Diabetic Eye Disease

Cui Y, Zhu Y, Wang JC, Lu Y, Zeng R, Katz R, Wu DM, Vavvas DG, Husain D, Miller JW, Kim LA, Miller JB. Imaging Artifacts and Segmentation Errors With Wide-Field Swept-Source Optical Coherence Tomography Angiography in Diabetic Retinopathy. Transl Vis Sci Technol 2019;8(6):18.Abstract
Purpose: To analyze imaging artifacts and segmentation errors with wide-field swept-source optical coherence tomography angiography (SS-OCTA) in diabetic retinopathy (DR). Methods: We conducted a prospective, observational study at Massachusetts Eye and Ear from December 2018 to March 2019. Proliferative diabetic retinopathy (PDR), nonproliferative diabetic retinopathy (NPDR), diabetic patients with no diabetic retinopathy (DR), and healthy control eyes were included. All patients were imaged with a SS-OCTA and the Montage Angio (15 × 9 mm) was used for analysis. Images were independently evaluated by two graders using the motion artifact score (MAS). All statistical analyses were performed using SPSS 25.0 and R software. Results: One hundred thirty-six eyes in 98 participants with the montage image were included in the study. Patients with more severe stages of DR had higher MAS by trend test analysis ( < 0.05). The occurrence of segmentation error was 0% in the healthy group, 10.53% in the no DR group, 10.00% in the NPDR group, and 50% in the PDR group. Multivariate regression analysis showed that the severity of DR and dry eye were the major factors affecting MAS ( < 0.05). There were some modifiable artifacts that could be corrected to improve image quality. Conclusions: Wide field SS-OCTA assesses retinal microvascular changes by noninvasive techniques, yet distinguishing real alterations from artifacts is paramount to accurate interpretations. DR severity and dry eye correlated with MAS. Translational Relevance: Understanding contributing factors and methods to reduce artifacts is critical to routine use and clinical trial with wide-field SS-OCTA.
Muqit MMK, Kourgialis N, Jackson-deGraffenried M, Talukder Z, Khetran ER, Rahman A, Chan WO, Chowdury FA, Nag D, Ahmad J, Friedman DS. Trends in Diabetic Retinopathy, Visual Acuity, and Treatment Outcomes for Patients Living With Diabetes in a Fundus Photograph-Based Diabetic Retinopathy Screening Program in Bangladesh. JAMA Netw Open 2019;2(11):e1916285.Abstract
Importance: Diabetic retinopathy (DR) is the leading cause of low vision among working-age adults. An estimated 6.9 million people in Bangladesh were living with diabetes in 2017, which is projected to increase to more than 10 million people in 2025. Currently, no standardized and/or large-scale DR screening program exists in Bangladesh. Objective: To develop a novel fundus photograph-based eye screening model for early detection of DR to prevent vision loss in Bangladeshi individuals with diabetes. Design, Setting, and Participants: In this cross-sectional study, 49 264 patients with diabetes underwent opportunistic eye screening at 2 eye hospitals and 1 diabetic hospital in Bangladesh between June 1, 2010, and September 30, 2017. The data set was analyzed from April 8 to December 30, 2018. Technicians were trained to obtain 2-field digital fundus photographs and to grade each according to a standardized DR severity scale. Each patient was counseled and triaged for treatment using defined DR referral criteria. Main Outcomes and Measures: Primary DR grading outcomes, visual acuity, and treatment outcomes. Results: A total of 49 264 patients (54.3% male; mean [SD] age, 50.8 [12.3] years) underwent DR screening during a 7-year period. The DR prevalence rate across all 3 sites was 33% (95% CI, 33%-33%). Prevalence rates varied by center (Chittagong, 64.6% [95% CI, 64.0%-65.0%]; Dhaka, 39.8% [95% CI, 39.0%-41.0%]; and Feni, 13.0% [95% CI, 13.0%-14.0%]). Across all age groups, male patients were at higher risk of prevalent DR than female patients (odds ratio, 1.99; 95% CI, 1.90-2.07). The prevalence was 3.9% for preproliferative DR, 7.8% for proliferative DR, and 19.2% for maculopathy. Individuals with DR had significantly worse visual acuity than those with no DR (best-corrected visual acuity, 0.35 vs 0.21 logMAR; P < .001). The rate of moderate visual impairment was 12.2%, and the rate of blindness was 2.5%. Primary treatments included laser photocoagulation (n = 1637), intravitreal injection (n = 1440), and vitrectomy (n = 309). Conclusions and Relevance: Screening Bangladeshi individuals known to have diabetes using fundus photography identified large numbers of patients with sight-threatening proliferative DR, maculopathy, and visual impairment or blindness. Expansion of eye screening services in Bangladesh is warranted as part of a national government eye care and diabetes health policy.
