Variability in Patients With Glaucomatous Visual Field Damage Is Reduced Using Size V Stimuli

Transcription

1 ariability in Patients With Glaucomatous isual Field Damage s Reduced Using Size Stimuli Michael Wall* Kim E. Kutzko,* and Balwantray C. Chauhanf Purpose. To test the hypothesis that variability of conventional automated perimetry can be reduced using size stimuli for patients with glaucomatous visual field damage. Methods. Ten patients with glaucoma and five age-matched control volunteers were tested with the Humphrey Field Analyzer program 24-2 or 0-2, after which the method of constant stimuli was used to measure frequency-of-seeing curves. This was done by controlling the perimeter with a custom program run by a personal computer. At two widely separated visual field locations on the program 24-2 or 0-2 grid, stimuli were presented in 2 db intervals to at least 10 db on either side of the estimated program 24-2 or 0-2 threshold. This protocol was performed for each of three stimulus sizes (Goldmann sizes,, and ). For die patients with glaucoma, one test location was chosen in an area of normal visual field sensitivity, the other in an area of 10 to 20 db loss. Control subjects were tested at the (, ) and ( 21, ) test locations. Fifteen repetitions were performed at each intensity. Results. Repeated measures analysis of variance showed that variability, as measured by die standard deviation of the cumulative Gaussian function of the fitted frequency-of-seeing curves, was lowest at the abnormal sensitivity test location in the subjects with glaucoma using a size stimulus. Differences between the results from die to and to stimuli were statistically significant (size = 2. db, = 10.1 db, = 10.1 db). The same trend in estimated standard deviations was present in tests of the area of normal sensitivity (size = 1.1 db, = 1.7 db, = 2.0 db) in subjects with glaucoma and for the control subjects' peripheral test locations, but not for the central location. The smaller reduction in variability between estimated standard deviations of the size and size stimuli was not statistically significant at any test location. ConclusioTis. Use of size stimuli in conventional automated perimetry reduces variability in tests of moderately damaged and normal sensitivity test locations in subjects with glaucoma. nvest Ophthalmol is Sci. 17; 8: A he automation of perimetry has produced many benefits. As a result of the delegation of stimulus presentation control and response recording to a computer, variability due to technique differences among perimetric examiners has been reduced. Automation has also allowed static perimetry, a more sensitive method of stimulus presentation, to be clinically practical. Sophisticated statistical indices have evolved, From the * Departments of Neurology and Ophthalmology, eterans Administration Hospital and University of owa College of Medicine, owa City, owa, and the ^Department of Ophthalmology, Dalhousie University, Halifax, Canada. Supported by a grant from A Merit Review, by an unrestricted grant to the Department of Ophthalmology, Research to Prevent Blindness (New York, NY) (MW), and by Medical Research Council of Canada grant MT-1157 (BCC). Submitted for publication May 0, 16; revised August 2, 16; accepted September 16, 16. Proprietary interest category: N. Reprint requests: Michael Wall, University of owa, College of Medicine, Department of Neurology, 200 Hawkins Drive #2007 RCP, owa City, A many in an attempt to determine what constitutes a change in visual function. n spite of control of more variables, variability of conventional automated perimetry in areas of visual field damage remains high. The magnitude of the variability is such that Werner has suggested that clinicians should not make judgments about progression of disease without at least six automated visual fields in hand. 1 The test-retest variability of conventional automated perimetry is in an acceptable range in control subjects. After a learning period of two to five visual fields, the variability of threshold measurements stabilizes. 2 The standard deviation of results of patients retested four to five times over a 1 mondi period ranges from 1 to 2 db centrally to 4 to 6 db at 27 degrees. ' 4 When examined with frequency-of-seeing curves, the transition zone in control subjects between 426 nvestigative Ophthalmology & isual Science, February 17, ol. 8, No. 2 Copyright Association for Research in ision and Ophthalmology

