Dr. Sabyasachi Sengupta S14805, Dr. ManaviDeokrismna Sindal, Dr. Prabu Baskaran, Dr. UtsabPan
Sabyasachi Sengupta, ManaviSindal, PrabuBaskaran, Utsab Pan, RengarajVenkatesh
Introduction
The global burden of diabetes is expected to rise from 285 million in 2010 to well over 439 million in 2030. As the prevalence of diabetes increases, the burden of diabetic retinopathy (DR) will also increase from its current prevalence of 93 million.Thirty percent, or 28 million, of those with DR have vision-threatening retinopathy (VTDR).Hence, in addition to ophthalmologists, diabetologists, optometrists and family care physicians would do well to employ effective techniques of screening for VTDR.
Because there are effective treatments for proliferative diabetic retinopathy (PDR) and diabetic macular edema (DME), blindness from diabetic eye disease should be preventable. A number of studies have established that the use of fundus photos is a reasonable way to improve access to high-quality diabetic retinopathy screening. All of these studies utilized table-top fundus cameras. In order to extend the reach of diabetic retinopathy screening into the rural areas not currently served by ophthalmologists, portable technologies must be developed and tested.
The objective of this current study is to evaluate the sensitivity and specificity of a smartphone based 45° mydriatic fundus camera (Remidio – Fundus on Phone (FOP)) compared to gold standard clinical evaluation and dilated Topcon fundus imaging to aid in the remote diagnosis of DR and VTDR.
Methods
The study was approved by the Institutional ethics committee and informed consent was obtained from all the participants. The study was performed according to the ICH-GCP guidelines and fulfilled the tenets of the declaration of Helsinki.
Participants
Participants >21 years of age were recruited from Aravind Eye Hospital, Pondicherry, India. Patients with clinically gradable DR were recruited from the retina clinic and patients without DR were recruited from the comprehensive clinic. Presence of vitreous hemorrhage and cataract were not exclusion criteria as long as the examining clinician felt confident of clinically visualizing the optic disc and determining presence of retinal features such as neovascularization of the disc and elsewhere. Patients were excluded if they had significant corneal or lenticular pathology precluding fundus examination, or underwent prior vitreo-retinal incisional or laser surgery.
Sample Size calculation
A Sample size of 200 eyes was chosen such that the measured rate of discordance in excellent images between the two imaging modalities (Remedio vs. TOPCON) would differ from the true rate of discordance by ≤7% assuming a true discordance rate ranging between 10% and 50%.
Clinical Assessment
Best-corrected visual acuity (BCVA), patient demographics and details of diabetes were recorded by a technician prior to dilation. All participants then underwent a comprehensive dilated examination with slit-lamp biomicroscopy and indirect ophthalmoscopy using a +90D and +20D lens by a single retina specialist (SS). The retina specialist noted all anterior segment findings, including corneal and lens status, along with an assessment of the diabetic retinopathy status i.e. mild, moderate, severe non-proliferative DR (NPDR), PDR and DME. VTDR was defined as severe NPDR or worse and/or the presence of DME.Cataracts were evaluated according to LOCS III grading. Those with nuclear sclerosis/opalescence > grade 3 (NS/NO3) and/or cortical cataract > C4 and/or posterior subcapsular cataract > P4 were categorized as “advanced immature” cataract. Those with less severe cataract were categorized as “early immature” cataract and those without any cataractous changes were categorized as “clear lens, pseudophakic or aphakic status.
Dilated Fundus Imaging
After the clinical examination, all participants had mydriatic 45° fundus photographs with the Remidio FOP and a standard tabletop fundus camera (Topcon TRC – 50DX, Tokyo, Japan) (mydriaticTopcon) including all three fields of view i.e. posterior pole (macula centered), nasal field and superotemporal field. All photographs were stored as JPEG files after removing all patient identifiers and assigning a randomly generated unique number identifier linked to the participant’s study ID number.
Remote Interpretation of the Fundus Photographs
Photographs were graded by two retina specialists (MS, PB) who were masked to the clinical examination status of the participant The retina specialists received the images in batches containing over 200 images taken from both photographic modalities: 1) Remidio FOP 2) mydriaticTopcon such that every three consecutive images were from the three fields of the same eye. Photographs from the same patient taken with different cameras were not included in the same reading batch so as to minimize bias from having seen a potentially “better” picture of the same fundus taken with a different type of camera. The two retina specialists graded the quality of each of the three fields according to predefined criteria of excellent, acceptable and ungradable. The two readers then graded the level of diabetic retinopathy (i.e. none, R1, R2, R3 level disease and the presence or absence of clinically significant diabetic macular edema, according to NHS guidelines for each eye after examining images from the three fields of view. VTDR was classified as > R2 level disease (severe NPDR or PDR) or the presence of macular edema.
Statistical Analysis:
Disease status was determined by the gold standard, a dilated fundus examination by a retina specialist (SS). Statistical analyses were performed using STATA software, version 12.0 (Texas, USA). Clinical variables were summarized overall and by disease status. The primary outcome, the sensitivity and specificity to detect DR, was estimated with a Wilson confidence interval. For each imaging modality type, inter- and intra-observer agreement was calculated for the primary outcome (presence of absence of DR) as well as a secondary outcome (quality of the photographs) using a kappa statistic and interpreted as previously described. Repeated measures logistic regression was used to predict the probability of a photo being gradable (excellent or acceptable) versus ungradable. Only the photographs where both readers agreed on the quality were included in this analysis.Univariate models were used to evaluate the following predictor variables: age, sex, presenting logMAR visual acuity, best corrected logMAR visual acuity, duration of DM, photographic modality (Remidio FOP vs. mydriatic Topcon), lens status, diabetic retinopathy status, the presence or absence of vitreous hemorrhage. A regression model with best – fit was used to create a multivariable logistic regression model.
