Retina
Localizing Functional Damage in the Neural Retina of Adolescents and Young Adults With Type 1 Diabetes Wylie Tan,1,2 Tom Wright,1 Annie Dupuis,3,4 Ekta Lakhani,1 and Carol Westall1,5 1
Department of Ophthalmology and Vision Sciences, The Hospital for Sick Children, Toronto, Canada School of Optometry and Vision Science, University of Waterloo, Waterloo, Canada 3Clinical Research Services, The Hospital for Sick Children, Toronto, Canada 4 Dalla Lana School of Public Health, University of Toronto, Toronto, Canada 5 Department of Ophthalmology and Vision Sciences, University of Toronto, Toronto, Canada 2
Correspondence: Carol Westall, Department of Ophthalmology and Vision Sciences, The Hospital for Sick Children, 555 University Avenue, Toronto, ON, Canada, M5G 1X8; [email protected]. Submitted: September 9, 2013 Accepted: February 1, 2014 Citation: Tan W, Wright T, Dupuis A, Lakhani E, Westall C. Localizing functional damage in the neural retina of adolescents and young adults with type 1 diabetes. Invest Ophthalmol Vis Sci. 2014;55:2432–2441. DOI: 10.1167/iovs.13-13232
PURPOSE. It is unknown which regions of the retina are most susceptible to damage by diabetes mellitus. We hypothesized that the standard and slow-flash (sf-) multifocal electroretinogram (mfERG) will localize retinal regions of greatest vulnerability. METHODS. A total of 55 adolescents and young adults with type 1 diabetes and without diabetic retinopathy (DR) or with mild nonproliferative DR and 54 typically-developing, age-similar control participants underwent mfERG and sf-mfERG testing. The amplitude and implicit time of the first order response of the standard mfERG and of three multifocal oscillatory potentials (mfOPs) of the sf-mfERG were compared between groups at the level of hexagons, quadrants, and rings using separate mixed model ANOVAs. Spatial mapping of the P values from post hoc pairwise comparisons illustrated patterns of retinal dysfunction. RESULTS. Delays in mfERG implicit times were evident across the tested retinal areas in the diabetes group. Delays in sf-mfERG implicit times were found at different eccentricities for each mfOP in the diabetes group. The greatest delays were noted in the periphery for mfOP1, in the midperiphery for mfOP2, and in the macular region for mfOP3. There were no significant group differences in amplitude for the mfERG and sf-mfERG protocols. CONCLUSIONS. Delays in mfERG and sf-mfERG responses suggest that the inner retina is particularly vulnerable to diabetes. Localizing regions of early dysfunction will help guide future studies to examine early structural damage associated with DR. Keywords: type 1 diabetes, multifocal electroretinogram, dysfunction, inner retina, oscillatory potentials
iabetic retinopathy (DR) is a common complication of diabetes mellitus and the leading cause of new blindness in young individuals in developed countries.1,2 Approximately 93 million people live with some form of DR worldwide.3 The prevalence of DR has been shown to be higher in individuals with type 1 diabetes (T1D) compared to type 2 diabetes (T2D), with a landmark study reporting 97.5% prevalence of DR among their participants with 15þ years of T1D.4 The DR prevalence rates have since declined,5,6 with latest global estimates reporting a prevalence of approximately 77% in this population.3 The DR will, however, continue to be a global health concern because of the expected increase in the number of people affected by diabetes.7 Hallmark diagnostic features of DR consist of clinically visible microvascular lesions identified through fundus photography or ophthalmoscopy. These lesions are characterized based on the modified Airlie House classification system8 and serve as the standard for diagnosing DR. Retinal dysfunction occurs before these vascular lesions are detected clinically,9–11 suggesting that early functional disturbances in the neural retina may serve as effective biomarkers for DR. It has long been known that oscillatory potentials of the electroretinogram (ERG) are one of the earliest electroretinographic components to be affected and they have been
suggested to be the most sensitive marker of neuroretinal dysfunction in diabetes.9–11 The global response properties of the ERG cannot identify localized regions of early functional damage. The standard and slow-flash (sf-) multifocal electroretinogram (mfERG) may be more ideal tools to localize early retinal damage. Several studies have assessed localized retinal dysfunction in participants with diabetes; the majority, however, focused on adult populations with T2D or a mixture of participants with T1D and T2D.12–16 The Wisconsin Epidemiological Study of Diabetic Retinopathy (WESDR)4 emphasized that duration of diabetes is a strong risk factor for DR and that prevalence of DR is much higher in individuals with T1D compared to T2D. Therefore, it is of interest to identify sensitive markers for DR as early as possible in the disease process, making adolescents and young adults with T1D an ideal study population. Few studies have investigated the retina’s vulnerability to diabetes topographically.17 This study will use the standard mfERG and sf-mfERG to assess the functional integrity of the outer and inner neural retina, respectively, in an adolescent and young adult population with T1D. The mfERG and sf-mfERG responses will be compared spatially between participants with and without T1D at the level of hexagons, quadrants, and rings.
Copyright 2014 The Association for Research in Vision and Ophthalmology, Inc. www.iovs.org j ISSN: 1552-5783
2432
D
IOVS j April 2014 j Vol. 55 j No. 4 j 2433
Localizing Functional Damage in the Diabetic Retina Findings from this study will identify regions of the retina that are most vulnerable to damage by the effects of diabetes.
