The prevalence of diabetes mellitus (DM) and the number of people affected by DM is increasing rapidly in all regions. The International Diabetes Federation estimated that 425 million people had diabetes in 2017, which will increase to 629 million in 2045 globally [1]. This increase is expected to be the highest in low-income and middle-income countries (LMICs) compared with high-income countries (HIC) [2]. Diabetic retinopathy (DR) is a common microvascular complication of DM, which can lead to visual impairment and blindness if not detected early and treated [3]. Many studies have reported that visual loss from DR can be largely prevented by early screening and appropriate treatment [4-6]. Diabetic retinopathy screening (DRS) can be done in 2 ways, systematic screening similar to national-level programs in HIC versus opportunistic screening and case detection, which is common in low-income settings. Most LMICs are unlikely to have full population-based screening program owing to resources constraints. The current method of DRS in most LMICs is direct ophthalmoscopy, which has a lower diagnostic accuracy and found to be ineffective even after training [7]. The mydriatic biomicroscopic examination by an ophthalmologist is practically not possible in these countries owing to a low number of ophthalmologists, and eye clinics are overburdened with highly prevalent blinding conditions such as cataract [8].

The reasons for the unavailability of DRS programs (DRSPs) in LMIC settings are mostly attributed to the lack of skilled human resources, financial resources, and evidence of what works in the local system [9-11]. Therefore, it would be important to understand the approaches for screening, especially in non-ophthalmological settings. Conventional digital cameras need a larger space, skilled photographers, and large image storage devices. In addition, systematic screening using sophisticated table-top imaging systems incur high capital investment though they are cost-effective [12]. Hand-held digital cameras are easy to move, require minimum space, minimum power consumption, and are user-friendly [13]. In addition, nonmydriatic hand-held cameras are less discomforting to participants and can be used while people with DM are waiting in front of a physician for consultation. The usage of a camera without pupil dilatation is comfortable to people with DM, as well as easy for providers. However, the latter depends on the quality of the image available for grading [14].

Various photographic studies have looked at the diagnostic test accuracy of DRS using digital imaging. Most of these studies used static table-top imaging systems and were conducted in HICs. These studies have shown a sensitivity of 68%-97% and a specificity of 71%-100% in nonmydriatic imaging using ophthalmic human resources as index graders [15-18]. Similarly, in mydriatic imaging, most of the studies have used table-top imaging systems, ophthalmic human resources as index test graders, and were conducted in HICs. These studies have shown a sensitivity of 77%-97% and a specificity of 76%-98% in mydriatic digital imaging [19-22]. There is a gap in evidence in digital retinal imaging in LMICs using non-ophthalmic human resources. In addition, the usage of context-specific imaging systems, such as hand-held digital retinal camera, in non-ophthalmic setting was not reported in the current literature.

Sri Lanka has achieved remarkable development in the health sector; however, there are public health concerns such as DR which have not been addressed to date [23]. The crude prevalence of DM in Sri Lanka was 12.6% (>20 years), being highest in the Western province (18.6%, 95% CI 15.8%-21.5%) [24]. In the Western province, there are approximately 750,000 (age>18 years) people with DM, 20% (150,000/750,000) of whom are likely to have nonproliferative DR (NPDR). A situational analysis conducted in this region has shown that the number of people undergoing opportunistic screening and free treatment in the public sector was far lower than the estimated need [25]. There is no systematic DRS in the Western province despite the high prevalence of DM [25]. In addition, there is no published data on this topic from Sri Lanka. The aim of this protocol is to describe the methods of validation of a DRS approach using digital imaging by physician graders in a tertiary-level public sector medical clinic. This study will demonstrate the functional and technical feasibility of using a hand-held digital camera in an LMIC non-ophthalmological setting and assess the diagnostic accuracy.

The physician graders have been trained, and currently, validation is being done in the Western province of Sri Lanka. The results of this study will be published in detail according to the Quality Assessment of Diagnostic Accuracy Study guidelines [26]. Data will be entered using a Microsoft Excel (2016) worksheet and transferred into STATA/IC-v14.2 analytical package following cleaning, consistency checks, and analysis. The sensitivity, specificity, and predictive values for each strategy and each level of DR will be presented using the same variables of 2 physician graders (nonmydriatic and mydriatic separately) compared with the reference standard, along with 95% CIs.

The level of skills acquired by physician graders is an important factor in the screening outcome. Different non-ophthalmologist graders have successfully conducted DRS in some settings [27-29]. We will describe the diagnostic accuracy of the detection of DR by physician graders. In addition, we will be able to study the effect of a range of population characteristics on the validity of detecting DR using imaging and understand the role of non-ophthalmic personnel to make recommendations for a systematic DRSP. In addition, we will describe the referral criterion applicable to this local context based on the validation study results. Defining a referable level DR at a non-ophthalmological setting, in a context where there is no systematic DRS, will filter out those not needing a referral and therefore reduce the workload at an ophthalmologist’s clinic. The 7-field imaging strategy used in early treatment diabetic retinopathy study is considered as the gold standard in DRS [30]. However, this technique is practically not feasible in this context owing to resources constraints. Therefore, we proposed to use the locally accepted reference standard of retinologists’ examination as the suitable reference standard. Digital retinal imaging has previously shown diagnostic accuracy levels that would comply with the accepted standards of established national-level screening programs [15,22,31].

A few studies (conducted in HICs) have used non-ophthalmologist human resources in DRS, with which we could compare our results. In Singapore, a nonmydriatic fundus camera showed a sensitivity of 69.8% (95% CI 61.3%-77.2%) and a specificity of 94.4% (95% CI 92.3%-96.1%) for nonphysician graders using a single field [32]. A study in the United Kingdom on DRS by general practitioners using 35-mm color images showed that detecting any level of DR increased from 62.6% (95% CI 55.9%-69.4%) with direct ophthalmoscopy to 79.2% (95% CI 73.6%-84.9%) using retinal photographs, and specificity remained unchanged (direct ophthalmoscopy 75.0% [95% CI 69.5%-80.5%] vs 73.5% [95% CI 68.0%-79.1%]) [33]. They concluded that retinal photography by trained general practitioners in primary care settings could attain an acceptable level of detection of sight-threatening DR (87%) [33]. In Thailand, the use of single-field digital nonmydriatic imaging showed a sensitivity of 80% and a specificity of 96% in a sample of people with DM, where 54.7% people with DM were aged 41-60 years and 45.3% people with DM had diabetes since 1-5 years [34].

Another important consideration in this study would be the gradeability of images. The image gradeability will depend on the lens opacity, media opacity, pupil size, and reflectivity of the fundus. We envisaged poor gradeability in nonmydriatic imaging considering the high prevalence of cataract in this local setting. Furthermore, iris color, age, and other population characteristics may affect the quality of images [14]. Scanlon et al showed that in the >80-years age group, the technical failure rates reduced from 41.6% to 16.9% following mydriasis. This study concluded that the odds of having one eye ungradable increases by 2.6% (95% CI 1.6%-3.7%) for each extra year since the diagnosis of DM and major cause of ungradability was having central cataract (57%) [35]. We will describe the factors affecting gradeability of images in addition to the diagnostic test accuracy results.

In this study, we will demonstrate the diagnostic accuracy of physician graders compared with the retinologist to make recommendations for developing an integrated DRSP in LMICs where there is no systematic DRS. The outcome of this study will be useful for the implementation of a systematic DRSP in this region and similar communities.