Retinal screening contributes to early detection of diabetic retinopathy and timely treatment. To facilitate the screening process, we develop a deep learning system, named DeepDR, that can detect early-to-late stages of diabetic retinopathy. DeepDR is trained for real-time image quality assessment, lesion detection and grading using 466,247 fundus images from 121,342 patients with diabetes. Evaluation is performed on a local dataset with 200,136 fundus images from 52,004 patients and three external datasets with a total of 209,322 images. The area under the receiver operating characteristic curves for detecting microaneurysms, cotton-wool spots, hard exudates and hemorrhages are 0.901, 0.941, 0.954 and 0.967, respectively. The grading of diabetic retinopathy as mild, moderate, severe and proliferative achieves area under the curves of 0.943, 0.955, 0.960 and 0.972, respectively. In external validations, the area under the curves for grading range from 0.916 to 0.970, which further supports the system is efficient for diabetic retinopathy grading.

Outdoor time is considered to reduce the risk of developing myopia. The purpose is to evaluate the evidence for association between time outdoors and (1) risk of onset of myopia (incident/prevalent myopia); (2) risk of a myopic shift in refractive error and c) risk of progression in myopes only. A systematic review followed by a meta‐analysis and a dose–response analysis of relevant evidence from literature was conducted. PubMed, EMBASE and the Cochrane Library were searched for relevant papers. Of the 51 articles with relevant data, 25 were included in the meta‐analysis and dose–response analysis. Twenty‐three of the 25 articles involved children. Risk ratio (RR) for binary variables and weighted mean difference (WMD) for continuous variables were conducted. Mantel–Haenszel random‐effects model was used to pool the data for meta‐analysis. Statistical heterogeneity was assessed using the I² test with I² ≥ 50% considered to indicate high heterogeneity. Additionally, subgroup analyses (based on participant's age, prevalence of myopia and study type) and sensitivity analyses were conducted. A significant protective effect of outdoor time was found for incident myopia (clinical trials: risk ratio (RR) = 0.536, 95% confidence interval (CI) = 0.338 to 0.850; longitudinal cohort studies: RR = 0.574, 95% CI = 0.395 to 0.834) and prevalent myopia (cross‐sectional studies: OR = 0.964, 95% CI = 0.945 to 0.982). With dose–response analysis, an inverse nonlinear relationship was found with increased time outdoors reducing the risk of incident myopia. Also, pooled results from clinical trials indicated that when outdoor time was used as an intervention, there was a reduced myopic shift of −0.30 D (in both myopes and nonmyopes) compared with the control group (WMD = −0.30, 95% CI = −0.18 to −0.41) after 3 years of follow‐up. However, when only myopes were considered, dose–response analysis did not find a relationship between time outdoors and myopic progression (R² = 0.00064). Increased time outdoors is effective in preventing the onset of myopia as well as in slowing the myopic shift in refractive error. But paradoxically, outdoor time was not effective in slowing progression in eyes that were already myopic. Further studies evaluating effect of outdoor in various doses and objective measurements of time outdoors may help improve our understanding of the role played by outdoors in onset and management of myopia.

To investigate the choroidal and retinal thickness in myopic, emmetropic, and hyperopic Chinese children by swept-source longer-wavelength optical coherence tomography. Cross-sectional study. Two-hundred and seventy-six schoolchildren aged 7-13 years underwent comprehensive ophthalmic examinations, including cycloplegic refraction, and swept-source optical coherence tomography measurements. The thickness of the choroid, retina, ganglion cell layer, and nerve fiber layer were compared among children of different refractive status. The topographic variation and factors related to the thickness of the choroid and retinal layers were analyzed. Compared to emmetropic subjects, those with myopia had a significantly thinner choroid in all regions (P < .01), and hyperopic subjects had a thicker choroid in most regions (P < .05). The myopic retinas were thinner than those of emmetropic or hyperopic subjects in the superior parafoveal and all 4 perifoveal subfields (P < .05), but no other subfields differed significantly among different refractive groups (P > .05). The axial length and refractive diopters were independently related to central foveal choroidal thickness (R(2) = 0.17, P < .01), while age and intraocular pressure were independently associated with central fovea retinal (R(2) = 0.15, P < .01) and ganglion cell layer thicknesses (R(2) = 0.10, P < .01) after adjustment for other systematic and ocular factors. Central foveal choroidal and retinal thickness were unrelated in children of different refractive status (P > .05). Choroidal thickness, but not retinal thickness, correlated closely with axial length and refractive diopters in Chinese children. Choroid thinning occurs before retina thinning early in myopic progression.

