The short-wavelength automated perimetry(SWAP) and multifocal electroretinogram (mfERG) stimuli. A, Short-wavelengthautomated perimetry central 24-2 display. B, Multifocal electroretinogramdisplay. The luminance for white, black, and background is 200 candelas persquare meter (cd/m2), less than 2 cd/m2, and 100 cd/m2, respectively. C, Spatial correspondence of SWAP and mfERG stimuluspattern. D, Example of interpolation procedure from a 45-year-old controlsubject showing implicit time in milliseconds. For each SWAP location, theclosest mfERG responses are used to interpolate an implicit time value correspondingto the SWAP location in Matlab program (The Mathworks, Natick, Mass). Locationslabeled “n” were not tested by mfERG and are excluded from ourdata analyses.
Template stretching measurement oflocal waveform. A local template (solid black line) representing the localmean waveform of the healthy subjects is independently scaled in the amplitudeand time dimensions so that the scaled template best fits the local responsefrom a patient with diabetes mellitus (gray dashed line). Stretching implicittime = stretching factor × template implicit time.
Short-wavelength automated perimetry(SWAP) (A) and multifocal electroretinogram (mfERG) (B) z score distributions for patients with diabetes mellitus (DM). z Score distributions were constructed for the patientswith DM and early nonproliferative diabetic retinopathy (NPDR) from a totalof 704 values (22 patients from 32 locations) and for those with DM withoutretinopathy from 576 values (18 patients from 32 locations). The SWAP andmfERG z score distributions in the NPDR and no retinopathygroups are shifted significantly away from the theoretical normal distribution(shaded areas).
z Scorescatterplot of short-wavelength automated perimetry (SWAP) and multifocalelectroretinogram (mfERG). A, The measurements were taken from the inferiornasal quadrant for patients with diabetes mellitus with early nonproliferativediabetic retinopathy (NPDR) (22 subjects with 8 responses per subject). Mostresponses from both measurements are worse than the mean normal z score of 0 (ie, most responses are in the top left quadrant). B, z Scores of SWAP and mfERG from the inferior nasal quadrantfor patients with diabetes mellitus without retinopathy (18 subjects with8 responses per subject). Most responses from both measurements are againworse than the mean normal z score.
The number of abnormal findings pereye for each individual with diabetes mellitus (DM) (abnormality is definedas z scores ≥2 for the multifocal electroretinogram[mfERG] measurement and z scores of −2 or lessfor the short-wavelength automated perimetry [SWAP] measurement). Each dotrepresents the number of abnormal findings detected by SWAP or mfERG measurementsfor each patient with DM. The shaded areas indicate the number of abnormalfindings per eye is less than 3. A, The SWAP results. B, The mfERG implicittime. NPDR indicates nonproliferative diabetic retinopathy.
Han Y, Adams AJ, Bearse MA, Schneck ME. Multifocal Electroretinogram and Short-Wavelength Automated PerimetryMeasures in Diabetic Eyes With Little or No Retinopathy. Arch Ophthalmol. 2004;122(12):1809-1815. doi:10.1001/archopht.122.12.1809
Copyright 2004 American Medical Association. All Rights Reserved.Applicable FARS/DFARS Restrictions Apply to Government Use.2004
To compare severity and locations of abnormalities detected by the multifocalelectroretinogram (mfERG) and short-wavelength automated perimetry (SWAP)in diabetic eyes with early or no retinopathy.
One eye from each of 22 patients with diabetes mellitus who had earlyretinopathy and 18 patients with diabetes mellitus who had no retinopathywere tested on mfERG and SWAP. The mfERG implicit times were interpolatedbased on SWAP stimulus locations and compared with normative values obtainedfrom 30 age-similar control subjects. The SWAP total threshold deviationswere analyzed using an age-based control data set from 255 healthy subjects.The z scores of both measures were derived to allowmeasurement comparisons.
Most responses for the 2 measurements were subnormal in both groupswith diabetes mellitus. The 2 measurements showed a similar number of significantabnormalities (z score ≥2), about 40% and 20%of responses for diabetic patients with retinopathy and diabetic patientswith no retinopathy, respectively. Local mfERG and SWAP results showed somespatial agreement for subjects with retinopathy (r = –0.38, P<.001) but not for those with no retinopathy.
Both mfERG and SWAP are sensitive measurements of diabetic dysfunction,even prior to retinopathy. The lack of spatial correspondence between mfERGand SWAP abnormalities in diabetic patients with no retinopathy reflects overlapping,but different, retinal anomalies in early diabetic eye disease.