METHODS Subjects A total of 55 adolescents and young adults with T1D and without DR, or with mild nonproliferative DR (NPDR) was recruited at The Hospital for Sick Children. Inclusion criteria were duration of T1D of ‡5 years and age 10 to 25 years. Participants with moderate or severe NPDR, or proliferative DR were excluded either by fundus examination by an ophthalmologist or based on seven-field, 308 stereoscopic fundus photographs graded by a retinal specialist according to the modified Airlie House classification system.8 A total of 54 typically-developing, age-similar participants without diabetes acted as control participants. All participants with other eye diseases, hemoglobinopathy, high refractive error (worse than 65 diopters [D]), poor visual acuity (worse than 0.3 logMAR), neurologic disorders, and those on medications affecting visual or retinal function were excluded. Informed consent was obtained from all participants after the purpose, protocol, and potential harms and benefits of the study were explained. All procedures were approved by the Research Ethics Board at The Hospital for Sick Children and were conducted in compliance with the tenets of the Declaration of Helsinki.
Data Acquisition All participants were tested at The Hospital for Sick Children. Since acute changes in ambient blood glucose are known to affect mfERG responses,18,19 blood glucose readings were measured with a glucose meter (OneTouch Ultra; LifeScan, Inc., Milpitas, CA, USA) at least three times: before psychophysical testing, before mfERG testing, and after mfERG testing. Participants with T1D completed light exercises, or were administered food or insulin, as advised by a registered nurse, to adjust and maintain blood glucose levels within a 4 to 10 mmol/L range. One eye was selected randomly for testing from each participant and the untested eye was occluded. All participants were assessed for visual acuity (Early Treatment of Diabetic Retinopathy Study [ETDRS], logMAR) and contrast sensitivity (Pelli-Robson). Color vision was assessed with the Hardy-RandRittler (HRR) pseudoisochromatic plates and the Mollon-Reffin Minimalist Test. The tested eye was anesthetized with a topical corneal anesthetic (0.5% proparacaine) and dilated pharmacologically (2.5% phenylephrine and 1% tropicamide). Refractive error was measured post dilation. Date of T1D diagnosis and hemoglobin A1c (HbA1c) values closest to the testing date were obtained from hospital records.
Multifocal ERG Participants were tested on the mfERG and sf-mfERG protocols with the VERIS FMSII Science System (Electro-Diagnostic Imaging, Inc., Redwood City, CA, USA) according to International Society for Clinical Electrophysiology of Vision (ISCEV) guidelines.20 A red cross was used as the fixation target and it was enlarged as required to ensure that the target was visible during recordings. The FMSII Stimulator was adjusted to ensure that the mfERG hexagons were in focus. The standard mfERG protocol consisted of a stimulus array with 103 hexagons that subtended a field diameter of 408 vertical 3 458 horizontal. Hexagons were scaled for eccentricity such that all responses had approximately equal amplitudes across a healthy retina. Each hexagon alternated between black
FIGURE 1. Amplitude (dashed arrow) and implicit time (solid arrow) measurements for the standard mfERG (A) and sf-mfERG with three mfOPs (B) and two mfOPs (C) from a control participant. Gray columns represent the 5-ms time span searched for an mfOP peak.
(0 cd/m2) and white (200 cd/m2) following an algorithmic pseudorandom m-sequence (215 1) with a base rate of 13.3 ms. For every frame, each hexagon had a 50% chance of being illuminated (mean luminance 100 cd/m2). The sf-mfERG protocol consisted of a stimulus array of 61 hexagons that subtended a similar field to the mfERG. Hexagons were again scaled for eccentricity. Each step of the m-sequence (212 1) consisted of six frames: in the first frame, each hexagon had a 50% chance of being white and in the subsequent five frames, all hexagons were black. The multifocal frames were separated by 79.8 ms, allowing for the development of inner retinal responses. A bipolar B¨ urian-Allen contact lens electrode (Hansen Ophthalmic Development Laboratory, Iowa City, IA, USA) was used for recordings with a gold-plated electrode (Grass Technologies, Warwick, RI) on the forehead to serve as ground. Incoming signals were amplified (350,000) and filtered with an analog filter (band-pass 10–300 Hz). Additional filtering was accomplished with digital filtering for mfERG (band stop filter at 60 Hz power) and sf-mfERG (75 Hz high pass filter). Two iterations of artifact removal were used and spatial averaging was not used as this option reduced the spatial accuracy of the data. Participant fixation was monitored with a built-in infrared fundus camera. Recording time was divided into 16 segments for participant comfort. Segments with fixation loss or noise artifacts were repeated. Total recording time for the mfERG (8 minutes) and sf-mfERG (~7 minutes) protocols was approximately 15 minutes.
Data Analysis The outcome measures for the mfERG recordings were the amplitude and implicit time of the first order response of each hexagon. The amplitude was measured from trough-to-peak (N1 to P1) and the implicit time was measured from the time of frame presentation to the time of the major peak (Fig. 1A). The outcome measures for the sf-mfERG recordings were the amplitude and implicit time of the three multifocal oscillatory potentials (mfOPs). Peaks and troughs were identified with a custom script written in R Statistical Analysis Software Version 2.13.1.21 The script identified peaks and
IOVS j April 2014 j Vol. 55 j No. 4 j 2434
Localizing Functional Damage in the Diabetic Retina
FIGURE 2. Stimulus array for mfERG (A–C) and sf-mfERG (D–F), divided into hexagons (A, D), quadrants (B, E), and rings (C, F).