Wang Y, Lin Z, Wen L, Rong SS, Ding XX, Li D, Feng KM, Wang FH, Liang YB, Zhai G. Rationale, Design, Methodology and Baseline Data of Fushun Diabetic Retinopathy Cohort Study (FS-DIRECT). Ophthalmic Epidemiol 2019;:1-10.Abstract
: To describe the rationale, design, methodology and baseline characteristics of Fushun Diabetic Retinopathy Cohort Study (FS-DIRECT), a community-based prospective cohort study in patients with type 2 diabetes mellitus (T2DM) living in northeast China.: Patients with T2DM, aged 30 years and above from communities of Fushun city, Liaoning province, China, were recruited. The presence and severity of the diabetic retinopathy (DR) were determined by a modified Early Treatment Diabetic Retinopathy Study (ETDRS) retinopathy scale of 6 fields fundus photographs. Detailed ocular examinations and questionnaires were collated, in addition to blood and urine sample collection.: Of the 2224 subjects eligible for the FS-DIRECT, 2033 (91.4%) participated in the study. The majority of participants were female (58.9%), the average age was 62.1 ± 9.1 years. The overall prevalence rates of DR, non-proliferative DR, proliferative DR, diabetic macular edema, and vision-threatening retinopathy were 44.3%, 40.0%, 4.3%, 15.2%, and 12.3%, respectively. Compared to the patients without DR, patients with DR had lower income, an earlier onset of diabetes, a longer duration of diabetes, higher proportion of insulin use, higher fasting plasma glucose, HbA1c, systolic blood pressure, total cholesterol and high density lipoprotein, as well as a higher level of urine protein (all < .05).: The baseline data of FS-DIRECT showed a high prevalence of DR in a community of northeast China. Further investigation will provide key information about the risk factors, impact, and trends of DR in this region.
McKay TB, Priyadarsini S, Karamichos D. Mechanisms of Collagen Crosslinking in Diabetes and Keratoconus. Cells 2019;8(10)Abstract
Collagen crosslinking provides the mechanical strength required for physiological maintenance of the extracellular matrix in most tissues in the human body, including the cornea. Aging and diabetes mellitus (DM) are processes that are both associated with increased collagen crosslinking that leads to increased corneal rigidity. By contrast, keratoconus (KC) is a corneal thinning disease associated with decreased mechanical stiffness leading to ectasia of the central cornea. Studies have suggested that crosslinking mediated by reactive advanced glycation end products during DM may protect the cornea from KC development. Parallel to this hypothesis, riboflavin-mediated photoreactive corneal crosslinking has been proposed as a therapeutic option to halt the progression of corneal thinning by inducing intra- and intermolecular crosslink formation within the collagen fibrils of the stroma, leading to stabilization of the disease. Here, we review the pathobiology of DM and KC in the context of corneal structure, the epidemiology behind the inverse correlation of DM and KC development, and the chemical mechanisms of lysyl oxidase-mediated crosslinking, advanced glycation end product-mediated crosslinking, and photoreactive riboflavin-mediated corneal crosslinking. The goal of this review is to define the biological and chemical pathways important in physiological and pathological processes related to collagen crosslinking in DM and KC.