Diabetic retinopathy is the leading cause of blindness among working-agedpeople in the United States.1 Visual loss isgenerally irreversible at stages when nonperfusion regions, neovascularization,or both are clearly identified by ophthalmoscopy and fluorescein angiography.Early diagnosis, treatment, and prevention of retinopathy are essential tosave sight.
The multifocal electroretinogram (mfERG) and short-wavelength perimetry(SWAP) are 2 promising measurements for early detection of visual functionalchanges in diabetes mellitus (DM). The mfERG is a powerful objective toolto study local retinal function,2- 4 allowingus to simultaneously and independently record the cone-driven activity atmore than 100 retinal locations within minutes. In eyes with diabetic retinopathy,local changes in retinal function are associated with the sites of retinopathiclesions.5 The mfERG detects functional abnormalitiesin patients with DM even before retinopathy is visible by fundus photography.5- 11 Theselective loss of short-wavelength (S-cone) visual pathway sensitivity hasalso been demonstrated psychophysically in diabetic patients with little orno retinopathy.12- 17 TheSWAP test, a subjective measure of local S-cone function across the retina,can identify the sensitivity loss in diabetic patients who have retinopathyand type 1 DM even without retinopathy.18- 24
Although both measurements are sensitive to diabetic retinopathy, theymay provide different results because they are mediated by different conesystems. The mfERG responses reflect function of longer-wavelength–sensitive(L- and M-) cone pathways, while SWAP thresholds result from stimulating onlythe S-cone pathway.25- 27 Todate there has not been a direct comparison of the 2 techniques in the samepatients. Since the 2 measurements reflect different mechanisms, one mightanticipate little correspondence in the 2 measurements, or one measurementmay be affected earlier than the other resulting in poor agreement.
This study has 2 aims. The first is to establish the extent to whichthe 2 noninvasive measurements, mfERG and SWAP, can detect functional lossin eyes with no or little diabetic retinopathy. The retinopathic lesions inour patients were much smaller than the stimulus sizes, distinguishing thisstudy from previous mfERG or SWAP studies whose diabetic subjects had broaderand/or advanced retinopathy. The second aim is to examine and compare theretinal locations and the retinal extent of the functional abnormalities detectedby the 2 measurements.
The left eyes of 22 patients with DM with early nonproliferative diabeticretinopathy (NPDR) (1 patient with type 1 DM and 21 patients with type 2 DM)and 18 patients with DM with no diabetic retinopathy (2 patients with type1 DM and 16 patients with type 2 DM) were tested on mfERG and SWAP on thesame day. Diabetic retinopathy was diagnosed by dilated eye examination andfundus photography. The severity of diabetic retinopathy was classified bya retinal ophthalmologist according to Early Treatment Diabetic RetinopathyStudy (ETDRS) criteria.28 In the NPDR group,2 subjects had moderate NPDR (each has a small patch of edema in the midperipheralretina); the other 20 patients with DM had only mild retinopathy. All eyesin both diabetic groups had 20/25 or better corrected visual acuity with refractiveerrors between –6.00 diopters (D) and +4.00 D. Patients with visiblemedia opacities or a history of other ocular disease or surgery were excludedfrom the study. The ages of NPDR subjects ranged from 32 to 59 years (meanage ± SD, 52.4 ± 6.0 years) with a durationof DM from 2 to 20 years (mean ± SD, 10.2 ± 6.2years). The ages of the patients with no retinopathy ranged from 26 to 64years (mean age ± SD, 43.5 ± 12.0 years)with a duration of DM from 3 to 20 years (mean ± SD, 7.8 ± 4.5years).
Thirty eyes of 30 healthy (free of ocular or systemic disease) nondiabeticsubjects were tested with the mfERG (19 right and 11 left eyes based on thesubject’s preference). The 14 men and 16 women ranged in age from 28to 60 years (mean age ± SD, 47.2 ± 9.5 years).All healthy eyes had 20/20 or better corrected visual acuity with refractiveerrors between –6.00 D and +4.00 D. Normal data for the SWAP test arebased on 255 eyes of 255 healthy adults who ranged in age from 25 to 65 years(mean age ± SD, 44.4 ± 11.3 years; SAFE[Structure and Function Evaluation] study in Portland, Ore).29
The purposes and potential risks of the study were explained and informedconsent was obtained from all subjects before testing. Procedures followedthe tenets of the Declaration of Helsinki, and the protocol was approved bythe University of California Committee for the Protection of Human Subjects,Berkeley.