troughs by comparing each data point with the two preceding data points and the two following data points; a data point was identified as a peak if it had the greatest positivity and a trough if it had the greatest negativity within a span of 5 ms (Fig. 1B). The amplitude of each mfOP was measured from trough-topeak, and amplitudes smaller than 10 nV were considered noise and omitted manually. Implicit time was measured from the time of stimulus presentation to the time of the peak of each mfOP. Since the timings of the mfOPs varied with hexagon location, all control data were averaged to find the approximate timing of mfOP1, mfOP2, and mfOP3 at each hexagon. These timings acted as guidelines to classify the peaks identified for all participants. Peaks were classified as mfOP1, mfOP2, or mfOP3 based on the proximity of the peak’s implicit time to the guidelines. In several cases, participants with and without diabetes had occasional hexagons that presented with two mfOPs rather than three. These two mfOPs were classified according to the guidelines and it was assumed that one of the three mfOPs was missing (Fig. 1C). It usually was mfOP2 or mfOP3 that was missing, and rarely mfOP1. Almost all participants had at least 1 of 61 hexagons with a missing mfOP and the maximum number of hexagons with a missing mfOP in a single participant was approximately 30 hexagons. All left eye recordings were converted to the right eye, such that temporal regions of the left eye were compared to temporal regions of the right eye and similarly for the nasal regions.
Mixed Model ANOVA To examine spatial differences in retinal function, betweengroup comparisons were made at the level of hexagons, quadrants, and rings with separate mixed model ANOVAs. The mfERG and sf-mfERG stimulus patterns were divided into four quadrants (along horizontal and vertical meridians from the fovea) and rings (six for mfERG and five for sf-mfERG, Fig. 2).
Quadrant and ring amplitude, and implicit time responses were comprised of the average of the amplitude and implicit time of all of the hexagons within each region. For mfERG and sfmfERG mixed model ANOVAs, the independent variable was group (T1D or control), the dependent variable was amplitude or implicit time, and the repeated measure variable was hexagon, quadrant, or ring. Mixed model analyses were performed using SAS version 9.2 (SAS Institute, Inc., Rockville, MD, USA). Group means were compared using post hoc pairwise comparisons. Bonferroni corrections were applied to account for the multiple comparisons: a ¼ 0.05 was divided by the number of comparisons or levels within the repeated measure variable. The Bonferroni adjustment is most appropriate when hypothesis tests are independent. As observations across hexagons, quadrants, and rings were correlated, the Bonferroni correction in this case is overly conservative. Thus, all results are presented as actual P values for the reader’s discretion.
RESULTS Demographic data and psychophysical testing results for all participants are shown in the Table. There were no significant between-group differences in age, visual acuity, or contrast sensitivity. Two participants with T1D had mild red-green color vision defects. Most participants with T1D (45/53) were diagnosed before the age of 10 and approximately 70% had been diagnosed with T1D for ‡8 years. Date of diagnosis was not available for two participants. The HbA1c values were not available for six participants because they were over the age of 18 and had left The Hospital for Sick Children. Over 89% of the participants with T1D had HbA1c levels above the recommended HbA1c target of 7% set by the Canadian Diabetes Association.22 Eight participants with diabetes had early NPDR. After qualitative comparisons between the fundus photographs and mfERG traces of these eight participants, no correlations
IOVS j April 2014 j Vol. 55 j No. 4 j 2435
Localizing Functional Damage in the Diabetic Retina
TABLE. Demographic Data and Psychophysical Testing Results for Participants with T1D and Control Participants T1D Participants, n ¼ 55 Sex, male/female NPDR, n Age at testing, y Age at diagnosis, y Duration of T1D, y HbA1c, % Visual acuity, logMAR Contrast sensitivity
16.6 7.2 9.5 8.6 0.01 1.68
6 6 6 6 6 6
26/29 8 2.3 (12.1–22.1) 3.9 (1.4–16.3) 3.1 (3.9–15.6) 1.3 (6.4–12.3) 0.12 (0.28–0.28) 0.13 (1.05–2.05)
Control Participants, n ¼ 54 19/35 – 17.1 6 3.6 (10.7–25.1) – – – 0.07 6 0.13 (0.60–0.24) 1.69 6 0.16 (1.05–2.20)
Data presented as mean 6 SD (range).
between morphologic NPDR findings and local mfERG delays were found. Data from most of the participants, with and without diabetes, have been included in prior reports.23–25 This study uses a different analysis technique to assess the spatial relationships of the mfERG and sf-mfERG data, and uses the peak-picking technique to analyze waveforms rather than waveform fitting techniques. The mixed model analyses were conducted for the mfERG and sf-mfERG data. Rerunning the models for the sf-mfERG data after removing the observations with two mfOPs only yielded similar results, possibly because the number of hexagons with only two mfOPs was relatively small (accounting for 15% and 13% of all data for participants with and without diabetes, respectively) compared to the hexagons with three mfOPs. Rerunning the mixed models after removing the participants with mild NPDR also returned the same results. The analyses presented herein included all of the mfERG and sf-mfERG data from participants with and without mild NPDR.
were delayed in participants with T1D compared to control participants at most locations (>60%, a ¼ 0.05). Implicit times of participants with T1D were significantly delayed at three hexagons only with the Bonferroni correction (a ¼ 0.0005). It is important to note, however, that the Bonferroni correction is the most conservative correction for multiple comparisons. Spatial mapping helps to visualize the number of hexagons that are approaching significance. Over 50% of the hexagons have a P < 0.02, indicating that implicit times of participants with T1D are delayed compared to controls across the majority of the retina. Figure 4 shows representative mfERG trace arrays from 2 participants with diabetes. Quadrant (Fig. 5) and ring (Fig. 6) analyses showed that implicit time of participants with T1D were significantly delayed compared to control participants in all quadrants and eccentricities except at the fovea. These comparisons were significant after the Bonferroni correction (a ¼ 0.01 for quadrants and a ¼ 0.0008 for rings).