Hutton DW, Stein JD, Glassman AR, Bressler NM, Jampol LM, Sun JK, Sun JK. Five-Year Cost-effectiveness of Intravitreous Ranibizumab Therapy vs Panretinal Photocoagulation for Treating Proliferative Diabetic Retinopathy: A Secondary Analysis of a Randomized Clinical Trial. JAMA Ophthalmol 2019;:1-9.Abstract
Importance: The DRCR Retina Network Protocol S randomized clinical trial suggested that the mean visual acuity of eyes with proliferative diabetic retinopathy (PDR) treated with ranibizumab is not worse at 5 years than that of eyes treated with panretinal photocoagulation (PRP). Moreover, the ranibizumab group had fewer new cases of diabetic macular edema (DME) with vision loss or vitrectomy but had 4 times the number of injections and 3 times the number of visits. Although 2-year cost-effectiveness results of Protocol S were previously identified, incorporating 5-year data from Protocol S could alter the longer-term cost-effectiveness of the treatment strategies from the perspective of the health care system. Objective: To evaluate 5- and 10-year cost-effectiveness of therapy with ranibizumab, 0.5 mg, compared with PRP for treating PDR. Design, Setting, and Participants: A preplanned secondary analysis of the Protocol S randomized clinical trial using efficacy, safety, and resource utilization data through 5 years of follow-up for 213 adults diagnosed with PDR and simulating results through 10 years. Interventions: Intravitreous ranibizumab, 0.5 mg, at baseline and as frequently as every 4 weeks based on a structured retreatment protocol vs PRP at baseline for PDR; eyes in both groups could receive ranibizumab for concomitant DME with vision loss. Main Outcomes and Measures: Incremental cost-effectiveness ratios (ICERs) of ranibizumab therapy compared with PRP were evaluated for those with and without center-involved DME (CI-DME) and vision loss (Snellen equivalent, 20/32 or worse) at baseline. Results: The study included 213 adults with a mean (SD) age of 53 (12) years, of whom 92 (43%) were women and 155 (73%) were white. The ICER of the ranibizumab group compared with PRP for patients without CI-DME at baseline was $582 268 per quality-adjusted life-year (QALY) at 5 years and $742 202/QALY at 10 years. For patients with baseline CI-DME, ICERs were $65 576/QALY at 5 years and $63 930/QALY at 10 years. Conclusions and Relevance: This study suggests that during 5 to 10 years of treatment, ranibizumab, 0.5 mg, as given in the studied trial compared with PRP may be within the frequently cited range considered cost-effective in the United States for eyes presenting with PDR and vision-impairing CI-DME, but not for those with PDR but without vision-impairing CI-DME. Substantial reductions in anti-vascular endothelial growth factor cost may make the ranibizumab therapy cost-effective within this range even for patients without baseline CI-DME. Trial Registration: ClinicalTrials.gov identifier: NCT01489189.
Sun JK, Jampol LM. The Diabetic Retinopathy Clinical Research Network (DRCR.net) and Its Contributions to the Treatment of Diabetic Retinopathy. Ophthalmic Res 2019;:1-6.Abstract
Over the past two decades, the Diabetic Retinopathy Clinical Research Network (now known as the DRCR Retina Network) has contributed to multiple and substantial advances in the clinical care of diabetic eye disease. Network studies helped establish anti-vascular endothelial growth factor (VEGF) agents as an effective alternative to panretinal photocoagulation for eyes with proliferative diabetic retinopathy (PDR) and as first-line therapy for eyes with visual impairment for diabetic macular edema (DME), defined treatment algorithms for the use of intravitreal medications in these conditions, and provided critical data to understand how to better evaluate the diabetic eye using optical coherence tomography and other imaging modalities. Ongoing DRCR.net studies will address whether anti-VEGF therapy is effective at preventing vision-threatening complications in eyes with severe non-proliferative diabetic retinopathy, if photobiomodulation has a beneficial effect in eyes with DME, and whether initiation of DME treatment with bevacizumab and rescue with aflibercept can provide visual outcomes as good as those achieved with aflibercept alone. Future plans for the Network also include the expansion into non-diabetic eye disease in areas such as age-related macular degeneration.