The SWAP visual fields (Humphrey Field Analyzer; Humphrey Systems, Dublin,Calif) were tested with undilated pupils using the 24-2 stimulus presentationpattern (Figure 1A) and full-thresholdstrategy. The 24-2 pattern was chosen because its testing field closely matchesthe stimulus area of the mfERG responses. Three minutes of adaptation to a100-candelas (cd)/m2 yellow background preceded testing.25 An optimal lens correction was used, and the felloweye was occluded with an eye patch. All of the subjects had fixation lossand false-positive and false-negative ratios less than 10%; most had ratiosthat were much less.
Multifocal ERGs were recorded using a stimulus-refractor unit (VERIS,version 4.3; Electro-Diagnostic Imaging, Inc, San Mateo, Calif). Pupils weredilated to 7 to 8 mm with a combination of 1.0% tropicamide and 2.5% phenylephrinehydrochloride. After the cornea was anesthetized with 0.5% proparacaine hydrochloride,a bipolar contact lens electrode (Hansen Ophthalmic, Solon City, Iowa) wasplaced on the test eye and a ground electrode clipped to the right earlobe.The fellow eye was occluded. The stimulus array of 103 hexagonal elements(Figure 1B) was delivered by an eyecamera display refractor unit (Electro-Diagnostic Imaging, Inc) driven ata frame rate of 75 Hz. The hexagons were modulated between white (200 cd/m2) and black (<2 cd/m2) according to a binary m-sequenceduring the 7.5-minute recordings. Before the test, observers adjusted thestimulus unit for best focus of the central fixation target. To improve thesubject’s ability to maintain fixation, the test was broken up into16 overlapping segments, each lasting approximately 30 seconds. The recordingsignals were filtered 10 to 100 Hz and amplified 100 000 times. The qualityof the recordings was controlled by a real-time display; eye movements weremonitored by the eye camera. Contaminated segments were discarded and repeated.The mfERGs were processed in the usual way with 1 iteration of artifact removaland spatial averaging with 1/6 of the surrounding responses.
Based on the mean ± SD from healthy subjects of theappropriate age group at each testing location, z scores (standard deviation units in decibel domain) of total deviationwere calculated for all diabetic participants. Total deviation representsthe difference in decibels between the subject’s test results and theage-corrected normal values at each tested point in the visual field.
A “template stretching method” described in detail by Hoodand Li30 was used to measure the implicit timeof the prominent peak (P1) of the first-order kernel (Figure 2). The 103 local mfERGs of each subject were compared withwaveform templates representing the mean local waveforms of the healthy subjects(right eye responses were converted to left eye orientation). Each templatewas independently scaled in the amplitude and time dimensions so that thebest least-square fit to each local response was obtained. Previous studieshave shown that measurement of amplitude is relatively insensitive to DM anddiabetic retinopathy.5,31,32 Wealso observed this in our study, therefore, herein we examined the relationshipbetween SWAP and mfERG P1 implicit time measures.
The spatial displays of mfERG and SWAP are different (Figure 1C). To compare the 2 measures point by point, one test’sresults have to be interpolated to match the other. Herein we interpolatedthe 103 mfERG results based on SWAP stimulus locations because the mfERG hasmore testing points and, subsequently, provides more information for the interpolation(Figure 1D). In the Matlab program (TheMathworks, Natick, Mass), the mfERG results were interpolated into a high-resolutionsurface using a linear algorithm, and the interpolated mfERG value at eachSWAP stimulus was determined by the coordinate of the SWAP stimulus.33,34 Because the extent of the testingfield of the retina for the mfERG is slightly smaller than that of the SWAPfield, the interpolated mfERG values beyond the actual testing area were notconsidered. The foveal test point was also excluded owing to the lack of availabilityof normal values at that location for the SWAP measurements. For the interpolatedmfERG implicit times, at each location z scores werederived for each subject with diabetes based on the normal interpolated mfERGdata. All further analyses use the z scores of theinterpolated values.
z Score distributions were constructed forboth SWAP and mfERG measurements for the NPDR group from a total of 704 values(22 subjects × 32 locations) and for the no retinopathy groupfrom 576 values (18 subjects × 32 locations). The SWAP andmfERG z score distributions in the no retinopathygroup are shifted significantly from the shaded theoretical normal distribution(Mann-Whitney test, P<.0001 for SWAP; P = .0003 for mfERG, Figure 3). As expected, both measurements show that patients with NPDR tendto have more abnormal findings than those with no retinopathy. In the NPDRgroup, 38.6% of all SWAP z scores are −2 orless (reduced sensitivity) and 36.8% of the mfERG z scoresare 2 or greater (implicit time delay, Figure3A). For those patients without retinopathy, although the distributionsof the 2 measurements are closer to the healthy subjects, 18.6% of SWAP z scores are −2 or less and 21.0% of the mfERG z scores are 2 or greater (Figure 3B). These z scores represent P≤.02.
