Maria A. Woodward

Sex, gender, race, ancestry, and ethnicity data are critical to ensuring an equitable, generalizable artificial intelligence (AI) framework for clinical applications in ophthalmology. The primary question is whether reporting demographic data in ophthalmology AI studies can be improved. This retrospective study investigated AI studies between June 2020 and June 2023. Reporting of sex, gender, race, and ethnicity data was stratified by publication year and institution. A sub-analysis of National Institutes of Health (NIH)-funded manuscripts assessed adherence to the 2020–2021 updated guidance for data reporting in science and medical journals. Adherence was assessed based on terminology used in the published manuscript; it does not indicate that the underlying construct or assessment method was verified. In total, 1005 articles were screened across 15 journals. Of the 268 included studies, 122 studies reported sex demographics, 35 studies reported gender demographics, and 96 studies reported race, ethnicity, or ancestry demographics. In 2020, 46% of studies reported either race, ethnicity, or ancestry. This frequency decreased from 46% in 2020 to 24% in 2021, followed by a modest increase in 2022 to 44%, and a slight decrease in 2023 to 35%. Sex and gender were reported for 53.57% and 10.71% of studies in 2020, 40% and 6% in 2021, 50% and 17% in 2022, and 44% and 17% in 2023, respectively. These results highlight gaps in demographic reporting. To ensure fair clinical tools, AI ophthalmic research can improve data transparency. Academic journals may require the reporting of demographic data per NIH guidelines to better understand AI tool generalizability.

Purpose: To detail the methodology for the Michigan Screening and Intervention for Glaucoma and Eye Health through Telemedicine Program Pragmatic Trial (MI-SIGHT 2) a follow-up to the original MI-SIGHT program that showed feasibility under ideal conditions and efficacy without a comparator. Patients and Methods: A pragmatic randomized controlled trial will compare rate of eye disease detected using the technology-based approach of the MI-SIGHT 2 program (intervention) vs standard optometric exam (control) in a Federally Qualified Health Center (FQHC) (n=900). Inclusion criteria are age ≥ 18 years and no acute eye symptoms. Exclusion criteria are pregnancy or cognitive impairment. The MI-SIGHT 2 intervention leverages technologies including autorefraction, pachymetry, optical coherence tomography, and external and fundus photography, in addition to care navigation support to access low-cost eyeglasses through an on-line retailer and support attendance at recommended follow-up. The primary outcome is detected eye disease as measured by an in-person optometrist exam (control) or by optometrist interpretation of the MI-SIGHT 2 technology-based protocol (intervention). The eye diseases targeted for detection include glaucoma/suspected glaucoma, visually significant cataract, and diabetic retinopathy. The secondary outcome is vision-related quality of life ascertained by survey six weeks after the initial study visit. The exploratory outcomes are program satisfaction and recommended follow-up completion. A cross-over design also will investigate within-person inconsistencies in diagnosis between arms by having participants undergo both screening approaches in randomized order. We will also determine accuracy of diagnoses by each protocol compared to a gold-standard diagnosis determined and adjudicated by glaucoma specialists. Two glaucoma specialist physicians will review all technology-based exams to assess the presence of eye diseases; a third glaucoma specialist physician will adjudicate when there is disagreement. Conclusion: This pragmatic trial will generate evidence to inform best practices for glaucoma and eye disease screening at community health centers.

