Search

Objective: Vision transformers (ViTs) have shown promising performance in various classification tasks previously dominated by convolutional neural networks (CNNs). However, the performance of ViTs in referable diabetic retinopathy (DR) detection is relatively underexplored. In this study, using retinal photographs, we evaluated the comparative performances of ViTs and CNNs on detection of referable DR. Design: Retrospective study. Participants: A total of 48 269 retinal images from the open-source Kaggle DR detection dataset, the Messidor-1 dataset and the Singapore Epidemiology of Eye Diseases (SEED) study were included. Methods: Using 41 614 retinal photographs from the Kaggle dataset, we developed 5 CNN (Visual Geometry Group 19, ResNet50, InceptionV3, DenseNet201, and EfficientNetV2S) and 4 ViTs models (VAN_small, CrossViT_small, ViT_small, and Hierarchical Vision transformer using Shifted Windows [SWIN]_tiny) for the detection of referable DR. We defined the presence of referable DR as eyes with moderate or worse DR. The comparative performance of all 9 models was evaluated in the Kaggle internal test dataset (with 1045 study eyes), and in 2 external test sets, the SEED study (5455 study eyes) and the Messidor-1 (1200 study eyes). Main Outcome Measures: Area under operating characteristics curve (AUC), specificity, and sensitivity. Results: Among all models, the SWIN transformer displayed the highest AUC of 95.7% on the internal test set, significantly outperforming the CNN models (all P

Abstract Background Artificial intelligence (AI) that utilizes deep learning (DL) has potential for systemic disease prediction using retinal imaging. The retina’s unique features enable non-invasive visualization of the central nervous system and microvascular circulation, aiding early detection and personalized treatment plans for personalized care. This review explores the value of retinal assessment, AI-based retinal biomarkers, and the importance of longitudinal prediction models in personalized care. Main text This narrative review extensively surveys the literature for relevant studies in PubMed and Google Scholar, investigating the application of AI-based retina biomarkers in predicting systemic diseases using retinal fundus photography. The study settings, sample sizes, utilized AI models and corresponding results were extracted and analysed. This review highlights the substantial potential of AI-based retinal biomarkers in predicting neurodegenerative, cardiovascular, and chronic kidney diseases. Notably, DL algorithms have demonstrated effectiveness in identifying retinal image features associated with cognitive decline, dementia, Parkinson’s disease, and cardiovascular risk factors. Furthermore, longitudinal prediction models leveraging retinal images have shown potential in continuous disease risk assessment and early detection. AI-based retinal biomarkers are non-invasive, accurate, and efficient for disease forecasting and personalized care. Conclusion AI-based retinal imaging hold promise in transforming primary care and systemic disease management. Together, the retina’s unique features and the power of AI enable early detection, risk stratification, and help revolutionizing disease management plans. However, to fully realize the potential of AI in this domain, further research and validation in real-world settings are essential.

ObjectiveThe aim of this study was to evaluate the accuracy, comprehensiveness, and safety of a publicly available large language model (LLM)—ChatGPT in the sub-domain of glaucoma.DesignEvaluation of diagnostic test or technology.Subjects, participants, and/or controlsWe seek to evaluate the responses of an artificial intelligence chatbot ChatGPT (version GPT-3.5, OpenAI).Methods, intervention, or testingWe curated 24 clinically relevant questions in the domain of glaucoma. The questions spanned four categories: pertaining to diagnosis, treatment, surgeries, and ocular emergencies. Each question was posed to the LLM and the responses obtained were graded by an expert grader panel of three glaucoma specialists with combined experience of more than 30 years in the field. For responses which performed poorly, the LLM was further prompted to self-correct. The subsequent responses were then re-evaluated by the expert panel.Main outcome measuresAccuracy, comprehensiveness, and safety of the responses of a public domain LLM.ResultsThere were a total of 24 questions and three expert graders with a total number of responses of n = 72. The scores were ranked from 1 to 4, where 4 represents the best score with a complete and accurate response. The mean score of the expert panel was 3.29 with a standard deviation of 0.484. Out of the 24 question-response pairs, seven (29.2%) of them had a mean inter-grader score of 3 or less. The mean score of the original seven question-response pairs was 2.96 which rose to 3.58 after an opportunity to self-correct (z-score − 3.27, p = 0.001, Mann–Whitney U). The seven out of 24 question-response pairs which performed poorly were given a chance to self-correct. After self-correction, the proportion of responses obtaining a full score increased from 22/72 (30.6%) to 12/21 (57.1%), (p = 0.026, χ2 test).ConclusionLLMs show great promise in the realm of glaucoma with additional capabilities of self-correction. The application of LLMs in glaucoma is still in its infancy, and still requires further research and validation.

Abstract X-chromosomal genetic variants are understudied but can yield valuable insights into sexually dimorphic human traits and diseases. We performed a sex-stratified cross-ancestry X-chromosome-wide association meta-analysis of seven kidney-related traits (n = 908,697), identifying 23 loci genome-wide significantly associated with two of the traits: 7 for uric acid and 16 for estimated glomerular filtration rate (eGFR), including four novel eGFR loci containing the functionally plausible prioritized genes ACSL4, CLDN2, TSPAN6 and the female-specific DRP2. Further, we identified five novel sex-interactions, comprising male-specific effects at FAM9B and AR/EDA2R, and three sex-differential findings with larger genetic effect sizes in males at DCAF12L1 and MST4 and larger effect sizes in females at HPRT1. All prioritized genes in loci showing significant sex-interactions were located next to androgen response elements (ARE). Five ARE genes showed sex-differential expressions. This study contributes new insights into sex-dimorphisms of kidney traits along with new prioritized gene targets for further molecular research.

Abstract Background Cataract diagnosis typically requires in-person evaluation by an ophthalmologist. However, color fundus photography (CFP) is widely performed outside ophthalmology clinics, which could be exploited to increase the accessibility of cataract screening by automated detection. Methods DeepOpacityNet was developed to detect cataracts from CFP and highlight the most relevant CFP features associated with cataracts. We used 17,514 CFPs from 2573 AREDS2 participants curated from the Age-Related Eye Diseases Study 2 (AREDS2) dataset, of which 8681 CFPs were labeled with cataracts. The ground truth labels were transferred from slit-lamp examination of nuclear cataracts and reading center grading of anterior segment photographs for cortical and posterior subcapsular cataracts. DeepOpacityNet was internally validated on an independent test set (20%), compared to three ophthalmologists on a subset of the test set (100 CFPs), externally validated on three datasets obtained from the Singapore Epidemiology of Eye Diseases study (SEED), and visualized to highlight important features. Results Internally, DeepOpacityNet achieved a superior accuracy of 0.66 (95% confidence interval (CI): 0.64–0.68) and an area under the curve (AUC) of 0.72 (95% CI: 0.70–0.74), compared to that of other state-of-the-art methods. DeepOpacityNet achieved an accuracy of 0.75, compared to an accuracy of 0.67 for the ophthalmologist with the highest performance. Externally, DeepOpacityNet achieved AUC scores of 0.86, 0.88, and 0.89 on SEED datasets, demonstrating the generalizability of our proposed method. Visualizations show that the visibility of blood vessels could be characteristic of cataract absence while blurred regions could be characteristic of cataract presence. Conclusions DeepOpacityNet could detect cataracts from CFPs in AREDS2 with performance superior to that of ophthalmologists and generate interpretable results. The code and models are available at https://github.com/ncbi/DeepOpacityNet ( https://doi.org/10.5281/zenodo.10127002 ).

