Serum and Urinary Metabolites Associated with Diabetic Retinopathy in Two Asian Cohorts

Serum and urinary metabolites associated with diabetic retinopathy in two Asian cohorts

AuthorBlock: Charumathi Sabanayagam3,1, Rehena Sultana1, Riswana Banu3, Zhou Lei3, Gavin SW Tan3, E Shyong Tai2, Ching-Yu Cheng1,3, Tien Yin Wong3,1 1Duke-NUS Medical School, Singapore, Singapore; 2National University of Singapore, , Singapore; 3Singapore Eye Research Institute, Singapore National Eye Center, Singapore, Singapore;

DisclosureBlock: Charumathi Sabanayagam, None; Rehena Sultana, None; Riswana Banu, None; Zhou Lei, None; Gavin SW Tan, None; E Shyong Tai, None; Ching-Yu Cheng, None; Tien Yin Wong, None;

Purpose

Serum metabolites have been shown to be associated with diabetes and its complications including diabetic retinopathy (DR). Metabolites in urine have recently been shown to provide information independent of serum metabolites. We examined the cross-sectional association of DR with a combination of serum and urinary metabolites in two Asian cohorts.Methods

Serum (n=225) and urinary metabolites (n=33) were quantified using nuclear magnetic resonance (NMR) spectroscopy in 1869 adults with diabetes who participated in the baseline visits (2007-2011, aged ≥40 years) of two independent population-based cohort studies (Chinese, n=581; Indians, n=1288) with similar methodology in Singapore. Diabetes-specific moderate DR was assessed from retinal photographs and defined as a level>43 using the Early Treatment Diabetic Retinopathy Study (ETDRS) scale. Multivariate logistic regression models were constructed for each metabolite (per SD increase) adjusted for age, sex, systolic blood pressure, duration of diabetes and HbA1c. Metabolites with the same direction of association in both cohorts were selected and those achieving significance corrected for multiple testing from both cohorts were meta-analysed.Results Prevalence of moderate DR was 11.2% in Chinese and 10.6% in Indians. Logistic regression identified 21 serum and 10 urinary metabolites associated with DR in one or both studies after Bonferroni correction, of which 6 serum (p<0.002) and 6 urine metabolites (p=0.005) were statistically significant in the meta-analysis. Among serum metabolites, higher levels of creatinine, total cholesterol and cholesterol esters in chylomicrons in extremely large VLDL (lipoprotein subclasses) were positively associated (odds ratios [ORs], 1.25, 1.11, 1.11) while higher levels of tyrosine, fatty acid saturation % and cholesterol esters in LDL, were inversely associated with DR (ORs 0.57, 0.60, 0.63) (Figure 1). Among urinary metabolites, higher levels of alanine, valine (aminoacids), formate (microbial), citrate (glycolysis), 3-hydroxyisovalerate, 3-hydroxy isobutyrate were inversely associated with DR with ORs ranging from 0.52 to 0.68 (Figure 2).Conclusions

Our results demonstrated that adults with DR had altered serum and urinary profiles mostly confined to aminoacids, lipoprotein, glycolysis, fluid and fatty acid metabolism.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details.

Risk Factors for Severe Diabetic Eye Disease in a Large Sample of Patients Newly-Diagnosed with Type 2 Diabetes

AuthorBlock: William S. Gange1, Khristina Lung2, Benjamin Xu1, Seth Seabury2, Brian C. Toy1 1Ophthalmology, USC Roski Eye Institute, Los Angeles, California, United States; 2Keck-Shaeﬀer Initiative for Population Health Policy, Keck School of Medicine, Los Angeles, California, United States;

DisclosureBlock: William S. Gange, None; Khristina Lung, None; Benjamin Xu, None; Seth Seabury, Precision Health Economics, LLC Code C (Consultant), Brian C. Toy, None;