Purpose: We assessed changes in age-specific prevalence of refractive error at the time of starting school, by comparing preschool and school age cohorts in Shanghai, China. Methods: A cross-sectional study was done in Jiading District, Shanghai during November and December 2013. We randomly selected 7 kindergartens and 7 primary schools, with probability proportionate to size. Chinese children (n = 8398) aged 3 to 10 years were enumerated, and 8267 (98.4%) were included. Children underwent distance visual acuity assessment and refraction measurement by cycloplegic autorefraction and subjective refraction. Results: The prevalence of uncorrected visual acuity (UCVA), presenting visual acuity, and best-corrected visual acuity in the better eye of ≤20/40 was 19.8%, 15.5%, and 1.7%, respectively. Among those with UCVA ≤ 20/40, 93.2% could achieve visual acuity of ≥20/32 with refraction. Only 28.7% (n = 465) of children with UCVA in the better eye of ≤20/40 wore glasses. Prevalence of myopia (spherical equivalent ≤−0.5 diopters [D] in at least one eye) increased from 1.78% in 3-year-olds to 52.2% in 10-year-olds, while prevalence of hyperopia (spherical equivalent ≥+2.0 D) decreased from 17.8% among 3-year-olds to 2.6% by 10 years of age. After adjusting for age, attending elite “high-level” school was statistically associated with greater myopia prevalence. Conclusions: The prevalence of myopia was lower or comparable to that reported in other populations from age 3 to 5 years, but increased dramatically after 6 years, consistent with a strong environmental role of schooling on myopia development.

Purpose: To systematically analyse the differences between cycloplegic and noncycloplegic refractive errors (RE) in children and to determine if the predictive value of noncycloplegic RE in categorizing RE can be improved.Methods: Random cluster sampling was used to select 6825 children aged 4–15 years. Autorefraction was performed under both noncycloplegic and cycloplegic (induced with 1% cyclopentolate drops) conditions. Paired differences between noncycloplegic and cycloplegic spherical equivalent (SE) RE were determined. A general linear model was developed to determine whether cycloplegic SE can be predicted using noncycloplegic SE, age and uncorrected visual acuity (UCVA).Results: Compared to cycloplegia, noncycloplegia resulted in a more myopic SE (paired difference: −0.63 ± 0.65D, 95% CI: −0.612 to −0.65D, 6017 eligible right eyes) with greater differences observed in younger participants and in eyes with more hyperopic RE and smaller AL. Using raw noncycloplegic data resulted in only 61% of the eyes being correctly classified as myopic, emmetropic or hyperopic. Using age and uncorrected VA in the model, the association improved and 77% of the eyes were classified correctly. However, predicted cycloplegic SE continued to show large residual errors for low myopic to hyperopic RE. Applying the model to only those eyes with uncorrected VA <6/6 resulted in an improvement (R² = 0. 93), with 80% of the eyes correctly classified. A higher VA cut‐off (i.e., ≤6/18) resulted in 97.5% of eyes classified correctly.Conclusion: Noncycloplegic assessment of RE in children overestimates myopia and results in a high error rate for emmetropic and hyperopic RE. Adjusting for age and applying uncorrected VA cut‐offs to noncycloplegic assessments improves detection of myopic RE and may help in identifying myopic RE in situations where cycloplegia is not available but does not help in identifying the magnitude of refractive error and therefore is of limited value.

To evaluate the association between the AL/CR ratio and refractive state and explore the effectiveness of this ratio in the assessment of myopia, especially when combined with uncorrected visual acuity in schoolchildren among whom myopia is common. Cross sectional study. 4686 children from 6 primary schools, aged from 6 to 12 years were selected using the clustered-stratified random sampling method. Uncorrected visual acuity (UCVA), axial length (AL), corneal radius of curvature (CR), and cycloplegic refraction were tested. Refraction was measured as the spherical equivalent (SE). 3922 children were included in the analysis. The mean AL/CR ratio was 2.973±0.002, increased with age, and different in gender. The coefficients of correlations of the SE with the AL/CR ratio, AL, and CR were -0.811, -0.657, and 0.095, respectively. Linear regression showed a 10.72 D shift towards myopia with every 1 unit increase in the AL/CR ratio (P<0.001, r² = 66.4%). The estimated SE values obtained by substituting the AL/CR ratio and gender back to the regression model that were within a difference of ±0.50 D in ATE/LER (allowable total error and limits for erroneous results) zones compared to the actual measured values was 51%. The area under the ROC curve of the AL/CR ratio, AL, and UCVA for myopia detection were 0.910, 0.822, and 0.889, respectively, and the differences between each pair were statistically significant (P<0.01). At a specificity of 90%, the sensitivities were 72.98%, 50.50%, 71.99%, and 82.96%, respectively, for the AL/CR ratio, AL, UCVA, and the combination of the AL/CR ratio and UCVA. The AL/CR ratio was found to explain the total variance in SE better than AL alone. The effectiveness of the AL/CR ratio was statistically significantly better than UCVA for detecting myopia in children, and combining the two produced increased sensitivity without significantly decreasing specificity.