Busch C, Fraser-Bell S, Iglicki M, Lupidi M, Couturier A, Chaikitmongkol V, Giancipoli E, Rodríguez-Valdés PJ, Gabrielle P-H, Laíns I, Santos AR, Cebeci Z, Amphornphruet A, Degenhardt V, Unterlauft J-D, Cagini C, Mané-Tauty V, Ricci GD'A, Hindi I, Agrawal K, Chhablani J, Loewenstein A, Zur D, Rehak M, Rehak M. Real-world outcomes of non-responding diabetic macular edema treated with continued anti-VEGF therapy versus early switch to dexamethasone implant: 2-year results. Acta Diabetol 2019;56(12):1341-1350.Abstract
AIMS: To provide 2-year follow-up data on eyes with diabetic macular edema (DME) that were non-responsive after three initial anti-vascular endothelial growth factor (VEGF) injections, comparing functional and anatomical outcomes under continued anti-VEGF therapy versus dexamethasone (DEX) implant. METHODS: Multicenter, retrospective chart review comparing eyes with treatment-naïve DME and a suboptimal response to a loading phase of anti-VEGF therapy (3 injections given monthly) which were then treated with (a) further anti-VEGF (n = 72) or (b) initially switched to DEX implant (n = 38). Main outcome measures were change in visual acuity (VA) and central subfield thickness (CST) from the end of the loading phase to 24 months. RESULTS: In 79% of the 12-month study population (87/110 eyes), 24-month data were available. One quarter of eyes in each group switched treatments during the second year. Eyes that were switched early to DEX implant maintained the functional and anatomical improvements at 24 months which were seen in the first year (from month 3: + 8.9 letters, - 214 µm). Eyes that were switched from anti-VEGF therapy to steroids in the second year improved VA and reduced CST at 24 months (from month 12: + 6.8 letters, p = 0.023; - 226 µm, p = 0.004). In eyes continued on anti-VEGF therapy, VA and CST were stable at 24 months (from month 3: + 2.8 letters, p = 0.254; - 24 µm, p = 0.243). Eyes that were non-responsive to anti-VEGF therapy for 12 months had similar chances to experience a VA gain from further therapy as eyes that were non-responsive for 3 months only (23.8 vs. 31.0%, p = 0.344). CONCLUSIONS: The beneficial effect of an early switch to DEX implant in DME non-responders seen at month 12 was maintained during the second year. A later switch from anti-VEGF to steroids still provided significant improvement. Eyes continued on anti-VEGF over a period of 24 months maintained vision. A quarter of eyes, which had not improved vision at 12 months, exhibited a delayed response to treatment.
Yokomizo H, Maeda Y, Park K, Clermont AC, Hernandez SL, Fickweiler W, Li Q, Wang C-H, Paniagua SM, Simao F, Ishikado A, Sun B, Wu I-H, Katagiri S, Pober DM, Tinsley LJ, Avery RL, Feener EP, Kern TS, Keenan HA, Aiello LP, Sun JK, King GL. Retinol binding protein 3 is increased in the retina of patients with diabetes resistant to diabetic retinopathy. Sci Transl Med 2019;11(499)Abstract
The Joslin Medalist Study characterized people affected with type 1 diabetes for 50 years or longer. More than 35% of these individuals exhibit no to mild diabetic retinopathy (DR), independent of glycemic control, suggesting the presence of endogenous protective factors against DR in a subpopulation of patients. Proteomic analysis of retina and vitreous identified retinol binding protein 3 (RBP3), a retinol transport protein secreted mainly by the photoreceptors, as elevated in Medalist patients protected from advanced DR. Mass spectrometry and protein expression analysis identified an inverse association between vitreous RBP3 concentration and DR severity. Intravitreal injection and photoreceptor-specific overexpression of RBP3 in rodents inhibited the detrimental effects of vascular endothelial growth factor (VEGF). Mechanistically, our results showed that recombinant RBP3 exerted the therapeutic effects by binding and inhibiting VEGF receptor tyrosine phosphorylation. In addition, by binding to glucose transporter 1 (GLUT1) and decreasing glucose uptake, RBP3 blocked the detrimental effects of hyperglycemia in inducing inflammatory cytokines in retinal endothelial and Müller cells. Elevated expression of photoreceptor-secreted RBP3 may have a role in protection against the progression of DR due to hyperglycemia by inhibiting glucose uptake via GLUT1 and decreasing the expression of inflammatory cytokines and VEGF.