2 ariability in Glaucoma Using Size Stimuli 427 seeing and not seeing with differential light sensitivity stimuli is in the range of 4 to 10 db, again increasing with eccentricity. 5 " 7 Therefore, in reliable trained control subjects, perimetry results are reproducible. The situation is different, however, for conventional automated perimetry results when optic nerve damage is present. A study by Heijl and colleagues 8 investigated the variability in 51 eyes of 51 experienced subjects with glaucoma representing all stages of optic nerve damage. The patients, all clinically stable, were tested four times in a 4 week period. Test locations, initially measured with 6 db loss, had a 0% prediction interval from 1 to 16 db. With 8 to 18 db loss initially, the 5% prediction interval nearly covered the full measurement range of the instrument (0 to 40 db). An important finding of that study, also found by others, " 14 is that pointwise interest variability increases dramatically with decreasing sensitivity of the test point. This finding has a major ramification; areas with the most visual loss have the highest variability and, therefore, the most clinically important regions are ones in which determination of change is most difficult. There are a limited number of strategies available to reduce this variability when using conventional automated perimetry. Attempts have been made to reduce the test time, but this has resulted in greater variability. 15 " 17 Johnson and colleagues, 18 using computer simulations, varied many conditions of the perimetric testing strategy, including the number of staircase reversals, staircase step size, and starting point value, without significant effect on the accuracy of the determination. They concluded that for these variables of the staircase algorithm, conventional automated perimetry was at or near optimal levels; however, response errors (mistakes in responding) did contribute significantly to the accuracy, as did variability. Using high pass resolution perimetry, it has been shown that variability in control subjects remains stable with increasing visual field eccentricity. 1 ' 20 Also, there is no significant increase in variability in patients with visual field damage. 20 ' 21 This method, which determines threshold by changing stimulus size rather than intensity, may reduce variability by using large stimuli in areas of sparse receptive coverage. Gilpin and colleagues 22 have also reported increased variability as stimulus size decreases. They found a decrease in variability with a size compared with a size stimulus, but found little change in total variability between the size and size stimulus. Because they only tested control subjects and did not include test locations outside 20 degrees, their data does not reflect the effect of low sensitivity or of visual field damage. Because variability of conventional automated perimetry has a major impact on clinical decisions, and because variability can be reduced using larger test stimuli in control subjects, we wished to determine whether using a large stimulus size could substantially reduce the variability in patients with glaucoma. Therefore, to test this hypothesis, we used an analysis of frequency-of-seeing curves on two groups of subjects. We tested subjects with Goldmann stimulus sizes,, and and compared control subjects with normal visual fields (central and peripheral test locations), patients with glaucoma in areas of normal visual field sensitivity, and patients with glaucoma in visual field areas with moderate visual field damage. METHODS Subjects Ten patients with well-established primary open angle glaucoma and five control volunteers gave informed consent to participate in the study. The protocol was approved by the University of owa nstitutional Review Board. The tenets of the Declaration of Helsinki were followed. The control subjects were paid volunteers who were hospital employees or friends or family members of eye clinic patients. The control subjects were matched to patients by age within 10 years. Control subjects were included if they had no histor)' of eye disease except refractive error and had a normal ophthalmologic examination. They all had normal automated perimetry results using the Humphrey Field Analyzer (HFA, Humphrey nstruments, San Leandro, CA), program f a potential control subject had three or more adjacent points with a total deviation score at the P < 0.05 level or two adjacent points with one at the P< 0.01 level with STATPAC (HFA), 2 they were excluded. Control subjects were also excluded if the mean deviation index was outside the 5% confidence bound for controls. The patients with glaucoma were recruited from past studies in which they had taken an HFA test that had shown a test location in the 10 to 20 db range in one eye with visual field damage (P < for pointwise total deviation with STATPAC analysis) and a test location with normal results (P > 0.05) with the total deviation plot. All patients with the clinical diagnosis of primary open angle glaucoma had open angles on gonioscopy. They had an intraocular pressure greater than 21 mm Hg during their course as well as glaucomatous optic disc changes, glaucomatous visual field defects, and no other mechanism of glaucoma apparent. All patients were receiving treatment for intraocular pressure control, but none were using a miotic. Primary open angle patients with glaucoma were excluded if they had any other disease causing or known to cause visual loss. Eight of the patients had been subjects in a study with a similar protocol using size stimuli. 