Comparison between binocular indirect ophthalmoscopy and digital retinography for diabetic retinopathy screening: the multicenter Brazilian Type 1 Diabetes Study

Diabetic retinopathy (DR) is one of the main causes of preventable blindness in the world, affecting 12.6 million and threatening the vision of 37.3 million in 2010 [1]. It is the main cause of preventable blindness in the economically active population in developed countries [2]. Diabetes mellitus (DM) accounts for more than 10 % of the health care budget of some countries, with DR being one of the main expenditures related to DM.

DR screening has been considered to have an excellent cost-effectiveness [3, 4], allowing the detection of early stages of the disease, preventing visual impairment and decreasing the economic burden of DR treatment [5‐9].

DR may impair vision in its advanced stages and also because of diabetic macular edema, which may occur at any stage of this diabetic chronic complication. DR can be diagnosed by binocular indirect ophthalmoscopy (BIO), with or without slit-lamp biomicroscopy, or through retinal photographs (retinographies) [10]. The Early Treatment Diabetic Retinopathy Study (ETDRS) established a gold standard of 7-field stereoscopic photographs for the diagnosis of DR [11].

Brazil, a country of continental dimensions, has approximately 12 million people between 20 and 79 years old with diabetes [12]. There are currently no nation-wide studies of DR prevalence in Brazil, but available data point to a prevalence ranging from 7.62 to 44.4 % of patients with diabetes, representing 0.9–5.3 million people [13‐19].

Brazil, along with several other developing countries, does not have a national screening DR program, except regional initiatives of detection of cases that mainly rely on BIO [13‐19].

Telemedicine has been established as a successful strategy for DR screening in various countries [6‐8]. Digital retinographies can be obtained by non-medical trained personnel and sent to reading centers, where specialists are able to classify DR and categorize patients in terms of observation or referral.

The aim of this study was to compare binocular indirect ophthalmoscopy (BIO) and digital retinography (mydriatic and non-mydriatic) for the screening of DR in a large multicenter type 1 DM patients study.

The number of participants who underwent at least two methods was 1266, with 56.8 % women. Comparison of the clinical outcome (observation or referral) between the group of patients who underwent all methods and the group of patients that missed at least one method showed homogeneity (p = 0.2471). The number of patients that underwent each method is displayed in Fig. 2. The number of participants who had both eyes evaluated by the four methods (BIO, NMR, 1FR, 2FR) was 1023.

The following data are given as mean (standard deviation) and median (range). Age (years), diabetes duration (years) and hemoglobin A1C (percentage) were respectively: 31 (12) and 29 (8–73); 16 (9) and 15 (0–56); and 10.6 (2.8) and 10.4 (5.0–18.6).

Quality was considered adequate in over 93, 74, 91 and 91 % of eyes for BIO, NMR, 1FR and 2FR, respectively. The main causes for poor quality are shown in Fig. 3.

DR and maculopathy classification in each center, in the four different Brazilian regions, is shown in Table 3. Classifications are based on the most severely affected eye and 2FR strategy, preferably.

Considering DR and maculopathy classification in the worst eye, 69 % of patients were “observable” and 31 % “referable” (see “Methods”).

When BIO and mydriatic retinography screening methods were compared regarding DR classification for the main outcome (observation versus referral), agreement was substantial (kappa 0.67–0.74). However, when BIO and non-mydriatic screening methods were compared regarding DR classification for the same outcome, agreement was only moderate (kappa 0.58). When BIO and the retinographies strategies were compared regarding maculopathy classification, agreement was fair to moderate (kappa 0.24–0.45). Agreement analysis is shown in Fig. 4.

Poor mydriasis was the main cause of poor quality in NMR, occurring in 11.92 % of right eyes and 16.96 % of left eyes. Poor mydriasis accounted for less than 1 % of poor quality in the mydriatic methods.

Poor mydriasis did not allow classification in 8.5 % of right eyes that underwent NMR, ranging 3.1–10.6 % in the various centers, and 12.1 % of left eyes, ranging 7.1–20 % in the various centers.

Poor mydriasis usually led readers to overestimate the DR stage in NMR, when compared to the classification for the same eyes with mydriatic methods. About 70 % of eyes with poor mydriasis in NMR that received a different classification in 1FR or 2FR were considered more severely affected in NMR (67.9 and 73.3 % of right eyes and 68.3 and 78.7 % of left eyes with 1FR for retinopathy and maculopathy, respectively; 67.7 and 73.1 % of right eyes and 68.1 and 78.0 % of left eyes with 2FR for retinopathy and maculopathy, respectively).

Patient age and retinopathy severity were associated with poor mydriasis or impossibility of classification in NMR. The average age of the patients with good mydriasis in NMR was 28 years old, whereas 36 years old for patients with poor mydriasis in NMR (p < 0.0001). The worse the severity of diabetic retinopathy, according to 2FR classification, the higher the percentage of eyes with poor mydriasis in NMR. For this sub-analysis, eyes were separated into two groups: the least severely affected and the most severely affected (see “Methods”). Eyes with poor mydriasis in NMR in the “least severe” group were 12 % right eyes and 15.1 % left eyes, whereas eyes with poor mydriasis in NMR in the “most severe” group were 62 % right eyes and 56 % left eyes. All differences were statistically significant (p < 0.0001).

BIO allowed for the classification of most eyes (58.1 %) that could not be properly classified by 2FR strategy.

We found a substantial agreement (kappa 0.67–0.74) between BIO and mydriatic retinographies for diabetic retinopathy classification regarding clinical outcome (observation vs referral), but only moderate agreement (kappa 0.58) between BIO and non-mydriatic retinography. The main factor that prevented diabetic retinopathy classification in the present study was poor mydriasis in NMR.