Abstract Neovascular age-related macular degeneration (nAMD), along with its clinical subtype known as polypoidal choroidal vasculopathy (PCV), are among the leading causes of vision loss in elderly Asians. In a genome-wide association study (GWAS) comprising 3,128 nAMD (1,555 PCV and 1,573 typical nAMD), and 5,493 controls of East Asian ancestry, we identify twelve loci, of which four are novel ( $$P \, < \, 1.19\times {10}^{-8}$$ P < 1.19 × 10 − 8 ). Substantial genetic sharing between PCV and typical nAMD is noted (r g = 0.666), whereas collagen extracellular matrix and fibrosis-related pathways are more pronounced for PCV. Whole-exome sequencing in 259 PCV patients revealed functional rare variants burden in collagen type I alpha 1 chain gene (COL1A1; $$P=1.05\times {10}^{-6}$$ P = 1.05 × 10 − 6 ) and potential enrichment of functional rare mutations at AMD-associated loci. At the GATA binding protein 5 (GATA5) locus, the most significant GWAS novel loci, the expressions of genes including laminin subunit alpha 5 (Lama5), mitochondrial ribosome associated GTPase 2 (Mtg2), and collagen type IX alpha 3 chain (Col9A3), are significantly induced during retinal angiogenesis and subretinal fibrosis in murine models. Furthermore, retinoic acid increased the expression of LAMA5 and MTG2 in vitro. Taken together, our data provide insights into the genetic basis of AMD pathogenesis in the Asian population.

Abstract Resting heart rate is associated with cardiovascular diseases and mortality in observational and Mendelian randomization studies. The aims of this study are to extend the number of resting heart rate associated genetic variants and to obtain further insights in resting heart rate biology and its clinical consequences. A genome-wide meta-analysis of 100 studies in up to 835,465 individuals reveals 493 independent genetic variants in 352 loci, including 68 genetic variants outside previously identified resting heart rate associated loci. We prioritize 670 genes and in silico annotations point to their enrichment in cardiomyocytes and provide insights in their ECG signature. Two-sample Mendelian randomization analyses indicate that higher genetically predicted resting heart rate increases risk of dilated cardiomyopathy, but decreases risk of developing atrial fibrillation, ischemic stroke, and cardio-embolic stroke. We do not find evidence for a linear or non-linear genetic association between resting heart rate and all-cause mortality in contrast to our previous Mendelian randomization study. Systematic alteration of key differences between the current and previous Mendelian randomization study indicates that the most likely cause of the discrepancy between these studies arises from false positive findings in previous one-sample MR analyses caused by weak-instrument bias at lower P-value thresholds. The results extend our understanding of resting heart rate biology and give additional insights in its role in cardiovascular disease development.

Purpose: To evaluate the relationships between chronic kidney disease (CKD) with retinal nerve fiber layer (RNFL) and ganglion cell-inner plexiform layer (GCIPL) thickness profiles of eyes in Asian and White populations. Design: Cross-sectional analysis. Participants: A total of 5066 Asian participants (1367 Malays, 1772 Indians, and 1927 Chinese) from the Singapore Epidemiology of Eye Diseases Study (SEED) were included, consisting of 9594 eyes for peripapillary RNFL analysis and 8661 eyes for GCIPL analysis. Additionally, 45 064 White participants (87 649 eyes) from the United Kingdom Biobank (UKBB) were included for both macular RNFL analysis and GCIPL analysis. Methods: Nonglaucoma participants aged ≥ 40 years with complete data for estimated glomerular filtration rate (eGFR) were included from both SEED and UKBB. In SEED, peripapillary RNFL and GCIPL thickness were measured by Cirrus HD-OCT 4000. In UKBB, macular RNFL and GCIPL were measured by Topcon 3D-OCT 1000 Mark II. Chronic kidney disease was defined as eGFR < 60 ml/min/1.73 m2 in both data sets. To evaluate the associations between kidney function status with RNFL and GCIPL thickness profiles, multivariable linear regression with generalized estimating equation models were performed in SEED and UKBB data sets separately. Main Outcome Measures: Average peripapillary and macular RNFL thickness and macular GCIPL thickness. Results: In SEED, after adjusting for age, gender, ethnicity, systolic blood pressure, antihypertensive medication, diabetes, hyperlipidemia, body mass index, smoking status, and intraocular pressure, presence of CKD (β = −1.31; 95% confidence interval [CI], −2.37 to −0.26; P = 0.015) and reduced eGFR (per 10 ml/min/1.73 m2; β = −0.32; 95% CI, −0.50 to −0.13; P = 0.001) were associated with thinner average peripapillary RNFL. Presence of CKD (β = −1.63; 95% CI, −2.42 to −0.84) and reduced eGFR (per 10 ml/min/1.73 m2; β = −0.30; 95% CI, −0.44 to −0.16) were consistently associated with thinner GCIPL in SEED (all P < 0.001). In UKBB, after adjusting for the above-mentioned covariates (except ethnicity), reduced eGFR (per 10 ml/min/1.73 m2; β = −0.06; 95% CI, −0.10 to −0.01; P = 0.008) was associated with thinner macular RNFL and CKD (β = −0.62; 95% CI, −1.16 to −0.08; P = 0.024) was associated with thinner average GCIPL. Conclusion: We consistently observed associations between CKD and thinning of RNFL and GCIPL across Asian and White populations' eyes. These findings further suggest that compromised kidney function is associated with RNFL and GCIPL thinning. Financial Disclosure(s): The author(s) have no proprietary or commercial interest in any materials discussed in this article.

Purpose: To examine the 6-year incidence of visual impairment (VI) and identify risk factors associated with VI in a multiethnic Asian population. Design: Prospective, population-based, cohort study. Participants: Adults aged ≥ 40 years were recruited from the Singapore Epidemiology of Eye Diseases cohort study at baseline. Eligible subjects were re-examined after 6 years. Subjects included in the final analysis had a mean age of 56.1 ± 8.9 years, and 2801 (50.5%) were female. Methods: All participants underwent standardized examination and interviewer-administered questionnaire at baseline. Incidences were standardized to the Singapore Population Census 2010. A Poisson binomial regression model was used to evaluate the associations between baseline factors and incident presenting VI. Main Outcome Measures: Incident presenting VI was assessed at the 6-year follow-up visit. Visual impairment (presenting visual acuity < 20/40), low vision (presenting visual acuity < 20/40 but ≥ 20/200), and blindness (presenting visual acuity < 20/200) were defined based on United States definition. Results: A total of 5551 subjects (2188 Chinese, 1837 Indians, and 1526 Malays) were evaluated, of whom 514 developed incident presenting VI over 6 years. Malays had a higher incidence of low vision and blindness (13.0%; 0.6%) than Indians (7.0%; 0.1%) and Chinese (7.7%; 0.2%). Among Malay individuals with VI at baseline, 52.8% remained visually impaired after 6 years, which was considerably higher than Chinese (32.4%) and Indians (37.2%). Older age (per decade; relative risk [RR] = 1.59), a history of cardiovascular disease (RR = 1.38), current smoking (RR = 1.31), smaller housing type (1- to 2-room public flat; RR = 2.01), and no formal education (RR = 1.63) at baseline were associated with a higher risk of incident VI (all P ≤ 0.027). Older age (> 60 years) contributed the highest population attributable risk to incident VI (27.1%), followed by lower monthly income (Singapore dollar < $2000; 26.4%) and smaller housing type (24.7%). Overall, undercorrected refractive error (49.1%) and cataract (82.6%) were leading causes for low vision and blindness, respectively. This was consistently observed across the 3 ethnicities. Conclusions: In this multiethnic Asian population, Malays had a higher VI incidence compared to Indians and Chinese. Leading causes of VI are mostly treatable, suggesting that more efforts are needed to further mitigate preventable visual loss. Financial Disclosure(s): The authors have no proprietary or commercial interest in any materials discussed in this article.