7 Examples of the types of visual field defects

3 428 nvestigative Ophthalmology & isual Science, February 17, ol. 8, No. 2 0": FGURE l. Gray-scaled results of threshold perimetry of the least damaged visual field (patient 1, top), typical amount of damage (patient 6, middle), and most damaged visual field (patient, bottom) of the 10 patients with glaucoma. of these patients are shown in Figure 1; the mean deviation and corrected pattern standard deviation of the patients with glaucoma is found in Table 1. All subjects had a corrected visual acuity of 20/25 or better, a pupil diameter of at least three millimeters when tested, a spectacle correction not exceeding 6.00 diopters (equivalent sphere), and previous experience with automated perimetric examinations. Testing Strategy Conventional automated perimetry was first performed with the HFA with program 24-2 or 0-2 according to the manufacturer's recommendations. We used a Goldmann size stimulus on a 1.5 apostilb background. The patients' appropriate near correction was used. Rest breaks were given when requested. Frequency-of-seeing curves was then measured by controlling the HFA with a custom program 5 run by a personal computer (BM 486/) using three different stimulus sizes (Goldmann sizes,, and, which correspond to sizes of 0.25, 4, and 64 mm 2 or 0.11, 0.4, and 1.72 of visual angle). Tests for the different sizes were conducted on the same day in random order; subjects were given at least 10 minutes between tests, depending on whether they felt fatigued. For the size stimuli at the two separated test locations, stimuli were presented 10 db on either side of the estimated (HFA program 24-2 or 0-2) threshold in 2 db intervals with fifteen repetitions at each stimulus intensity. For the size stimulus, we subtracted 10 from the size threshold and used that to estimate the 50% correct threshold, and for the size we started at 42 db and went to 20 db in 2 db steps. All presentations of stimulus intensity and location were randomized. To determine false positive and false negative responses, a 60 db and 0 db stimulus were also presented twenty times at each location (because the 0 db stimulus was so intense with size, the 20 db stimulus was substituted). Consequently, each location was tested 205 times. To prevent subjects from concentrating their attention on the two tested points, eight random additional locations of normal or near normal vision were tested with three repetitions of the 0 db or 20 db stimulus for a total of 24 extra trials. These 44 trials produced a test duration of approximately 0 minutes. This constraint was added to simulate the usual clinical testing time of two 0-2 tests. f the subject did not respond within two seconds, the stimulus was considered "not seen." Fixation was monitored by the visual field technician using the video display of the instrument. Fixation did not differ between tests and was well-maintained except for one subject with glaucoma who needed to be reminded to look at the fixation stimulus between trials. Breaks of approximately 5 minutes were provided to every subject after the 150th and 00th trials. Two main locations were tested in both the control subjects and the patients with glaucoma. n the control subjects, the test locations were at the Cartesian coordinates (,) and ( 21,-). n patients with glaucoma, using thresholds from their 24-2 or 0-2 test, one of the two points tested came from an area of normal visual field sensitivity as determined by the total deviation plot (P > 0.05); the other point from an area of the patient's visual field where there was 10 to 20 db of visual field loss. All damaged test locations had a P < total deviation score. We tried to use test locations that were doubly determined and thresholds that were not greatiy different from neighboring test points. n other words, we avoided testing locations at steep edges of visual field defects because of their known high test-retest variability. 24 Also, we tried to choose high enough sensitivities in the patients with glaucoma' normal sensitivity test location

