Acute Bilateral Periorbital Edema in a Psychiatric Patient.

An older patient with history of alcohol use disorder was admitted for altered mental status and received valproic acid and olanzapine from the psychiatric service. On hospital day 2, the patient acutely developed bilateral periorbital edema. What would you do next?

Assessment of Missing Data on Glaucoma Severity among Participants in the NIH All of Us Research Program of the United States.

PRECIS: There were statistically significant differences across multiple socioeconomic characteristics and self-reported barriers to care among primary glaucoma patients with severity staging data versus those missing this data in the NIH All of Us database. PURPOSE: To characterize missing data among glaucoma patients within All of Us. PATIENTS AND METHODS: We used diagnosis codes to define cohorts of primary glaucoma patients with and without severity staging specified. Descriptive analyses were conducted by presence of disease severity stage. Analysis of missing data was conducted using a set intersection plot and Little's Test of Missing Completely at Random. T-tests were performed to evaluate differences. RESULTS: Of 2982 participants, 1714 (57%) did not have glaucoma severity stage specified, and 11 of 23 analyzed variables had missing data. Little's Test indicated data was not missing completely at random (P<0.001). Significant differences existed between the two cohorts with respect to age, age of first glaucoma diagnosis, gender, ethnicity, education, income, insurance, history of glaucoma surgery and medication use, and answers regarding ability to afford eyeglasses and having seen an eye care provider in the last 12 months (all P values≤0.01). CONCLUSION: There were significant differences between glaucoma participants with glaucoma severity stage specified versus those with unstaged disease across multiple socioeconomic characteristics and self-reported barriers to care. Glaucoma severity staging data was not missing completely at random. The unstaged cohort included higher rates of multiple underrepresented communities, which may potentially contribute to bias in ophthalmology research as participants from vulnerable populations may be disproportionately excluded from electronic health records or claims data studies where diagnosis codes with severity/staging levels are used to examine risk factors for disease, progression, and treatment efficacy.

A review of ophthalmology education in the era of generative artificial intelligence.

PURPOSE: To explore the integration of generative AI, specifically large language models (LLMs), in ophthalmology education and practice, addressing their applications, benefits, challenges, and future directions. DESIGN: A literature review and analysis of current AI applications and educational programs in ophthalmology. METHODS: Analysis of published studies, reviews, articles, websites, and institutional reports on AI use in ophthalmology. Examination of educational programs incorporating AI, including curriculum frameworks, training methodologies, and evaluations of AI performance on medical examinations and clinical case studies. RESULTS: Generative AI, particularly LLMs, shows potential to improve diagnostic accuracy and patient care in ophthalmology. Applications include aiding in patient, physician, and medical students' education. However, challenges such as AI hallucinations, biases, lack of interpretability, and outdated training data limit clinical deployment. Studies revealed varying levels of accuracy of LLMs on ophthalmology board exam questions, underscoring the need for more reliable AI integration. Several educational programs nationwide provide AI and data science training relevant to clinical medicine and ophthalmology. CONCLUSIONS: Generative AI and LLMs offer promising advancements in ophthalmology education and practice. Addressing challenges through comprehensive curricula that include fundamental AI principles, ethical guidelines, and updated, unbiased training data is crucial. Future directions include developing clinically relevant evaluation metrics, implementing hybrid models with human oversight, leveraging image-rich data, and benchmarking AI performance against ophthalmologists. Robust policies on data privacy, security, and transparency are essential for fostering a safe and ethical environment for AI applications in ophthalmology.

American Society of Retina Specialists Artificial Intelligence Task Force Report.

Since the Artificial Intelligence Committee of the American Society of Retina Specialists developed the initial task force report in 2020, the artificial intelligence (AI) field has seen further adoption of US Food and Drug Administration-approved AI platforms and significant development of AI for various retinal conditions. With expansion of this technology comes further areas of challenges, including the data sources used in AI, the democracy of AI, commercialization, bias, and the need for provider education on the technology of AI. The overall focus of this committee report is to explore these recent issues as they relate to the continued development of AI and its integration into ophthalmology and retinal practice.