Purpose Diabetic eye disease is the leading cause of blindness in working-age adults in the United States. As rates of diabetes continue to rise in the United States, adequately screening all patients for diabetic retinopathy at guideline-recommended intervals remains a public health challenge. We performed a retrospective study of medical claims data for patients newly-diagnosed with type 2 diabetes (DM2), in order to better identify high risk populations who may benefit most from improved rates of screening.Methods Patients age 18 or older with newly-diagnosed DM2 from 2007-2015 were recruited from a commercial claims database. All patients were required to have continuous enrollment for 6 years: 1 year prior to and 5 years after the index diabetes diagnosis. 72,067 patients with newly-diagnosed DM2 were identified. Demographic data were collected. Patients were identified as having severe diabetic eye disease if they had an ICD-9 diagnosis code for proliferative diabetic retinopathy, vitreous hemorrhage, neovascular glaucoma, rubeosis iridis, tractional retinal detachment, or blindness, or a CPT code for pars plana vitrectomy, glaucoma drainage device, or panretinal photocoagulation within 5 years of diagnosis. Multivariable logistic regression was used to test associations between demographic factors and identification of severe disease.Results At 5 years following diagnosis with diabetes, 3.97% of patients had severe diabetic eye disease. Black patients were more likely than white patients to have severe disease (OR 1.26, p<.001). Age at diagnosis was positively correlated with severe diabetic eye disease, with patients diagnosed at ages 18-34 having the lowest risk (OR 0.75, p<.01) and those diagnosed after age 75 having the highest risk of severe disease (OR 1.63, p<.001). Smokers were more likely to have severe disease than non- smokers (OR 1.15, p<.02). Income and education level were not associated with early development of severe diabetic eye disease in this insured population.Conclusions A subset of patients with DM2 develop severe diabetic eye disease within the first 5 years after diagnosis. Black patients, older patients, and smokers were more likely to have severe diabetic eye disease, which suggests either a longer duration of undiagnosed disease, or poorer control of disease. These patients may require increased screening efforts.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details. Individualised Screening for Diabetic Retinopathy: the ISDR study. A Randomised Controlled Trial of Safety, Efficacy and Cost Effectiveness

AuthorBlock: Simon P. Harding1,2, Christopher Cheyne3, James Lathe7, Amu Wang1, Pilar Vasquez Arango1, Irene Mary Stratton6, Jiten Vora5, Mark Gabbay4, Marta Garcia Finana3, Marilyn James7, Deborah Broadbent2,1 1Eye and Vision Science, University of Liverpool, Liverpool, ENGLAND, United Kingdom; 2St. Paul's Eye Unit, Royal Liverpool University Hospital, Liverpool, United Kingdom; 3Department of Biostatistics, University of Liverpool, Liverpool, United Kingdom; 4Health Services Research, University of Liverpool, , United Kingdom; 5Diabetes and Endocrinology, Royal Liverpool University Hospital, , United Kingdom; 6Gloucestershire Retinal Research Group, Cheltenham General Hospital, , United Kingdom; 7Division of Rehabilitation, Ageing and Wellbeing, School of Medicine, University of Nottingham, , United Kingdom;

DisclosureBlock: Simon P. Harding, None; Christopher Cheyne, None; James Lathe, None; Amu Wang, None; Pilar Vasquez Arango, None; Irene Mary Stratton, None; Jiten Vora, None; Mark Gabbay, None; Marta Garcia Finana, None; Marilyn James, None; Deborah Broadbent, None;