Bressler NM, Odia I, Maguire M, Glassman AR, Jampol LM, MacCumber MW, Shah C, Rosberger D, Sun JK, Sun JK. Association Between Change in Visual Acuity and Change in Central Subfield Thickness During Treatment of Diabetic Macular Edema in Participants Randomized to Aflibercept, Bevacizumab, or Ranibizumab: A Post Hoc Analysis of the Protocol T Randomized Clinic. JAMA Ophthalmol 2019;Abstract
Importance: The determination of optical coherence tomography (OCT) central subfield thickness (CST) is an objective measure, and visual acuity (VA) is a subjective measure. Therefore, using OCT CST changes as a surrogate for VA changes in diabetic macular edema seems reasonable. However, studies suggest that change in OCT CST following anti-vascular endothelial growth factor (anti-VEGF) treatment for diabetic macular edema is correlated with changes in VA but varies substantially among individuals, and so may not be a good surrogate for changes in VA. Objective: To determine associations between changes in VA and changes in OCT CST across 3 anti-VEGF agents (aflibercept, bevacizumab, or ranibizumab) used in a randomized clinical trial for diabetic macular edema. Design, Setting, and Participants: Post hoc analyses were conducted of DRCR Retina Network Protocol T among 652 of 660 participants (98.8%) meeting inclusion criteria for this investigation. The study was conducted between August 22, 2012, and September 23, 2015. The post hoc data collection and analysis were performed from May 29 to July 11, 2018. Interventions: Six monthly intravitreous anti-VEGF injections (unless success was achieved after 3-5 months) were administered; subsequent injections or focal/grid laser photocoagulation treatments were given as needed per protocol to achieve stability. Main Outcomes and Measures: Association between changes in VA letter score with changes in CST at 12, 52, and 104 weeks after randomization to aflibercept, bevacizumab, or ranibizumab. Results: Of the 652 participants, 304 were women (46.6%); median age was 61 years (interquartile range, 54-67 years). The correlation between CST and VA at the follow-up visits was 0.24 (95% CI, 0.16-0.31) in 616 patients at 12 weeks, 0.31 (95% CI, 0.24-0.38) in 609 patients at 52 weeks, and 0.23 (95% CI, 0.15-0.31) in 566 patients at 104 weeks. The correlation coefficients of change in VA vs change in OCT CST for these time intervals were 0.36 (95% CI, 0.29-0.43) at 12 weeks, 0.36 (95% CI, 0.29-0.43) at 52 weeks, and 0.33 (95% CI, 0.26-0.41) at 104 weeks. Conclusions and Relevance: Changes in CST appear to account for only a small proportion of the total variation in changes in VA. These findings do not support using changes in OCT CST as a surrogate for changes in VA in phase 3 clinical trials evaluating anti-VEGF for diabetic macular edema or as a guide to inform the physician or patient about changes in VA after anti-VEGF treatment. Trial Registration: ClinicalTrials.gov identifier: NCT01627249.
AbdelAl O, Ashraf M, Sampani K, Sun JK. "For Mass Eye and Ear Special Issue" Adaptive Optics in the Evaluation of Diabetic Retinopathy. Semin Ophthalmol 2019;:1-9.Abstract
Retinal imaging is a fundamental tool for clinical and research efforts in the evaluation and management of diabetic retinopathy. Adaptive optics (AO) is an imaging technique that enables correction of over 90% of the optical aberrations of an individual eye induced primarily by the tear film, cornea and lens. The two major tasks of any AO system are to measure the optical imperfections of the eye and to then compensate for these aberrations to generate a corrected wavefront of reflected light from the eye. AO scanning laser ophthalmoscopy (AOSLO) provides a theoretical lateral resolution limit of 1.4 μm, allowing the study of microscopic features of the retinal vascular and neural tissue. AOSLO studies have revealed irregularities of the photoreceptor mosaic, vascular loss, and details of vascular lesions in diabetic eyes that may provide new insight into development, regression, and response to therapy of diabetic eye disease.
Laville V, Kang JH, Cousins CC, Iglesias AI, Nagy R, Cooke Bailey JN, Igo RP, Song YE, Chasman DI, Christen WG, Kraft P, Rosner BA, Hu F, Wilson JF, Gharahkhani P, Hewitt AW, Mackey DA, Hysi PG, Hammond CJ, van Duijn CM, Haines JL, Vitart V, Fingert JH, Hauser MA, Aschard H, Wiggs JL, Khawaja AP, Macgregor S, Pasquale LR, UK Biobank, International Glaucoma Genetics Consortium NEIGHBORHOODC. Genetic correlations between diabetes and glaucoma: an analysis of continuous and dichotomous phenotypes. Am J Ophthalmol 2019;Abstract
PURPOSE: A genetic correlation is the proportion of phenotypic variance between traits that is shared on a genetic basis. Here we explore genetic correlations between diabetes- and glaucoma-related traits. DESIGN: Cross-sectional study. METHODS: We assembled genome-wide association study summary statistics from European-derived participants regarding diabetes-related traits like fasting blood sugar (FBS) and type 2 diabetes (T2D) and glaucoma-related traits (intraocular pressure (IOP), central corneal thickness (CCT), corneal hysteresis (CH), corneal resistance factor (CRF), cup-disc ratio (CDR), and primary open-angle glaucoma (POAG)). We included data from the National Eye Institute Glaucoma Human Genetics Collaboration Heritable Overall Operational Database, the UK Biobank and the International Glaucoma Genetics Consortium. We calculated genetic correlation (r) between traits using linkage disequilibrium score regression. We also calculated genetic correlations between IOP, CCT and selected diabetes-related traits based on individual level phenotype data in two Northern European population-based samples using pedigree information and Sequential Oligogenic Linkage Analysis Routines (SOLAR). RESULTS: Overall, there was little r between diabetes- and glaucoma-related traits. Specifically, we found a non-significant negative correlation between T2D and POAG (r=-0.14; p=0.16). Using SOLAR, the genetic correlations between measured IOP, CCT, FBS, fasting insulin and hemoglobin A1c, were null. In contrast, genetic correlations between IOP and POAG (r ≥0.45; p≤3.0E-04) and between CDR and POAG were high (r =0.57; p=2.8E-10). However, genetic correlations between corneal properties (CCT, CRF and CH) and POAG were low (r range: -0.18 - 0.11) and non-significant (p≥0.07). CONCLUSION: These analyses suggest there is limited genetic correlation between diabetes- and glaucoma-related traits.
Baker CW, Glassman AR, Beaulieu WT, Antoszyk AN, Browning DJ, Chalam KV, Grover S, Jampol LM, Jhaveri CD, Melia M, Stockdale CR, Martin DF, Sun JK, Sun JK. Effect of Initial Management With Aflibercept vs Laser Photocoagulation vs Observation on Vision Loss Among Patients With Diabetic Macular Edema Involving the Center of the Macula and Good Visual Acuity: A Randomized Clinical Trial. JAMA 2019;Abstract
Importance: Intravitreous injections of antivascular endothelial growth factor agents are effective for treating diabetic macular edema (DME) involving the center of the macula (center-involved DME [CI-DME]) with visual acuity impairment (20/32 or worse). The best approach to treating patients with CI-DME and good visual acuity (20/25 or better) is unknown. Objective: To compare vision loss at 2 years among eyes initially managed with aflibercept, laser photocoagulation, or observation. Design, Setting, and Participants: Randomized clinical trial conducted at 91 US and Canadian sites among 702 adults with type 1 or type 2 diabetes. Participants had 1 study eye with CI-DME and visual acuity of 20/25 or better. The first participant was randomized on November 8, 2013, and the final date of follow-up was September 11, 2018. Interventions: Eyes were randomly assigned to 2.0 mg of intravitreous aflibercept (n = 226) as frequently as every 4 weeks, focal/grid laser photocoagulation (n = 240), or observation (n = 236). Aflibercept was required for eyes in the laser photocoagulation or observation groups that had decreased visual acuity from baseline by at least 10 letters (≥ 2 lines on an eye chart) at any visit or by 5 to 9 letters (1-2 lines) at 2 consecutive visits. Main Outcomes and Measures: The primary outcome was at least a 5-letter visual acuity decrease from baseline at 2 years. Antiplatelet Trialists' Collaboration adverse events (defined as myocardial infarction, stroke, or