4 ariability in Glaucoma Using Size Stimuli 42 TABLE l. Data for Patients with Glaucoma From Conventional Automated Perimetry Results Using Program 24-2 or 0-2 and Test Locations for the Frequency-of-Seeing Curves* Normal Sensitivity Location Abnormal Sensitivity Location Patient Number Mean Deviation CPSD rti ft 24-2 Threshold 24-2 Loss FOS Location X y 24-2 threshold r\ A o 24-2 loss FOS Location X y Mean SD i CPSD = corrected pattern standard deviation; FOS = frequency-of-seeing test, 24-2 refers to results on the HFA 24-2 or 0-2 full threshold test; SD = standard deviation; x and y = Cartesian coordinates of the locations for the frequency-of-seeing test. * Threshold results and indices are in decibel units. to match the thresholds of the control subjects' peripheral test location. To balance the location of the test presentations, these two points were chosen at opposite sides of either the vertical or horizontal meridians. Although we did not pair eccentricities individually, our mean distances from fixation in the normal sensitivity test location of patients with glaucoma were 1.2, and in the abnormal test location, 12.6 (Table 1). Our controls were tested at the x =, y = and the x = 21, y = locations, giving distances from fixation of 4.2 and 22.8 (mean = 1.5 ). Testing of patients with glaucoma was different from control subjects in two ways. First, the locations were selected using the criteria stated above: one point from an area of normal sensitivity and one point from an area with 10 to 20 db visual loss. Second, due to the greater variability in patients with glaucoma, when a range wider than 10 db on either side of the threshold was needed to construct the frequency-of-seeing curves, these additional points were tested at a separate sitting during the same day. Seven of the patients with glaucoma required this extra testing at their abnormal test location because of failure to reach 0% or seen. Three patients needed extra testing for all three sizes, two for only size, one for only size, and one for size and ; the r 2 values for the curves of these patients were size = 0.82, range of 0.68 to 0.6, size = 0.86, range of 0.66 to 0.8 and size = 0.. Data Analysis Frequency-of-seeing curves for each subject was constructed as cumulative Gaussian functions and a least squares fit was calculated using the Microsoft Excel solver function (Microsoft, Redmond, WA). This function changed the mean and standard deviation of the curve fit until the r 2 value was maximized. The resultant mean of the fitted distribution corresponds to the 50% correct threshold for the test location (defined as the stimulus intensity corresponding to the 50% frequency-of-seeing point of the fitted curve). The standard deviation of the cumulative Gaussian function (an index of the maximum slope of the frequency-of-seeing function) and the coefficient of determination or goodness of fit (r 2 ) were calculated for each frequency-of-seeing curve. We also estimated the transition zone of the frequency-of-seeing curve (the range of intensities from seeing 0% of stimuli to seeing of stimuli) by inspection of the mean cumulative Gaussian functions for each group and stimulus condition. Statistical Analysis The subjects' data files were imported into Sigmastat (San Rafael, CA) for further statistical analysis. For outcome variables not normally distributed using the Kolmogorov-Smirnov test (P > 0.05) or having variances dissimilar using the Levene Median test (P > 0.05), we used a Kruskal-Wallis test and performed post-hoc pairwise comparisons using Dunn's method. For variables meeting criteria for parametric statistics, we used a one-way repeated measures analysis of variance with post-hoc pairwise comparisons using Tukey's test. Because the size results of one subject with glaucoma at the abnormal sensitivity test location could not be fit with a cumulative Gaussian function, we used nine subjects for the corresponding repeated

5 Sizel Size T Sizem Size in -r Size j 80% - 60% - 40% - 20% - 0% % 60% 40% 20% 0% B Size Size Size 80% 60% 40% 20% 0% T SizeH Sizem 60% - 40% - Size T Size FGURE 2. Frequency-of-seeing curves of individual subjects. (A) Representative results from a central test location of a control subject. (B) Results from a peripheral test location of the same control subject. (C) Results from a normal sensitivity test location, x =, y =, of a subject with glaucoma (patient 6; see middle gray scale, Fig. 1, for this patient's conventional automated perimetry results). (D) Normal sensitivity test location of the same patient with glaucoma (see middle gray scale, Fig. 1, x = 15, y = - ).