Purpose Systematic screening to detect sight threatening diabetic retinopathy is widely established but with major variations in models of delivery. Varying screen intervals informed by personal risk offers better targeting including for high risk groups, addresses the increasing prevalence of diabetes and allows reallocation of resources. However safety data especially on extended intervals is minimal. We aimed to evaluate the safety, efficacy and cost effectiveness of individualised variable-interval risk-based screening in a population setting compared to usual annual screening.Methods Masked, 2 arm, parallel assignment, equivalence RCT (largest to date in screening) with independent trials unit monitoring in people with diabetes aged ≥12 years attending screening in a single English programme. Randomisation was 1:1 to individualised screening (active group; 6, 12 or 24 months for high, medium and low risk) determined by a risk calculation engine using real-time local demographic, retinal and clinical data, compared with annual screening (control). Cost effectiveness analysis measuring payer (NHS) and societal costs took a 2 year time horizon.Results 4534 participants entered the study - after withdrawals/loss to follow-up: active 2097; control 2224. Attendance rates at first follow up visit (primary outcome, safety) were equivalent (per protocol analysis, 5% margin): active 83.6%, control 84.7% (difference 1.0, 95% CI -3.2, 1.2). STDR detection rates (secondary safety) were non-inferior (1.5% margin): active 1.4% control 1.7% (difference -0.3, CI -1.1, 0.5). 43.2% fewer screening appointments (secondary, efficacy) were required in the individualised arm. Worsening of glyceamic control was not detected. Quality of life (EQ5D5L, HUI3) was not significantly different between the groups. Incremental cost savings per person were: NHS perspective £19.73 (18.28 to 21.16); societal perspective £26.19 (24.41 to 27.87).Conclusions All parties involved in diabetes care can be reassured that extended and personalised screening intervals can safely and cost-effectively be introduced in established screening programmes. Scale-up with further validation outside a research setting is recommended. Funding: NIHR UK. Views expressed are solely those of the authors. Trial registration ISRCTN 87561257Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details. The number of people worldwide with diabetes is rising rapidly. Screening to detect the associated retinopathy at a stage when visual impairment can be prevented is established in many countries and is typically done annually. A personalised approach varying the screening interval calculated from an individual's own risk factors offers longer intervals for people at low risk of sight threatening retinopathy and improved cost effectiveness.

Our randomised controlled study of 4534 people with diabetes, the largest to date in ophthalmology and screening, compared individualised variable-interval risk-based screening to annual screening in one programme in England. Individualised screening showed equivalent attendance rates, with no worsening of diabetes control, giving reassurance to all those involved in diabetes care that extending intervals is safe, at least in a research setting. More frequent screening in high risk groups improved detection rates. The large reductions in number of required appointments will reduce the burden for people with diabetes. Substantial cost savings can be made releasing resources to target high risk and hard to reach groups. Gene set enrichment analysis implicates vascular endothelial growth factor pathway in genetic risk of developing proliferative diabetic retinopathy

AuthorBlock: Gayatri Susarla5, Richard Jensen6, Albert V. Smith14, Mary Frances Cotch7, Brian Yaspan8, Donald Bowden9,10, Ronald Klein1, Barbara Klein1, Tien Y. Wong11,12, Jie Jin Wang11,13, Kathryn P. Burdon2, Xiaohui Li3, Sudha K. Iyengar4, Lucia Sobrin5, Ayellet Segrè5 1Department of Opthalmology and VIsual Sciences, University of Wisconsin, Wisconsin, United States; 2Menzies Institute for Medical Research, University of Tasmania, Hobart, Tasmania, Australia; 3Institute for Translational Genomics and Population Sciences, LABioMed and Department of Pediatrics at Harbor-UCLA Medical Center, Torrance, California, United States; 4Department of Population and Quantitative Health Sciences, Case Western University, Cleveland, Cleveland, Ohio, United States; 5Department of Ophthalmology, Harvard Medical School, Massachusetts Eye and Ear Inﬁrmary, Boston, Massachusetts, United States; 6Cardiovascular Health Research Unit, Department of Medicine, Epidemiology and Health Services, University of Washington, Seattle, Washington, United States; 7Division of Epidemiology and Clinical Applications, National Eye Institute, National Institutes of Health, Bethesda, Maryland, United States; 8Genentech Inc., South San Francisco, California, United States; 9Wake Forest School of Medicine, Center for Genomics and Personalized Medicine Research, Winston-Salem, North Carolina, United States; 10Wake Forest School of Medicine, Department of Biochemistry, Winston-Salem, North Carolina, United States; 11Duke-NUS Medical School, , Singapore; 12Department of Ophthalmology, Yong Loo Lin School of Medicine, National University of Singapore, , Singapore; 13Centre for Vision Research, Westmead Institute for Medical Research, University of Sydney, Sydney, New South Wales, Australia; 14University of Michigan, Michigan, United States;

