Identifying Glaucomatous Vision Loss with Visual-Function–Specific

Identifying Glaucomatous Vision Loss with Visual-Function–Speciﬁc Perimetry in the Diagnostic Innovations in Glaucoma Study

Pamela A. Sample, Felipe A. Medeiros, Lyne Racette, John P. Pascual, Catherine Boden, Linda M. Zangwill, Christopher Bowd, and Robert N. Weinreb

PURPOSE. To compare the diagnostic results of four perimetric rameters at suggested criterion values provided good sensitiv- tests and to identify useful parameters from each for determin- ity and specificity. FDT showed the highest sensitivity overall, ing abnormality. with SAP performing better than in prior reports. Of note, the METHODS. One hundred eleven eyes with glaucomatous optic same area of the retina was identified as damaged in all tests. neuropathy (GON), 31 with progressive optic neuropathy (Invest Ophthalmol Vis Sci. 2006;47:3381–3389) DOI: (PGON) 53 with ocular hypertension, and 51 with no disease 10.1167/iovs.05-1546 were included (N ϭ 246). Visual field results were not used to classify the eyes. Short-wavelength automated perimetry ver the past several years, psychophysical tests of specific (SWAP), frequency-doubling technology perimetry (FDT), Ovisual functions have been used to measure visual perfor- high-pass resolution perimetry (HPRP), and standard auto- mance and to understand the underlying glaucomatous mated perimetry (SAP) were performed. Receiver operating changes in retinal ganglion cell function. Testing vision with characteristic (ROC) curves were used to compute the areas standard automated perimetry (SAP) is not selective for a par- under the curves (AUC) and sensitivity levels at given specific- ticular ganglion cell type. Any of the primary ganglion cell ities for a variety of abnormality criteria. The agreement among subtypes can respond to an achromatic incremental stimulus tests for abnormality, location, and extent of visual field deficit presented on an achromatic background. In contrast, each were assessed. visual-function–specific perimetric test attempts to isolate a RESULTS. AUC analysis: When the normal group was compared subpopulation of ganglion cells by evaluating a specific visual with the GON group, the FDT pattern SD (PSD) area was larger function characteristically processed by that cell subtype. As than the HPRP PSD (P ϭ 0.020), and the FDT area of total an example, short-wavelength automated perimetry (SWAP) deviation (TD) Ͻ5% was larger than the HPRP mean deviation elicits detection by the short-wavelength cones. The stimulus (MD; P ϭ 0.004). When the normal group was compared with information is then processed through the blue–yellow gan- the PGON group, the FDT area of pattern deviation (PD) Ͻ5% glion cells. The amount of isolation is unknown for each of was larger than the SWAP PSD (P ϭ 0.020). A difference from these function-specific tests, with the exception of SWAP, previous work was that AUCs for PSD or the best SAP were not which provides approximately 15 dB of isolation. This means significantly poorer than those in the function-specific tests. At the blue–yellow ganglion cell system would have to lose 15 dB set specificities, FDT yielded higher sensitivities than all other of sensitivity before another cell type could assist in respond- 1 tests for all parameters. The agreement among tests for abnor- ing to the SWAP stimulus. mality was fair to moderate (␬ ϭ 247–0.563). When loss was Initially, it was hypothesized that the blue–yellow ganglion 2,3 present on more than one test, the quadrant of the visual field cells tested by SWAP were parvocellular. It was later learned 4 5,6 affected was the same in 95% (79/83) of eyes. The number of from Dacey and others that the blue–yellow cells are small eyes identified and number of abnormal quadrants increased bistratified ganglion cells that project their axons to the konio- across groups with increasing certainty of glaucoma. cellular (interlaminar) layers of the lateral geniculate nucleus (LGN) of the thalamus, rather than to the parvocellular layers.7 CONCLUSIONS. At equal specificity, no single perimetric test was Frequency-doubling technology perimetry (FDT)8,9 and vari- always affected, whereas others remained normal. Several pa- ous forms of motion perimetry10–14 attempt to target the magnocellular (also known as parasol) ganglion cells that project to the magnocellular layers of LGN, and high-pass From the Visual Function Laboratory and Hamilton Glaucoma resolution perimetry (HPRP)15,16 is thought to assess the par- Center, Department of Ophthalmology, University of California at San vocellular (also known as midget) ganglion cells that project to Diego, La Jolla, California. the parvocellular layers of LGN. Recent reviews detail the Supported by National Eye Institute Grants EY 08208 (PAS) and evidence supporting anatomic and functional segregation of EY11008 (LMZ) and participant retention incentive grants in the form three primary pathways (parvocellular, magnocellular, and ko- of glaucoma medication at no cost: Alcon Laboratories Inc, Allergan, 17,18 Pfizer Inc, and SANTEN Inc. niocellular) through the LGN. Anatomic projections to Submitted for publication December 5, 2005; revised March 17 cortex and functional preferences within cortical layers are 18,19 and April 10, 2006; accepted June 9, 2006. much less segregated. In addition, the relationship of vi- Disclosure: P.A. Sample, Carl Zeiss Meditec, Inc., Welch-Allyn, sual function to underlying visual pathways is based primarily and Haag-Streit (F); F.A. Medeiros, Carl Zeiss Meditec, Inc. (F); L. on electrophysiology in healthy primates and on lesion stud- Racette, None; J.P. Pascual, None; C. Boden, None; L.M. Zangwill, ies,17 and so some caveats apply in the application of visual- Carl Zeiss Meditec, Inc. (F, R); C. Bowd, None; R.N. Weinreb, Carl function–specific perimetry to ganglion cell assessment (see Zeiss Meditec, Inc. (F, R) the Discussion section). The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertise- Several studies comparing one visual-function–specific test ment” in accordance with 18 U.S.C. §1734 solely to indicate this fact. to SAP have shown that function-specific tests are superior to Corresponding author: Pamela A. Sample, Department of Ophthal- SAP for early detection of vision loss associated with glau- 20–23 mology, University of California at San Diego, 9500 Gilman Drive, La coma. There are some problems with these studies (see Jolla, CA 92093-0946; [email protected]. the Discussion section). Very few have compared more than

