Progression of Early Retinal Dysfunction in Diabetes Over Time: Results of a Long-term Prospective Clinical Study

Diabetic retinopathy has classically been considered to be a microvascular complication caused by elevated blood glucose levels and metabolic pathways triggered by hyperglycemia. Vascular lesions have been characterized in detail, and guidelines for treatment based on various grading scales have been developed to help preserve visual acuity. However, the retina is not primarily a vascular tissue; rather, it is a neuronal tissue with a vascular supply in which the retinal neurons, glia, and retinal vasculature are interconnected to form a functional neurovascular unit with intricate molecular interactions (1). Hyperglycemia likely affects not only the vasculature per se but also the neuroretina, resulting in dysfunctions other than impairment and loss of visual acuity.

There is increasing evidence of early retinal neurodegeneration in diabetes, which may even precede the vascular changes (2). Neuronal degeneration patterns have been observed in various animal models of diabetes, and early retinal dysfunction has been demonstrated (3,4). Postmortem studies in humans have revealed neuronal degeneration and apoptosis in retinas, mainly in the ganglion cell layer (5,6). Furthermore, use of the modern technique of spectral domain optical coherence tomography for retinal imaging (7) has shown thinning of the retinal nerve fiber (8,9) and photoreceptor layers (10) in diabetic subjects in vivo before any visible vascular lesions could be detected. In addition, thinning of the ganglion cell layer and inner plexiform layers has been reported in patients with mild diabetic retinopathy (11).

The most commonly used methods for detecting retinal dysfunction in humans are psychophysical (e.g., perimetry) or electrophysiological (e.g., electroretinogram [ERG]). Several of those techniques have been applied to evaluate retinal dysfunction in diabetic subjects, but so far, very few long-term longitudinal studies have been performed to establish the usefulness of those methods.

The most widely used test of retinal dysfunction is standard automated perimetry (SAP), which has rendered results indicating a reduction of retinal sensitivity in diabetic subjects without retinopathy (12,13) as well as in those with mild/moderate (14) or moderate/severe retinopathy (15,16). Moreover, reduction of retinal sensitivity revealed by SAP was found to correlate with stepwise increases in the severity of retinopathy (17,18). That SAP can be used to predict the development of diabetic retinopathy has also been suggested (19).

Microperimetry is another perimetric method in which a small area of the central field is tested using white stimuli but on a darker background than in SAP. This test was recently reported to show reduced sensitivity in subjects who did or did not have various types of retinopathy (20,21). A drawback of microperimetry is that the short dynamic range of the stimulus presented results in truncation of threshold sensitivities (22), and hence, normal or nearly normal function cannot be accurately measured.

Frequency doubling technology and short wavelength automated perimetry (SWAP) are two examples of selective perimetry, the latter designed to expose blue stimuli on an intense yellow background. The stimuli used in these methods aim at testing specific subpopulations of retinal ganglion cells, which increases the sensitivity of the tests. Reduced SWAP sensitivity has been reported in diabetic subjects without retinopathy (12,23) as well as in individuals with mild, moderate, or more severe diabetic retinopathy (17,18,24). SWAP is not optimal for analysis of longitudinal data. Compared with SAP, SWAP entails considerably larger test–retest variability (18), making it difficult to detect subtle changes, and SWAP is also much more sensitive to cataract development. Perimetry using frequency doubling technology stimuli has reported reduced retinal sensitivity in diabetic subjects without as well as with retinopathy (13,16,25), but so far this has only been noted in a limited number of cross-sectional studies.

The ERG is an electrophysiological test that can provide objective and quantitative information on retinal dysfunction in diabetic subjects (26). There is evidence that ERG abnormalities can be detected very early in the course of retinopathy and that special techniques can be used to demonstrate local responses. To our knowledge, changes in ERG responses over time have only been investigated in insulin-dependent diabetic patients and have yielded contradictory results (27,28), but it has been reported that a delayed ERG response can predict the onset of diabetic retinopathy (29). Those results could indicate that diabetes may affect the retinal neurons ahead of the retinal vascular network.