Summary: In low- and middle-income countries (LMICs), the fields of medicine and public health grapple with numerous challenges that continue to hinder patients' access to healthcare services. ChatGPT, a publicly accessible chatbot, has emerged as a potential tool in aiding public health efforts in LMICs. This viewpoint details the potential benefits of employing ChatGPT in LMICs to improve medicine and public health encompassing a broad spectrum of domains ranging from health literacy, screening, triaging, remote healthcare support, mental health support, multilingual capabilities, healthcare communication and documentation, medical training and education, and support for healthcare professionals. Additionally, we also share potential concerns and limitations associated with the use of ChatGPT and provide a balanced discussion on the opportunities and challenges of using ChatGPT in LMICs.

Summary: In light of growing interest in using emerging large language models (LLMs) for self-diagnosis, we systematically assessed the performance of ChatGPT-3.5, ChatGPT-4.0, and Google Bard in delivering proficient responses to 37 common inquiries regarding ocular symptoms. Responses were masked, randomly shuffled, and then graded by three consultant-level ophthalmologists for accuracy (poor, borderline, good) and comprehensiveness. Additionally, we evaluated the self-awareness capabilities (ability to self-check and self-correct) of the LLM-Chatbots. 89.2% of ChatGPT-4.0 responses were ‘good’-rated, outperforming ChatGPT-3.5 (59.5%) and Google Bard (40.5%) significantly (all p

IntroductionOur study aimed to examine the relationship between cardiovascular diseases (CVD) with peripapillary retinal fiber layer (RNFL) and macular ganglion cell-inner plexiform layer (GCIPL) thickness profiles in a large multi-ethnic Asian population study.Methods6,024 Asian subjects were analyzed in this study. All participants underwent standardized examinations, including spectral domain OCT imaging (Cirrus HD-OCT; Carl Zeiss Meditec). In total, 9,188 eyes were included for peripapillary RNFL analysis (2,417 Malays; 3,240 Indians; 3,531 Chinese), and 9,270 eyes (2,449 Malays, 3,271 Indians, 3,550 Chinese) for GCIPL analysis. History of CVD was defined as a self-reported clinical history of stroke, myocardial infarction, or angina. Multivariable linear regression models with generalized estimating equations were performed, adjusting for age, gender, ethnicity, diabetes, hypertension, hyperlipidaemia, chronic kidney disease, body mass index, current smoking status, and intraocular pressure.ResultsWe observed a significant association between CVD history and thinner average RNFL (β = −1.63; 95% CI, −2.70 to −0.56; p = 0.003). This association was consistent for superior (β = −1.79, 95% CI, −3.48 to −0.10; p = 0.038) and inferior RNFL quadrant (β = −2.14, 95% CI, −3.96 to −0.32; p = 0.021). Of the CVD types, myocardial infarction particularly showed significant association with average (β = −1.75, 95% CI, −3.08 to −0.42; p = 0.010), superior (β = −2.22, 95% CI, −4.36 to −0.09; p = 0.041) and inferior (β = −2.42, 95% CI, −4.64 to −0.20; p = 0.033) RNFL thinning. Among ethnic groups, the association between CVD and average RNFL was particularly prominent in Indian eyes (β = −1.92, 95% CI, −3.52 to −0.33; p = 0.018). CVD was not significantly associated with average GCIPL thickness, albeit a consistent negative direction of association was observed (β = −0.22, 95% CI, −1.15 to 0.71; p = 0.641).DiscussionIn this large multi-ethnic Asian population study, we observed significant association between CVD history and RNFL thinning. This finding further validates the impact of impaired systemic circulation on RNFL thickness.

Summary: Background: Large language models (LLMs) are garnering wide interest due to their human-like and contextually relevant responses. However, LLMs’ accuracy across specific medical domains has yet been thoroughly evaluated. Myopia is a frequent topic which patients and parents commonly seek information online. Our study evaluated the performance of three LLMs namely ChatGPT-3.5, ChatGPT-4.0, and Google Bard, in delivering accurate responses to common myopia-related queries. Methods: We curated thirty-one commonly asked myopia care-related questions, which were categorised into six domains—pathogenesis, risk factors, clinical presentation, diagnosis, treatment and prevention, and prognosis. Each question was posed to the LLMs, and their responses were independently graded by three consultant-level paediatric ophthalmologists on a three-point accuracy scale (poor, borderline, good). A majority consensus approach was used to determine the final rating for each response. ‘Good’ rated responses were further evaluated for comprehensiveness on a five-point scale. Conversely, ‘poor’ rated responses were further prompted for self-correction and then re-evaluated for accuracy. Findings: ChatGPT-4.0 demonstrated superior accuracy, with 80.6% of responses rated as ‘good’, compared to 61.3% in ChatGPT-3.5 and 54.8% in Google Bard (Pearson's chi-squared test, all p ≤ 0.009). All three LLM-Chatbots showed high mean comprehensiveness scores (Google Bard: 4.35; ChatGPT-4.0: 4.23; ChatGPT-3.5: 4.11, out of a maximum score of 5). All LLM-Chatbots also demonstrated substantial self-correction capabilities: 66.7% (2 in 3) of ChatGPT-4.0's, 40% (2 in 5) of ChatGPT-3.5's, and 60% (3 in 5) of Google Bard's responses improved after self-correction. The LLM-Chatbots performed consistently across domains, except for ‘treatment and prevention’. However, ChatGPT-4.0 still performed superiorly in this domain, receiving 70% ‘good’ ratings, compared to 40% in ChatGPT-3.5 and 45% in Google Bard (Pearson's chi-squared test, all p ≤ 0.001). Interpretation: Our findings underscore the potential of LLMs, particularly ChatGPT-4.0, for delivering accurate and comprehensive responses to myopia-related queries. Continuous strategies and evaluations to improve LLMs’ accuracy remain crucial. Funding: Dr Yih-Chung Tham was supported by the National Medical Research Council of Singapore (NMRC/MOH/HCSAINV21nov-0001).