Our results agree with previous reports [28‐30] that also observed problems with NMR. Left eyes were most affected by poor mydriasis (Fig. 3): after the first retinography taken on the right eye, both pupils contract because of the flash, making it more difficult to obtain enough mydriasis on the second (left) eye. Poor mydriasis in NMR promoted classification errors, because of shadows or reflexes on the image, usually leading the reader to overestimate the severity in about 70 % of eyes. Poor mydriasis was observed more often in older patients or in eyes with more severe stages of retinopathy. Poor mydriasis was the only modifiable factor that prevented a readable image, since it dropped considerably in the mydriatic methods, whereas all other factors persisted (Fig. 3). Our findings advise against a diabetic retinopathy screening strategy with conventional non-mydriatic retinography. Ultra-wide field retinography is not currently a possible screening strategy in developing countries because of high costs [31‐33].

Mydriatic 1-field and 2-field retinography strategies had an almost perfect agreement (kappa 0.91–0.93). However, a small percentage (5.08 %) of observable patients according to 1FR were considered referable according to 2FR. These patients probably had significant changes in the nasal retina. As it is extremely easy and not time-consuming to obtain the second photo once the pupils are dilated, we believe that our findings support a 2-field protocol screening strategy [29].

Our results showed that BIO agreed substantially with digital retinography only regarding the clinical outcome (observation or referral), but they did not show a good agreement in DR classification if all stages were considered. BIO and retinographies did not show a good agreement regarding diabetic maculopathy (Fig. 4), either. BIO usually does not allow for identification of minor changes such as microaneurysms, retinal neovessels or mild macular edema [34], and probably that was the reason why BIO did not show a good agreement with retinographies regarding retinopathy and maculopathy classification. According to BIO, only 4 % of eyes had maculopathy, whereas maculopathy was present in 9 % of eyes according to 2FR. Additionally, BIO can only be performed by a specialist, precluding a telemedicine-based strategy.

The ETDRS protocol is still the gold standard for diabetic retinopathy classification, but it is very time-consuming and requires a highly trained professional for image acquisition [35]. Protocols with a smaller number of fields [6, 36, 37], as well as protocols with digital retinographies [38, 39], have already been validated for DR screening. In the present study, we used a 2-field digital retinography protocol that required less than 5 min per patient.

Besides poor mydriasis, some documentation for the present study was lacking because of improper technique. We believe this problem may be resolved with better training of the personnel, since there is a learning curve for retinography acquisition [40]. Some anatomical features of the eye may also render the examination or the reading especially difficult, for example, high myopic eyes with a hypopigmented fundus [41].

The referral criteria for the present study encompassed patients with moderate non-proliferative diabetic retinopathy or worse [21]. Patients with moderate non-proliferative retinopathy are clinically followed and not treated, and the inclusion of this stage might have overestimated the referral burden. However, our study still concluded that such classification would decrease the referral burden by about 70 %.

ETDRS classified diabetic macular edema into clinically significant (treatable) and non-clinically significant (observable). Our study evaluated only indirect signs of macular edema, such as hard exudates, but not retinal thickening itself. Hence, instead of using ETDRS classification, we employed the classification “apparently present” or “apparently absent” [10], with the former corresponding to both “observable maculopathy” and “referable maculopathy” categories from the Scottish grading system [10]. All patients with an “apparently present” macular edema were considered referable in the present study.

The global increase in the prevalence of diabetes mellitus calls for the optimization of human and economic resources. One of the main advantages of telemedicine is that it allows for a decreased proportion of patients that otherwise would have been referred to a specialist [5, 7‐9, 42]. Our results showed that around 70 % of examined patients would be observable, thereby alleviating the demand for the specialist.

BIO and retinography are both valid strategies for DR screening. The main advantages of BIO are its ease of handling and low cost [13]. BIO may perform superiorly in certain circumstances in comparison to retinographies, as in cases of media opacities or poor patient collaboration, for example. In our study, BIO allowed for the classification of most eyes that could not have been properly classified by 2FR strategy. BIO has some disadvantages, such as not being sensitive enough to detect minor signs of DR [34]. The main disadvantages of the retinography strategy are its cost and its poor performance in unfavorable anatomical features, as stated above. However, the central issue regarding the comparison of BIO and retinographies is that the former depends on the presence of a specialist, whereas the latter is compatible with telemedicine strategies. The current and future demands for retinal evaluation due to the global diabetes epidemics is not compatible with the presence of a specialist for every fundus exam [43]; telemedicine also is a feasible screening strategy for remote locations or areas with an uneven distribution of specialists.

This is the first Brazilian multicenter study that compared several DR screening methods in type 1 DM patients; we believe that our results could be extrapolated to other developing countries with remote areas and an uneven distribution of specialists.

This study had limitations. First, it was not designed to compare DR strategies to the gold-standard, ETDRS 7-field protocol. However, the ETDRS protocol is not suitable as a screening technique since it requires expert photographers and is time-consuming [44]. Additionally, the present study included only patients enrolled in referral urban centers. Further studies should also enroll patients from other settings, such as rural areas or primary care units. Future research on telemedicine and DR screening should also evaluate the practical impact of such measures in the decline of diabetic retinopathy as a major cause of blindness, as was the case in England and Wales [44]. Automatic detection of DR combined with telemedicine is also a promising field of research [44], as well as the study of other layers of interventions, such as telemedicine with health education and promotion, in order to improve DR screening coverage [45].

Our study showed that BIO and a telemedicine-based mydriatic retinography protocol are equivalent methods for DR screening. Because of its cost-effectiveness and its role in regions where specialists are unavailable, we recommend the telemedicine-based mydriatic strategy for DR screening.