The American Telemedicine Association (ATA) suggested that telehealth programs for DR should demonstrate an ability to compare favorably with ETDRS film photography as reflected in kappa values for agreement of diagnosing levels of retinopathy and macular edema . However, we always found it difficult to conduct ETDRS film photography in Chinese diabetic residents, mainly because of limited ocular health care resources in Chinese communities. For this reason, we used the Chinese traditional DR screening method (in-person funduscopic examination after pupil dilation) instead, in the validation study. The high kappa value calculated herein demonstrated that our telehealth system was as valid as the traditional method for DR screening. These results are similar to that observed in former studies: there was an agreement of 85% to 94% between DR stages graded by both ophthalmologists by direct examination and by inspecting the digital images [10, 20]. In a previous epidemiology study of 795 diabetic residents in the same community in 2007, we found that the prevalence rates of DR and PDR were 27.09% and 1.13%, respectively . The proportions of DR and PDR (24.42% and 2.12%) revealed in this study were similar, and the small difference was plausible because of different study design, population and period. Thus, we regard BCDRT as a reliable and valid system for DR screening in local diabetic residents.
One characteristic of our DR telehealth system is that the internet-transmitted data encompass not only the digital retinographs but also visual acuity, and the latter was seldom analyzed in other DR telemedicine systems based on our MEDLINE literature search. In comparisons to other DR telemedicine systems [2, 6, 9, 10], the BCDRT tried a simplified protocol that just acquires two 45-degree digital retinographs per eye without pupil dilation via a non-mydriatic funduscopic camera in order to reduce the screening time required and achieve greater patient acceptability. However, we did have concern that the sensitivity of detecting macular edema may decrease. For this reason, we added the visual acuity information into the digital data of BCDRT, and hoped this additional inexpensive measurement tool would trigger referral to macular edema with this system. According to the data in our report, if the diagnoses from ophthalmologists are assumed to be the "gold standard," the sensitivity and specificity of BCDRT for detecting macular edema are 88% (22/25) and 94% (29/31), respectively. These results are similar to those of Kim and associates, where the sensitivity and specificity of two non-mydriatic digital fundus image assessment of diabetic macular edema, by a retinal fellow, was 0.80 and 0.93 respectively with a 35-mm fundus image assessment as the reference standard . We admit that with some stereoscopic retinal imaging systems, such as the Joslin Vision Network , the concordance in the gradable eyes for macular edema could be as high as 100%. However, we consider the accuracy of BCDRT diagnoses of diabetic macular edema to be acceptable, and one can expect that very few cases of macular edema will be missed using BCDRT. This system is in agreement with the Guidelines for Ocular Telehealth for DR from ATA, and belongs to the "category 2" system, which is defined as the ability to distinguish patients in two categories: (1) those who have non-sight-threatening DR or (2) sight-threatening DR . In addition, the BCDRT system also guarantees the low rate of unreadable photographs for DR screening (≤ 5%) recommended by the British Diabetic Association . There is good evidence that pupillary dilation reduces the proportion of patients with ungradable retinal images. For example, in the Vine Hill study, ungradable retinal images only occurred in 0.5% of the 201 screened diabetic patients under mydriasis . However, pupillary dilation requires a higher level of medical supervision in many settings, and places some constraints on patients in performing certain visual tasks until the pupillary dilation dissipates . The proportions of ungradable retinal images differ between DR telehealth screening systems. In the study by Ahmed and associates , 35% of the total images were judged inadequate to be graded fully, and the patients with ungradable eyes were significantly older than those with gradable eyes. Scanlon and associates19 have previously shown that for patients 70 years and older, approximately 25% of non-mydriatic images are unreadable. A total of 74.5% (126/169) of ungradable eyes had early to obvious central cataract, and 10 eyes (6%) had a corneal scar . In the present study, the proportion of media opacity in ungradable images is lower: 73.3% (11/15) of the 15 patients with ungradable images, and who were later referred to in-person examination, were diagnosed with moderate to severe cataract or corneal scar. Furthermore, we seldom encountered small pupil size and poor focus or poor patient fixation cases. Therefore, the ungradable rate in the present study (4.88%) seems to be much less than Scanlon's report. In the study by Gomez-Ulla and associates, the ungradable rate (5.2%) was similar to our result . Cavallerano and associates demonstrated an even lower ungradable rate of 1.3% . In the OPHDIAT study, the rate of non-gradable photographs dropped from 12.2% to 8% between September 2004 and December 2006, because technicians received additional training in an effort to improve the quality of future images . Similarly, we hope the rate of non-gradable photographs in BCDRT will decrease as the PCPs have increasingly more experience with non-mydriatic image capturing.
As is well understood, the common barriers to high annual DR screening rates in most underdeveloped countries are the insufficiencies in ophthalmic equipment and experienced ophthalmologists, particularly in rural areas where access to medical resources is hindered by geographic distance. It is believed that telehealth systems help to achieve a significant increase in regular DR surveillance in underdeveloped countries, based on two characteristics inherent in these systems: (1) interactions between patients and doctors are not limited by geographic distance, (2) a well-established telehealth system facilitates centralized DR screening . For example, at least 40 participants can be screened during half a day with BCDRT, which helps to achieve the maximum effects offered by medical resources. To overcome the aforementioned barriers to DR telemedicine system implementation in underdeveloped countries, we initially introduced the BCDRT system to the general residents via the mass media. As a result, although not specially queried, most of the 471 diabetic residents in this study mentioned that they had heard about BCDRT before they came to participate. The cost of necessary accessories, including a digital funduscopic camera and essential software, will impact wide-spread implementation of a DR telehealth system in some regions with limited resources. Most commercial telemedicine systems in developed countries are proprietary in that the end users (primary care centers) must use licensed software built into the imaging system for storage and transmission of images . Because we did not have an adequate budget for purchase and maintenance of specific software packages after expenditures for a non-mydriatic funduscopic camera, we finally used commercially available software to construct our DR telehealth system instead, and considered it appropriate for primary care settings that serve indigent residents in underdeveloped countries. Wei and his collaborators have described some other deficiencies with the proprietary systems, including poor support for telemedicine workflow, a dependence on local area networks or dedicated computer network connections and, hence, a lack of true internet scalability and, lastly, a lack of interoperability between components from different vendors . Massin and associates suggested that DR telemedicine screening centers should ideally be located in areas with a high rate of poverty and a low number of ophthalmologists . Our experience with BCDRT implementation may help construction of efficient DR telehealth systems in a majority of communities and villages in underdeveloped countries, which meets the criterion of Massin's "ideal location".
A shortcoming of our study should be mentioned. First, because of limited resources, we were not able to include detailed blood laboratory studies. Second, because patients identified as normal by BCDRT were not referred to in-person examination, it is impossible to verify that the patients with normal fundus photographs indeed do not have DR on in-person examination. Therefore, we were not able to calculate additional important statistics such as sensitivity and specificity for detecting DR, because of lack of data from patients whose digital images were graded as normal. Third, the sample size is small, which leads to a high mean age and a single diabetes mellitus type in the patients studied. However, the number of participants in the BCDRT will no doubt increase with time, due to satisfactory patient preferences at the next screening (96.82%). Thus, it is promising to anticipate larger scale studies that better represent the general diabetic population, in the future.