Abstract Background Currently in the United Kingdom, cardiovascular disease (CVD) risk assessment is based on the QRISK3 score, in which 10% 10-year CVD risk indicates clinical intervention. However, this benchmark has limited efficacy in clinical practice and the need for a more simple, non-invasive risk stratification tool is necessary. Retinal photography is becoming increasingly acceptable as a non-invasive imaging tool for CVD. Previously, we developed a novel CVD risk stratification system based on retinal photographs predicting future CVD risk. This study aims to further validate our biomarker, Reti-CVD, (1) to detect risk group of ≥ 10% in 10-year CVD risk and (2) enhance risk assessment in individuals with QRISK3 of 7.5–10% (termed as borderline-QRISK3 group) using the UK Biobank. Methods Reti-CVD scores were calculated and stratified into three risk groups based on optimized cut-off values from the UK Biobank. We used Cox proportional-hazards models to evaluate the ability of Reti-CVD to predict CVD events in the general population. C-statistics was used to assess the prognostic value of adding Reti-CVD to QRISK3 in borderline-QRISK3 group and three vulnerable subgroups. Results Among 48,260 participants with no history of CVD, 6.3% had CVD events during the 11-year follow-up. Reti-CVD was associated with an increased risk of CVD (adjusted hazard ratio [HR] 1.41; 95% confidence interval [CI], 1.30–1.52) with a 13.1% (95% CI, 11.7–14.6%) 10-year CVD risk in Reti-CVD-high-risk group. The 10-year CVD risk of the borderline-QRISK3 group was greater than 10% in Reti-CVD-high-risk group (11.5% in non-statin cohort [n = 45,473], 11.5% in stage 1 hypertension cohort [n = 11,966], and 14.2% in middle-aged cohort [n = 38,941]). C statistics increased by 0.014 (0.010–0.017) in non-statin cohort, 0.013 (0.007–0.019) in stage 1 hypertension cohort, and 0.023 (0.018–0.029) in middle-aged cohort for CVD event prediction after adding Reti-CVD to QRISK3. Conclusions Reti-CVD has the potential to identify individuals with ≥ 10% 10-year CVD risk who are likely to benefit from earlier preventative CVD interventions. For borderline-QRISK3 individuals with 10-year CVD risk between 7.5 and 10%, Reti-CVD could be used as a risk enhancer tool to help improve discernment accuracy, especially in adult groups that may be pre-disposed to CVD.

Abstract Studies using machine learning (ML) approaches have reported high diagnostic accuracies for glaucoma detection. However, none assessed model performance across ethnicities. The aim of the study is to externally validate ML models for glaucoma detection from optical coherence tomography (OCT) data. We performed a prospective, cross-sectional study, where 514 Asians (257 glaucoma/257 controls) were enrolled to construct ML models for glaucoma detection, which was then tested on 356 Asians (183 glaucoma/173 controls) and 138 Caucasians (57 glaucoma/81 controls). We used the retinal nerve fibre layer (RNFL) thickness values produced by the compensation model, which is a multiple regression model fitted on healthy subjects that corrects the RNFL profile for anatomical factors and the original OCT data (measured) to build two classifiers, respectively. Both the ML models (area under the receiver operating [AUC] = 0.96 and accuracy = 92%) outperformed the measured data (AUC = 0.93; P

A large-scale GWAS provides insight on diabetes-dependent genetic effects on the glomerular filtration rate, a common metric to monitor kidney health in disease.

Objective: Lipid dysregulation and complement system (CS) activation are 2 important pathophysiology pathways for age-related macular degeneration (AMD). We hypothesized that the relationship between lipids and AMD may also differ according to CS genotype profile. Thus, the objective was to investigate the relationships between lipid-related metabolites and AMD according to CS genotypes. Design: Population-based cross-sectional study. Participants: A total of 6947 participants from Singapore Epidemiology of Eye Diseases study with complete relevant data were included. Methods: We investigated a total of 32 blood lipid–related metabolites from nuclear magnetic resonance metabolomics data including lipoproteins and their subclasses, cholesterols, glycerides, and phospholipids, as well as 4 CS single nucleotide polymorphisms (SNPs): rs10922109 (complement factor H), rs10033900 (complement factor I), rs116503776 (C2-CFB-SKIV2L), and rs2230199 (C3). We first investigated the associations between AMD and the 32 lipid-related metabolites using multivariable logistic regression models. Then, to investigate whether the effect of lipid-related metabolites on AMD differ according to the CS SNPs, we tested the possible interactions between the CS SNPs and the lipid-related metabolites. Main Outcome Measures: Age-related macular degeneration was defined using the Wisconsin grading system. Results: Among the 6947 participants, the prevalence of AMD was 6.1%, and the mean age was 58.3 years. First, higher levels of cholesterol in high-density lipoprotein (HDL) and medium and large HDL particles were associated with an increased risk of AMD, and higher levels of serum total triglycerides (TG) and several very-low–density lipoprotein subclass particles were associated with a decreased risk of AMD. Second, these lipids had significant interaction effects on AMD with 2 CS SNPs: rs2230199 and rs116503776 (after correction for multiple testing). For rs2230199, in individuals without risk allele, higher total cholesterol in HDL2 was associated with an increased AMD risk (odds ratio [OR] per standard deviation increase, 1.20; 95% confidence interval (CI), 1.06–1.37; P = 0.005), whereas, in individuals with at least 1 risk allele, higher levels of these particles were associated with a decreased AMD risk (OR, 0.69; 95% CI, 0.45–1.05; P = 0.079). Conversely, for rs116503776, in individuals without risk allele, higher serum total TG were associated with a decreased AMD risk (OR, 0.84; 95% CI, 0.74–0.95; P = 0.005), whereas, in individuals with 2 risk alleles, higher levels of these particles were associated with an increased risk of AMD (OR, 2.3, 95% CI, 0.99–5.39, P = 0.054). Conclusions: Lipid-related metabolites exhibit opposite directions of effects on AMD according to CS genotypes. This indicates that lipid metabolism and CS may have synergistic interplay in the AMD pathogenesis.

PurposeTo develop a deep learning (DL) algorithm for predicting anterior chamber depth (ACD) from smartphone-acquired anterior segment photographs.MethodsFor algorithm development, we included 4,157 eyes from 2,084 Chinese primary school students (aged 11–15 years) from Mojiang Myopia Progression Study (MMPS). All participants had with ACD measurement measured with Lenstar (LS 900) and anterior segment photographs acquired from a smartphone (iPhone Xs), which was mounted on slit lamp and under diffuses lighting. The anterior segment photographs were randomly selected by person into training (80%, no. of eyes = 3,326) and testing (20%, no. of eyes = 831) dataset. We excluded participants with intraocular surgery history or pronounced corneal haze. A convolutional neural network was developed to predict ACD based on these anterior segment photographs. To determine the accuracy of our algorithm, we measured the mean absolute error (MAE) and coefficient of determination (R2) were evaluated. Bland Altman plot was used to illustrate the agreement between DL-predicted and measured ACD values.ResultsIn the test set of 831 eyes, the mean measured ACD was 3.06 ± 0.25 mm, and the mean DL-predicted ACD was 3.10 ± 0.20 mm. The MAE was 0.16 ± 0.13 mm, and R2 was 0.40 between the predicted and measured ACD. The overall mean difference was −0.04 ± 0.20 mm, with 95% limits of agreement ranging between −0.43 and 0.34 mm. The generated saliency maps showed that the algorithm mainly utilized central corneal region (i.e., the site where ACD is clinically measured typically) in making its prediction, providing further plausibility to the algorithm's prediction.ConclusionsWe developed a DL algorithm to estimate ACD based on smartphone-acquired anterior segment photographs. Upon further validation, our algorithm may be further refined for use as a ACD screening tool in rural localities where means of assessing ocular biometry is not readily available. This is particularly important in China where the risk of primary angle closure disease is high and often undetected.