Purpose: To investigate the association between clinical findings and vision-related quality of life in patients with microbial keratitis (MK) in the United States. Methods: Data from a prospective cohort of individuals with newly diagnosed MK were obtained at baseline and 90 days, including demographics, best-corrected visual acuity (BCVA), morphology, infection type, and the 25-item National Eye Institute Visual Function Questionnaire (VFQ-25). Spearman correlations (r _s ) were used to assess associations between clinical measures and VFQ-25 scores; group differences were tested using Wilcoxon tests with Holm adjustment. Primary outcomes were 90-day VFQ-25 composite and subscale scores. Results: Seventy-nine participants (mean age: 53.6 ± 18.0 years, 47.4% male) completed the VFQ-25 with a mean composite score of 72.4 ± 23.1. Lower composite scores were associated with worse baseline affected-eye BCVA (r _s = −0.44, P = 0.0027) and fluffy infiltrate (median 64.6 vs. 83.1, P = 0.018). At 90 days, lower scores were associated with worse affected-eye BCVA (r _s = −0.68, P < 0.0001), worse better-eye BCVA (r _s = −0.42, P = 0.006), persistent epithelial defect (36.1 vs. 83.1, P = 0.0098), persistent stromal infiltrate (42.0 vs. 83.1, P = 0.0240), and lack of healing (42.0 vs. 83.3, P < 0.0001). Worse 90-day affected-eye BCVA was associated with lower subscale scores for general vision, ocular pain, mental health, role difficulties, and driving (all P ≤ 0.0005). VFQ-25 scores did not differ by infection type ( P = 1.0000). Conclusions: Vision-related impairment in patients with MK is associated not only with BCVA but also with clinical markers of disease activity and healing, including persistent epithelial defect, stromal infiltrate, and nonhealed status. These findings suggest that functional impairment is related to factors beyond visual acuity.

Background/Objectives: To investigate the utility of using a census tract (CT) measure to assess neighbourhood disadvantage and to explore its effects on CT measures of visual difficulty and blindness (VDB). Subjects/Methods: This cross-sectional study used census tract data from the 2018 to 2022 American Community Survey and the National Neighbourhood Data Archive Neighbourhood Socioeconomic Status and Demographic Characteristics. CT measures of neighbourhood disadvantage, VDB, and demographics were summarized with descriptive statistics (mean, standard deviation (SD)). The main outcome was the number of CTs residents reporting VDB and the association with neighbourhood disadvantage (an aggregate measure ranging from 0 to 1 with higher scores indicating more disadvantage) assessed using logistic regression. Results: In total, 83,388 CTs were included, with a mean of 2.53% of the population reporting VDB (SD = 2.62) and a mean neighbourhood disadvantage of 0.18 (SD = 0.12). Neighbourhood disadvantage was associated with 2.9% increased odds of VDB (Odds Ratio (OR): 1.029; 95% Confidence Interval (CI): 1.029,1.029; p < 0.001), after adjusting for neighbourhood demographics including median age, percentage of the CT that were female sex, percentage of the CT that identified as a person of colour, population size of the CT, and the state in which the CT was located. Conclusion: This study identified that neighbourhood disadvantage was associated with a greater number of residents reporting VDB. Clinicians could use this measure to identify neighbourhoods with both higher levels of visual impairment and greater neighbourhood deprivation, allowing for targeted interventions that provide eye disease screening and eye care to those most at risk for poor vision outcomes.

Purpose: To survey United States (US) cornea specialists on clinical factors that distinguish between microbial keratitis (MK) organism groups. Methods: This cross-sectional survey of US cornea specialists assessed whether 61 clinical factors, including history factors, ocular symptoms, and exam findings, distinguished a specific organism group in the typical patient with MK. MK organism groups included gram-positive bacterial keratitis (GPBK), gram-negative bacterial keratitis (GNBK), mold fungal keratitis (MFK), yeast fungal keratitis (YFK), parasitic keratitis (PK), and viral keratitis (VK). The survey was distributed via the Cornea Society's email listserv from March 11, 2024 to May 7, 2024. Descriptive statistical analysis was performed to identify the percentage of specialists identifying a clinical factor as distinguishing a specific MK organism group. Results: A total of 67 specialists completed the survey. There were 20 factors (32.8% of 61) that >75% of specialists identified as distinguishing a specific MK group, including 9 factors for BK (eg copious ocular discharge, loose ocular sutures), 3 factors for FK (eg feathery infiltrate, ocular trauma with organic matter), 5 factors for PK (eg pain out of proportion, perineuritis), and 3 factors for VK (eg reduced/absent corneal sensation, prodromal symptoms). The remaining 41 factors had ≤75% of specialists identify them as distinguishing. Conclusions: Specialists agreed that 32.5% of clinical factors distinguished different MK organism groups, aligned with prior work demonstrating difficulty with interorganism differentiation. These 20 factors may aid clinicians in their initial MK diagnosis; however, more work is needed to validate their salience in diagnosis.