The purpose of our study was to demonstrate the usefulness of SAP for detecting early retinal dysfunction over time in patients with type 1 and type 2 diabetes. In our interim report published after 18 months of follow-up of subjects with and without mild/moderate diabetic retinopathy (30), we described how our previously defined limits of significant change for SAP in diabetes (18) provided promising results regarding the monitoring of perimetric change. Here, we confirm those findings in an extended longitudinal study with an average follow-up time of 4 years.

The subjects and study design have previously been described in detail (30) and are summarized here. From September 2006 to May 2009, patients who had type 1 (13 of 81) or type 2 (68 of 81) diabetes and were between the ages of 18 and 75 years were consecutively recruited from the screening program for diabetic retinopathy at the Department of Ophthalmology, Karlstad, County Council of Värmland. This program uses fundus photography to grade vascular lesions and is open to all individuals with diabetes who are older than 10 years of age. Type 1 diabetes was defined as having a diagnosis of diabetes at age <30 years and receiving insulin treatment within 1 year of diagnosis. One randomly selected eye per subject was examined and included in the study. Patients were not included if they had previously received laser treatment or any other local treatment for diabetic retinopathy, were in need of such treatment, or had other conditions that were likely to affect the visual field. Intraocular pressure measurements ranging between 11 and 22 mmHg and normal optic discs on baseline photographs ruled out the presence of undiagnosed glaucoma. To be included, patients had to be able to perform reliable visual fields defined as ≤15% false-positive responses. The study was approved by the regional ethical review board of Lund University, Sweden, and all patients gave written informed consent.

Patients were scheduled for follow-up visits every 6 months for the first 3 years and thereafter annually until 5 years from baseline. All visits included an examination of visual acuity, SAP, fundus photography, and measurement of HbA_1c and blood pressure.

Visual acuity was tested using Early Treatment of Diabetic Retinopathy Study (ETDRS) charts (31) and was expressed as the number of correctly read letters. Visual acuity was measured after refraction with manual adjustment of autorefractor (KR-8100P; Topcon, Tokyo, Japan) values in sphere and cylinder to 0.25-diopter accuracy. We considered a difference of five or more letters as a change in visual acuity based on our earlier results of measurements of short-term variation in visual acuity in diabetic patients (32).

Visual fields were tested using a Humphrey Field Analyzer 750 (Carl Zeiss Meditec, Inc., Dublin, CA) with the SITA Standard 24-2 program, including 54 test point locations within the central 24° visual field. Normal limits are based on a multicenter collection of perimetric data from 330 healthy subjects between 19 and 84 years of age (33). Before the initial visit during the study period, all subjects performed one visual field test to avoid perimetric learning effects during follow-up. The perimetric interpretation tool “single field analysis” provides probability maps that flag test locations with sensitivities that are significantly depressed compared with age-corrected normal values (34) (Fig. 1). The global perimetric index mean deviation (MD) describes the global status of the visual field; a value of 0 dB corresponds to a normal field, and approximately –30 dB corresponds to a blind field.

Longitudinal visual field change was assessed by comparing each follow-up field with the baseline field representing an average of the two tests performed at the first two visits during the study period. Differences in age-corrected threshold values were calculated for each test point, except for the two located in the blind spot area, and thereafter were compared with the limits for significant improvement and deterioration that we had previously defined for diabetic subjects (18). This change analysis was developed in analogy with the glaucoma change probability maps used to assess glaucomatous progression (35). Test locations with significant change (deterioration or improvement) at the P < 0.05 level (Fig. 1) were counted. As in our previous study (30), fields with five or more significantly improved or deteriorated test locations were considered as showing a possible change from baseline, and repeated improvement or deterioration at five or more test locations in two or more consecutive field tests was regarded as indicating likely change. The risk for having at least five test points falsely flagged as deteriorated or improved at the P < 0.05 level is about 10% in a single test, and by requiring at least the same number in the consecutive test, the risk for false change diminishes considerably to 1%, assuming the tests are independent.

To confirm the results obtained by our method, we applied the established pointwise linear regression analysis (36,37) but used age-corrected threshold values over time to detect significant changes in visual fields. Eyes were considered deteriorated if five or more test points showed significantly negative regression slopes at the P < 0.05 level.

Stereo fundus photography of seven 35° fields was performed using a TRC 50IX retinal camera (Topcon) and Fujichrome Sensia 100 slide film (Fujifilm, Tokyo, Japan). The degree of retinopathy was graded according to the ETDRS final severity scale (38). The grader (E.A.) was masked to patient characteristics, visit number, and outcome of the functional tests. Progression or regression of retinopathy was defined as a two-step change according to the ETDRS final severity scale.

HbA_1c was analyzed by high-performance liquid chromatography initially using a Variant II Hemoglobin A1c Program (Bio-Rad, Hercules, CA) and from 2 February 2011 using a TOSOH G8 analyzer (Medinor, Tokyo, Japan): normal ranges are 27–42 and 31–46 mmol/mol (4.6–6.0% and 5.0–6.3% according to the National Glycohemoglobin Standardization Program) for individuals aged <50 and ≥50 years, respectively. Blood pressure was expressed as the mean of two measurements performed using an aneroid sphygmomanometer with the patient sitting.

Depending on the type of data to be analyzed, parametric (paired t test) or nonparametric (McNemar, Wilcoxon sign rank, and Mann-Whitney) tests were used to compare changes in patient characteristics and perimetric MD values between baseline and the last follow-up visit. We also used κ statistics to analyze agreement between the two methods applied to assess changes in visual fields.

Ninety-three diabetic subjects received information on the study and participated in the training session, and 12 were excluded due to proliferative retinopathy (n = 1), unreliable perimetry (n = 3), or individual choice (n = 8). Of the remaining 81 subjects who met the inclusion criteria, 74 completed the minimum follow-up of 3 years. The reasons for dropout were death (n = 2), severe illness (n = 2), and unwillingness to continue (n = 3). The median follow-up time was 4 years, and 22, 18, and 34 of the subjects completed 5, 4, and 3 years of follow-up, respectively. Compliance with the visit schedule was high: only three of all possible visits were missed by three subjects who did not attend the 12-month visit.

Of the 74 subjects who completed follow-up, 12 had type 1 diabetes. Subject characteristics at baseline are summarized in Table 1. Very few changes occurred in their characteristics during follow-up: at the last study visit, fewer subjects were being treated by diet alone, and diastolic blood pressure was slightly lower.

After 3 to 5 years of follow-up, visual acuity had decreased five or more letters compared with baseline in 20 of the 74 patients, in 2 of 12 type 1 diabetic subjects, and in 18 of 62 type 2 diabetic subjects. In the worst case, 15 letters were lost due to development of clinically significant macular edema, and another patient with cataract at the last visit lost 11 letters. We have no explanation for the decrease in visual acuity in the remaining 18 patients. Visual acuity improved by five or more letters in six patients, one of whom underwent cataract surgery during the follow-up period.

Subjective assessment of “single field analysis” and its probability maps revealed no obvious field loss at baseline or in follow-up tests. For the entire cohort, the global MD value improved slightly, from an average of –0.55 dB at baseline to –0.39 dB at the last visit (P = 0.31).

Our change analysis revealed considerably more test locations with significant deterioration than with significant improvement at the last follow-up visit. The number of test locations showing deterioration increased over time, whereas the number exhibiting improvement remained at essentially the same level (Fig. 2). At the last visit, 27 eyes showed significant deterioration at five or more test locations, and only 3 eyes had significant improvement at five or more locations. Only 1 subject had five or more test points showing significant improvement in the last two tests, whereas 16 subjects had five or more test points with significant deterioration at least at the last two visits in 2 of 12 with type 1 and in 14 of 62 with type 2 diabetes. The global MD value for those 16 individuals decreased from –0.19 dB to –1.26 dB (P = 0.003). Test points with repeated deterioration appeared in all parts of the visual field without any typical pattern during the study period. The proportions of deteriorated test points detected within the central 10° of the field and outside that area were similar, at 21% and 18%, respectively. The corresponding proportions in the upper and lower hemifields were also similar. According to the pointwise linear regression method, 9 of the 16 eyes showed deterioration. There was moderate agreement between the results provided by our method and those obtained using the linear regression method (κ = 0.60; Fig. 3). Individual fluctuation of significantly deteriorated test points over time is shown in Fig. 4.

Loss of visual acuity by five or more letters was noted in 20 eyes at the final test, and 6 of these were among the 16 eyes with repeated deterioration of the visual field indicated by our method.