Abstract This study aimed to determine COVID-19-related awareness, knowledge, impact and preparedness among elderly Asians; and to evaluate their acceptance towards digital health services amidst the pandemic. 523 participants (177 Malays, 171 Indians, 175 Chinese) were recruited and underwent standardised phone interview during Singapore’s lockdown period (07 April till 01 June 2020). Multivariable logistic regression models were performed to evaluate the associations between demographic, socio-economic, lifestyle, and systemic factors, with COVID-19 awareness, knowledge, preparedness, well-being and digital health service acceptance. The average perception score on the seriousness of COVID-19 was 7.6 ± 2.4 (out of 10). 75.5% of participants were aware that COVID-19 carriers can be asymptomatic. Nearly all (≥ 90%) were aware of major prevention methods for COVID-19 (i.e. wearing of mask, social distancing). 66.2% felt prepared for the pandemic, and 86.8% felt confident with government’s handling and measures. 78.4% felt their daily routine was impacted. 98.1% reported no prior experience in using digital health services, but 52.2% felt these services would be helpful to reduce non-essential contact. 77.8% were uncomfortable with artificial intelligence software interpreting their medical results. In multivariable analyses, Chinese participants felt less prepared, and more likely felt impacted by COVID-19. Older and lower income participants were less likely to use digital health services. In conclusion, we observed a high level of awareness and knowledge on COVID-19. However, acceptance towards digital health service was low. These findings are valuable for examining the effectiveness of COVID-19 communication in Singapore, and the remaining gaps in digital health adoption among elderly.

Abstract The coronavirus disease 2019 (COVID-19) pandemic has overwhelmed healthcare services, faced with the twin challenges in acutely meeting the medical needs of patients with COVID-19 while continuing essential services for non-COVID-19 illnesses. The need to re-invent, re-organize and transform healthcare and co-ordinate clinical services at a population level is urgent as countries that controlled initial outbreaks start to experience resurgences. A wide range of digital health solutions have been proposed, although the extent of successful real-world applications of these technologies is unclear. This study aims to review applications of artificial intelligence (AI), telehealth, and other relevant digital health solutions for public health responses in the healthcare operating environment amidst the COVID-19 pandemic. A systematic scoping review was performed to identify potentially relevant reports. Key findings include a large body of evidence for various clinical and operational applications of telehealth (40.1%, n = 99/247). Although a large quantity of reports investigated applications of artificial intelligence (AI) (44.9%, n = 111/247) and big data analytics (36.0%, n = 89/247), weaknesses in study design limit generalizability and translation, highlighting the need for more pragmatic real-world investigations. There were also few descriptions of applications for the internet of things (IoT) (2.0%, n = 5/247), digital platforms for communication (DC) (10.9%, 27/247), digital solutions for data management (DM) (1.6%, n = 4/247), and digital structural screening (DS) (8.9%, n = 22/247); representing gaps and opportunities for digital public health. Finally, the performance of digital health technology for operational applications related to population surveillance and points of entry have not been adequately evaluated.

Abstract We evaluated the 6-year incidence and risk factors of pterygium in a multi-ethnic Asian population. Participants who attended the baseline visit of the Singapore Epidemiology of Eye Diseases Study (year 2004–2011) and returned six years later, were included in this study. Pterygium was diagnosed based on anterior segment photographs. Incident pterygium was defined as presence of pterygium at 6-year follow-up in either eye, among individuals without pterygium at baseline. Multivariable logistic regression models were used to determine factors associated with incident pterygium, adjusting for baseline age, gender, ethnicity, body mass index, occupation type, educational level, income status, smoking, alcohol consumption, presence of hypertension, diabetes and hyperlipidemia. The overall age-adjusted 6-year incidence of pterygium was 1.2% (95% confidence interval [CI] 1.0–1.6%); with Chinese (1.9%; 95% CI 1.4%-2.5%) having the highest incidence rate followed by Malays (1.4%; 95% CI 0.9%-2.1%) and Indians (0.3%; 95% CI 0.3–0.7%). In multivariable analysis, Chinese (compared with Indians; odds ratio [OR] = 4.21; 95% CI 2.12–9.35) and Malays (OR 3.22; 95% CI 1.52–7.45), male (OR 2.13; 95% CI 1.26–3.63), outdoor occupation (OR 2.33; 95% CI 1.16–4.38), and smoking (OR 0.41; 95% CI 0.16–0.87) were significantly associated with incident pterygium. Findings from this multi-ethnic Asian population provide useful information in identifying at-risk individuals for pterygium.

Here, the authors report a multi-ethnic genome wide association meta-analysis of 12 studies from the International Cataract Genetics Consortium. They find six new loci associated with age-related nuclear cataract, in addition to replicating the association at CRYAA, and suggest a strong genetic link between age-related nuclear and congenital cataracts.

BackgroundDeep learning algorithms have been built for the detection of systemic and eye diseases based on fundus photographs. The retina possesses features that can be affected by gender differences, and the extent to which these features are captured via photography differs depending on the retinal image field. ObjectiveWe aimed to compare deep learning algorithms’ performance in predicting gender based on different fields of fundus photographs (optic disc–centered, macula-centered, and peripheral fields). MethodsThis retrospective cross-sectional study included 172,170 fundus photographs of 9956 adults aged ≥40 years from the Singapore Epidemiology of Eye Diseases Study. Optic disc–centered, macula-centered, and peripheral field fundus images were included in this study as input data for a deep learning model for gender prediction. Performance was estimated at the individual level and image level. Receiver operating characteristic curves for binary classification were calculated. ResultsThe deep learning algorithms predicted gender with an area under the receiver operating characteristic curve (AUC) of 0.94 at the individual level and an AUC of 0.87 at the image level. Across the three image field types, the best performance was seen when using optic disc–centered field images (younger subgroups: AUC=0.91; older subgroups: AUC=0.86), and algorithms that used peripheral field images had the lowest performance (younger subgroups: AUC=0.85; older subgroups: AUC=0.76). Across the three ethnic subgroups, algorithm performance was lowest in the Indian subgroup (AUC=0.88) compared to that in the Malay (AUC=0.91) and Chinese (AUC=0.91) subgroups when the algorithms were tested on optic disc–centered images. Algorithms’ performance in gender prediction at the image level was better in younger subgroups (aged

Inbreeding depression has been observed in many different species, but in humans a systematic analysis has been difficult so far. Here, analysing more than 1.3 million individuals, the authors show that a genomic inbreeding coefficient (FROH) is associated with disadvantageous outcomes in 32 out of 100 traits tested.

Urinary albumin-to-creatinine ratio (UCAR) is associated with various clinical outcomes such as kidney disease and cardiovascular disease. Here, the authors report genome-wide meta-analysis in over 500,000 individuals and find 68 UACR loci, followed by statistical fine-mapping, gene prioritization and experimental validation in flies.