6 ariability in Glaucoma Using Size Stimuli 41 TABLE 2. Estimates of the 50% Frequency-of-Seeing Thresholds (in db) for the Different Groups by Size* Glaucoma Normal,,, 22,- -21,- -21,- Mean Median SD Transition zone = abnormal; = test location of normal sensitivity for patient with glaucoma; SD = standard deviation. * Estimated mean transition zone from seeing to nonseeing in decibels is also shown. measures analysis. Statistical significance was set at P < RESULTS The mean age of the patients was 62.6 ±. years and that of the control subjects was 5.8 ±.4 years. The age difference between the groups was not statistically significant (P= 0.12). Representative examples of the frequency-of-seeing curves of the subjects are found in Figure 2. As expected, the frequency-of-seeing curve thresholds of the control subjects' central test location was lower than the peripheral location for each size (Table 2). The control subjects' peripheral test location was 2.7 db less than the patients with glaucoma' normal sensitivity test location with the size stimulus. This difference was not statistically significant (P 0.07). False negatives, as measured by failure to respond to the 0 or 20 db stimulus, were uncommon (four total) except at the abnormal sensitivity test location for the patients with glaucoma. They occurred here because the brightest stimulus (0 db) was not seen of the time (for five subjects with the size stimulus and four with the size stimulus). False positives, which were responses to the 60 db stimulus, were also rare (five total). Repeated measures analysis of variance showed that intratest variability, illustrated by the standard deviation of the cumulative Gaussian function of the frequency-of-seeing curves, was lowest at the abnormal sensitivity test location in the subjects with glaucoma using a size stimulus (size = 2. db, = 10.1 db, = 10.1 db, P = 0.004, with the to and to differences being significant; Table 2, Figure ). n one subject with glaucoma, thresholds were so high in abnormal sensitivity test location with the size stimulus that a cumulative Gaussian function could not be fit. There was a lower magnitude of the variability, but a similar trend in frequency-of-seeing curves standard deviations was present in the area of normal sensitivity of subjects with glaucoma, with the size to size difference being significant (P 0.00) (size = 1.1 db, = 1.7 db, = 2.0 db). Again, a similar trend was present for the control subjects' peripheral test location (see Table, Figure ), but the smaller reduction in variability between the frequency-ofseeing curves standard deviations of these test locations was not statistically significant for any size comparison. The frequency-of-seeing curves estimated standard deviations for the control subjects' central test location were all close to 1.0. The difference in r 2 values, representing the goodness of fit of the frequency-of-seeing curves, among the groups (normal and abnormal sensitivity test locations of patients and central and peripheral locations of control subjects) was significant; however, only the abnormal sensitivity test location for the size to size T 2 comparison was significantly different (P = 0.02) (Table 4). Lastly, we found a direct linear relationship between threshold and goodness offit for the individual groups (normal and abnormal sensitivity test locations of patients and central and peripheral'locations of control subjects) and for combining data for all the groups. For the latter, the least squares linear regression equation was: threshold = 1.2 db X r , r 2 = 0.26, P< 0.001). DSCUSSON We found very high variability with conventional automated perimetry using Goldmann stimulus sizes and in areas of moderate visual field damage. A modest decrease in variability with larger stimulus sizes has been reported in control subjects. 22 ' 25 This is the first report of reduction of variability using large stimulus sizes in patients with visual loss. We found a three-fold reduction in the estimated standard deviations of the frequency-of-seeing function with the size stimulus. This reduction was accompanied by a tighter fit (higher r 2 ) to the psychometric differential light sensitivity function. This reduction in variability of the frequency-ofseeing curves implies that more reproducible values can be achieved in conventional automated perimetry

7 42 nvestigative Ophthalmology & isual Science, February 17, ol. 8, No. 2 Size J S 40% 20%-- Size.( f Size» 60%-- S 40%-- Size H h Size Size «J 60% - - a 20% - i - «80% Size B 60% -- 40% - 20% -- H Size Size U 1 20% 4-0% -r Size Size in Size FGURE. Graph of all subjects'best fits of the frequency-of-seeing data for each stimulus size for (A) control subjects' central test location, (B) control subjects' peripheral test location, (C) normal sensitivity test location of patients with glaucoma and (D) abnormal sensitivity test location of patients with glaucoma. The mean of the all 10 subjects' frequency-of-seeing data best fits is represented by the bold dashed curve. Note low estimated standard deviation (an index of the maximum slope) of the frequencyof-seeing curves in the control subjects and the high standard deviation in most subjects with glaucoma. Also, note the low standard deviation in the subjects with glaucoma with the size stimulus.