Mixed Models for sf-mfERG Mixed Models for mfERG Mixed model analyses for mfERG implicit time yielded main effects for group at the level of hexagons (P < 0.0001), quadrants (P < 0.0001), and rings (P < 0.0001), but not for amplitude. The P values for the post hoc pairwise comparisons from the hexagon analysis are presented in a spatial color map to illustrate the distribution of dysfunction (Fig. 3). Positive (þ) and negative () signs denote whether participants with T1D had earlier or delayed implicit times, respectively, compared to control participants. Implicit times
Mixed model analyses for mfOP1, mfOP2, and mfOP3 of the sfmfERG yielded significant between-group differences for implicit time, but not for amplitude. Significant main effects were found for mfOP1 at the level of hexagons (P ¼ 0.0003) and quadrants (P < 0.0001). Significant ring differences were found for mfOP2 (P ¼ 0.007) and mfOP3 (P ¼ 0.0008), mfOP1 approached significance (P ¼ 0.06). The P values for the post hoc pairwise comparisons at the level of hexagons were mapped spatially in color plots (Fig. 7). The areas of interest are those in which the implicit time of
FIGURE 3. Spatial color map representing the mfERG implicit time differences between groups at the level of hexagons (illustrated on an unscaled hexagon array).
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2436
FIGURE 4. Examples of representative mfERG trace arrays from two participants (A, B) with diabetes. Blue traces represent individual patient data and dashed black traces represent average control data. Shaded hexagons indicate responses that are significantly delayed (>2 SD from controls) (illustrated on unscaled hexagon arrays).
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2437
FIGURE 5. Comparison of mfERG implicit time between groups at the level of quadrants. Data presented as mean 6 95% confidence limits. Bonferroni correction, P < 0.01.
participants with T1D was delayed significantly compared to control participants at a ¼ 0.05 (color toward the red end of the spectrum with negative sign). The distribution of these delayed regions was different for each mfOP: the red hexagons were concentrated in the periphery for mfOP1, in the midperiphery for mfOP2, and in the macular region for mfOP3. Although there are minimal areas with significant
differences with the Bonferroni correction (a ¼ 0.0008), it is important to note the pattern in the distribution of P values. Ring analysis showed a similar distribution of delays (Fig. 7): the greatest implicit time delays were observed in Ring 4 for mfOP1 (P < 0.0001), Ring 3 for mfOP2 (P < 0.0001), and Ring 2 for mfOP3 (P < 0.0001). These differences are significant with the Bonferroni correction (a ¼ 0.01). Figure 8 shows
FIGURE 6. Comparison of mfERG implicit time between groups at the level of rings. Data presented as mean 6 95% confidence limits. Bonferroni correction, P < 0.0008.
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2438
FIGURE 7. Spatial color maps illustrating the distribution of mfOP implicit time differences between groups: both hexagon and ring analyses show that each mfOP is most delayed at different eccentricities (illustrated on unscaled hexagon arrays).
representative sf-mfERG trace arrays from two participants with diabetes. Quadrant analysis showed that mfOP1 is consistently more delayed in participants with T1D compared to control participants in all four retinal quadrants: superior temporal (P < 0.0001), superior nasal (P ¼ 0.007), inferior temporal (P ¼ 0.004), inferior nasal (P ¼ 0.001).
DISCUSSION This study found no change in mfERG amplitude, but mfERG implicit times were delayed across the retina in the diabetes group. This is consistent with mfERG literature, where there has been no consensus as to whether mfERG amplitude is affected by diabetes,14,26,27 but multifocal ERG implicit times has been reported consistently as delayed in participants with diabetes,14,23,27 and delayed to greater extents with increasing DR severity.26,28 Quadrant and ring analyses of the mfERG data revealed that the majority of the retina is affected by diabetes, as participants with T1D showed response delays across all quadrants and eccentricities. Bearse et al.12 similarly studied and mapped retinal dysfunction in participants with diabetes, and found a spatial preference of vulnerability to the effects of diabetes. They used an sf-mfERG protocol, rather than the mfERG protocol, to evaluate first-order N1, P1, and N2 implicit times, and found that P1 implicit time delays are most frequent in the
inferior and peripheral retina, and, thus, concluded these regions are most vulnerable to damage. A few key factors differentiate our studies, including differences in the protocol used, the population studied, and the analysis technique. Bearse et al.12 studied an adult population with a mix of participants with T1D and T2D, whereas our study examined an adolescent and young adult population with T1D only. The advantages of studying biomarkers with a younger rather than an older adult population have been discussed previously.4 Bearse et al.12 plotted the frequency of delays at each hexagon and defined the region most frequently associated with abnormalities as the region of greatest vulnerability. In our study, the region of greatest vulnerability was defined as the region in which the mean implicit time of participants with T1D differed most significantly with the mean implicit time of control participants. A comprehensive review conducted by Hood29 examined studies that used the mfERG to assess different retinal diseases. He derived a guide to locate the site of damage within the retina based on characteristic changes to the mfERG waveform. No amplitude changes and a small implicit time delay (