Are z-scores of SWAP and mfERG measurementslocally spatially correlated across the retina? To answer this, each retinalquadrant is examined (22 subjects × 8 responses per subjectfor the NPDR group and 18 subjects × 8 responses per subjectfor the group without retinopathy). Figure 4 showsthe results for the 2 diabetic groups in the inferior nasal quadrant whichis representative of the results of all quadrants. Most responses from bothmeasurements are worse than the average healthy population in both diabeticgroups (ie, most responses are in the top left quadrant in each plot) despitethe fact that more than 90% of the locations had no detectable retinopathy.The correlation coefficients between the 2 measurements were –0.38 forthe patients with NPDR (P<.001, Figure 4A) and -0.20 (P = .10, Figure 4B) for the group with DM without retinopathy.The correlation coefficients for superior temporal, superior nasal, and inferiortemporal quadrant are -0.38, -0.34, and -0.35 in the NPDR group and -0.24,-0.23, and -0.21 in the group with DM without retinopathy, respectively.
The correlation analysis we did above assumed a linear relationshipbetween the 2 measurements, an assumption that may or may not be correct.Therefore, we next performed an agreement analysis, which assumed no specificform for the underlying relationship between the 2 measurements. Two categorieswere defined, agreement and disagreement of the direction in which both SWAPand mfERG measurements differ from the mean of the healthy group (z score = 0) at each retinal location. The 2 measurementsare in agreement if the SWAP and mfERG z scores ata specific location are both worse or both better than the control means,and in disagreement if one of the z scores is betterand the other is worse. By chance, we expect 16 of 32 common testing locationsto disagree (ie, one measurement better than the control population, the otherworse), and the remainder of the 16 locations to agree (ie, 8 better thanthe control population and 8 worse on both measurements).
For the NPDR group, however, SWAP and mfERG results agree on averageat 30 (93.8%) of the testing locations (P<.001,Wilcoxon signed rank test) and very few of them are better than the healthypopulation (median values, Table). Onthe other hand, for the subjects with DM but no retinopathy, there is littlecorrespondence between the 2 measurements (P = .12,Wilcoxon signed rank test). The number of locations where the 2 measurementsdisagreed is similar to our prediction (median values, Table). Although for both groups most responses are worse than thoseof the healthy subjects for each measurement, only in the NPDR group doesthe direction of SWAP and mfERG results agree.
Furthermore, the correspondence of the 2 measurements was examined whenlocal responses were classified as normal or abnormal by a criterion z score of 2. For the NPDR group, 20.6% (145/704) of theresponses were defined as abnormal by both SWAP and mfERG measurements atthe same retinal locations, compared with only 8.7% (50/576) for the groupwith DM with no retinopathy. The abnormal results for mfERG and SWAP correspondconsiderably better for NPDR group than for the group with DM with no retinopathy.
Finally, we examined how well the 2 measurements could distinguish diabeticeyes from the healthy eyes. For each measurement an eye was classified asabnormal if more than 2 local responses were abnormal. This strict criterionis chosen on the basis of the following rationale. Since we chose z scores of ±2 as the criteria for abnormal results, each testedlocation has a 2.3% probability to be labeled as abnormal. As a result, basedon the binomial distribution, the probability of more than 2 abnormal locationsper eye is 3.7%.35 Therefore, classificationof normal or abnormal eyes by this criterion makes it unlikely that a normalretina will be labeled as abnormal by chance. Nevertheless, we find that theSWAP test identifies 14 (63.6%) of the NPDR group and 9 (50.0%) of the groupwith DM with no retinopathy as having abnormal findings (Figure 5A) and the mfERG measurement classifies 13 (59.1%) of theNPDR group and 6 (33.3%) of the group with DM with no retinopathy as havingabnormal findings (Figure 5B). The 2measurements generally classified the same individuals as having abnormalfindings; 11 eyes in the NPDR group and 5 eyes in the group with DM with noretinopathy are classified as having abnormal findings by both measurements.
Both the SWAP and mfERG implicit time measures show local functionalabnormalities in the 2 groups of patients with diabetes we examined. Moresevere functional loss is seen in patients with DM with retinopathy than inthose with DM with no retinopathy. There is a significant spatial correlationof the 2 measurements in the NPDR group. However, this is not evident in eyeswith DM without retinopathy, consistent with the fact that the 2 measurementstap different but overlapping mechanisms.
Foveal measures have shown that the S-cone pathway is selectively susceptibleearly in diabetic retinopathy.12- 15 Inour study, the SWAP measurement, which probes S-cone pathway sensitivity acrossthe extrafoveal visual field, is also affected in early diabetic eye disease.Previous studies using SWAP reported abnormalities in diabetic eyes with retinopathy,18,20,21 especially when clinicallysignificant macular edema was present.22- 24 However,for diabetic eyes with no retinopathy, significant SWAP defects have beenpreviously reported in patients with type 1 DM but not in patients with type2 DM.18- 20 Inthis study, the distribution of SWAP z scores ineyes with DM with no retinopathy is significantly different from healthy eyes,and 50% (8/16) of the eyes of patients with type 2 DM with no retinopathyhave more than 2 local SWAP defects (and are, therefore, classified as beingabnormal). The possible explanation is that the SWAP analysis we used considersthe influence of age on ocular media (crystalline lens yellowing).29 Moreover, the larger normal database (n = 255)used may have sufficiently reduced the normal confidence interval so thatSWAP sensitivity is significantly improved.29
In contrast with previous mfERG study findings of local retinal abnormalitiesin diabetic patients,5,6,31,32 90%of our patients with NPDR have only mild diabetic retinopathy, and all ofthe visible lesion sizes are considerably smaller than the stimulus patches.Despite this minimal retinopathy, we find the mfERG implicit time to be asensitive measurement of diabetic retinal function loss. We also find a highproportion of mfERG defects in diabetic patients who have yet to develop anyretinopathy.
Previous studies provide some anatomical basis for the detection ofmfERG abnormality in diabetic patients. Diabetic retinopathy is caused bydefects of retinal capillaries lying mainly at the inner nuclear layer36; most types of retinopathic lesion, such as microaneurysms,hard exudates, and retinal edema, occur in the middle layer of the retina,close to the inner nuclear layer. Pharmacological experiments on rhesus monkeys(Macaca mulatta) show that the major responses ofmfERGs are generated by the bipolar cells,37,38 locatedin the inner nuclear layer of the retina.
In the NPDR group, mfERG and SWAP measurements at more than 90% of testlocations agree in the direction of deviation from the normal mean (most towardabnormal). Thus, the 2 measurements agree well qualitatively for eyes withDM and retinopathy. However, the local correlation (r = −0.38),while highly statistically significant (P<.001),is quantitatively less robust. It could be the case that the 2 measurementswould show better correlation if different scales of measurements were used.
Although both mfERG and SWAP measurements are sensitive to DM even inthe absence of retinopathy, we found little local correspondence of the 2measurements for patients with DM who have yet to develop retinopathy. Forthe eyes of patients with DM with no retinopathy, the 2 measurements do notagree qualitatively and they are not significantly correlated. This resultmight be related to the different mechanisms of response generation for the2 measurements. The mfERGs mainly reflect L- and M-cone pathway activity priorto the nerve fiber layer. On the other hand, SWAP specifically tests the isolatedfunction of the entire S-cone pathway, so abnormality can reflect disruptionanywhere in the pathway. In the early stages of DM (prior to the retinopathy),the 2 systems may be differentially affected among different patients and/orretinal locations, resulting in the poor local correlation. However, afterthe development of diabetic retinopathy, all cone pathways might be affected.In this case both mfERG and SWAP are more likely to show abnormalities inthe same or similar retinal locations.
It will be of interest to examine in the future whether SWAP or mfERG(or both) abnormalities in a particular retinal location might precede thedevelopment of retinopathy in that location. Our patients are participatingin a 3-year longitudinal study that should reveal whether one test is a betterpredictor than the other or whether a combination of the 2 tests might havegreater predictive power than either test alone.
Correspondence: Ying Han, PhD, 360 MinorHall, School of Optometry, University of California at Berkeley, Berkeley,CA 94720-2020 (firstname.lastname@example.org).
Submitted for Publication: August 11, 2003;final revision received May 18, 2004; accepted May 18, 2004.
Financial Disclosure: None.
Funding/Support: This study was supported bygrant EY-02271 from the National Eye Institute, National Institutes of Health,Bethesda, Md (Dr Adams).