AIM: To identify the most-cited articles in ophthalmic epidemiology over the last decade. METHODS: We performed a cited reference search on articles that were included in the ISI Web of Science database using the terms “Epidemi*” AND “ophthalm*” AND “population*” during year 2006 to 2016. TOP 100 most cited articles (T100) in ophthalmic epidemiology were short listed and analysed using bibliometrics. RESULTS: These top 100 articles in ophthalmic epidemiology were cited between 61 to 333 times. Of these T100 articles, 36% originated from United States, and 34% were published in the Ophthalmology journal. The three major topics identified were age-related macular degeneration (AMD, n=23), glaucoma (n=16) and visual impairment (n=12). The top-cited article was a study on outdoor activities and its association with the prevalence of myopia in school-aged children, published in 2008. CONCLUSION: This bibliometric analysis provides useful insights into the current development in ophthalmic epidemiology in the past decade and can help recognizing the quality of the researches, discoveries, and trends steering ophthalmic epidemiology.

Summary: Background: By 2050, almost 5 billion people globally are projected to have myopia, of whom 20% are likely to have high myopia with clinically significant risk of sight-threatening complications such as myopic macular degeneration. These are diagnoses that typically require specialist assessment or measurement with multiple unconnected pieces of equipment. Artificial intelligence (AI) approaches might be effective for risk stratification and to identify individuals at highest risk of visual loss. However, unresolved challenges for AI medical studies remain, including paucity of transparency, auditability, and traceability. Methods: In this retrospective multicohort study, we developed and tested retinal photograph-based deep learning algorithms for detection of myopic macular degeneration and high myopia, using a total of 226 686 retinal images. First we trained and internally validated the algorithms on datasets from Singapore, and then externally tested them on datasets from China, Taiwan, India, Russia, and the UK. We also compared the performance of the deep learning algorithms against six human experts in the grading of a randomly selected dataset of 400 images from the external datasets. As proof of concept, we used a blockchain-based AI platform to demonstrate the real-world application of secure data transfer, model transfer, and model testing across three sites in Singapore and China. Findings: The deep learning algorithms showed robust diagnostic performance with areas under the receiver operating characteristic curves [AUC] of 0·969 (95% CI 0·959–0·977) or higher for myopic macular degeneration and 0·913 (0·906–0·920) or higher for high myopia across the external testing datasets with available data. In the randomly selected dataset, the deep learning algorithms outperformed all six expert graders in detection of each condition (AUC of 0·978 [0·957–0·994] for myopic macular degeneration and 0·973 [0·941–0·995] for high myopia). We also successfully used blockchain technology for data transfer, model transfer, and model testing between sites and across two countries. Interpretation: Deep learning algorithms can be effective tools for risk stratification and screening of myopic macular degeneration and high myopia among the large global population with myopia. The blockchain platform developed here could potentially serve as a trusted platform for performance testing of future AI models in medicine. Funding: None.

Summary: Background: Coronary artery calcium (CAC) score is a clinically validated marker of cardiovascular disease risk. We developed and validated a novel cardiovascular risk stratification system based on deep-learning-predicted CAC from retinal photographs. Methods: We used 216 152 retinal photographs from five datasets from South Korea, Singapore, and the UK to train and validate the algorithms. First, using one dataset from a South Korean health-screening centre, we trained a deep-learning algorithm to predict the probability of the presence of CAC (ie, deep-learning retinal CAC score, RetiCAC). We stratified RetiCAC scores into tertiles and used Cox proportional hazards models to evaluate the ability of RetiCAC to predict cardiovascular events based on external test sets from South Korea, Singapore, and the UK Biobank. We evaluated the incremental values of RetiCAC when added to the Pooled Cohort Equation (PCE) for participants in the UK Biobank. Findings: RetiCAC outperformed all single clinical parameter models in predicting the presence of CAC (area under the receiver operating characteristic curve of 0·742, 95% CI 0·732–0·753). Among the 527 participants in the South Korean clinical cohort, 33 (6·3%) had cardiovascular events during the 5-year follow-up. When compared with the current CAC risk stratification (0, >0–100, and >100), the three-strata RetiCAC showed comparable prognostic performance with a concordance index of 0·71. In the Singapore population-based cohort (n=8551), 310 (3·6%) participants had fatal cardiovascular events over 10 years, and the three-strata RetiCAC was significantly associated with increased risk of fatal cardiovascular events (hazard ratio [HR] trend 1·33, 95% CI 1·04–1·71). In the UK Biobank (n=47 679), 337 (0·7%) participants had fatal cardiovascular events over 10 years. When added to the PCE, the three-strata RetiCAC improved cardiovascular risk stratification in the intermediate-risk group (HR trend 1·28, 95% CI 1·07–1·54) and borderline-risk group (1·62, 1·04–2·54), and the continuous net reclassification index was 0·261 (95% CI 0·124–0·364). Interpretation: A deep learning and retinal photograph-derived CAC score is comparable to CT scan-measured CAC in predicting cardiovascular events, and improves on current risk stratification approaches for cardiovascular disease events. These data suggest retinal photograph-based deep learning has the potential to be used as an alternative measure of CAC, especially in low-resource settings. Funding: Yonsei University College of Medicine; Ministry of Health and Welfare, Korea Institute for Advancement of Technology, South Korea; Agency for Science, Technology, and Research; and National Medical Research Council, Singapore.

Summary: Background: The application of deep learning to retinal photographs has yielded promising results in predicting age, sex, blood pressure, and haematological parameters. However, the broader applicability of retinal photograph-based deep learning for predicting other systemic biomarkers and the generalisability of this approach to various populations remains unexplored. Methods: With use of 236 257 retinal photographs from seven diverse Asian and European cohorts (two health screening centres in South Korea, the Beijing Eye Study, three cohorts in the Singapore Epidemiology of Eye Diseases study, and the UK Biobank), we evaluated the capacities of 47 deep-learning algorithms to predict 47 systemic biomarkers as outcome variables, including demographic factors (age and sex); body composition measurements; blood pressure; haematological parameters; lipid profiles; biochemical measures; biomarkers related to liver function, thyroid function, kidney function, and inflammation; and diabetes. The standard neural network architecture of VGG16 was adopted for model development. Findings: In addition to previously reported systemic biomarkers, we showed quantification of body composition indices (muscle mass, height, and bodyweight) and creatinine from retinal photographs. Body muscle mass could be predicted with an R2 of 0·52 (95% CI 0·51–0·53) in the internal test set, and of 0·33 (0·30–0·35) in one external test set with muscle mass measurement available. The R2 value for the prediction of height was 0·42 (0·40–0·43), of bodyweight was 0·36 (0·34–0·37), and of creatinine was 0·38 (0·37–0·40) in the internal test set. However, the performances were poorer in external test sets (with the lowest performance in the European cohort), with R2 values ranging between 0·08 and 0·28 for height, 0·04 and 0·19 for bodyweight, and 0·01 and 0·26 for creatinine. Of the 47 systemic biomarkers, 37 could not be predicted well from retinal photographs via deep learning (R2≤0·14 across all external test sets). Interpretation: Our work provides new insights into the potential use of retinal photographs to predict systemic biomarkers, including body composition indices and serum creatinine, using deep learning in populations with a similar ethnic background. Further evaluations are warranted to validate these findings and evaluate the clinical utility of these algorithms. Funding: Agency for Science, Technology, and Research and National Medical Research Council, Singapore; Korea Institute for Advancement of Technology.