Localizing Functional Damage in the Neural Retina of Adolescents and Young Adults With Type 1 Diabetes Wylie Tan,1,2 Tom Wright,1 Annie Dupuis,3,4 Ekta Lakhani,1 and Carol Westall1,5 1
Department of Ophthalmology and Vision Sciences, The Hospital for Sick Children, Toronto, Canada School of Optometry and Vision Science, University of Waterloo, Waterloo, Canada 3Clinical Research Services, The Hospital for Sick Children, Toronto, Canada 4 Dalla Lana School of Public Health, University of Toronto, Toronto, Canada 5 Department of Ophthalmology and Vision Sciences, University of Toronto, Toronto, Canada 2
Correspondence: Carol Westall, Department of Ophthalmology and Vision Sciences, The Hospital for Sick Children, 555 University Avenue, Toronto, ON, Canada, M5G 1X8; [email protected]. Submitted: September 9, 2013 Accepted: February 1, 2014 Citation: Tan W, Wright T, Dupuis A, Lakhani E, Westall C. Localizing functional damage in the neural retina of adolescents and young adults with type 1 diabetes. Invest Ophthalmol Vis Sci. 2014;55:2432–2441. DOI: 10.1167/iovs.13-13232
PURPOSE. It is unknown which regions of the retina are most susceptible to damage by diabetes mellitus. We hypothesized that the standard and slow-flash (sf-) multifocal electroretinogram (mfERG) will localize retinal regions of greatest vulnerability. METHODS. A total of 55 adolescents and young adults with type 1 diabetes and without diabetic retinopathy (DR) or with mild nonproliferative DR and 54 typically-developing, age-similar control participants underwent mfERG and sf-mfERG testing. The amplitude and implicit time of the first order response of the standard mfERG and of three multifocal oscillatory potentials (mfOPs) of the sf-mfERG were compared between groups at the level of hexagons, quadrants, and rings using separate mixed model ANOVAs. Spatial mapping of the P values from post hoc pairwise comparisons illustrated patterns of retinal dysfunction. RESULTS. Delays in mfERG implicit times were evident across the tested retinal areas in the diabetes group. Delays in sf-mfERG implicit times were found at different eccentricities for each mfOP in the diabetes group. The greatest delays were noted in the periphery for mfOP1, in the midperiphery for mfOP2, and in the macular region for mfOP3. There were no significant group differences in amplitude for the mfERG and sf-mfERG protocols. CONCLUSIONS. Delays in mfERG and sf-mfERG responses suggest that the inner retina is particularly vulnerable to diabetes. Localizing regions of early dysfunction will help guide future studies to examine early structural damage associated with DR. Keywords: type 1 diabetes, multifocal electroretinogram, dysfunction, inner retina, oscillatory potentials
iabetic retinopathy (DR) is a common complication of diabetes mellitus and the leading cause of new blindness in young individuals in developed countries.1,2 Approximately 93 million people live with some form of DR worldwide.3 The prevalence of DR has been shown to be higher in individuals with type 1 diabetes (T1D) compared to type 2 diabetes (T2D), with a landmark study reporting 97.5% prevalence of DR among their participants with 15þ years of T1D.4 The DR prevalence rates have since declined,5,6 with latest global estimates reporting a prevalence of approximately 77% in this population.3 The DR will, however, continue to be a global health concern because of the expected increase in the number of people affected by diabetes.7 Hallmark diagnostic features of DR consist of clinically visible microvascular lesions identified through fundus photography or ophthalmoscopy. These lesions are characterized based on the modified Airlie House classification system8 and serve as the standard for diagnosing DR. Retinal dysfunction occurs before these vascular lesions are detected clinically,9–11 suggesting that early functional disturbances in the neural retina may serve as effective biomarkers for DR. It has long been known that oscillatory potentials of the electroretinogram (ERG) are one of the earliest electroretinographic components to be affected and they have been
suggested to be the most sensitive marker of neuroretinal dysfunction in diabetes.9–11 The global response properties of the ERG cannot identify localized regions of early functional damage. The standard and slow-flash (sf-) multifocal electroretinogram (mfERG) may be more ideal tools to localize early retinal damage. Several studies have assessed localized retinal dysfunction in participants with diabetes; the majority, however, focused on adult populations with T2D or a mixture of participants with T1D and T2D.12–16 The Wisconsin Epidemiological Study of Diabetic Retinopathy (WESDR)4 emphasized that duration of diabetes is a strong risk factor for DR and that prevalence of DR is much higher in individuals with T1D compared to T2D. Therefore, it is of interest to identify sensitive markers for DR as early as possible in the disease process, making adolescents and young adults with T1D an ideal study population. Few studies have investigated the retina’s vulnerability to diabetes topographically.17 This study will use the standard mfERG and sf-mfERG to assess the functional integrity of the outer and inner neural retina, respectively, in an adolescent and young adult population with T1D. The mfERG and sf-mfERG responses will be compared spatially between participants with and without T1D at the level of hexagons, quadrants, and rings.
Copyright 2014 The Association for Research in Vision and Ophthalmology, Inc. www.iovs.org j ISSN: 1552-5783
2432
D
IOVS j April 2014 j Vol. 55 j No. 4 j 2433
Localizing Functional Damage in the Diabetic Retina Findings from this study will identify regions of the retina that are most vulnerable to damage by the effects of diabetes.