Downloaded from iovs.arvojournals.org on 01/31/2019 3382 Sample et al. IOVS, August 2006, Vol. 47, No. 8

one visual-function–specific test in the same patient popula- Participants. Two hundred forty-six eyes from 246 participants tion.24–26 We first made such a comparison 6 years ago and were evaluated on all three visual-function–specific perimetry tests as evaluated SAP, SWAP, motion automated perimetry (MAP), and well as on SAP. Criteria for classification are detailed later. FDT in 71 eyes with glaucomatous optic neuropathy, 37 ocular Diagnostic categories were based on simultaneous stereoscopic hypertensive eyes, and 28 age-matched normal control eyes.24 color photographs (TRC-SS camera; Topcon America Corp., Paramus, For detection of functional loss, it was found that (1) stan- NJ), obtained after maximal pupil dilation. All photograph evaluations dard visual field testing was not optimal, missing 54% of eyes were taken using a stereoscopic viewer (Stereo Viewer II; Asahi Pen- with glaucomatous optic neuropathy (GON); (2) a combina- tax, Golden, CO) illuminated with color-corrected fluorescent lighting. tion of two or more tests improved detection of functional loss; Two trained and masked graders from the UCSD Optic Disc Reading (3) in an individual, the same retinal location was damaged, Center assessed each photograph independently. Inconsistencies be- regardless of the visual function tested; and (4) SWAP, MAP, tween the two graders’ evaluations were resolved by consensus or and FDT showed promise as early indicators of function loss in through adjudication by a third evaluator. GON was identified by glaucoma. evidence of any of the following: excavation, neuroretinal rim thinning One limitation of this previous study was that the normal or notching, nerve fiber layer defects, or asymmetry of the vertical eyes had normal SAP visual fields and we had to adjust speci- cup-to-disc ratio Ն0.2 between the two eyes. ficities based on other published reports. This limitation does To identify progressive glaucomatous optic neuropathy, two not exist in the present study. A second limitation was the trained graders independently compared the first and last photographs testing of only the koniocellular and magnocellular pathways in each participant’s series. Graders were masked to all other partici- and the lack of a function-specific test preferred by the parvo- pant information, photograph date and test results. Photographs were cellular pathway. We have since obtained sufficient data with graded A equal to B, A worse than B, or A better than B. Inconsistencies HPRP, and in this current study, we to compared it to SWAP, between the two graders’ evaluations were resolved through adjudi- FDT, and SAP in the same individuals. cation by a third evaluator for each pair of photographs. After consensus was reached, the temporal sequence of the photographs was unmasked. Progression was defined based on evidence of increasing METHODS excavation, rim thinning or enlarging of notches or nerve fiber layer defects in the later photograph. Changes in rim color and the presence All participants were selected from the ongoing longitudinal Diagnos- of disc hemorrhage or progressive parapapillary atrophy were not tic Innovations in Glaucoma Study (DIGS), conducted at the Hamilton sufficient for characterization of progression. The time frame for pro- Glaucoma Center at the University of California at San Diego (UCSD). gression on photographs ranged from 2 to 12 years. This study is prospectively designed to assess structure and function in glaucoma. DIGS patients are followed annually. Informed consent was Classification of Study Groups obtained from all participants after the nature and procedures of the study were explained. The Institutional Review Board of the University The 246 participants were placed in one of four diagnostic categories of California at San Diego approved the study, which adheres to the based on the consensus appearance of the optic disc on stereophoto- tenets of the Declaration of Helsinki. graphs, the ocular examination results, and their IOP history. Visual Inclusion Criteria for DIGS. Participants underwent com- fields were not used to classify the participants. This classification plete ophthalmic examinations including slit lamp biomicroscopy, resulted in 111 eyes with GON, 31 with progressive GON (PGON), 53 intraocular pressure (IOP) measurement, and dilated stereoscopic fun- with ocular hypertension (OHT), and 51 with normal eyes, serving as dus examination. Simultaneous stereoscopic photographs were ob- the control. We included PGON as it provides strong evidence of tained for all participants and had to be of adequate quality for the glaucoma, when a nonfunctional measure is used.27 subject to be included. All participants had open angles, a best cor- Normal Control Subjects. Normal eyes had IOP Ͻ23 mm Hg rected acuity of 20/40 or better, a spherical refraction within and with no history of increased IOP, normal findings in an ocular exami- inclusive of Ϯ5.0 D (transposition allowed), and cylinder correction nation, and normal optic discs according to the criteria. within Ϯ3.0 D. A family history of glaucoma was allowed. Ocular Hypertension. Ocular hypertensive eyes had normal Exclusion Criteria for DIGS. Normal and ocular hypertensive optic discs and highest IOPs Ն23 mm Hg on at least two separate participants were excluded if they had a history of intraocular surgery occasions. (except for uncomplicated cataract surgery). We also excluded all Glaucomatous Optic Neuropathy. These participants had participants with nonglaucomatous secondary causes of elevated IOP evidence of glaucomatous-appearing optic discs on stereophotographs (e.g., iridocyclitis, trauma), other intraocular eye disease, other dis- with or without elevated IOP. eases affecting the visual field (e.g., pituitary lesions, demyelinating Progressive Glaucomatous Optic Neuropathy. These diseases, HIVϩ or AIDS, or diabetic retinopathy), with medications participants showed progressive GON on evaluation of serial photo- known to affect visual field sensitivity, or with problems other than graphs sometime during their follow-up in DIGS before the visual field glaucoma affecting color vision (as assessed by the Farnsworth D-15 dates used in this analysis. color vision test). DIGS participants are not chosen based on any criteria other than Psychophysical Tests of Function those specified herein. Most of the patient participants came to the study through the Glaucoma Service at the UCSD Department of Four perimetric procedures were used to test visual function. All Ophthalmology. Normal subjects were recruited from the general procedures were tested within the central 30° of visual field and population through advertisement as well as from the staff and em- necessitated fixation by the patient. Proper refraction was provided for ployees of the University of California San Diego. each device. All required a 3-mm or larger pupil. Dilation was used if The Current Study. All eligible DIGS participants with reliable necessary. Lids of eyes with potential ptosis were taped to reduce visual field results on all four tests defined as Յ25% false positives, false artifacts. negatives, and fixation losses were included. All tests were performed Standard Achromatic Automated Perimetry. In SAP, a in randomized order and completed within a 3-month period. All small (0.47°) 200-ms flash of white light is presented as the target on possible orders of the four tests were determined, and participants a dim background (10 cd/m2 or 31.5 asb). The target was randomly were assigned to an order sequentially on arrival for study. One eye presented to 54 locations within the central 24° of visual field using the was selected randomly from each subject, except in cases in which Humphrey Visual Field Analyzer II (Carl Zeiss Meditec, Dublin, CA), only one eye met study criteria, and then that eye was included. program 24-2, software version 3.4.7, and the SITA testing algorithm.