BackgroundClinical trials have shown beneficial effects of blood pressure (BP) control in reducing the risk of diabetic retinopathy (DR). However, association between BP control and DR in population-based studies is not clear. We aimed to examine the association of hypertension and BP control with DR.MethodsWe analysed data from a population-based cross-sectional study of Chinese, Malay and Indians adults with diabetes and hypertension (2004-2011, n = 2189, aged 40-80 years) in Singapore. DR severity was assessed from retinal photographs and graded for any- and vision-threatening DR (VTDR) using the modified Airlie House classification. Hypertension status was classified into (1) good control: on treatment (SBP < 130 and DBP < 80 mm Hg), (2) moderate control: on treatment, with BP levels other than group 1 and 3, (3) poor control: on treatment (SBP ≥140 and DBP ≥ 90 mm Hg), (4) untreated hypertension, any BP level. SBP, DBP and pulse pressure (PP) were analyzed as categories and as continuous variables. The association between BP and DR was assessed using multivariable logistic regression models.ResultsThe prevalence of any-DR and VTDR in the study population was 33.8% and 9.0% respectively. Both poorly controlled and untreated hypertension were significantly associated with any-DR with odds ratio (OR) (95% confidence interval [CI]) of 1.97 (1.39-2.83), and 2.01 [1.34-3.05]. Among BP components, SBP and PP were associated with both any-DR and VTDR with OR (95% CI) of 1.45 (1.28-1.65) and 1.61 (1.41-1.84) for any-DR, and 1.44 (1.19-1.76) and 1.67 (1.37-2.06) for VTDR.ConclusionIn a population-based sample of Asian adults with diabetes and hypertension, treated but poorly controlled as well as untreated hypertension were significantly associated with any-DR. Among the BP components, higher SBP and PP levels were associated with both any-DR and VTDR. Further longitudinal studies are necessary to confirm our findings.

Abstract Population-based data investigating generational differences in the risk of incident cardiovascular disease (CVD) and its risk determinants are rare. We examined the 6-year incidence of CVD and its risk factors in first- and second-generation ethnic Indians living in Singapore. 1749 participants (mean age [SD]: 55.5 [8.8] years; 47.5% male) from a population-based, longitudinal study of Indian adults were included for incident CVD outcome. Incident CVD was defined as self-reported myocardial infarction, angina pectoris or stroke which developed between baseline and follow-up. CVD-related risk factors included incident diabetes, hypertension, hyperlipidemia, obesity and chronic kidney disease (CKD). For incident CVD outcome, of the 1749 participants, 406 (23.2%) and 1343 (76.8%) were first and second-generation Indians, respectively. Of these, 73 (4.1%) reported incident CVD. In multivariable models, second-generation individuals had increased risk of developing CVD (RR = 2.04; 95% CI 1.04, 3.99; p = 0.038), hyperlipidemia (RR = 1.27; 95% CI 1.06, 1.53; p = 0.011), and CKD (RR = 1.92; 95% CI 1.22, 3.04; p = 0.005), compared to first-generation Indians. Second-generation Indians have increased risk of developing CVD and its associated risk factors such as hyperlipidemia and CKD compared to first-generation immigrants, independent of traditional CVD risk factors. More stratified and tailored CVD prevention strategies on second and subsequent generations of Indian immigrants in Singapore are warranted.

AIMS:To determine the diagnostic performance of macular ganglion cell-inner plexiform layer (GCIPL) thickness measured by spectral-domain optical coherence tomography (SD-OCT) for glaucoma detection in a Chinese population in comparison with optic nerve head (ONH) and retinal nerve fiber layer (RNFL) parameters measured by both SD-OCT and Heidelberg Retina Tomography 3 (HRT-3). METHODS:Adults aged 40 to 80 years were recruited from the population-based study (n = 3353, response rate 72.8%). Macular cube 200x200 scan was performed with Cirrus SD-OCT (version 6.0, Carl Zeiss Meditec Inc, Dublin, CA) for GCIPL thickness measurement. ONH and RNFL imaging was performed with Cirrus SD-OCT and HRT-3 (Heidelberg Engineering, Heidelberg, Germany). Glaucoma was defined according to International Society for Geographical and Epidemiological Ophthalmology criteria. RESULTS:In total, 86 eyes of 60 subjects with glaucoma and 1709 eyes of 1001 non-glaucoma participants were included. The best performing parameters for Cirrus SD-OCT GCIPL, Cirrus SD-OCT ONH and HRT-3 were minimum GCIPL thickness (Area under receiver-operating curve [AUC] = 0.89, 95% CI 0.83-0.95), vertical cup-disc ratio (CDR) (AUC = 0.94, 0.91-0.98) and vertical CDR (AUC = 0.86, 0.81-0.92), respectively. At 85% specificity, vertical CDR measured using Cirrus OCT ONH scan showed the highest sensitivity (88.64%, 95% CI 75.4-96.2) compared to minimum GCIPL thickness with sensitivity of 60.53% (95% CI 46.4-73.0) (p

Abnormal durations of sleep have been associated with risk of diabetes. However, it is not clear if sleep duration is associated with diabetic retinopathy (DR).In a cross-sectional study, we included 1,231 (Malay, n = 395; Indian, n = 836) adults (mean age 64.4 ± 9.0 years, 50.4% female) with diabetes from the second visit of two independent population-based cohort studies (2011-15) in Singapore. Self-reported habitual sleep duration was categorized as short (

BACKGROUND/AIMS:To determine the association between aspirin use and diabetic retinopathy (DR) among persons with diabetes, in a population-based, cross-sectional study. METHODS:Subjects with diabetes aged >40 years from the Singapore Epidemiology of Eye Diseases Study were enrolled in this study. Retinal photographs were graded for DR according to the modified Airlie House classification system. Vision threatening diabetic retinopathy (VTDR) was defined as the presence of severe non-proliferative DR, or proliferative DR, or clinically significant macular oedema. The association between aspirin use and the presence of DR or VTDR was assessed using multivariable logistic regression models including age, gender, ethnicity, socioeconomic status, HbA1c, systolic blood pressure, anti-hypertension medicine, total cholesterol, anti-cholesterol medicine, BMI, current smoking status, diabetes duration, history of cardiovascular disease (CVD) and chronic kidney disease (CKD.). RESULTS:A total of 2,061 participants with diabetes and complete record of relevant systemic and DR data were included. Of these, 711 (34.5%) had any stage of DR, and among these 177 (8.6%) had VTDR. After adjusting for co-variables listed, the association between aspirin use and VTDR was significant (OR = 1.69, P = 0.019), while the association between aspirin use and any DR was borderline (OR = 1.31, P = 0.063). Aspirin use was not associated with either DR or VTDR after additional adjustment of CVD and CKD. Further stratification by history of CVD or CKD showed no association between aspirin use and DR/VTDR in either subgroup. CONCLUSION:Aspirin use was not significantly associated with DR but might be an indicator of diabetic complications (CVD, CKD) that were co-present with more severe DR type. Future longitudinal studies are warranted to confirm our findings.