METHODS Subjects A total of 55 adolescents and young adults with T1D and without DR, or with mild nonproliferative DR (NPDR) was recruited at The Hospital for Sick Children. Inclusion criteria were duration of T1D of ‡5 years and age 10 to 25 years. Participants with moderate or severe NPDR, or proliferative DR were excluded either by fundus examination by an ophthalmologist or based on seven-field, 308 stereoscopic fundus photographs graded by a retinal specialist according to the modified Airlie House classification system.8 A total of 54 typically-developing, age-similar participants without diabetes acted as control participants. All participants with other eye diseases, hemoglobinopathy, high refractive error (worse than 65 diopters [D]), poor visual acuity (worse than 0.3 logMAR), neurologic disorders, and those on medications affecting visual or retinal function were excluded. Informed consent was obtained from all participants after the purpose, protocol, and potential harms and benefits of the study were explained. All procedures were approved by the Research Ethics Board at The Hospital for Sick Children and were conducted in compliance with the tenets of the Declaration of Helsinki.
Data Acquisition All participants were tested at The Hospital for Sick Children. Since acute changes in ambient blood glucose are known to affect mfERG responses,18,19 blood glucose readings were measured with a glucose meter (OneTouch Ultra; LifeScan, Inc., Milpitas, CA, USA) at least three times: before psychophysical testing, before mfERG testing, and after mfERG testing. Participants with T1D completed light exercises, or were administered food or insulin, as advised by a registered nurse, to adjust and maintain blood glucose levels within a 4 to 10 mmol/L range. One eye was selected randomly for testing from each participant and the untested eye was occluded. All participants were assessed for visual acuity (Early Treatment of Diabetic Retinopathy Study [ETDRS], logMAR) and contrast sensitivity (Pelli-Robson). Color vision was assessed with the Hardy-RandRittler (HRR) pseudoisochromatic plates and the Mollon-Reffin Minimalist Test. The tested eye was anesthetized with a topical corneal anesthetic (0.5% proparacaine) and dilated pharmacologically (2.5% phenylephrine and 1% tropicamide). Refractive error was measured post dilation. Date of T1D diagnosis and hemoglobin A1c (HbA1c) values closest to the testing date were obtained from hospital records.
Multifocal ERG Participants were tested on the mfERG and sf-mfERG protocols with the VERIS FMSII Science System (Electro-Diagnostic Imaging, Inc., Redwood City, CA, USA) according to International Society for Clinical Electrophysiology of Vision (ISCEV) guidelines.20 A red cross was used as the fixation target and it was enlarged as required to ensure that the target was visible during recordings. The FMSII Stimulator was adjusted to ensure that the mfERG hexagons were in focus. The standard mfERG protocol consisted of a stimulus array with 103 hexagons that subtended a field diameter of 408 vertical 3 458 horizontal. Hexagons were scaled for eccentricity such that all responses had approximately equal amplitudes across a healthy retina. Each hexagon alternated between black
FIGURE 1. Amplitude (dashed arrow) and implicit time (solid arrow) measurements for the standard mfERG (A) and sf-mfERG with three mfOPs (B) and two mfOPs (C) from a control participant. Gray columns represent the 5-ms time span searched for an mfOP peak.
(0 cd/m2) and white (200 cd/m2) following an algorithmic pseudorandom m-sequence (215 1) with a base rate of 13.3 ms. For every frame, each hexagon had a 50% chance of being illuminated (mean luminance 100 cd/m2). The sf-mfERG protocol consisted of a stimulus array of 61 hexagons that subtended a similar field to the mfERG. Hexagons were again scaled for eccentricity. Each step of the m-sequence (212 1) consisted of six frames: in the first frame, each hexagon had a 50% chance of being white and in the subsequent five frames, all hexagons were black. The multifocal frames were separated by 79.8 ms, allowing for the development of inner retinal responses. A bipolar B¨ urian-Allen contact lens electrode (Hansen Ophthalmic Development Laboratory, Iowa City, IA, USA) was used for recordings with a gold-plated electrode (Grass Technologies, Warwick, RI) on the forehead to serve as ground. Incoming signals were amplified (350,000) and filtered with an analog filter (band-pass 10–300 Hz). Additional filtering was accomplished with digital filtering for mfERG (band stop filter at 60 Hz power) and sf-mfERG (75 Hz high pass filter). Two iterations of artifact removal were used and spatial averaging was not used as this option reduced the spatial accuracy of the data. Participant fixation was monitored with a built-in infrared fundus camera. Recording time was divided into 16 segments for participant comfort. Segments with fixation loss or noise artifacts were repeated. Total recording time for the mfERG (8 minutes) and sf-mfERG (~7 minutes) protocols was approximately 15 minutes.
Data Analysis The outcome measures for the mfERG recordings were the amplitude and implicit time of the first order response of each hexagon. The amplitude was measured from trough-to-peak (N1 to P1) and the implicit time was measured from the time of frame presentation to the time of the major peak (Fig. 1A). The outcome measures for the sf-mfERG recordings were the amplitude and implicit time of the three multifocal oscillatory potentials (mfOPs). Peaks and troughs were identified with a custom script written in R Statistical Analysis Software Version 2.13.1.21 The script identified peaks and
IOVS j April 2014 j Vol. 55 j No. 4 j 2434
Localizing Functional Damage in the Diabetic Retina
FIGURE 2. Stimulus array for mfERG (A–C) and sf-mfERG (D–F), divided into hexagons (A, D), quadrants (B, E), and rings (C, F).
troughs by comparing each data point with the two preceding data points and the two following data points; a data point was identified as a peak if it had the greatest positivity and a trough if it had the greatest negativity within a span of 5 ms (Fig. 1B). The amplitude of each mfOP was measured from trough-topeak, and amplitudes smaller than 10 nV were considered noise and omitted manually. Implicit time was measured from the time of stimulus presentation to the time of the peak of each mfOP. Since the timings of the mfOPs varied with hexagon location, all control data were averaged to find the approximate timing of mfOP1, mfOP2, and mfOP3 at each hexagon. These timings acted as guidelines to classify the peaks identified for all participants. Peaks were classified as mfOP1, mfOP2, or mfOP3 based on the proximity of the peak’s implicit time to the guidelines. In several cases, participants with and without diabetes had occasional hexagons that presented with two mfOPs rather than three. These two mfOPs were classified according to the guidelines and it was assumed that one of the three mfOPs was missing (Fig. 1C). It usually was mfOP2 or mfOP3 that was missing, and rarely mfOP1. Almost all participants had at least 1 of 61 hexagons with a missing mfOP and the maximum number of hexagons with a missing mfOP in a single participant was approximately 30 hexagons. All left eye recordings were converted to the right eye, such that temporal regions of the left eye were compared to temporal regions of the right eye and similarly for the nasal regions.
Mixed Model ANOVA To examine spatial differences in retinal function, betweengroup comparisons were made at the level of hexagons, quadrants, and rings with separate mixed model ANOVAs. The mfERG and sf-mfERG stimulus patterns were divided into four quadrants (along horizontal and vertical meridians from the fovea) and rings (six for mfERG and five for sf-mfERG, Fig. 2).
Quadrant and ring amplitude, and implicit time responses were comprised of the average of the amplitude and implicit time of all of the hexagons within each region. For mfERG and sfmfERG mixed model ANOVAs, the independent variable was group (T1D or control), the dependent variable was amplitude or implicit time, and the repeated measure variable was hexagon, quadrant, or ring. Mixed model analyses were performed using SAS version 9.2 (SAS Institute, Inc., Rockville, MD, USA). Group means were compared using post hoc pairwise comparisons. Bonferroni corrections were applied to account for the multiple comparisons: a ¼ 0.05 was divided by the number of comparisons or levels within the repeated measure variable. The Bonferroni adjustment is most appropriate when hypothesis tests are independent. As observations across hexagons, quadrants, and rings were correlated, the Bonferroni correction in this case is overly conservative. Thus, all results are presented as actual P values for the reader’s discretion.
RESULTS Demographic data and psychophysical testing results for all participants are shown in the Table. There were no significant between-group differences in age, visual acuity, or contrast sensitivity. Two participants with T1D had mild red-green color vision defects. Most participants with T1D (45/53) were diagnosed before the age of 10 and approximately 70% had been diagnosed with T1D for ‡8 years. Date of diagnosis was not available for two participants. The HbA1c values were not available for six participants because they were over the age of 18 and had left The Hospital for Sick Children. Over 89% of the participants with T1D had HbA1c levels above the recommended HbA1c target of 7% set by the Canadian Diabetes Association.22 Eight participants with diabetes had early NPDR. After qualitative comparisons between the fundus photographs and mfERG traces of these eight participants, no correlations
IOVS j April 2014 j Vol. 55 j No. 4 j 2435
Localizing Functional Damage in the Diabetic Retina
TABLE. Demographic Data and Psychophysical Testing Results for Participants with T1D and Control Participants T1D Participants, n ¼ 55 Sex, male/female NPDR, n Age at testing, y Age at diagnosis, y Duration of T1D, y HbA1c, % Visual acuity, logMAR Contrast sensitivity
16.6 7.2 9.5 8.6 0.01 1.68
6 6 6 6 6 6
26/29 8 2.3 (12.1–22.1) 3.9 (1.4–16.3) 3.1 (3.9–15.6) 1.3 (6.4–12.3) 0.12 (0.28–0.28) 0.13 (1.05–2.05)
Control Participants, n ¼ 54 19/35 – 17.1 6 3.6 (10.7–25.1) – – – 0.07 6 0.13 (0.60–0.24) 1.69 6 0.16 (1.05–2.20)
Data presented as mean 6 SD (range).
between morphologic NPDR findings and local mfERG delays were found. Data from most of the participants, with and without diabetes, have been included in prior reports.23–25 This study uses a different analysis technique to assess the spatial relationships of the mfERG and sf-mfERG data, and uses the peak-picking technique to analyze waveforms rather than waveform fitting techniques. The mixed model analyses were conducted for the mfERG and sf-mfERG data. Rerunning the models for the sf-mfERG data after removing the observations with two mfOPs only yielded similar results, possibly because the number of hexagons with only two mfOPs was relatively small (accounting for 15% and 13% of all data for participants with and without diabetes, respectively) compared to the hexagons with three mfOPs. Rerunning the mixed models after removing the participants with mild NPDR also returned the same results. The analyses presented herein included all of the mfERG and sf-mfERG data from participants with and without mild NPDR.
were delayed in participants with T1D compared to control participants at most locations (>60%, a ¼ 0.05). Implicit times of participants with T1D were significantly delayed at three hexagons only with the Bonferroni correction (a ¼ 0.0005). It is important to note, however, that the Bonferroni correction is the most conservative correction for multiple comparisons. Spatial mapping helps to visualize the number of hexagons that are approaching significance. Over 50% of the hexagons have a P < 0.02, indicating that implicit times of participants with T1D are delayed compared to controls across the majority of the retina. Figure 4 shows representative mfERG trace arrays from 2 participants with diabetes. Quadrant (Fig. 5) and ring (Fig. 6) analyses showed that implicit time of participants with T1D were significantly delayed compared to control participants in all quadrants and eccentricities except at the fovea. These comparisons were significant after the Bonferroni correction (a ¼ 0.01 for quadrants and a ¼ 0.0008 for rings).