Results
Demographics and baseline characteristics
One hundred and fifty-five subjects were recruited for the study; 233 eyes were included in the analysis. Thirty-five eyes were excluded from the analysis because all three fields of view were not obtained for each imaging modality according to the study protocol. Subjects had a median age of 54 years (interquartile range=48-61years) and 58% were men. The average duration of DM was10.7 years (stddev=5.1 years). As per clinical exam, there were 68 eyes with R1 disease (Mild + moderate NPDR), 44 with R2 (severe NPDR), 62 with R3 (PDR) and 59 with R0 (No DR)
Sensitivity and Specificity to detect VTDR
Compared to the gold standard clinical exam by a retina specialist (SS) using images from the Remidio FOP, graders 1 and 2 had a sensitivity of 95.9% (95% CI= 91.7, 98.3) and 96% (95% CI= 91.8, 98.3) and a specificity of 89.1 (95% CI= 77.8, 95.9) and 94.5 (95% CI= 84.9, 98.9) respectively, in identifying DR (ĸ=0.75, 95% CI=0.69, 0.77). With mydriatic images from the Topcon camera, both graders had equally high sensitivities and specificities in identifying VTDR (sensitivity = 97% (95% CI, 92-99) and 95% (95% CI, 90-98), specificity = 89% (95% CI, 83-93) and 90% (95% CI, 84-94)) (k=0.88, 95%CI=0.82-0.93).Sensitivity increases with worsening DR status (88% for R1 and 96% for R3) in both Remidio and Topcon.
Inter and Intra-Observer Reliability
Overall, the inter-observer reliability between the two retina specialists grading the images was quite high. The kappa statistic values ranged from κ=0.82-0.91 for the two imaging modalities. Intra-observer reliability indices for detecting VTDR using any of the three imaging modalities were also high, ranging from 0.82-1.00.There is greater agreement while grading VTDR as compared to individual levels of DR
Image quality and gradability
Images from the Remidio FOP has fewer ungradable images compared to Topcon, especially in the macula (5 vs 10 eyes) and nasal quadrants (4 vs. 10 eyes). In all 3 fields, the number of excellent images were consistently greater in Remidio FOP.
Predictors of Photo Gradability
A multivariable repeated measures logistic regression analysis showed that progressively worsening BCVA was associated with reduced odds of a gradable image. Using Topcon (vs. Remidio) had reduced odds of image being gradable but this was not statistically significant (Odds ratio=2.5, 95%CI=0.9-7.1, p=0.09)Images:
Topcon
Remidio fop
Discussion
The British Diabetic Association states that any new imaging device must have a sensitivity ≥80% to be considered for use in DR screening.(18) In this study, we found that the Remidio FOP fundus camera met this criterion with sensitivity between 93-95%, depending on the grader. We also found that the specificity for detecting DR using the Remidio FOP camera was excellent. Additionally, this smartphone based mydriatic camera acquired images of reasonable quality enough to enable remote diagnosis of DR and VTDR with good grader agreement.
The burden of diabetes and DR is increasing worldwide, and DR is a leading cause of blindness among working-age people.Our study echoes these findings, as our patients with VTDR were relatively young. Timely screening is an important strategy to prevent blindness from DR. There have been many studies assessing tabletop non-mydriatic fundus cameras in diabetic retinopathy screening programs. When remote readers grade the images from these table-top cameras, the range of sensitivities to detect VTDR is between 48% to 91% and the specificities range between 76% to 91%. Our results using a smartphone based portable fundus camera fall at the top of these ranges. A recent study by Ryan and colleagues evaluated smartphone based fundus photography and non-mydriatic fundus photography compared to standard 7-field dilated fundus photography in 300 Indian diabetic subjects attending a diabetologists clinic.(24) They found 59% sensitivity with the non-mydriatictabletop camera and a 54% sensitivity with the smartphone based camera for detecting VTDR. Our smartphone adapter showed significantly better results.
The need of the hour is to have an ideal portable imaging modality that is lightweight, non–mydriatic, have no issues with alignment, enable visualization through hazy media, and have easy user-interface and image transfer features. Many new and exciting portable technologies including smartphone-based technologies have been recently introduced to the market for DR screening but, to the best of our knowledge, none of the existing devices meet all the above-mentioned criteria. The Remidio FOP fundus camera tested in this study has many of these ideal characteristics. However, there are still many obstacles to widespread use of the Remidio FOP. It does not have any internal fixation lights, so patient alignment was difficult. Advantages are itprovides high quality images for instant transfer of images through the cellular data networks for remote grading.
There were a number of limitations to this study. It was conducted in a tertiary care eye hospital in India, and so the results may not be applicable to other populations. We used three images from the same eye to evaluate for VTDR so we do not yet know whether grading a single image of the posterior pole would yield equivalent sensitivity and specificity for detecting VTDR.
The strengths of our study are the relatively large sample of eyes with all grades of DR, the use of masked retina specialists as graders, and the use of standard photographs and guidelines for grading quality and severity of disease. Employing a single, well trained photographer as the imager also reduced the variability of our results. We also included eyes with cataract and vitreous hemorrhage in order to better generalize our results to a larger patient population.
Future research is required to replicate our results in the outreach camp settings where a dark environment for imaging may not always be available, electricity shortages are common and subjects oftentimes have dense cataract and other media opacities. In an outreach camp setting, hundreds of people are screened in a morning, so there will be a need to find new ways to structure work flow, from image acquisition to image analysis. New techologies continue to emerge that will disrupt the way care is provided for diabetic patients, and there is a need to continue to rigorously evalute these technologites before deploying them in large scale screening effort.