Retinal structural changes after cessation of pentosan polysulfate sodium.

PURPOSE: To quantify baseline and longitudinal structural changes post-cessation in pentosan polysulfate sodium (PPS) retinopathy patients. METHODS: This is a retrospective cohort study. Retinal thickness and volume of choroidal and hyperreflective retinal pigment epithelium (RPE) excrescences were manually segmented from optical coherence tomography (OCT) volume scans. Baseline measurements were compared against age-matched controls. Longitudinal measurements were performed on patients with follow-up data. RESULTS: Twenty-four eyes of 13 patients were included. At baseline, the mean total retinal thickness was lower in the PPS retinopathy cohort than in age and sex-matched controls (269.1 µm vs. 290.2 µm, p = 0.006).The median (range) of follow-up was 18.6 (4.1 to 34.7) months, with the mean last follow-up of 35.2 months after cessation. During the follow-up period, the thickness of the retina decreased significantly by 11.3 µm (CI: 16.8, 5.8) (p<0.001), with an annual mean decrease of 6.70 µm. However, the mean hyperreflective RPE excrescence volume did not change significantly (p = 0.140) over the follow-up period. CONCLUSIONS: Following PPS discontinuation, although RPE excrescence volumes do not change significantly in volume, there continues to be a progressive long-term thinning of the retina which continues at a rate greater than that associated with normal aging. Consequently, long-term follow-up is suggested to monitor patients with PPS maculopathy.

Gap Analysis of Glaucoma Examination Concept Representations within Standard SNOMED Clinical Terms.

PURPOSE: Standardization of eye care data is important for clinical interoperability and research . We aimed to address gaps in the representations of glaucoma examination concepts within Systemized Nomenclature of Medicine - Clinical Terms (SNOMED-CT), the preferred terminology of the American Academy of Ophthalmology. DESIGN: Study of data elements. METHODS: Structured eye exam data fields from two electronic health records (EHR) systems (Epic Systems and Medisoft) were compared against existing SNOMED-CT codes for concepts representing glaucoma examination findings3. Glaucoma specialists from multiple institutions were surveyed to identify high-priority gaps in representation, which were discussed among the SNOMED International Eye Care Clinical Reference Group. Proposals for new codes to address the gaps were formulated and submitted for inclusion in SNOMED-CT. MAIN OUTCOME MEASURES: Gaps in SNOMED-CT glaucoma examination concept representations RESULTS: We identified several gaps in SNOMED-CT regarding glaucoma examination concepts. A survey of glaucoma specialists identified high-priority data elements within the categories of tonometry and gonioscopy. For tonometry, there was consensus that we need to define new codes related to maximum intraocular pressure (IOP) and target IOP, and to delineate all methods of measuring IOP. These new codes were proposed and successfully added to SNOMED-CT for future use. Regarding gonioscopy, the current terminology did not include the ability to denote the gonioscopic grading system used (e.g., Shaffer or Spaeth), degree of angle pigmentation, iris configuration (except for plateau iris), and iris approach. There was also no ability to specify eye laterality or angle quadrant for gonioscopic findings. We proposed a framework for representing gonioscopic findings as observable entities in SNOMED-CT. DISCUSSION: There are existing gaps in the standardized representation of findings related to tonometry and gonioscopy within SNOMED-CT. These are important areas for evaluating clinical outcomes and enabling secondary use of EHR data for glaucoma research. This international, multi-institutional collaborative process enabled identification of gaps, prioritization, and development of data standards to address these gaps. CONCLUSION: Addressing these gaps and augmenting SNOMED-CT coverage of glaucoma examination findings could enhance clinical documentation and future research efforts related to glaucoma.

