatment have not improved survival.