Background: Chronic ocular surface pain (COSP), occurring either in isolation or as part of numerous ocular conditions, such as dry eye syndrome, is a leading cause of eye care visits in the United States. Conventional treatments directed at the ocular surface—the perceived pain source—are often inadequate for pain relief. We hypothesize that some individuals with COSP are experiencing symptoms driven by central nervous system (CNS) dysfunction, similar to chronic overlapping pain conditions, rather than solely pathological problems in the eye. Some individuals with chronic overlapping pain conditions (eg, fibromyalgia) show evidence of nociplastic pain mechanisms, where the pain results from amplified or dysregulated CNS signaling and sensory processing. Although data exist suggesting the presence of nociplastic pain features in COSP, there is a need for comprehensive studies. Objective: Our aim is to rigorously define the role of nociplastic pain in COSP with a large, representative cohort of 200 participants with COSP using established clinical, neurobiological, and treatment response features. We propose that as sign and symptom discordance increases, features indicative of nociplastic pain will also increase. Methods: In aim 1, we will clinically phenotype participants with COSP, using validated patient-reported outcome measures and standard ocular exams. In aim 2, we will compare the neurobiological features of nociplastic pain across the discordance spectrum among a subset of aim 1 participants using multimodal evoked sensory testing (pressure, thermal, and visual testing) at sites both local to and remote from the eye. Participants will also complete structural and functional brain magnetic resonance imaging to assess regions important for pain perception and modulation. In aim 3, we will examine and validate predictors of COSP pain responses before and after application of a topical anesthetic to the ocular surface, which should block peripherally induced discomfort to allow for clarification of pain origination. Results: We received funding for this project in August 2024. Recruitment and enrollment began in January 2025 after protocol development and piloting were completed. This study is ongoing with 59 participants enrolled and 51 participants completing the study visits as of January 2026. Conclusions: Findings from the study have the potential to fundamentally change the way ocular pain syndromes are conceptualized, diagnosed, and treated. This work will not only help identify new CNS-directed pain treatments for a subset of patients with COSP but also help us better understand why many peripherally directed therapies are destined to be ineffective in a subset of individuals experiencing COSP. International Registered Report Identifier (IRRID): DERR1-10.2196/84240

Purpose: To extract microbial keratitis (MK) descriptors from clinician notes in electronic health records using Large Language Models (LLMs) with a zero-shot prompting approach and compare the descriptors with those identified by expert human annotators. Methods: Two hundred fifteen patients with culture-proven MK seen between 2019 and 2023 at Aravind Eye Hospital, Salem, India was gathered. Free-text clinical notes from each patient's first encounter corneal examination were obtained. Each of the 3 MK descriptors—centrality, infiltrate depth, and thinning—was annotated by expert consensus and coded as 1 (present), 0 (absent), or 9 (details unavailable). GPT-4o and GPT-4o mini were prompted to extract the 3 MK descriptors. LLM responses were compared with human annotations using agreement measures such as Cohen Kappa scores at 95% confidence interval, in addition to sensitivity and specificity. Results: GPT-4o demonstrated mean sensitivity of 92%, 86%, and 97%, for centrality, depth, and thinning, respectively, and 96%, 93%, and 99% mean specificity. Cohen Kappa scores calculated for centrality, depth, and thinning were 0.88, 0.73, and 0.88, respectively, indicating good agreement. The overall sensitivity, specificity, and Cohen Kappa of GPT-4o mini were lower compared with GPT-4o, but the difference was not statistically significant. Conclusions: Both GPT-4o and GPT-4o mini showed good agreement with human annotations in extracting MK descriptors. Detection of MK descriptors was influenced by limitations in the quality and consistency of electronic health record documentation. The trade-off between efficiency and performance among LLM architectures remains a consideration for implementation of LLM for large-scale MK data analysis.