Hidden flaws behind expert-level accuracy of multimodal GPT-4 vision in medicine.

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges-an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Natural Language Processing in medicine and ophthalmology: A review for the 21st-century clinician.

Natural Language Processing (NLP) is a subfield of artificial intelligence that focuses on the interaction between computers and human language, enabling computers to understand, generate, and derive meaning from human language. NLP's potential applications in the medical field are extensive and vary from extracting data from Electronic Health Records -one of its most well-known and frequently exploited uses- to investigating relationships among genetics, biomarkers, drugs, and diseases for the proposal of new medications. NLP can be useful for clinical decision support, patient monitoring, or medical image analysis. Despite its vast potential, the real-world application of NLP is still limited due to various challenges and constraints, meaning that its evolution predominantly continues within the research domain. However, with the increasingly widespread use of NLP, particularly with the availability of large language models, such as ChatGPT, it is crucial for medical professionals to be aware of the status, uses, and limitations of these technologies.

Social Factors Associated with the Risk of Glaucoma Suspect Conversion to Glaucoma: Analysis of the Nationwide All of Us Program.

PURPOSE: To examine social factors associated with the 5-year risk of glaucoma suspects (GS) converting to open-angle glaucoma (OAG). DESIGN: Retrospective cohort analysis. SUBJECTS: We screened for participants diagnosed with GS in the All of Us database. Cases that converted to OAG within 5 years of GS diagnosis (the "conversion group") were compared with control cases that did not convert. METHODS: Demographic, socioeconomic and health-care utilization data of the cases were extracted and compared between the conversion group and the control group. Multivariable Cox proportional hazards modeling was used to identify potential factors associated with the risk of conversion. MAIN OUTCOME MEASURES: Hazard ratios (HRs) of significant factors associated with the risk of conversion. RESULTS: A total of 5274 GS participants were identified, and 786 (15%) cases converted to OAG within 5-year follow-up. The 2 groups showed significant differences in age, race, gender, employment status, income/education level, history of intraocular surgery, and health-care utilization patterns. In the multivariable model, African American/Black race (HR : 1.70 [95% confidence interval (CI) 1.44-2.00]), older age at GS diagnosis (1.17 [95% CI 1.09-1.25]), male gender (1.30 [95% CI 1.13-1.50], no history of recreational drug use (1.23 [1.07-1.42]), history of intraocular surgery (1.60 [95% CI 1.02-1.53]), and having more reasons for delayed health-care access (2.27 [95% CI 1.23-4.18]) were associated with a greater hazard of conversion, while being employed (0.71 [95% CI 0.60-0.86]) was associated with a smaller hazard of conversion (P < 0.05 for all). CONCLUSIONS: Several social factors were associated with the conversion from GS to OAG, which may help to identify patients at higher risk of disease progression. Future studies are needed to examine the basis for these findings and the potential interventions that could address them. FINANCIAL DISCLOSURES: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Long-Term Systemic Use of Calcium Channel Blockers and Incidence of Primary Open-Angle Glaucoma.