8 ariability in Glaucoma Using Size Stimuli 4 TABLE. Estimated Standard Deviations of the Calculated Frequency-of-Seeing Responses (in db) for the Different Groups by Size Mean Median SD Glaucoma Normal,,, 21,- -21, = abnormal; = test location of normal sensitivity for patient with glaucoma; SD = standard deviation in areas of moderate visual field damage with the size stimulus. n our study, the transition zone from seeing to nonseeing for the mean frequency-of-seeing curves at the abnormal test location was 0 db for the size 1 stimulus, 40 db for the size stimulus, and 12 db for the size stimulus (Table 2). The apparent 10 db reduction for the size stimulus comes at the 0 db end of the function and is likely due to shift of the frequency-of-seeing curves to the left, with a resultant small dynamic range of the size stimulus at the bright end of the range. The size stimulus is associated with a higher frequency-of-seeing curve estimated standard deviation at the normal sensitivity test location of subjects with glaucoma and in the control subjects' peripheral test location. The lower variability in areas of visual field damage using the size stimulus may be due to a reduction in the signal-to-noise ratio. With visual field damage, retinal ganglion cell receptive fields are sparser, leading to undersampling of the stimulus. Perception in the area of damage requires that either a brighter small stimulus or a larger stimulus of the same brightness be used to be perceived. Because fewer receptive fields are covered by the small fixed-size luminance increment stimulus, stimulus movements due to head tilting or microfixation shifts are more likely to cause different receptive field coverage on retest with the smaller sized stimuli. Swanson and Felius 26 have modeled the stimulus-response function of conventional automated perimetry based on Croner and Kaplan's electrophysiologic data 27 on macaque receptive fields and Dacey's anatomic data. 28 Assuming that the response is mediated primarily by the midget ganglion cells, they calculated that a size stimulus covers one midget cell receptive field, a size stimulus covers about 7 receptive fields, and a size stimulus about 100 receptive fields. This model would predict that spatially irregular loss of ganglion cell receptive fields would increase the signal-to-noise ratio for small stimuli and have much less effect on larger stimuli. A reduction in the signal-to-noise ratio has also been used to explain the results from two studies with high-pass resolution perimetry 1 ' 21 that suggest that obtaining thresholds by changing stimulus size has lower variability than results using a fixed-size (Goldmann size ) luminance increment stimulus. Using highpass resolution perimetry, stimuli much larger than the size stimulus are usually needed to obtain a threshold in areas of visual field damage. Using Swanson and Felius' 26 model and assuming an irregular receptive field array, the likelihood of similar receptive field coverage on retest is greater with larger stimuli. Although the size stimulus appears to be superior when considering variability, this comes at the cost of loss of spatial resolution. 2 That is, small scotomas may be missed using larger stimulus sizes. Therefore, the size stimulus is not optimal for routine use in patients with normal visual field examinations or fields with mild damage. Once the damage is moderate or severe, however, and it is important to be confident about visual field change, the size stimulus appears to be preferable. n addition to lower variability, it increases the dynamic range, thereby giving more opportunity to document progression. The size TABLE 4. Goodness of Fit for the Frequency-of-Seeing Functions of the Different Groups by Size Glaucoma Normal,,, 21,- -21,- H -21,- Mean Median SD = abnormal; = test location of normal sensitivity for patient with glaucoma; SD = standard deviation.