DisclosureBlock: Gayatri Susarla, None; Richard Jensen, None; Albert V. Smith, None; Mary Frances Cotch, None; Brian Yaspan, Genentech Inc. Code E (Employment), Donald Bowden, None; Ronald Klein, None; Barbara Klein, None; Tien Y. Wong, None; Jie Jin Wang, None; Kathryn P. Burdon, None; Xiaohui Li, None; Sudha K. Iyengar, None; Lucia Sobrin, None; Ayellet Segrè, None;

Purpose To gain statistical discovery power and biological insight into the genetic architecture of diabetic retinopathy (DR), we tested for enrichment of multiple genetic variant associations with DR in gene sets or pathways related to DR pathophysiology, rather than inspecting variants individually.Methods We applied the gene set enrichment analysis (GSEA) method, Meta-Analysis Gene-set Enrichment of variaNT Associations (MAGENTA), to four previously published DR genome-wide association study (GWAS) analyses adjusted for duration of diabetes and glycemic control. These GWAS were performed separately in patients of African American (AA) and European (EU) ancestry, using one of two DR case-control definitions: (1) PDR analysis: cases (n=1495) had proliferative DR (PDR) and controls (n=4362) had any level of non-proliferative DR or no DR and (2) Extremes of DR analysis: cases (n=1495) had PDR and controls (n=2911) had no DR. From a literature review, we identified key DR- associated biological processes including angiogenesis, glial dysregulation, neuronal dysfunction, and inflammation. Gene sets representing these pathways were extracted from four gene set databases: Gene Ontology (n=144 pathways extracted), Kyoto Encyclopedia of Genes and Genomes (KEGG, n=13 pathways extracted), Reactome (n=10 pathways extracted), and Mouse Phenotype Ontology (MGI, n=79 pathways extracted). Multiple hypothesis correction based on false discovery rate (FDR) was applied to gene sets from each database separately.Results MAGENTA identified one statistically significant pathway from KEGG, the vascular endothelial growth factor (VEGF) signaling pathway, for the PDR analysis in the AA population. Significance was based on a 75th percentile enrichment cutoff (Gene set enrichment P=0.0015, FDR=0.0159). This analysis suggests 11 new DR-associated genes (1.6 fold-enrichment) based on their confounder-adjusted gene association P values, with the variant DR P values ranging between 2x10-5 and 0.001. No significant enrichment was found among these genes in the EU extremes of DR analysis.Conclusions Genetic variation in the VEGF signaling pathway may be associated with PDR risk in the AA population. Replication of this finding with alternate GSEA analysis methods and in independent DR GWAS are needed to confirm these results and to determine if the significance is AA-specific.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details. A Multi-Task Deep Learning System to Detect Diabetic Macular Edema for Different Optical Coherence Tomography Devices

AuthorBlock: Carol Yim-lui Cheung1, Fangyao Tang1, Xi Wang2, Shibo Tang3, Hongjie Ma3, Gerald Liew4, Bamini Gopinath4, Theodore Leng5, Schahrouz Kakavand5, Suria Mannil5, Robert Chang5, Haoyu Chen6, Haifang Huang6, Hao Chen2, Pheng Ann Heng2 1Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong, Hong Kong; 2Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, China; 3Aier School of Ophthalmology,, Central South University, , China; 4Ophthalmology, University of Sydney, New South Wales, Australia; 5Byers Eye Institute at Stanford, Stanford University School of Medicine, California, United States; 6Joint Shantou International Eye Center, , China;

DisclosureBlock: Carol Yim-lui Cheung, None; Fangyao Tang, None; Xi Wang, None; Shibo Tang, None; Hongjie Ma, None; Gerald Liew, None; Bamini Gopinath, None; Theodore Leng, None; Schahrouz Kakavand, None; Suria Mannil, None; Robert Chang, None; Haoyu Chen, None; Haifang Huang, None; Hao Chen, None; Pheng Ann Heng, None;