PURPOSE: To evaluate the association between the systemic use of calcium channel blockers (CCBs) and primary open-angle glaucoma (POAG) using a diverse nationwide dataset. DESIGN: Retrospective cohort study. SUBJECTS: 213 424 individuals aged 40 years and older in the National Institutes of Health All of Us dataset, notable for its demographic, geographic, and medical diversity and inclusion of historically underrepresented populations. Patients with a diagnosis of POAG prior to use of any kind of antihypertensive medication were excluded. METHODS: Bivariate and multivariable regression analyses were performed to evaluate associations between CCB use and POAG. Calcium channel blocker use was further divided into exposure to dihydropyridine CCBs and nondihydropyridine CCBs, and subgroup analyses were performed using chi-square and Fisher tests. MAIN OUTCOME MEASURES: Diagnosis of POAG. RESULTS: Within our cohort, 2772 participants (1.3%) acquired a diagnosis of POAG, while 210 652 (98.7%) did not. Among patients who developed POAG, the mean age was 73.3 years, 52.5% were female, and 48.2% identified as White. Among patients with POAG, 32.6% used 1 or more CCB, 28.2% used a dihydropyridine CCB, and 2.2% used a nondihydropyridine CCB. In bivariate analysis, use of any CCBs was associated with an increased risk of POAG (odds ratio [OR]: 1.29, 95% confidence interval [CI]: 1.27-1.31, P < 0.001). In multivariable analysis adjusting for age, gender, race, ethnicity, and comorbidities such as diabetes, hyperlipidemia, and hypertension, use of any CCBs remained associated with an increased risk of developing POAG (OR: 1.52, 95% CI: 1.33-1.74, P < 0.001). When stratified by type of CCB, the use of dihydropyridine CCBs (OR: 1.31, 95% CI: 1.14-1.50, P < 0.001) was associated with increased POAG risk. CONCLUSIONS: Use of dihydropyridine CCBs was associated with a significantly higher risk of developing POAG, both before and while adjusting for demographic factors and comorbid medical conditions. FINANCIAL DISCLOSURE(S): Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Association between metformin use with circumpapillary retinal nerve fibre layer thickness and capillary vessel density in glaucoma.

BACKGROUND/AIMS: To investigate the association between use of metformin and circumpapillary retinal nerve fibre layer (cpRNFL) thickness, as well as whole image capillary density (wiCD), in patients with glaucoma. METHODS: This cross-sectional study included patients with glaucoma suspect or primary open-angle glaucoma (POAG) underwent optical coherence tomography angiography imaging. Use and duration of antidiabetic medications were assessed at the time of imaging. Multivariable linear mixed-effect modelling was used to estimate the effect of diabetes medication on wiCD and cpRNFL while controlling for covariates including age, race, body mass index, diagnosis, 24-2 visual field mean deviation, and intraocular pressure, average signal strength index as well as any variables that showed a p <0.1 in the univariable analysis. RESULTS: A total of 577 eyes (330 POAG and 247 glaucoma suspect) of 346 patients were included. Sixty-five patients (23%) had diabetes, of whom 55 (78.5%) used metformin, and 17 (26.2%) used insulin. After adjusting for covariates, the association between metformin use and wiCD (1.56 (95% CI 0.40 to 2.71); p=0.008), duration of metformin use and wiCD (0.12 (95% CI 0.02 to 0.22) per 1 year longer; p=0.037), and metformin use and cpRNFL thickness (5.17 (95% CI 1.24 to 9.10) µm; p=0.010) had statistically significant associations in each model. CONCLUSIONS: Metformin use was associated with higher wiCD and thicker cpRNFL. These findings indicate a potential association, underscoring the need for longitudinal studies to determine if metformin plays a role in the retinal conditions of patients with glaucoma. TRIAL REGISTRATION NUMBER: NCT00221897.

Association between Glucagon-Like Peptide 1 (GLP-1) Receptor Agonists Exposure and Intraocular Pressure Change.

Objective: This study aims to provide data on the effects of glucagon-like peptide 1 receptor (GLP-1R) agonists on intraocular pressure (IOP). Design: Retrospective cohort study. Subjects Participants and/or Controls: 1247 glaucoma surgery and treatment naïve eyes of 626 patients who were initiated on GLP-1R agonists compared to 1083 glaucoma surgery and treatment naïve eyes of 547 patients who were initiated on other oral antidiabetics. Methods Intervention or Testing: The University of California Health Data Warehouse was queried for patients exposed to GLP-1R agonists or other oral antidiabetics. Index date was defined as the date of first exposure to the medication. Eyes with at least one pre-exposure and one post-exposure tonometry record within 365 days of the index date were included in the analysis. Clinical and laboratory data elements were extracted from the database. Eyes were censored from the analysis upon exposure to glaucoma hypotensive medication or glaucoma surgery. ΔIOP was analyzed using a paired t-test. Regression analysis was conducted using generalized estimating equations (GEE) accounting for inter-eye correlation. Sensitivity analyses were performed to assess the robustness of the findings. Main Outcome Measures: Primary outcome measure was ΔIOP after exposure to the medication. Results: The median age of all included subjects was 66.2 years [IQR=18.3]; 607 (51.7%) were female, and 667 (56.9%) were Caucasian. Median pre-exposure IOP, HbA1c, and BMI were 15.2 mmHg [IQR=3.8], 7.5 [IQR=2.4], and 29.8 [IQR=9.4], respectively. 776 individuals (66.1%) had diabetes, with the median number of active oral antidiabetics being 1.0 [IQR=1.0], and 441 (37.5%) being insulin users. Several pre-exposure characteristics significantly differed between the GLP-1R agonist and the control group. The mean ΔIOP was -0.4±2.8 mmHg (paired t-test p<0.001) and -0.2±3.3 mmHg (paired t-test p = 0.297) in the GLP-1R agonist and other antidiabetics groups, respectively. Pre-exposure IOP was the only independent predictor of ΔIOP in multivariable GEE. Sensitivity analyses yielded similar results. Conclusions: Although GLP-1R agonists were significantly associated with a decrease in IOP in the paired analysis, they were not associated with ΔIOP in multivariable GEE. Moreover, the difference between the ΔIOP in the two groups was small. Future prospective studies following a standardized dose and delivery method may provide further insights.

Breaking Barriers in Behavioral Change: The Potential of Artificial Intelligence-Driven Motivational Interviewing.

Patient outcomes in ophthalmology are greatly influenced by adherence and patient participation, which can be particularly challenging in diseases like glaucoma, where medication regimens can be complex. A well-studied and evidence-based intervention for behavioral change is motivational interviewing (MI), a collaborative and patient-centered counseling approach that has been shown to improve medication adherence in glaucoma patients. However, there are many barriers to clinicians being able to provide motivational interviewing in-office, including short visit durations within high-volume ophthalmology clinics and inadequate billing structures for counseling. Recently, Large Language Models (LLMs), a type of artificial intelligence, have advanced such that they can follow instructions and carry coherent conversations, offering novel solutions to a wide range of clinical problems. In this paper, we discuss the potential of LLMs to provide chatbot-driven MI to improve adherence in glaucoma patients and provide an example conversation as a proof of concept. We discuss the advantages of AI-driven MI, such as demonstrated effectiveness, scalability, and accessibility. We also explore the risks and limitations, including issues of safety and privacy, as well as the factual inaccuracies and hallucinations to which LLMs are susceptible. Domain-specific training may be needed to ensure the accuracy and completeness of information provided in subspecialty areas such as glaucoma. Despite the current limitations, AI-driven motivational interviewing has the potential to offer significant improvements in adherence and should be further explored to maximally leverage the potential of artificial intelligence for our patients.

Generative artificial intelligence responses to patient messages in the electronic health record: early lessons learned.

Background: Electronic health record (EHR)-based patient messages can contribute to burnout. Messages with a negative tone are particularly challenging to address. In this perspective, we describe our initial evaluation of large language model (LLM)-generated responses to negative EHR patient messages and contend that using LLMs to generate initial drafts may be feasible, although refinement will be needed. Methods: A retrospective sample (n = 50) of negative patient messages was extracted from a health system EHR, de-identified, and inputted into an LLM (ChatGPT). Qualitative analyses were conducted to compare LLM responses to actual care team responses. Results: Some LLM-generated draft responses varied from human responses in relational connection, informational content, and recommendations for next steps. Occasionally, the LLM draft responses could have potentially escalated emotionally charged conversations. Conclusion: Further work is needed to optimize the use of LLMs for responding to negative patient messages in the EHR.

AI-Generated Draft Replies Integrated Into Health Records and Physicians' Electronic Communication.

Importance: Timely tests are warranted to assess the association between generative artificial intelligence (GenAI) use and physicians' work efforts. Objective: To investigate the association between GenAI-drafted replies for patient messages and physician time spent on answering messages and the length of replies. Design, Setting, and Participants: Randomized waiting list quality improvement (QI) study from June to August 2023 in an academic health system. Primary care physicians were randomized to an immediate activation group and a delayed activation group. Data were analyzed from August to November 2023. Exposure: Access to GenAI-drafted replies for patient messages. Main Outcomes and Measures: Time spent (1) reading messages, (2) replying to messages, (3) length of replies, and (4) physician likelihood to recommend GenAI drafts. The a priori hypothesis was that GenAI drafts would be associated with less physician time spent reading and replying to messages. A mixed-effects model was used. Results: Fifty-two physicians participated in this QI study, with 25 randomized to the immediate activation group and 27 randomized to the delayed activation group. A contemporary control group included 70 physicians. There were 18 female participants (72.0%) in the immediate group and 17 female participants (63.0%) in the delayed group; the median age range was 35-44 years in the immediate group and 45-54 years in the delayed group. The median (IQR) time spent reading messages in the immediate group was 26 (11-69) seconds at baseline, 31 (15-70) seconds 3 weeks after entry to the intervention, and 31 (14-70) seconds 6 weeks after entry. The delayed group's median (IQR) read time was 25 (10-67) seconds at baseline, 29 (11-77) seconds during the 3-week waiting period, and 32 (15-72) seconds 3 weeks after entry to the intervention. The contemporary control group's median (IQR) read times were 21 (9-54), 22 (9-63), and 23 (9-60) seconds in corresponding periods. The estimated association of GenAI was a 21.8% increase in read time (95% CI, 5.2% to 41.0%; P = .008), a -5.9% change in reply time (95% CI, -16.6% to 6.2%; P = .33), and a 17.9% increase in reply length (95% CI, 10.1% to 26.2%; P < .001). Participants recognized GenAI's value and suggested areas for improvement. Conclusions and Relevance: In this QI study, GenAI-drafted replies were associated with significantly increased read time, no change in reply time, significantly increased reply length, and some perceived benefits. Rigorous empirical tests are necessary to further examine GenAI's performance. Future studies should examine patient experience and compare multiple GenAIs, including those with medical training.

Variations in Electronic Health Record-Based Definitions of Diabetic Retinopathy Cohorts: A Literature Review and Quantitative Analysis.

Purpose: Use of the electronic health record (EHR) has motivated the need for data standardization. A gap in knowledge exists regarding variations in existing terminologies for defining diabetic retinopathy (DR) cohorts. This study aimed to review the literature and analyze variations regarding codified definitions of DR. Design: Literature review and quantitative analysis. Subjects: Published manuscripts. Methods: Four graders reviewed PubMed and Google Scholar for peer-reviewed studies. Studies were included if they used codified definitions of DR (e.g., billing codes). Data elements such as author names, publication year, purpose, data set type, and DR definitions were manually extracted. Each study was reviewed by ≥ 2 authors to validate inclusion eligibility. Quantitative analyses of the codified definitions were then performed to characterize the variation between DR cohort definitions. Main Outcome Measures: Number of studies included and numeric counts of billing codes used to define codified cohorts. Results: In total, 43 studies met the inclusion criteria. Half of the included studies used datasets based on structured EHR data (i.e., data registries, institutional EHR review), and half used claims data. All but 1 of the studies used billing codes such as the International Classification of Diseases 9th or 10th edition (ICD-9 or ICD-10), either alone or in addition to another terminology for defining disease. Of the 27 included studies that used ICD-9 and the 20 studies that used ICD-10 codes, the most common codes used pertained to the full spectrum of DR severity. Diabetic retinopathy complications (e.g., vitreous hemorrhage) were also used to define some DR cohorts. Conclusions: Substantial variations exist among codified definitions for DR cohorts within retrospective studies. Variable definitions may limit generalizability and reproducibility of retrospective studies. More work is needed to standardize disease cohorts. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.

Novel Technologies in Artificial Intelligence and Telemedicine for Glaucoma Screening.

PURPOSE: To provide an overview of novel technologies in telemedicine and artificial intelligence (AI) approaches for cost-effective glaucoma screening. METHODS/RESULTS: A narrative review was performed by summarizing research results, recent developments in glaucoma detection and care, and considerations related to telemedicine and AI in glaucoma screening. Telemedicine and AI approaches provide the opportunity for novel glaucoma screening programs in primary care, optometry, portable, and home-based settings. These approaches offer several advantages for glaucoma screening, including increasing access to care, lowering costs, identifying patients in need of urgent treatment, and enabling timely diagnosis and early intervention. However, challenges remain in implementing these systems, including integration into existing clinical workflows, ensuring equity for patients, and meeting ethical and regulatory requirements. Leveraging recent work towards standardized data acquisition as well as tools and techniques developed for automated diabetic retinopathy screening programs may provide a model for a cost-effective approach to glaucoma screening. CONCLUSION: Leveraging novel technologies and advances in telemedicine and AI-based approaches to glaucoma detection show promise for improving our ability to detect moderate and advanced glaucoma in primary care settings and target higher individuals at high risk for having the disease.

Evaluation of bias and gender/racial concordance based on sentiment analysis of narrative evaluations of clinical clerkships using natural language processing.

There is increasing interest in understanding potential bias in medical education. We used natural language processing (NLP) to evaluate potential bias in clinical clerkship evaluations. Data from medical evaluations and administrative databases for medical students enrolled in third-year clinical clerkship rotations across two academic years. We collected demographic information of students and faculty evaluators to determine gender/racial concordance (i.e., whether the student and faculty identified with the same demographic). We used a multinomial log-linear model for final clerkship grades, using predictors such as numerical evaluation scores, gender/racial concordance, and sentiment scores of narrative evaluations using the SentimentIntensityAnalyzer tool in Python. 2037 evaluations from 198 students were analyzed. Statistical significance was defined as P < 0.05. Sentiment scores for evaluations did not vary significantly by student gender, race, or ethnicity (P = 0.88, 0.64, and 0.06, respectively). Word choices were similar across faculty and student demographic groups. Modeling showed narrative evaluation sentiment scores were not predictive of an honors grade (odds ratio [OR] 1.23, P = 0.58). Numerical evaluation average (OR 1.45, P < 0.001) and gender concordance between faculty and student (OR 1.32, P = 0.049) were significant predictors of receiving honors. The lack of disparities in narrative text in our study contrasts with prior findings from other institutions. Ongoing efforts include comparative analyses with other institutions to understand what institutional factors may contribute to bias. NLP enables a systematic approach for investigating bias. The insights gained from the lack of association between word choices, sentiment scores, and final grades show potential opportunities to improve feedback processes for students.

Hidden Flaws Behind Expert-Level Accuracy of Multimodal GPT-4 Vision in Medicine.

Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V's rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges - an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V's high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

Social determinants of health and health disparities in glaucoma: A review.

Social determinants of health and barriers to care can significantly impact patients' access to glaucoma care and treatment, resulting in disparities within disease presentation, progression, management, and treatment outcomes. The widespread adoption of electronic health record systems has allowed researchers and clinicians to further explore these relationships, identifying factors such as race, ethnicity, and socioeconomic status to be risk factors for more severe disease and lower treatment adherence. These disparities highlight potential targets for interventions to combat these disparities and improve overall patient outcomes. This article provides a summary of the available data on health disparities within glaucoma disease presentation, progression, management, treatment, and outcomes and discusses interventions to improve care delivery and outcomes among patients with glaucoma.