Purpose To determine whether combining measures of retinal structure and function predicts need for intervention for diabetic retinopathy (DR) better than either modality alone. Methods The study sample consisted of 279 diabetic patients who participated in an earlier cross-sectional study. Patients were excluded if they were previously treated for macular edema or proliferative DR or if they had other retinopathies. Medical records were reviewed for ocular interventions including vitrectomy, intravitreal injection, and laser treatment. Need for intervention was analyzed using Kaplan-Meier analyses and Cox proportional hazards. Baseline electroretinograms and fundus photographs were obtained. Two definitions of structural positive findings were as follows: 1. Early Treatment of Diabetic Retinopathy Study diabetic retinopathy severity scale (ETDRS-DR) severity ≥ level 53 (ETDRS-DR+) and 2. ETDRS-DR+ or clinically significant macular edema (VTDR+). A positive function finding corresponded to a RETeval DR Score >23.5 (RETeval+). Results For patients with VTDR+ the incidence of intervention was 19%, 31%, and 53% after 1, 2, and 3 years of follow-up. In these patients, intervention incidence increased to 34%, 54%, and 74% the subsequent 1, 2, and 3 years if function was above criterion (RETeval+), whereas RETeval−  results reduced the risk to 3%, 4%, and 29%, respectively, reducing risk to similar levels seen for patients with VTDR− results at baseline. Conclusions Prediction of subsequent intervention was best when combining structural and functional information. Translational Relevance This study demonstrates that clinical management of diabetic retinopathy is improved by adding electroretinography to fundus photographic information in assessing the risk of the need for intervention.

Diabetic retinopathy (DR) is diagnosed clinically by directly viewing retinal vascular changes during ophthal-moscopy or through fundus photographs. However, electroretinography (ERG) studies in humans and rodents have revealed that retinal dysfunction is demonstrable prior to the development of visible vascular defects. Specifically, delays in dark-adapted ERG oscillatory potential (OP) implicit times in response to dim-flash stimuli (<21.8 log cd $ s/m2) occur prior to clinically recognized DR. Animal studies suggest that retinal dopamine deficiency underlies these early functional deficits. In this study, we randomized individuals with diabetes, without clinically detectable retinopathy, to treatment with either low-or high-dose Sinemet (levodopa plus carbidopa) for 2 weeks and compared their ERG findings with those of control subjects (no diabetes). We assessed dim-flash–stimulated OP delays using a novel handheld ERG system (RETeval) at baseline and 2 and 4 weeks. RETeval recordings identified significant OP implicit time delays in individuals with diabetes without retinopathy compared with age-matched control subjects (P < 0.001). After 2 weeks of Sinemet treatment, OP implicit times were restored to control values, and these improvements persisted even after a 2-week washout. We conclude that detection of dim-flash OP delays could provide early detection of DR and that Sinemet treatment may reverse retinal dysfunction.

Aims: to evaluate the performance of the RETeval device, a handheld instrument using flicker electroretinography (ERG) and pupillography on undilated subjects with diabetes, to detect vision-threatening diabetic retinopathy (VTDR). Methods: performance was measured using a cross-sectional, single armed, non-interventional, multi-site study with Early Treatment Diabetic Retinopathy Study 7-standard field, stereo, color fundus photography as the gold standard. The 468 subjects were randomized to a calibration phase (80%), whose ERG and pupillary waveforms were used to formulate an equation correlating with the presence of VTDR, and a validation phase (20%), used to independently validate that equation. The primary outcome was the prevalence-corrected area under the receiver operating characteristic (ROC) curve for the detection of VTDR. Results: the area under the ROC curve was 0.86 for VTDR. With a sensitivity of 83%, the specificity was 78% and the negative predictive value was 99%. The average testing time was 2.3 min. Conclusions: with a VTDR prevalence similar to that in the US, the RETeval device will identify about 75% of the population as not having VTDR with 99% accuracy. The device is simple to use, does not require pupil dilation, and has a short testing time.

OBJECTIVE With rising global prevalence of diabetic retinopathy (DR), automated DR screening is needed for primary care settings. Two automated artificial intelligence (AI)–based DR screening algorithms have U.S. Food and Drug Administration (FDA) approval. Several others are under consideration while in clinical use in other countries, but their real-world performance has not been evaluated systematically. We compared the performance of seven automated AI-based DR screening algorithms (including one FDA-approved algorithm) against human graders when analyzing real-world retinal imaging data. RESEARCH DESIGN AND METHODS This was a multicenter, noninterventional device validation study evaluating a total of 311,604 retinal images from 23,724 veterans who presented for teleretinal DR screening at the Veterans Affairs (VA) Puget Sound Health Care System (HCS) or Atlanta VA HCS from 2006 to 2018. Five companies provided seven algorithms, including one with FDA approval, that independently analyzed all scans, regardless of image quality. The sensitivity/specificity of each algorithm when classifying images as referable DR or not were compared with original VA teleretinal grades and a regraded arbitrated data set. Value per encounter was estimated. RESULTS Although high negative predictive values (82.72–93.69%) were observed, sensitivities varied widely (50.98–85.90%). Most algorithms performed no better than humans against the arbitrated data set, but two achieved higher sensitivities, and one yielded comparable sensitivity (80.47%, P 5 0.441) and specificity (81.28%, P 5 0.195). Notably, one had lower sensitivity (74.42%) for proliferative DR (P 5 9.77 x 104) than the VA teleretinal graders. Value per encounter varied at $15.14–$18.06 for ophthalmologists and $7.74–$9.24 for optometrists. CONCLUSIONS The DR screening algorithms showed significant performance differences. These results argue for rigorous testing of all such algorithms on real-world data before clinical implementation.

To adapt to the changing field of ophthalmology, graduate medical education (GME) must teach trainees to harness the power of information technology. The Accreditation Council for Graduate Medical Education (ACGME) Ophthalmology Milestones 2.0 includes the use of information technology, such as the integration of multimodal data into clinical diagnosis for patient-centered care, effective communication and documentation utilizing the electronic health records (EHR), and analysis of EHR data for quality improvement.

Retinal fundus photographs can be used to detect a range of retinal conditions. Here we show that deep-learning models trained instead on external photographs of the eyes can be used to detect diabetic retinopathy (DR), diabetic macular oedema and poor blood glucose control. We developed the models using eye photographs from 145,832 patients with diabetes from 301 DR screening sites and evaluated the models on four tasks and four validation datasets with a total of 48,644 patients from 198 additional screening sites. For all four tasks, the predictive performance of the deep-learning models was significantly higher than the performance of logistic regression models using self-reported demographic and medical history data, and the predictions generalized to patients with dilated pupils, to patients from a different DR screening programme and to a general eye care programme that included diabetics and non-diabetics. We also explored the use of the deep-learning models for the detection of elevated lipid levels. The utility of external eye photographs for the diagnosis and management of diseases should be further validated with images from different cameras and patient populations.