Purpose To develop and validate an artificial intelligence (AI) based approach that can effectively identify diabetic macular edema (DME) and non-DME retinal abnormalities imaged by different OCT devices and protocols.Methods Three retrospective macular OCT datasets (Cirrus HD-OCT, DRI-OCT Triton, Spectralis OCT) from individuals with diabetes were used to develop the AI deep learning system (DLS). The scanning protocols used for Cirrus, Triton and Spectralis were 6mm*6mm 3D cube (128 B-scans, 512 A-scans each over 1024 samplings), radial 9mm*30degree (12 B-scans, 1024 A-scans each), and 6.3mm*6.3mm (25 B-scans, 1024 A-scans each) or 6.5mm*4.9mm (19 B-scan, 1024 A-scan each), respectively. All OCT scans were labeled as yes/no any DME by professional graders. Among the DME cases, the OCT scans were further labelled as center-involved/non-center involved DME, according to DRCR.net protocol-defined thresholds. The scans were also labeled as yes/no non-DME retinal abnormalities. We used 3D ResNet34 and 2D ResNet18 to build multi-task deep networks to perform two tasks simultaneously for different types of OCT, one for classifying OCT images into normal, non- center involved DME or center-involved DME (ternary classification), and the other for classifying yes/no non-DME retinal abnormalities (binary classification). Five external independent datasets from different countries were further used to test the performance of the DLS independently.Results A total of 3,790 Cirrus cubes, 38,976 Triton B-scans and 30,468 Spectralis B-scans were used for training (60%), validation (20%) and primary testing (20%), and a total of 856 Cirrus cubes, 2331 Triton B-scans and 8,806 Spectralis B-scans were used for external testing. In the primary testing, the accuracies of the ternary classification were 95.5%, 95.0% and 95.7%, the AUCs of the binary classification were 0.881, 0.846 and 0.975, for Cirrus, Triton and Spectralis OCTs, respectively. In the external testing, the accuracies of the ternary classification were >87%, 94%, and >95%, the AUCs of the binary classification were >0.8, 0.93 and >0.97, for Cirrus, Triton and Spectralis OCTs, respectively.Conclusions The proposed multi-task DLS could detect DME and non-DME retinal abnormalities from different OCT devices with high accuracy, which may be a very useful DME screening tool that could save resources and speed up workflow substantially.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details. Accurate identification of diabetic macular edema (DME) and timely treatment can prevent vision loss. Optical coherence tomography (OCT) is widely used as a tool to detect and manage DME. Nevertheless, current OCT classification does not provide information regarding specific eye diseases. Interpretation of OCT data for DME assessment still requires expertise from ophthalmologists, particular from retinal specialists. This is prohibitively cost-ineffective, and also practically impossible under present circumstances due to the lack of experienced expert manpower to perform screening. Deep learning, a branch of artificial intelligence (AI) machine learning algorithms, has recently been optimized to handle image classification problems at or above human level performance once trained on big data. We developed and validateed an AI deep learning algorithm using OCT images for the detection of DME with further classification into center-involved DME versus noncenterl-involved DME. Our proposed algorithm will be used as a screening tool to identify DME more accurately under current screening workflow which will tremendously reduce the amount of manpower required in reading OCT images. Prevalence and Predictors of Depression in Patients with Diabetic Retinopathy in a Nationally Representative Sample

AuthorBlock: Yicheng Bao1, Skyler Cope1, Monica Gaddis1, Betty Drees1 1UMKC School of Medicine, Kansas City, Missouri, United States;

DisclosureBlock: Yicheng Bao, None; Skyler Cope, None; Monica Gaddis, None; Betty Drees, None;