vascular or unknown death) were reported. Results: Among 702 randomized participants (mean age, 59 years; 38% female [n=264]), 625 of 681 (92% excluding deaths) completed the 2-year visit. For eyes with visual acuity that decreased from baseline, aflibercept was initiated in 25% (60/240) and 34% (80/326) in the laser photocoagulation and observation groups, respectively. At 2 years, the percentage of eyes with at least a 5-letter visual acuity decrease was 16% (33/205), 17% (36/212), and 19% (39/208) in the aflibercept, laser photocoagulation, and observation groups, respectively (aflibercept vs laser photocoagulation risk difference, -2% [95% CI, -9% to 5%]; relative risk, 0.88 [95% CI, 0.57-1.35; P = .79]; aflibercept vs observation risk difference, -3% [95% CI, -11% to 4%]; relative risk, 0.83 [95% CI, 0.55-1.27; P = .79]; laser photocoagulation vs observation risk difference, -1% [95% CI, -9% to 6%]; relative risk, 0.95 [95% CI, 0.64-1.41; P = .79]). Antiplatelet Trialists' Collaboration vascular events occurred in 15 (7%), 13 (5%), and 8 (3%) participants in the aflibercept, laser photocoagulation, and observation groups. Conclusions and Relevance: Among eyes with CI-DME and good visual acuity, there was no significant difference in vision loss at 2 years whether eyes were initially managed with aflibercept or with laser photocoagulation or observation and given aflibercept only if visual acuity worsened. Observation without treatment unless visual acuity worsens may be a reasonable strategy for CI-DME. Trial Registration: ClinicalTrials.gov Identifier: NCT01909791.
Sayres R, Taly A, Rahimy E, Blumer K, Coz D, Hammel N, Krause J, Narayanaswamy A, Rastegar Z, Wu D, Xu S, Barb S, Joseph A, Shumski M, Smith J, Sood AB, Corrado GS, Peng L, Webster DR. Using a Deep Learning Algorithm and Integrated Gradients Explanation to Assist Grading for Diabetic Retinopathy. Ophthalmology 2019;126(4):552-564.Abstract
PURPOSE: To understand the impact of deep learning diabetic retinopathy (DR) algorithms on physician readers in computer-assisted settings. DESIGN: Evaluation of diagnostic technology. PARTICIPANTS: One thousand seven hundred ninety-six retinal fundus images from 1612 diabetic patients. METHODS: Ten ophthalmologists (5 general ophthalmologists, 4 retina specialists, 1 retina fellow) read images for DR severity based on the International Clinical Diabetic Retinopathy disease severity scale in each of 3 conditions: unassisted, grades only, or grades plus heatmap. Grades-only assistance comprised a histogram of DR predictions (grades) from a trained deep-learning model. For grades plus heatmap, we additionally showed explanatory heatmaps. MAIN OUTCOME MEASURES: For each experiment arm, we computed sensitivity and specificity of each reader and the algorithm for different levels of DR severity against an adjudicated reference standard. We also measured accuracy (exact 5-class level agreement and Cohen's quadratically weighted κ), reader-reported confidence (5-point Likert scale), and grading time. RESULTS: Readers graded more accurately with model assistance than without for the grades-only condition (P < 0.001). Grades plus heatmaps improved accuracy for patients with DR (P < 0.001), but reduced accuracy for patients without DR (P = 0.006). Both forms of assistance increased readers' sensitivity moderate-or-worse DR: unassisted: mean, 79.4% [95% confidence interval (CI), 72.3%-86.5%]; grades only: mean, 87.5% [95% CI, 85.1%-89.9%]; grades plus heatmap: mean, 88.7% [95% CI, 84.9%-92.5%] without a corresponding drop in specificity (unassisted: mean, 96.6% [95% CI, 95.9%-97.4%]; grades only: mean, 96.1% [95% CI, 95.5%-96.7%]; grades plus heatmap: mean, 95.5% [95% CI, 94.8%-96.1%]). Algorithmic assistance increased the accuracy of retina specialists above that of the unassisted reader or model alone; and increased grading confidence and grading time across all readers. For most cases, grades plus heatmap was only as effective as grades only. Over the course of the experiment, grading time decreased across all conditions, although most sharply for grades plus heatmap. CONCLUSIONS: Deep learning algorithms can improve the accuracy of, and confidence in, DR diagnosis in an assisted read setting. They also may increase grading time, although these effects may be ameliorated with experience.