Background/aims: To compare the performance of generative versus retrieval-based chatbots in answering patient inquiries regarding age-related macular degeneration (AMD) and diabetic retinopathy (DR). Methods: We evaluated four chatbots: generative models (ChatGPT-4, ChatGPT-3.5 and Google Bard) and a retrieval-based model (OcularBERT) in a cross-sectional study. Their response accuracy to 45 questions (15 AMD, 15 DR and 15 others) was evaluated and compared. Three masked retinal specialists graded the responses using a three-point Likert scale: either 2 (good, error-free), 1 (borderline) or 0 (poor with significant inaccuracies). The scores were aggregated, ranging from 0 to 6. Based on majority consensus among the graders, the responses were also classified as 'Good', 'Borderline' or 'Poor' quality. Results: Overall, ChatGPT-4 and ChatGPT-3.5 outperformed the other chatbots, both achieving median scores (IQR) of 6 (1), compared with 4.5 (2) in Google Bard, and 2 (1) in OcularBERT (all p =8.4×10-3). Based on the consensus approach, 83.3% of ChatGPT-4's responses and 86.7% of ChatGPT-3.5's were rated as 'Good', surpassing Google Bard (50%) and OcularBERT (10%) (all p =1.4×10-2). ChatGPT-4 and ChatGPT-3.5 had no 'Poor' rated responses. Google Bard produced 6.7% Poor responses, and OcularBERT produced 20%. Across question types, ChatGPT-4 outperformed Google Bard only for AMD, and ChatGPT-3.5 outperformed Google Bard for DR and others. Conclusion: ChatGPT-4 and ChatGPT-3.5 demonstrated superior performance, followed by Google Bard and OcularBERT. Generative chatbots are potentially capable of answering domain-specific questions outside their original training. Further validation studies are still required prior to real-world implementation. [ABSTRACT FROM AUTHOR]

Background: Large language models (LLMs) are fast emerging as potent tools in healthcare, including ophthalmology. This systematic review offers a twofold contribution: it summarises current trends in ophthalmology-related LLM research and projects future directions for this burgeoning field. Methods: We systematically searched across various databases (PubMed, Europe PMC, Scopus and Web of Science) for articles related to LLM use in ophthalmology, published between 1 January 2022 and 31 July 2023. Selected articles were summarised, and categorised by type (editorial, commentary, original research, etc) and their research focus (eg, evaluating ChatGPT's performance in ophthalmology examinations or clinical tasks). Findings: We identified 32 articles meeting our criteria, published between January and July 2023, with a peak in June (n=12). Most were original research evaluating LLMs' proficiency in clinically related tasks (n=9). Studies demonstrated that ChatGPT-4.0 outperformed its predecessor, ChatGPT-3.5, in ophthalmology exams. Furthermore, ChatGPT excelled in constructing discharge notes (n=2), evaluating diagnoses (n=2) and answering general medical queries (n=6). However, it struggled with generating scientific articles or abstracts (n=3) and answering specific subdomain questions, especially those regarding specific treatment options (n=2). ChatGPT's performance relative to other LLMs (Google's Bard, Microsoft's Bing) varied by study design. Ethical concerns such as data hallucination (n=27), authorship (n=5) and data privacy (n=2) were frequently cited. Interpretation: While LLMs hold transformative potential for healthcare and ophthalmology, concerns over accountability, accuracy and data security remain. Future research should focus on application programming interface integration, comparative assessments of popular LLMs, their ability to interpret image-based data and the establishment of standardised evaluation frameworks. [ABSTRACT FROM AUTHOR]

Background Deep learning (DL) is promising to detect glaucoma. However, patients' privacy and data security are major concerns when pooling all data for model development. We developed a privacy-preserving DL model using the federated learning (FL) paradigm to detect glaucoma from optical coherence tomography (OCT) images. Methods This is a multicentre study. The FL paradigm consisted of a 'central server' and seven eye centres in Hong Kong, the USA and Singapore. Each centre first trained a model locally with its own OCT optic disc volumetric dataset and then uploaded its model parameters to the central server. The central server used FedProx algorithm to aggregate all centres' model parameters. Subsequently, the aggregated parameters are redistributed to each centre for its local model optimisation. We experimented with three three-dimensional (3D) networks to evaluate the stabilities of the FL paradigm. Lastly, we tested the FL model on two prospectively collected unseen datasets. Results We used 9326 volumetric OCT scans from 2785 subjects. The FL model performed consistently well with different networks in 7 centres (accuracies 78.3%-98.5%, 75.9%-97.0%, and 78.3%-97.5%, respectively) and stably in the 2 unseen datasets (accuracies 84.8%-87.7%, 81.3%-84.8%, and 86.0%-87.8%, respectively). The FL model achieved non-inferior performance in classifying glaucoma compared with the traditional model and significantly outperformed the individual models. Conclusion The 3D FL model could leverage all the datasets and achieve generalisable performance, without data exchange across centres. This study demonstrated an OCT-based FL paradigm for glaucoma identification with ensured patient privacy and data security, charting another course toward the real-world transition of artificial intelligence in ophthalmology. [ABSTRACT FROM AUTHOR]

Aims To determine axial length (AL) elongation profiles in children aged 3-6 years in an Asian population. Methods Eligible subjects were recruited from the Growing Up in Singapore Towards Healthy Outcomes birth cohort. AL measurement was performed using IOLMaster (Carl Zeiss Meditec, Jena, Germany) at 3 and 6 years. Anthropometric measurements at birth, cycloplegic refraction at 3 and 6 years, questionnaires on the children's behavioural habits at 2 years and parental spherical equivalent refraction were performed. Multivariable linear regression model with generalised estimating equation was performed to determine factors associated with AL elongation. Results 273 eyes of 194 children were included. The mean AL increased from 21.72±0.59 mm at 3 years to 22.52±0.66 mm at 6 years (p<0.001). Myopic eyes at 6 years had greater AL elongation (1.02±0.34 mm) compared with emmetropic eyes (0.85±0.25 mm, p=0.008) and hyperopic eyes (0.74±0.16 mm, p<0.001). The 95th percentile limit of AL elongation was 1.59 mm in myopes, 1.34 mm in emmetropes and 1.00 mm in hyperopes. Greater birth weight (per 100 g, ß=0.010, p=0.02) was significantly associated with greater AL elongation from 3 to 6 years, while parental and other behavioural factors assessed at 2 years were not (all p=0.08). Conclusion In this preschool cohort, AL elongates at an average length of 0.80 mm from 3 to 6 years, with myopes demonstrating the greatest elongation. The differences in 95th percentile limits for AL elongation between myopes, emmetropes and hyperopes can be valuable information in identifying myopia development in preschool children. [ABSTRACT FROM AUTHOR]

With the rise of generative artificial intelligence (AI) and AI‑powered chatbots, the landscape of medicine and healthcare is on the brink of significant transformation. This perspective delves into the prospective influence of AI on medical education, residency training and the continuing education of attending physicians or consultants. We begin by highlighting the constraints of the current education model, challenges in limited faculty, uniformity amidst burgeoning medical knowledge and the limitations in ‘traditional’ linear knowledge acquisition. We introduce ‘AI‑assisted’ and ‘AI‑integrated’ paradigms for medical education and physician training, targeting a more universal, accessible, high‑quality and interconnected educational journey. We differentiate between essential knowledge for all physicians, specialised insights for clinician‑scientists and mastery‑level proficiency for clinician‑computer scientists. With the transformative potential of AI in healthcare and service delivery, it is poised to reshape the pedagogy of medical education and residency training. [ABSTRACT FROM AUTHOR]