Purpose Vision loss from diabetic eye disease exerts significant psychosocial stress, including disrupted roles and relationships, social isolation, and dependence. As a result, the rate of depression in those with diabetic retinopathy (DR) is higher than those without. However, the predictors of depression in DR have not been well characterized. Here, we investigate the prevalence and predictors of depression in participants with DR.Methods Data was gathered from the National Health and Nutrition Examination Survey 2005-2008. This study included participants age ≥40, who underwent fundus photography and Patient Health Questionnaire (PHQ)-9. PHQ-9 score ≥10 determined the presence of depression. Self-reported measures of visual function were measured from the Visual Function Questionnaire (VFQ-25). Multivariable logistic regression was used to evaluate whether DR was a significant risk factor for depression and to evaluate the risk factors for depression in those with DR.Results A total of 5704 participants, 47% male, 523% female, and mean age 56.5 years were included in this study. Persons with moderate, severe non-proliferative diabetic retinopathy (NPDR), or proliferative retinopathy (PDR) had higher prevalence of depression than participants with mild retinopathy or no retinopathy (14.3% vs 6.9% vs 7.0%). In a multivariable logistic regression model controlling for age, ethnicity, and gender, moderate to severe NPDR or PDR (OR: 2.21 (95% CI: 1.02, 4.78), p=0.04) was associated with greater odds of depression when compared with persons without retinopathy. Among persons with DR, best corrected visual acuity (BCVA) (p=0.67) and HbA1c (p=0.97) were not associated with depression after controlling for age, ethnicity, and gender. However, self-reported measures of vision were predictive of depression: poor or very poor visual function (OR: 9.45 (95% CI: 2.34, 37.24), p=0.002), vision limits activities most of the time or all the time (OR: 7.09 (95% CI: 1.35 37.20), p=0.022).Conclusions A significant proportion of patients with DR in the NHANES population had co-morbid major depression. Objective measures of visual function were not predictive of depression in those with DR, while subjective, self-reported measures of visual function were highly predictive of depression, suggesting that subjective measures are a better determinant of low mood and poor functional status than BCVA.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details. Diabetic Eye Disease of American Indians/Alaska Natives, 2016-2019, as Found with Ultra-Wide-Field Imaging

AuthorBlock: Stephanie Fonda2, Drew Lewis2, Sven-Erik Bursell1 1Telehealth Research Institute, John A. Burns School of Medicine, University of Hawaii, Honolulu, Hawaii, United States; 2Estenda Solutions, Pennsylvania, United States;

DisclosureBlock: Stephanie Fonda, None; Drew Lewis, None; Sven-Erik Bursell, None;

Purpose Recently the Indian Health Service-Joslin Vision Network (IHS-JVN) teleophthalmology program has transitioned to macula-centered, 200o field-of-view (FOV), ultra-wide-field imaging (UWFI) from multi- field, nonmydriatic, 45o FOV fundus photography (NMFP). This project analyzed the prevalence rates of diabetic retinopathy (DR) and macula edema (DME) from UWFI and compared these rates to those previously reported when NMFP was the predominant technology.Methods The sample was composed of American Indian and Alaskan Natives (AI/AN) with diabetes served by the nationally-distributed (n=100 clinics) and clinically validated IHS-JVN from 01Nov2016 to 31Oct2019 (n=53,000). Patients were recruited during primary care visits. All patients were imaged using UWFI. Patients’ first available imaging study over the time period was included in this analysis. All images were evaluated by a certified optometrist or ophthalmologist using validated protocols to identify the severity levels of non-proliferative diabetic retinopathy (NPDR), proliferative diabetic retinopathy (PDR), and DME. The analysis included sight-threatening retinopathy (STR), defined as presence of severe NPDR, any PDR, or any DME.Results Gradable images were achieved in 95% of patients. Prevalence of any NPDR, PDR, any DME, and STR was 28.6%, 2.8%, 3.0% and 5.3%, respectively. These rates of any DR and STR are higher than those from a recent previous report predominantly based on NMFP (DR=17.7% and STR=4.2%); however, the present rates are consistent with a subset analysis of UWFI images in the recent previous report (DR=28.2% and STR=5.4%). The higher rate of any DR from UWFI was associated with the clinicians’ ability to more easily identify lesions found in mild and moderate DR; e.g., in the present analysis, 22.3% of diagnosed mild NPDR had findings outside standard ETDRS fields. Notably, findings peripheral to ETDRS were noted for 2.2% of patients for whom DR was still clinically considered ‘absent’.Conclusions Replacement of NMFP with UWFI in the IHS-JVN has led to higher gradable rates and more complete prevalence data on diabetic eye disease in AI/AN. The present analysis in this high-risk population indicates prevalence is higher than found by NMFP alone.Layman Abstract (optional): Provide a 50-200 word description of your work that non-scientists can understand. Describe the big picture and the implications of your findings, not the study itself and the associated details.