Downloaded from iovs.arvojournals.org on 01/31/2019 IOVS, August 2006, Vol. 47, No. 8 Visual Function in Glaucoma 3383

FIGURE 1. Examples of visual ﬁeld pattern deviation display for SAP, SWAP, FDT, and the small and deep dent display for HPRP in a patient with GON. Each plot shows the location and number of stimulus test locations as designated by either a box or a dot. Dot: within normal limits. The shading in the boxes denotes the probability of abnormality relative to the internal normative database of each device. Probabilities are shown in the corresponding key.

The two locations just above and below the blind spot were not Statistical Analyses included in the analysis. Figure 1 shows all target locations in all four tests. Also shown are the abnormal locations for each test in the same Continuous variables were compared by using ANOVA overall, with example eye with GON. between-test comparisons conducted with the Tukey honestly significantly different (HSD) test with ␣ set at 0.05. The ␹2 analysis was used Short-Wavelength Automated Perimetry. SWAP was mea- to assess agreement of categorical variables. sured with the same HFA perimeter as SAP (software version 3.4.7, and 1 Visual field parameter data were derived by comparison to the the 24-2 program), but with the full threshold (FT) strategy. In SWAP, manufacturer’s internal normative database for SAP, FDT, and HPRP a 440-nm, narrow-band, 1.8° target is presented at 200-ms duration on and to our laboratory’s normative database for SWAP (n ϭ 345, one eye 2 a bright 100 cd/m yellow background and selectively tests the short- per subject), which was developed in collaboration with Chris Johnson wavelength–sensitive cones and their connections. for the SWAP ancillary arm of the Ocular Hypertension Treatment Frequency-Doubling Technology Perimetry. FDT was Study after it was suspected that the manufacturer’s internal normative measured with the frequency-doubling visual field instrument (Carl database for SWAP-full threshold testing was inaccurate. Soliman et Zeiss Meditec, Dublin, CA) using Welch-Allyn technology (Skaneate- al.28 have confirmed the inaccuracy. The inclusion–exclusion criteria les Falls, NY) and the N-30 program, software version 3.00.1. The were the same as for our normal control group, but none of our control targets consist of a 0.25-cyc/deg sinusoidal grating that undergoes a subjects were included in the SWAP normative database. 25-Hz counterphase flicker. The test involves a modified binary Abnormality. Receiver operator characteristic (ROC) curves search staircase threshold procedure with stimuli presented for a were generated for the visual field parameters shown in Table 1, by maximum of 720 ms. FDT measures the contrast needed for detec- comparing normal and patient eyes with GON and then again compar- tion of the stimulus. Each grating target is a square subtending ing normal and patient eyes with PGON. The area under these curves approximately 10° in diameter. Targets are presented in one of 18 (AUCs) were compared statistically for the best-performing parameter test areas located within the central 20° radius of the visual field from each test and for the pattern SD (PSD), which performed well in 29 temporally and 30° nasally. all tests, using the method of DeLong et al. Theirs is a nonparametric method used to compare correlated AUC (that is, AUC for different High-Pass Resolution Perimetry. In HPRP, ring-shaped van- tests obtained in the same group of individuals). It is based on a ishing optotypes which vary in size are used to assess resolution ability Mann-Whitney statistic and has been widely used for this purpose in in the central 30° of the visual field.15 The optotypes used in HPRP are the medical literature.27,30 high-spatial-frequency filtered targets where the inner and outer por- The ROC results were then used to select criterion values for 2 tions of the rings are darker (15 cd/m ), whereas the center portion of various parameters to provide specificities of 80% and 90% in the 2 the rings is brighter (25 cd/m ). The space-averaged luminance of the normal control group, to equate somewhat the tests for diagnosing 2 entire ring is equal to the luminance of the background (20 cd/m ). abnormality, because each test presents different stimuli and targets Therefore, when the edges of the ring cannot be resolved, the rings different test locations to assess different visual functions. Sensitivity blend into the background, that is, the targets are either resolved for diagnosis of glaucoma was then computed based on these criterion (seen) or they are invisible. The target consists of rings of different values. The percentage of OHT eyes identified as abnormal by each test sizes, presented at 50 locations within the central 30°. No stimuli are according to the specified criterion values for the most sensitive presented within the central 5° of the visual field (Fig. 1). The subject parameters was also determined. responds when the target is large enough to resolve. We measured Agreement among Tests. We evaluated the agreement among HPRP with the Ophthimus High-Pass Resolution Perimeter, version 2.0, the tests to determine the percentage of eyes that were classified the software version 2.51 (HighTech Vision, Malmo¨, Sweden). same, by the various combinations of tests, based on the pattern

Downloaded from iovs.arvojournals.org on 01/31/2019 3384 Sample et al. IOVS, August 2006, Vol. 47, No. 8

TABLE 1. AUC and Standard Error for Groups with GON and PGON

GON PGON

Parameter AUC SE AUC SE Criterion Sens/Spec (%) Criterion Sens/Spec (%)

SAP PSD 0.713 0.041 0.762 0.056 2.31 48/90 1.93 dB 52/80 SWAP PSD 0.733 0.041 0.775 0.052 4.48 45/90 3.75 dB 48/80 FDT PSD 0.770 0.036 0.875 0.041 4.76 68/90 4.36 dB 71/80 HPRP PSD 0.661 0.043 0.780 0.054 0.87 52/90 0.8 dB 65/80 SAP TD 5% 0.708 0.041 0.758 0.060 17 45/90 9 68/80 SAP TD 1% 0.711 0.036 0.797 0.051 4 55/90 2 65/80 SWAP TD 5% 0.641 0.046 0.646 0.062 23 23/90 14 42/80 SWAP TD 1% 0.659 0.044 0.696 0.061 11 32/92 4 48/82 FDT TD 5% 0.795 0.033 0.880 0.044 4 71/90 3 84/78 FDT TD 1% 0.763 0.028 0.820 0.047 1 68/90 — N/A HPRP Small dent 0.583 0.047 0.604 0.063 8 16/94 7 23/88 HPRP Deep dent 0.652 0.041 0.759 0.054 4 32/92 3 42/86 SAP PD 5% 0.704 0.041 0.734 0.058 11 39/92 9 55/80 SAP PD 1% 0.669 0.039 0.733 0.058 5 39/92 3 39/86 SWAP PD 5% 0.710 0.044 0.727 0.057 11 32/90 7 42/84 SWAP PD 1% 0.684 0.041 0.731 0.059 5 36/92 3 48/84 FDT PD 5% 0.763 0.036 0.891 0.043 5 74/98 4 84/86 FDT PD 1% 0.741 0.031 0.818 0.048 2 58/96 1 71/84 SAP MD 0.601 0.045 0.731 0.069 Ϫ2.22 55/90 Ϫ1.73 dB 65/80 SWAP MD 0.601 0.045 0.587 0.069 Ϫ7.07 29/90 Ϫ5.42 dB 42/80 FDT MD 0.755 0.037 0.813 0.054 Ϫ2.37 61/90 Ϫ1.48 dB 74/80 HPRP MD 0.670 0.044 0.638 0.064 2.67 19/90 1.82 dB 39/80

Also shown are criterion values (Ն) derived for various parameters at speciﬁcities near 90% and 80%, based on the normal group with resultant sensitivities for the PGON group. Criterion values for PD and TD probabilities and HPRP dents are number of points meeting the criterion. PD at 1% criterion values used for abnormality and location of defect are shaded. Best criterion for each test shown in bold (GON) and italics (PGON). N/A, not applicable (no criterion values near 80% speciﬁcity for this parameter).

deviation (PD) Ͻ1% (or deep dent for HPRP) criterion number of size again that SAP fields were not used for the classification of points that yielded 80% specificity. The developer of HPRP, Lars Frisen any study group, to prevent bias. However, for the reader’s supplied the following explanation of dents (personal communication, information, Table 2 gives the descriptive results for the par- 2005). “Dents for HPRP were identified by sets of linear regressions. ticipants in each of the four groups. The means, standard One set ran over test locations that could be viewed as members of one deviation, and range of mean defect (MD) and pattern SD (PSD) and the same field radius. The other set ran over test points that are given for SAP to provide an idea of the range of disease for normally share much the same value, i.e., they normally belong to one study participants. None of the PGON and only 7% of the GON and the same isopter. Normally, test results should rise monotonically were in advanced stages based on a modification of the criteria along radii, and should stay constant along isopters. The regressions for visual field severity of Hodapp et al.32 The GON and PGON helped identify test locations that deviated by 1-l.9 dB from expected groups showed comparable early and moderate levels of sever- (ϭ shallow dent), or by 2 dB or more (ϭ deep dent). Hence, the ity. The normal subjects were significantly younger than the analysis essentially aimed to highlight deviations from the normal patients with GON. For this reason, only the age-corrected shape of the threshold surface.” parameters from Statpac (Carl Zeiss Meditec, Inc.) were eval- A ␹2 analysis was used to determine the statistical significance of uated. All four tests were completed within a median time of this agreement in pairs of tests. Kappa statistics were used and signif- 0.25 months (range, 0–6.33). icance assessed by using the method of Landis and Koch.31 Abnormality. Table 1 gives the AUCs comparing normal Location of Field Defect. The stimuli and test locations differ subjects with patients having GON and with those having among the four tests. For this reason, we assessed the agreement in the PGON. The more stringent criteria for glaucoma, PGON, re- location of the defect by quadrant. We first determined whether the sulted in larger AUCs, but the relationships among the four field was abnormal based on the criterion number of PD points at 1% tests were comparable for the PGON and GON groups. To determined from the ROC analysis for 80% specificity (shaded criteria avoid numerous multiple comparisons, statistical comparisons in Table 1). Once it was determined that a field had a sufficient number of AUC shown in Table 1 were made (1) comparing the AUC of PD locations at Յ1% to be called abnormal, a quadrant was called for PSD from each test and (2) comparing the best performing abnormal if it contained any one of these PD points (deep dents for parameter from each of the four tests for both the GON (bold) HPRP). HPRP presents stimuli centered on the vertical and horizontal and PGON (italic) definitions of glaucoma. The results of these meridians. When one of these points (which overlap two quadrants) comparisons are shown in Table 3. FDT consistently showed was abnormal, we attributed it to the most defective quadrant. the largest AUC regardless of the visual field parameter. For Extent of Defect. The number of quadrants affected on each PSD in the GON group, the FDT AUC was 0.770, followed by perimetric test gave the extent of the defect. Overall significant differ- SWAP (0.733), SAP (0.713), and HPRP (0.661). In the PGON ences among all tests were further assessed with paired comparisons group, FDT AUC was 0.875, followed by HPRP (0.780), SWAP using the Tukey HSD post hoc test. (0.775), and SAP (0.762). The FDT AUC was significantly larger than only the HPRP AUC (P ϭ 0.020) and only in the GON RESULTS group. Using the best parameter from each test, FDT TD at 5% AUC The major purpose of this study was to evaluate each of the (0.795), SWAP PSD (0.733), SAP PSD (0.713), and HPRP MD visual field procedures against the others. We want to empha- (0.670), showed that the FDT AUC was significantly larger than

Downloaded from iovs.arvojournals.org on 01/31/2019 IOVS, August 2006, Vol. 47, No. 8 Visual Function in Glaucoma 3385

TABLE 2. Summary Data for Each of the Four Participant Groups

(31 ؍ PGON (n (111 ؍ GON (n (53 ؍ OHT (n (51 ؍ Normal (n

Age (y) Mean Ϯ SD 51.81 Ϯ 13.70O,G,P 60.27 Ϯ 11.61N,G,P 65.59 Ϯ 11.42N,O 66.85 Ϯ 10.57N,O Range 25.75 to 81.33 23.29 to 79.96 33.21 to 87.47 40.01 to 81.96 Highest IOP (mm Hg) Mean Ϯ SD 16.08 Ϯ 3.03O,G,P 29.87 Ϯ 7.16N,G 25.44 Ϯ 7.36N,O,P 30.26 Ϯ 9.47N,G Range 10 to 23 13 to 59 14 to 54 16 to 56 SAP MD (dB) Mean Ϯ SD Ϫ0.72 Ϯ 1.17G,P Ϫ0.36 Ϯ 1.38G,P Ϫ2.89 Ϯ 3.97N,O Ϫ3.19 Ϯ 3.36N,O Range Ϫ4.05 to 0.95 Ϫ4.08 to 3.13 Ϫ20.14 to 2.04 Ϫ12.41 to 1.52 SAP PSD (dB) Mean Ϯ SD 1.72 Ϯ 0.68G,P 1.61 Ϯ 0.69G,P 3.62 Ϯ 3.58N,O 4.16 Ϯ 3.97N,O Range 1.05 to 4.46 0.90 to 4.2 1.02 to 16.11 1.33 to 14.19

ANOVA was performed on the continuous variables (all were signiﬁcant at P ϭ 0.001) and the Tukey HSD test was used to determine which groups signiﬁcantly differed from each other (see superscripts within the table). N, normal O, OHT; G, GON; P, PGON.

was the HPRP MD (P ϭ 0.004) in the GON group and than the classified as abnormal by only one test was high in the normal SWAP PSD (P ϭ 0.020) in the PGON group. No other compar- (39%) and OHT groups (36%) using the PD plot criterion (top), isons reached significance, including comparisons between but quite reasonable when two or more tests are required, 4% FDT and SAP or SWAP and SAP. and 8%, respectively. The percentage of eyes determined to be In addition to the AUC, the shape of the ROC curve is normal by all tests decreased in the expected direction from important. There is a tradeoff between sensitivity and specific- 53% in the normal subjects through 49% in OHT, 27% in GON ity. For this reason, we chose two different specificities for to only 16% in the PGON group. These relationships were analysis. The ROC curve results were used to determine crite- similar for the criteria PSD (Table 5; bottom). rion values for various parameters at specificities near 80% and Agreement among Tests. Figures 2 (PSD) and 3 (PD) 90%. Using this step-wise approach proceeding from the ROC show the agreement among the four tests in identifying abnor- analysis to determining the desired specificity can assist in mality in eyes with GON and PGON combined (n ϭ 142), using developing comparable criteria for abnormality for each of the the 80% specificity criterion values. Table 6 gives the ␬ statis- four tests. The PGON results are shown in Table 1. Sensitivities tics, proportion agreement, and strength of agreement for were slightly lower in the GON group, but again the relation- 31 ship among the test results was comparable to that in the these relationships. Agreement is fair to moderate with better PGON group. FDT showed the highest sensitivity in all cases. agreement found for some pairs of tests (SWAP and FDT; SWAP With this approach, a possible advantage to FDT testing for and HPRP) when using the PD plot criteria compared with the detection of vision loss in participants with PGON or GON PSD. emerges at both specificities and for all parameters. Table 4 Location of Field Defect. Eighty-three participants were shows the percentage of OHT eyes classified as abnormal when considered abnormal on two or more tests based on the PD using these same criterion values. criteria at 80% specificity: 4 (8%) of 51 normal, 8 (15%) of 53 Table 5 shows how the tests compared in separating par- OHT, 54 (49%) of 111 GON, and 17 (55%) of 31 PGON sub- ticipant groups when specificity was set to 80% for the param- jects. At least one quadrant was found to be commonly defec- eters PSD (top) or number of PD plot locations triggered at P Ͻ tive in 79 (95%) of the 83 eyes, regardless of whether two, 1% (bottom). It is important to note that the percentage of eyes three, or four tests had abnormal quadrants.

TABLE 3. Statistical Comparison of ROC Area for PSD and for the Best Parameter from Each Test

A. PSD

GON GON P PGON P SAP vs. HPRP 0.291 0.762 SAP vs. FDT 0.197 0.076 SAP vs. SWAP 0.633 0.813 HPRP vs. FDT 0.020 0.135 HPRP vs. SWAP 0.193 0.939 FDT vs. SWAP 0.432 0.067

B. Best parameter from each test

GON P PGON P

SAP PSD vs. HPRP MD 0.434 SAP TD 1% vs. HPRP PSD 0.795 SAP PSD vs. FDT TD 5% 0.069 SAP TD 1% vs. FDT PD 5% 0.090 SAP PSD vs. SWAP PSD 0.633 SAP TD 1% vs. SWAP PSD 0.677 HPRP MD vs. FDT TD 5% 0.004 HPRP PSD vs. FDT PD 5% 0.097 HPRP MD vs. SWAP PSD 0.240 HPRP PSD vs. SWAP PSD 0.939 FDT TD 5% vs. SWAP PSD 0.182 FDT PD 5% vs. SWAP PSD 0.020

Signiﬁcant results are shown in bold.

Downloaded from iovs.arvojournals.org on 01/31/2019 3386 Sample et al. IOVS, August 2006, Vol. 47, No. 8

TABLE 4. Percentage of Ocular Hypertensives Called Abnormal by Each Test

Speciﬁcity 80% Speciﬁcity 90%

PSD PD < 1% PSD PD < 1%

SAP 18.9% 15.1% 9.4% 9.4% HPRP 24.5% 20.8% 15.0% 11.3% FDT 45.3% 30.2% 26.4% 15.1% SWAP 18.9% 13.2% 15.1% 9.4%

n ϭ 53.

Extent of Defect. Table 7 shows the number of abnormal quadrants for each participant group and each test based on the PD criteria at 80% specificity. Within diagnostic groups, a significant difference in the normal group was attributed to FDT compared with each of the other three tests. Within a perimetric test the differences were significant (P Ͻ 0.0001) and were attributable to differences between normal subjects and the GON and PGON groups for all tests and between the OHT and the GON and PGON groups for all tests. No differences were found between normal and OHT or between GON and PGON.

DISCUSSION In this study, we found that no one test type always resulted in abnormal findings, whereas others remained normal in eyes with glaucoma. We also found that although FDT consistently showed the largest AUC, the areas were not significantly dif- FIGURE 2. A Venn diagram showing the overall agreement among ferent from the other tests, except that the FDT area was larger SAP, SWAP, FDT, and HPRP in the GON and PGON groups combined (n ϭ 142). Abnormality was based on the PSD and a specificity of 80%. than that in HPRP (PSD comparison) in the GON group, FDT TD at 5% area was larger than HPRP MD (best parameter from each) in the GON group and FDT PD5% was larger than SWAP 84% compared with SWAP at 42%, SAP at 68%, and HPRP small PSD (best parameter from each) in PGON group. FDT showed dent at 23%, consistent with the results from our earlier higher sensitivities at set specificities for all parameters. An study.24 example can be seen in Table 1 looking at the TD values at 5% There were several important improvements in this study and specificities near 80%. In this case, FDT has a sensitivity of compared with those in our original study.24 First, visual fields

TABLE 5. The Number of Eyes and Percentage with Number of Visual Field Types Showing Abnormal Results

Number of PGON GON OHT Normal (51 ؍ n) (53 ؍ n) (111 ؍ n) (31 ؍ Abnormal Results (n

PD Plot All 4 8 (26) 24 (22) 3 (6) 0 (0) 3 of 4 3 (10) 15 (14) 1 (2) 2 (4) 2 of 4 6 (19) 15 (14) 4 (8) 2 (4) 1 of 4 9 (29) 27 (24) 19 (36) 20 (39) SAP only 1 7 1 4 SWAP only 1 3 3 6 FDT only 6 14 10 6 HPRP only 1 3 5 4 None 5 (16) 30 (27) 26 (49) 27 (53) PSD All 4 10 (32) 31 (28) 4 (7) 1 (2) 3 of 4 3 (10) 17 (15) 2 (4) 2 (4) 2 of 4 8 (26) 16 (15) 7 (13) 6 (12) 1 of 4 8 (26) 30 (27) 21 (40) 18 (35) SAP only 0 5 0 3 SWAP only 1 6 3 4 FDT only 4 12 13 5 HPRP only 3 7 5 6 None 2 (6) 17 (15) 19 (36) 24 (47)

Data are based on the PD at 1% plot (top) and PSD (bottom) for each patient group at 80% speciﬁcity for each test.

Downloaded from iovs.arvojournals.org on 01/31/2019 IOVS, August 2006, Vol. 47, No. 8 Visual Function in Glaucoma 3387

The second improvement in this study was the inclusion of HPRP, a possible parvocellular cell test. Although magnocellular and parvocellular cell functions overlap greatly,33 there is a range for both cell types where one is significantly more likely to mediate detection than the other if the parameters of the visual stimulus are properly designed.18 The high-resolution nature of HPRP is most likely handled by the parvocellular cells, especially in the central visual field. The amount of parvocellular cell damage needed before magnocellular cells can take over detection of the HPRP stimulus (the amount of parvocellular cell isolation), however, is unknown. The third improvement was the inclusion of a group of eyes with progressive GON. Glaucoma is defined as a progressive optic neuropathy. By design we did not use visual field loss to help verify the presence of glaucoma in eyes with recently identified or long-term stable GON. Therefore, evidence of PGON provided a subgroup that most surely had glaucoma.27 The results of this group, even though few in number and therefore lower in power, were consistent with those in the larger GON group, thereby strengthening the conclusions drawn. When new tests are developed, it is problematic to determine what criteria are best for separating healthy from dis- eased eyes. A variety of methods have been used. For example, the modified Statpac criteria (Carl Zeiss Meditec, Inc.) for SWAP are based on those developed for SAP.34 Often, different criteria are developed and used by different investigators. An example of this is the number of different criteria for evaluating abnormality on FDT (reviewed in Muskens et al.35). A strength of the present study was the use of ROC analysis to set FIGURE 3. A Venn diagram showing the overall agreement among SAP, SWAP, FDT, and HPRP in the GON and PGON groups combined criteria for each perimetric procedure and to equate the tests (n ϭ 142). Abnormality is based on the PD plot results at the 1% level for specificity. In clinical practice, each instrument uses its and a specificity of 80%. own internal normative database. It seemed most clinically relevant to use these databases and to assess specificities for comparisons among tests using an independent group of were not used to classify any of the study participants. This healthy eyes. avoids bias in two ways. For example, if SAP is used as a gold A limitation to our study was that we did not have confirm- standard along with GON to define normal and glaucoma, no ing visual field results for all test types, and so this criterion was other test could ever exceed it in sensitivity and specificity by not included. If a test is overcalling abnormality, it suggests a definition. In our original study, normal participants were re- higher sensitivity for that test than is the actual case. In this real quired to have normal SAP examinations causing difficulties in false-positive situation, a repeat would most likely not confirm setting all tests for equal specificities. We had to use other the abnormal result. Therefore, requiring two abnormal test published data to set criterion values. Not requiring SAP de- results in a row should improve specificity and give a truer fects to classify participants in the present study allowed SAP measure of the test’s sensitivity to glaucomatous loss. In our deficits to precede those of the function specific tests, or vice study, this could influence the results if one test is more likely versa, without constraint. to overcall abnormality than the others. For example, the

TABLE 6. The ␬ Statistics and Levels of Agreement for All Combinations of Test Pairings in the GON and PGON Groups Combined

Proportion Strength of ␬ SE Agreement Agreement

PD Plot SAP and SWAP 0.586 0.07 0.79 Moderate SAP and FDT 0.393 0.07 0.69 Fair SAP and HPRP 0.375 0.08 0.69 Fair SWAP and FDT 0.411 0.07 0.69 Moderate SWAP and HPRP 0.470 0.07 0.74 Moderate FDT and HPRP 0.326 0.07 0.64 Fair PSD SAP and SWAP 0.563 0.07 0.78 Moderate SAP and FDT 0.349 0.07 0.67 Fair SAP and HPRP 0.394 0.08 0.69 Fair SWAP and FDT 0.366 0.07 0.68 Fair SWAP and HPRP 0.268 0.08 0.63 Fair FDT and HPRP 0.247 0.08 0.62 Fair

Top: based on the PSD; bottom: PD plot criterion values at P Ͻ 1%.

Downloaded from iovs.arvojournals.org on 01/31/2019 3388 Sample et al. IOVS, August 2006, Vol. 47, No. 8

TABLE 7. Number of Abnormal Quadrants for Each Group and Test

Normal OHT GON PGON ANOVA P (31 ؍ n) (111 ؍ n) (53 ؍ n) (51 ؍ n)

SAP 0.76 Ϯ 0.89 0.55 Ϯ 0.89 1.38 Ϯ 1.25 1.61 Ϯ 1.20 Ͻ0.001 SWAP 0.69 Ϯ 0.99 0.49 Ϯ 0.78 1.21 Ϯ 1.05 1.39 Ϯ 1.15 Ͻ0.001 FDT 0.20 Ϯ 0.49 0.42 Ϯ 0.75 1.09 Ϯ 1.13 1.32 Ϯ 1.11 Ͻ0.001 HPRP 0.57 Ϯ 0.88 0.66 Ϯ 0.92 1.13 Ϯ 1.14 1.42 Ϯ 0.96 Ͻ0.001 P 0.004 0.4889 0.2445 0.7569

For the normal subjects, a Tukey HSD test shows that the FDT had fewer abnormal quadrants than did SAP and SWAP, but not HPRP. Within each test, the Tukey showed that the normal subjects differed from the GON and PGON groups and the OHT differed from the GON and PGON groups. Mean Ϯ SD.

percentage of OHT eyes classified as abnormal was higher for ity among tests was only fair to moderate. These two findings FDT (45%) at 80% specificity than for the other tests. This together are consistent with the conclusion that no one gan- percentage is higher than the number that would be expected glion cell subtype is always affected first in glaucoma.24,26,37 to convert to glaucoma, a result similar to our earlier finding.24 All optic nerve fibers are damaged, but tests that favor detec- At 90% specificity, the result (26%) is more reasonable. There tion of a stimulus by one visual pathway or processing sub- has been a case report showing that FDT results fluctuate with system reduce the ability of the visual system to use other 36 changes in IOP, but none of our participants had the high pathways to compensate for the damaged retinal ganglion cell 26 IOP noted in that study. In FDT’s favor, Spry et al. found little type under test.2,3 These findings are consistent with those difference in confirmation of field results for FDT (47%) com- from an elegant study in primate animal models of glaucoma. pared with SWAP (42%), and the small difference seen was in Harwerth et al.38 combined psychophysics, electrophysiology, the direction opposite that expected if FDT was simply over- anatomy, and histochemistry to show that glaucomatous atro- calling abnormality in these OHT eyes. To address partially the phy causes a nonselective reduction of metabolism of magno- lack of confirmation on a specific test type, results from two cellular and parvocellular neurons in the afferent visual path- different test types can be used to verify the presence of way. Yucel et al.39 also find no evidence for selective cell loss glaucomatous vision loss. In this situation we found that only a in glaucoma within the magnocellular, parvocellular, or konio- small number of normal eyes were identified as abnormal when cellular layers of the LGN.39 However, not all eyes with pri- at least two tests were required (Table 5). mary open-angle glaucoma or those at risk of the disease are Our study also does not address progressive visual field loss. affected in the same way. Perhaps blue–yellow ganglion cell Additional study is needed to determine the relative ability of each test type to follow the disease once visual field loss is function is reduced first in one individual, parvocellular in another individual, and magnocellular ganglion cell function in already present. 24,40 The results of the present study point out a difference from another. The inclusion of HPRP did not alter this conclu- earlier findings,24 where we found that SWAP-FT and FDT-N30 sion. The study also replicated our earlier finding that when outperformed SAP-FT. In the present study, this difference was two or more test results are abnormal, the same quadrant of the not found. It may be that the introduction of the SITA thresh- visual field is affected on all. This finding is very important for olding algorithm for SAP, with its tighter confidence limits, has using two different tests to verify the presence of abnormality improved SAP’s ability to detect abnormality relative to the and for targeting follow-up by careful monitoring of specific longer and more variable SWAP full threshold. It remains to be areas of the visual field. seen if the newly released SITA version of SWAP will improve In summary, no one test type was always affected in pa- its performance in a similar manner. This result also highlights tients with GON or PGON, whereas the other test types re- one of the ongoing problems in clinical research. The technol- mained normal. Several parameters were identified that pro- ogies are changing so rapidly that it is difficult to obtain data vided good sensitivity at set specificities. The FDT N-30 test and report results on a device before it has moved on to the showed the highest sensitivity for all parameters. SAP perfor- next generation with improvements. For glaucoma, this prob- mance was equal to or slightly better than SWAP and not lem exists both with measures of visual function and with significantly different from FDT, a finding that differs from assessment of the optic nerve and retinal nerve fiber layer. those in previous studies. Consistent with previous findings, Ongoing re-evaluation with patients transitioning from one the same quadrant of the retina is damaged for all affected tests. version or test to another is critical to giving a complete picture A combination of test types may be most efficient in identifying of how these new instruments or procedures will best alter early loss and confirming the area of the retina affected by clinical practice. glaucoma. An additional reason for the improvement in SAP performance relative to SWAP may be that none of the participants in this study were selected based on their SAP visual field results. References As we have mentioned, our previous study included only 1. Sample PA, Johnson CA, Haegerstrom-Portnoy G, Adams AJ. Opti- normal control subjects with normal SAP fields, and specificity mum parameters for short-wavelength automated perimetry. J criterion values were derived from the literature. SAP was not Glaucoma. 1996;5:375–383. used for this in the present study and the same normal subjects 2. Johnson CA. Selective versus non-selective losses in glaucoma. J were tested on all four tests and used to set the specificity Glaucoma. 1994;3:S32–S44. criterion values, removing any possible bias due to different 3. Sample PA, Madrid M, Weinreb RN. Evidence for a variety of normal subjects used for the specificity of each test. functional defects in glaucoma suspect eyes. J Glaucoma. 1994; This study found that patients could show deficits on any 3:S5–S18. one of the four perimetric procedures, while remaining normal 4. Dacey DM. Morphology of a small-field bistratified ganglion cell on some or all the other three. In addition, the results were type in the macaque and human retina. Vis Neurosci. 1993;10: similar to those in earlier work in that agreement on abnormal- 1081–1098.

Downloaded from iovs.arvojournals.org on 01/31/2019 IOVS, August 2006, Vol. 47, No. 8 Visual Function in Glaucoma 3389

5. Dacey DM, Lee BB. The ‘blue-on’ opponent pathway in primate ferent ganglion cell populations in glaucoma. Invest Ophthalmol retina originates from a distinct bistratified ganglion cell type. Vis Sci. 2000;41:1783–1790. Nature. 1994;367:731–735. 25. Sample PA, Chauhan BC, Araie M, Johnson CA. Comparison of 6. Dacey DM, Packer OS. Colour coding in the primate retina: diverse functional methods. In: Weinreb RN, Greve EL, eds. Glaucoma cell types and cone-specific circuitry. Curr Opin Neurobiol. 2003; Diagnosis: Structure and Function. The Hague, The Netherlands: 13:421–427. Kugler Publications; 2004:119–126. 7. Martin PR, White AJ, Goodchild AK, Wilder HD, Sefton AE. Evi- 26. Spry PG, Johnson CA, Mansberger SL, Cioffi GA. Psychophysical dence that blue-on cells are part of the third geniculocortical investigation of ganglion cell loss in early glaucoma. J Glaucoma. pathway in primates. Proc Natl Acad Sci USA. 1997;27:5900– 2005;14:11–19. 5905. 27. Medeiros FA, Zangwill LM, Bowd C, Sample PA, Weinreb RN. Use 8. Maddess T, Henry GH. Performance of nonlinear visual units in of progressive glaucomatous optic disk change as the reference ocular hypertension and glaucoma. Clin Vis Sci. 1992;7:371–383. standard for evaluation of diagnostic tests in glaucoma. Am J Ophthalmol. 2005;139:1010–1018. 9. Johnson CA, Demirel S. The role of spatial and temporal factors in frequency-doubling perimetry. In: Wall M, Heijl A, eds. Perimetry 28. Soliman MA, de Jong LA, Ismaeil AA, van den Berg TJ, de Smet MD. Update 1996/97: Proceedings of the XIIth International Perimet- Standard achromatic perimetry, short wavelength automated perimetry, and frequency doubling technology for detection of glau- ric Society Meeting. Amsterdam: Kugler; 1997:13–19. coma damage. Ophthalmology. 2002;109:444–454. 10. Fitzke FW, Poinoosawmy D, Nagasubramanian S, Hitchings RA. 29. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas Peripheral displacement thresholds in glaucoma and ocular hyper- under two or more correlated receiver operating characteristic tension. In: Heijl A, ed. Perimetry Update 1988/89. Amsterdam: curves: a nonparametric approach. Biometrics. 1988;44:837–845. Kugler; 1989:399–405. 30. Goldbaum MH, Sample PA, Chan K, et al. Comparing machine 11. Trick GL, Steinman BS, Amyot M. Motion perception deficits in learning classifiers for diagnosing glaucoma from standard auto- glaucomatous optic neuropathy. Vision Res. 1995;25:2225–2233. mated perimetry. Invest Ophthalmol Vis Sci. 2002;43:162–169. 12. Wall M, Jennisch CS, Munden PM. Motion perimetry identifies 31. Landis JR, Koch GG. The measurement of observer agreement for nerve fiber bundle-like defects in ocular hypertension. Arch Oph- categorical data. Biometrics. 1977;33:159–174. thalmol. 1997;115:26–33. 32. Budenz DL, Rhee P, Feuer WJ, McSoley J, Johnson CA, Anderson 13. Bosworth CF, Sample PA, Weinreb RN. Perimetric motion thresh- DR. Comparison of glaucomatous visual field defects using stan- olds are elevated in glaucoma suspects and glaucoma patients. dard full threshold and Swedish interactive threshold algorithms. Vision Res. 1997;37:1989–1997. Arch Ophthalmol. 2002;120:1136–1141. 14. Silverman SE, Trick GL, Hart WM Jr. Motion perception is abnor- 33. Anderson RS. The psychophysics of glaucoma: improving the mal in primary open-angle glaucoma. Invest Ophthalmol Vis Sci. structure/function relationship. Prog Retin Eye Res. 2005;25:79– 1990;31:722–729. 97. 15. Frisen L. High-pass resolution perimetry: evidence for parvocellu- 34. Johnson CA, Sample PA, Cioffi GA, Liebmann JR, Weinreb RN. lar neural channel dependence. Neuro-ophthalmology. 1992;4: Structure and function evaluation (SAFE): I. Criteria for glaucoma- 257–264. tous visual field loss using standard automated perimetry (SAP) and 16. Frisen L. High-pass resolution perimetry: central-field neuroretinal short wavelength automated perimetry (SWAP). Am J Ophthal- correlates. Vis Res. 1995;35:293–301. mol. 2002;134:177–185. 17. Callaway EM. Structure and function of parallel pathways in the 35. Muskens RP, Heeg GP, Jansonius NM. An evaluation of algorithms primate early visual system. J Physiol. 2005;566:13–19. designed to classify the results from frequency doubling perimetry. 18. Kaplan E. The M, P, K pathways of the primate visual system. In: Ophthalmic Physiol Opt. 2004;24:498–503. Chalupa LM, Werner JS, eds. The Visual Neurosciences. Cam- 36. Fujimoto N, Zhang D, Sato E, Miyauchi O, Mizota A. Reversible bridge, MA: The MIT Press; 2004:481–493. changes in frequency-doubling perimetry with transiently elevated intraocular pressure. Arch Ophthalmol. 2004;122:1540–1543. 19. Dobkins KR, Albright TD. Merging processing streams: color cues for motion detection and interpretation. In: Chalupa LM, Werner 37. Martin L, Wanger P, Vancea L, Gothlin B. Concordance of high- JS, eds. The Visual Neurosciences. Cambridge, MA: The MIT Press; pass resolution perimetry and frequency-doubling technology pe- 2004:1217–1228. rimetry results in glaucoma: no support for selective ganglion cell damage. J Glaucoma. 2003;12:40–44. 20. Anderson AJ, Johnson CA. Frequency-doubling technology perim- 38. Harwerth RS, Crawford ML, Frishman LJ, Viswanathan S, Smith EL, etry. Ophthalmol Clin North Am. 2003;16:213–225. Carter-Dawson L. Visual field defects and neural losses from ex- 21. Chauhan BC. The value of high-pass resolution perimetry in glau- perimental glaucoma. Prog Retin Eye Res. 2002;21:91–125. coma. Curr Opin Ophthalmol. 2000;11:85–89. 39. Yucel YH, Zhang Q, Weinreb RN, Kaufman PL, Gupta N. Effects of 22. Racette L, Sample PA. Short-wavelength automated perimetry. retinal ganglion cell loss on magno-, parvo-, koniocellular path- Ophthalmol Clin North Am. 2003;16:227–236. ways in the lateral geniculate nucleus and visual cortex in glau- 23. Sample PA. Short-wavelength automated perimetry: its role in the coma. Prog Retin Eye Res. 2003;22:465–481. clinic and for understanding ganglion cell function. Prog Retin Eye 40. Sample PA, Bosworth CF, Weinreb RN. Short-wavelength auto- Res. 2000;19:369–383. mated perimetry and motion automated perimetry in patients with 24. Sample PA, Bosworth CF, Blumenthal EZ, Girkin C, Weinreb RN. glaucoma (see comments). Arch Ophthalmol. 1997;115:1129– Visual function-specific perimetry for indirect comparison of dif- 1133.

Downloaded from iovs.arvojournals.org on 01/31/2019