9 44 nvestigative Ophthalmology & isual Science, February 17, ol. 8, No. 2 stimulus is currently used in patients with very advanced damage; our results suggest that this stimulus may be useful before this "end stage" of glaucoma. Spatial summation is the decrease in threshold related to an increase in stimulus size. t likely plays a role in the capacity of size stimuli to have a low variability in areas of visual damage. Spatial summation gradually increases with increasing distance from the fovea in control subjects. 01 n areas of glaucomatous visual field damage, Fellman and colleagues 2 found thresholds decreased much more than expected using a size stimulus than with a size stimulus. They concluded that spatial summation and recruitment of neighboring areas of nearly normal sensitivity account for most of this improvement in threshold. We believe a similar explanation holds for our improvement in thresholds using the size stimulus in areas of moderate glaucomatous visual field damage. The average age of our patients with glaucoma was approximately years younger than that of the control subjects. Although there was no significant difference in the ages, this does not eliminate age difference as an effect in the control subjects. This difference would not affect our main finding: within the various groups, variability decreases with increasing stimulus size. n addition, some error may have been introduced by retesting subjects at a separate sitting because of failure to reach 0% or seen; however, because the r 2 values were all above 0.66 and similar to the r 2 values of those not needing retesting, this error is not likely to be great. Although variability is less using the size stimulus, it is unclear whether using this stimulus size will allow earlier detection of change in areas of moderate visual field damage. t is possible that this stimulus may be more resistant to change from visual loss for the same reason (undersampling) it gives a reduction in variability. A longitudinal study of a progressive optic neuropathy like glaucoma that compares results from size with size stimuli will be needed to answer this question. Our results confirm those of others that test result variability is high in patients with glaucoma with moderate visual field loss. 8 ' ' 121 ' 4 We found a reduction in this variability using a size stimulus. n addition, our results showing an increase in variability in normal areas of the visual field in patients with glaucoma are similar to those of other reports. 5 ' ' 5 Although use of a size stimulus comes with a reduction in variability, this is at the expense of loss of resolution. The best compromise may be strategies that obtain thresholds by changing stimulus size rather than stimulus intensity. Key Words glaucoma, perimetric variability, perimetry, visual field, visual testing References 1. Werner EB. Errors in the diagnosis of visual field progression in normal tension glaucoma. Ophthalmology. 14; 101: Heijl A, Lindgren G, Olsson J. The effect of perimetric experience in normal subjects. Arch Ophthalmol. 18; 107: Parrish RK, Schiffman J, Anderson DR. Static and kinetic visual field testing. Reproducibility in normal volunteers. Arch Ophthalmol. 184; 102: Heijl A, Lindgren G, Olsson J. Normal variability of static perimetric threshold values across the central visual field. Arch Ophthalmol. 187; 105: Chauhan BC, Tompkins JD, LeBlanc RP, McCormick TA. Characteristics of frequency-of-seeing curves in normal subjects, patients with suspected glaucoma, and patients with glaucoma. nvest Ophthalmol is Sci. 1; 4: Weber J, Rau S. The properties of perimetric thresholds in normal and glaucomatous eyes. German J Ophthalmol. 12; 1: Wall M, Maw RJ, Stanek KE, Chauhan BC. The psychometric function and reaction times of automated perimetry in normal and abnormal areas of the visual field in glaucoma patients. nvest Ophthalmol is Sci. 16; 7: Heijl A, Lindgren A, Lindgren G. Test-retest variability in glaucomatous visual fields. Am J Ophthalmol. 