Mixed Models for sf-mfERG Mixed Models for mfERG Mixed model analyses for mfERG implicit time yielded main effects for group at the level of hexagons (P < 0.0001), quadrants (P < 0.0001), and rings (P < 0.0001), but not for amplitude. The P values for the post hoc pairwise comparisons from the hexagon analysis are presented in a spatial color map to illustrate the distribution of dysfunction (Fig. 3). Positive (þ) and negative () signs denote whether participants with T1D had earlier or delayed implicit times, respectively, compared to control participants. Implicit times
Mixed model analyses for mfOP1, mfOP2, and mfOP3 of the sfmfERG yielded significant between-group differences for implicit time, but not for amplitude. Significant main effects were found for mfOP1 at the level of hexagons (P ¼ 0.0003) and quadrants (P < 0.0001). Significant ring differences were found for mfOP2 (P ¼ 0.007) and mfOP3 (P ¼ 0.0008), mfOP1 approached significance (P ¼ 0.06). The P values for the post hoc pairwise comparisons at the level of hexagons were mapped spatially in color plots (Fig. 7). The areas of interest are those in which the implicit time of
FIGURE 3. Spatial color map representing the mfERG implicit time differences between groups at the level of hexagons (illustrated on an unscaled hexagon array).
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2436
FIGURE 4. Examples of representative mfERG trace arrays from two participants (A, B) with diabetes. Blue traces represent individual patient data and dashed black traces represent average control data. Shaded hexagons indicate responses that are significantly delayed (>2 SD from controls) (illustrated on unscaled hexagon arrays).
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2437
FIGURE 5. Comparison of mfERG implicit time between groups at the level of quadrants. Data presented as mean 6 95% confidence limits. Bonferroni correction, P < 0.01.
participants with T1D was delayed significantly compared to control participants at a ¼ 0.05 (color toward the red end of the spectrum with negative sign). The distribution of these delayed regions was different for each mfOP: the red hexagons were concentrated in the periphery for mfOP1, in the midperiphery for mfOP2, and in the macular region for mfOP3. Although there are minimal areas with significant
differences with the Bonferroni correction (a ¼ 0.0008), it is important to note the pattern in the distribution of P values. Ring analysis showed a similar distribution of delays (Fig. 7): the greatest implicit time delays were observed in Ring 4 for mfOP1 (P < 0.0001), Ring 3 for mfOP2 (P < 0.0001), and Ring 2 for mfOP3 (P < 0.0001). These differences are significant with the Bonferroni correction (a ¼ 0.01). Figure 8 shows
FIGURE 6. Comparison of mfERG implicit time between groups at the level of rings. Data presented as mean 6 95% confidence limits. Bonferroni correction, P < 0.0008.
Localizing Functional Damage in the Diabetic Retina
IOVS j April 2014 j Vol. 55 j No. 4 j 2438
FIGURE 7. Spatial color maps illustrating the distribution of mfOP implicit time differences between groups: both hexagon and ring analyses show that each mfOP is most delayed at different eccentricities (illustrated on unscaled hexagon arrays).
representative sf-mfERG trace arrays from two participants with diabetes. Quadrant analysis showed that mfOP1 is consistently more delayed in participants with T1D compared to control participants in all four retinal quadrants: superior temporal (P < 0.0001), superior nasal (P ¼ 0.007), inferior temporal (P ¼ 0.004), inferior nasal (P ¼ 0.001).
DISCUSSION This study found no change in mfERG amplitude, but mfERG implicit times were delayed across the retina in the diabetes group. This is consistent with mfERG literature, where there has been no consensus as to whether mfERG amplitude is affected by diabetes,14,26,27 but multifocal ERG implicit times has been reported consistently as delayed in participants with diabetes,14,23,27 and delayed to greater extents with increasing DR severity.26,28 Quadrant and ring analyses of the mfERG data revealed that the majority of the retina is affected by diabetes, as participants with T1D showed response delays across all quadrants and eccentricities. Bearse et al.12 similarly studied and mapped retinal dysfunction in participants with diabetes, and found a spatial preference of vulnerability to the effects of diabetes. They used an sf-mfERG protocol, rather than the mfERG protocol, to evaluate first-order N1, P1, and N2 implicit times, and found that P1 implicit time delays are most frequent in the
inferior and peripheral retina, and, thus, concluded these regions are most vulnerable to damage. A few key factors differentiate our studies, including differences in the protocol used, the population studied, and the analysis technique. Bearse et al.12 studied an adult population with a mix of participants with T1D and T2D, whereas our study examined an adolescent and young adult population with T1D only. The advantages of studying biomarkers with a younger rather than an older adult population have been discussed previously.4 Bearse et al.12 plotted the frequency of delays at each hexagon and defined the region most frequently associated with abnormalities as the region of greatest vulnerability. In our study, the region of greatest vulnerability was defined as the region in which the mean implicit time of participants with T1D differed most significantly with the mean implicit time of control participants. A comprehensive review conducted by Hood29 examined studies that used the mfERG to assess different retinal diseases. He derived a guide to locate the site of damage within the retina based on characteristic changes to the mfERG waveform. No amplitude changes and a small implicit time delay (