The number of test point locations showing deterioration varied considerably in patients older than 40 years of age, whereas almost no deteriorated points were seen in the few patients who were younger than 40 (Fig. 5A). The difference in age between those with and those without likely deterioration of visual fields (median ages 66 and 63 years, respectively) was not significant (P = 0.23). At the last visit, deterioration of visual fields was not explained by diabetes duration (Fig. 5B), mean HbA_1c levels during the study period (Fig. 5C), or diastolic blood pressure (Fig. 5D). HbA_1c levels at baseline did not correlate with visual field change at any follow-up assessment. On a group basis, HbA_1c values were the same at baseline as at the last visit.

Thirty-five eyes (49%) had no or questionable (n = 1) signs of vascular lesions (ETDRS level 10–15) at baseline, and the remaining 38 eyes had mild to moderate lesions (ETDRS level 20–43). Twenty-four eyes were graded as level 10 at all visits. The level of retinopathy was stable in 68 eyes, whereas the predefined two-step change occurred in only 6 eyes: progression in two of them and regression in the other four (from level 10 to 20, or conversely). The occurrence or disappearance of one or a few microaneurysms represented the only sign of structural vascular change.

No association was found between progression of retinopathy and likely deterioration of the visual field. In the last visual field test, the two patients with progression had, respectively, only one test point and two test points with significant deterioration. No test points showed significant improvement in the four patients with regression of retinopathy, but two of them had five or more test points exhibiting deterioration. Furthermore, the number of deteriorated test locations noted at the last visit did not differ between eyes without signs of vascular lesions and eyes with an ETDRS level >10 at any visit (median 2.5 [range 0–21] vs. 3 [range 0–37]).

To our knowledge, this is the first prospective longitudinal study to use SAP to evaluate diabetic subjects with no or mild/moderate diabetic retinopathy. This approach detected progression of early retinal dysfunction, even though the ETDRS final severity scale indicated stable retinopathy. Because visual field deterioration did not differ between subjects without and those with any signs of microvascular abnormalities and because the ETDRS level of retinopathy was stable over time, we propose that the neuronal dysfunction could represent an early feature of diabetic retinopathy due to primary neurodegeneration. A limitation of the study is that the number of subjects with progression of retinopathy during the 4 years of follow-up was insufficient to explore whether SAP is able to predict the development of microvascular impairment.

We used empirically derived limits for change based on test–retest variability previously measured in an independent sample of diabetic patients (18). Knowledge of such variability, together with longitudinal comparisons within individuals, allows more sensitive monitoring of visual function than is provided by cross-sectional comparisons of individuals. Had the limits been too narrow, a substantial proportion of significantly changed—both deteriorated and improved—test points would have been found. Our analysis yielded a substantial number of test locations showing significant deterioration but very few locations exhibiting improvement. In 27 eyes (36%), at least five test points showed deterioration in the last test. Furthermore, deterioration of at least five test points was noted in 16 eyes (22%) in the two last tests, which indicated a likely change, and those 16 eyes also had decreased MD values. Agreement was good between our method and the pointwise linear regression technique, although the latter identified fewer eyes with deterioration, indicating that our method is more sensitive. Our approach has long been applied to assess the eyes of patients with glaucoma (35), and in that disease has been reported to identify significant deterioration earlier than is possible with the regression method (39–41). Therefore, as expected, in the current study we found the regression method revealed fewer deteriorated eyes than our method.

To evaluate sensitive methods for detecting change over time, it is necessary to use a longitudinal study design. However, such an approach always entails the risk of dropouts, which can weaken the interpretation of the results. Fortunately, the dropout rate in our study was low: more than 90% of the subjects completed the 3 years of follow-up.

Several cross-sectional investigations using various perimetric and electrophysiological methods have suggested slight dysfunction in eyes with no diabetic retinopathy (12,13,16,20,21,23–26), but few longitudinal studies have considered this issue. One longitudinal investigation was conducted by Di Leo et al. (28) in 1994, who found that patients with type 1 diabetes, but without retinopathy, exhibited significant changes in ERG responses 3 years after baseline, and two studies demonstrated that slight depression of SAP (19) and delayed ERG response (29) predicted later onset of diabetic retinopathy. The strength of our study is that we performed a longitudinal collection of more functional data. By using SAP to test the visual field every 6 months for up to 3 years and thereafter annually, we were able to identify consistent deterioration in 22% of the eyes that were evaluated.

Factors other than diabetes-induced retinal dysfunction (e.g., cataract) can explain decreased perimetric sensitivity. Although cataract is known to increase with age (42), we found no significant difference in age between patients with and those without a likely visual field change. Cataract typically leads to general depression of the visual field and might also have a more pronounced effect on paracentral than on peripheral test points (43), but such a pattern was not discerned in our 16 subjects with likely deterioration. Although we did not specifically grade lens changes, that the functional changes detected by our method were caused by cataract is unlikely. Furthermore, we did not find any difference in change in perimetric foveal (central) threshold values between the 16 eyes with repeated deterioration and those without. The mean change in foveal threshold was −0.6 dB among the 16 eyes and −1.1 dB among the other 58 eyes. Thus, cataract development is unlikely to explain our finding of repeated deterioration in some eyes.

There was a poor correlation between likely visual field deterioration and visual acuity change, and most eyes with a likely visual field deterioration did not lose visual acuity. A limitation of the study is the lack of explanation for vision loss in a subset of subjects. The present five-letter cutoff for visual acuity change can be discussed, but test–retest variability is lower in eyes with good visual acuity, as was the case for the eyes in our study. Had the cutoff been set at nine letters, which has been reported as the overall coefficient of repeatability for diabetic subjects with various visual acuity (44), six subjects would have had a worsening of visual acuity at follow-up, and merely one of those had a likely visual field deterioration.

We found no association between SAP deterioration and duration of diabetes or HbA_1c levels in our cohort of diabetic subjects with stable metabolic control, which agrees with results obtained by Nitta et al. (45). Di Leo et al. (28) and Verrotti et al. (19) have also reported a lack of correlation between ERG and SAP, respectively, and metabolic control. We previously demonstrated that quite extensive intraindividual blood glucose fluctuations on a 1-month basis had no or little influence on SAP deterioration in a different cohort of diabetic subjects (18). Whether fluctuating HbA_1c levels on a long-term basis may affect visual function, including SAP, can only be speculated about.

This prospective longitudinal study is the first of its kind, and our results demonstrate that SAP with an analysis tailored for monitoring the eyes of patients with diabetes over time can reveal repeated deterioration of retinal neuronal function in this disease. We believe our method provides reliable assessments of visual function early in the course of diabetic retinopathy. Moreover, it can serve as an alternative end point for investigation of early visual disturbances, which can be useful when evaluating new treatment strategies for diabetic retinal disease, including neuroprotection.

Acknowledgments. The authors thank Dr. F. Lundin, Centre for Clinical Research, County Council of Värmland, for technical support in obtaining the change maps; J. Appelgren and D. Larsson, Karlstad University, for technical support in performing the pointwise linear regression analysis; and C. Byhlén and A. Lundh, Department of Ophthalmology, Karlstad, County Council of Värmland, for assessment of the high-quality fundus photographs.

Funding. This study was funded by the Department of Ophthalmology, Karlstad and Centre for Clinical Research, County Council of Värmland, the Järnhardt Foundation, the Foundation for Visually Impaired in Former Malmöhus län, the Swedish Medical Society, the Swedish Eye Foundation, Skåne University Hospital Foundation, the Cronqvist Foundation, and Lund University.

Duality of Interest. B.B. has received research funding from Carl Zeiss Meditec, Inc. and is a consultant of Carl Zeiss Meditec, Inc. The Department of Clinical Sciences–Ophthalmology at Lund University in Malmö, has received research funding from Carl Zeiss Meditec, Inc. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. K.-J.H. organized the project, recruited the subjects, performed all measurements, analyzed the data, wrote a manuscript draft, edited the manuscript, and approved the final draft. E.A. and B.B. originated the project, edited the manuscript, and approved the final draft. K.-J.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 20th Meeting of the European Association for the Study of Diabetes Eye Complications Study Group, Paris, France, 21–22 May 2010; at the Annual Meeting of the Association for Research in Vision and Ophthalmology (ARVO), Fort Lauderdale, FL, 6–10 May 2012; and at the Annual Meeting of the Swedish Ophthalmological Society, Malmö, Sweden, 29–31 August 2013.