18;108:10.. Boeglin RJ, Caprioli J, Zulauf M. Long-term fluctuation of the visual field in glaucoma. AmJ Ophthalmol. 12; 11: Flammer J, Drance SM, Fankhauser F, Augustiny L. Differential light threshold in automated static perimetry. Factors influencing short-term fluctuation. Arch Ophthalmol. 184; 102: Werner EB, Saheb N, Thomas D. ariability of static visual threshold responses in patients with elevated OPs. Arch Ophthalmol. 182; 100: Werner EB, Petrig B, Krupin T, Bishop K. ariability of automated visual fields in clinically stable glaucoma patients. nvest Ophthalmol is Sci. 18; 0: Flammer J. Fluctuations in the isual Field. n: Drance SM, Anderson DR, eds. Automatic Perimetiy in Glaucoma. A Practical Guide. Orlando: Grune 8c Stratton, 185: Flammer J, Drance SM, Zulauf M. Differential light threshold. Short- and long-term fluctuation in patients with glaucoma, normal controls, and patients with suspected glaucoma. Arch Ophthalmol. 184; 102: Flanagan JG, WildJM, Trope GE. Evaluation of fastpac, a new strategy for threshold estimation with the humphrey field analyzer, in a glaucomatous population. Ophthalmology. 1; 100: Schaumberger M, Schafer B, Lachenmayr BJ. Glaucomatous visual fields. FASTPAC versus full threshold strategy of the Humphrey Field Analyzer. nvest Ophthalmol is Set. 15;6: Mills RP, Barnebey HS, Migliazzo C, Li Y. Does saving time using FASTPAC or suprathreshold testing reduce

10 ariability in Glaucoma Using Size Stimuli 45 quality of visual fields? Ophthalmology. 14; 101: Johnson CA, Chauhan BC, Shapiro LR. Properties of staircase procedures for estimating thresholds in automated perimetry. nvest Ophthalmol is Sci. 12; : Chauhan BC, House PH. ntratest variability in conventional and high-pass resolution perimetry. Ophthalmology. 11;8: Wall M, Lefante J, Conway M. ariability of high-pass resolution perimetry in normals and patients with idiopathic intracranial hypertension. nvest Ophthalmol is Sci. 11;2: House P, Schulzer M, Drance S, Douglas G. Characteristics of the normal central visual field measured with resolution perimetry. Graefes Arch ClinExp Ophthalmol. 11;22: Gilpin LB, Stewart WC, Hunt HH, Broom CD. Threshold variability using different Goldmann stimulus sizes. Acta Ophthalmol. 10;68: Heijl A, Lindgren G, Olsson J; Asman P. isual field interpretation with empiric probability maps. Arch Ophthalmol. 18; 107: Haefliger O, Flammer J. Fluctuation of the differential light threshold at the border of absolute scotomas. Comparison between glaucomatous visual field defects and blind spots. Ophthalmology. 11;8: Wall M, Kardon R, Moore P. Effects of stimulus size on test-retest variability. Perimetry Update 12/1, Proceedings of the Xth nternational Perimetric Society Meeting 1; Swanson WH, Felius J. Spatial summation for discrete ganglion cell mosaics. nvest Ophthalmol is Sci. (suppl) 16;7:S Croner LJ, Purpura K, Kaplan E. Response variability in retinal ganglion cells of primates. Proc Nat Acad Sci USA. 1;0: Dacey DM. Morphology of a small-field bistratified ganglion cell type in the macaque and human retina. w Neurosci. 1; 10: Gramer E, Kontic D, Krieglstein GK. Computer perimetry of glaucomatous visual field defects at different stimulus sizes. Ophthalmologica. 181; 18: Sloan LL. Area and luminance of test object as variables in examination of the visual field by projection perimetry. ision Res. 161; 1: HalletPE. Spatial Summation. ision Res. 16;: Fellman RL, Lynn JR, Starita RJ, Swanson WH. Clinical importance of spatial summation in glaucoma. Perimetry Update 188/18. Proceedings of the lllth nternational Perimetric Society Meeting. 18; Dannheim F, Drance SM. Psychovisual disturbances in glaucoma. A study of temporal and spatial summation. Arch Ophthalmol. 174;1: Flammer J, Drance SM, Schulzer M. Covariates of the long-term fluctuation of the differential light threshold. Arch Ophthalmol. 184; 102: Werner EB, Drance SM. ncreased scatter of responses as a precursor of visual field changes in glaucoma. Can] Ophthalmol. 177; 12: