ACTA O P H T H A L M O L O G ICA
70 (1992)665-670
Association of reliability with reproducibility of the glaucomatous visual field Tod A. McMillan’, William C. Stewart’ and Hurshell H. Hunt’ Glaucoma Service at Department of Ophthalmology’ and the Department of Biostatistics, Epidemiology and System Sciences*at the Medical University of South Carolina, Charleston, South Carolina, USA
Abstract. We studied 207 glaucoma patients (207 eyes) who had two successive visual tests on the Humphrey Field Analyzer to determine the association of unreliability (> 10%incidence of a missed catch trial test) with threshold variation. We found that in patients where one visual field showed > 30%false negative or positive errors, and the other visual field also was unreliable, a significant increase in the variance of the mean defect existed from reliable patients (p < 0.05). Of patients who had only one of two visual fields which was unreliable (from false negative errors), only those with marked glaucomatous visual field loss (< -15 dB depth of defect) showed a greater variance in mean defect between examinations over reliable patients (p < 0.05). In total, 41 patients (19.8%)had unreliable visual fields associated with statistically increased threshold variation. This study suggests that patient reliability should be considered when interpreting changes in threshold between automated visual field examinations. Key words: false negative errors - false positive errors - fixation loss - catch trial tests - threshold fluctuation - glaucoma.
A variation in repeat threshold measurements exists when testing glaucoma patients with automated static perimetry. This variation is important because it decreases the accuracy of individual measurements for threshold. Furthermore, unreliable patient responses to static threshold testing potentially may increase retest variability. Greater understanding of this effect would help the clini-
cian better interpret changes between successive automated visual field examinations We studied the visual fields of glaucoma and ocular hypertensive patients tested on the Humphrey Field Analyzer to determine the association of reliability with threshold ’reproducibility in automated perimetry.
Materialsand Methods We included in this study patients with ocular hypertension (increased intraocular pressure without optic nerve or visual field damage) or glaucoma (glaucomatous optic nerve damage with increased intraocular pressure with or without visual field loss) who had performed at least two automated visual fields within one year between 1988 and 1991. Visual fields were acquired consecutively in alphabetical order from patient files at the Medical University of South Carolina. We excluded patients who showed progressive glaucomatous optic disc changes or who had an intraocular pressure > 21 mmHg between glaucomatous visual field examinations. Only one randomly selected eye from each patient was evaluated. Ifa patient had completed multiple visual field examinations the last two were evaluated in this study to limit any influence of a learning effect. The patient’s best corrected refraction for a 330 mil665
limeter (mm) distance was used for each test. Each patient was tested on the Humphrey Field Analyzer, which is a single unit projection perimeter with a 330 rmn bowl. Each patient was tested with the program 30-2 which examines 76 locations within the central 30" with a static full-threshold strategy. Locations are spaced 6" apart and offset symmetrically across the vertical and horizontal midlines. The Humphrey uses three catch trial tests to es-
timate reliability. A fixation loss is recorded when a projected stimulus into the expected area of the patient's blind spot is perceived. A false positive is determined if a patient responds to an imagined stimulus when none is presented. A false negative is recorded when a patient fails to perceive a more intense stimulus than seen earlier at the same location. Reliability was evaluated for individual visual fields by determining for each catch trial the per-
Table 1 Association of increasing unreliability and the standard deviation of the differences in the mean defect and pattern standard deviation between tests.
Mean defect Percent Unreliability
I
I
Fixation Loss
False Positives
I
False Negatives
One unreliable test: 10 - 20 20 - 30 >30
2.73, n = 12 3.37, n = 11 2.49, n = 8
1.37, n = 4 O+ O+
1.35, n=13 1.80, n = 8 3.51, n = 6
2.24, n = 10 1.79, n = l l 3.95, n = 50
O+ O+ 8.13*, n = 5
3.28, n = 6 1.95, n = l l 6.15*, n = 27
Two unreliable tests: 10 - 20 20 - 30 >30 Reliable fields
=
2.63, n = 25 Pattern standard deviation ~
Percent Unreliability
Fixation Loss
False Positives
False Negatives
0.94, n = 12 1.70, n = l l 1.87, n = 8
0.25, n = 4 O+ O+
2.71, n = 13 1.11,n= 8 1.83, n = 6
1.83, n = 10 1.18, n = 11 2.15, n = 50
O+ O+ 4.25, n = 5
2.43, n = 6 2.19, n = 11 2.78, n = 27
One unreliable test: 10 - 20 20 - 30 >30 Two unreliable tests: 10 - 20 20 - 30 >30 Reliable fields
=
1.68, n = 25
Numbers in decibels +
Insufficient data to determine variance
* Signifies significant difference between visual field examinations compared to reliable patients (p < 0.05) 666
centage of positive responses to the number of times tested. Patients then were categorized as reliable (all three catch trials with less than 10%positive responses for both tests), unreliable on one exam (one or more catch trial tests with greater than 10%positive responses on one visual field examination, but the other examination was reliable), or unreliable on both examinations (a catch trial test with greater than 10%positive responses on both visual field examinations).Unreliable patients were grouped also by the individual catch trial (fixation losses, false positive or negative errors) with the greatest percentage of missed questions. Ten percent was used as a figure for unreliability because of previous evidence that statistically significant alterations may occur above this level compared to visual fields with fewer than 10% missed catch trial questions (Cascairo et al. 1991). As a measure for threshold several global indices were used as determined by Statpac. The mean defect is a weighted average deviation for the normal reference field and estimates the uniform part of the deviation (Heijl et al. 1987).The pattern standard deviation is a weighted standard deviation of the pointwise differences between the measured and normal reference fields and estimates the nonuniform part of the deviation (Heijl et al. 1987). We determined the association of unreliability upon threshold fluctuation by comparing the variances of the mean defect and pattern standard deviation from reliable to unreliable visual fields. Unreliable visual fields were categorized by the catch trial test (fixation losses, false positive and false negative errors) which had the greatest percentage of missed questions, then subcategorized by the percentage (10-20%,20-30%,and > 30%)of missed questions within each category. The variances of the mean defect and pattern standard deviation for each level of unreliability for each catch trial were compared to those of reliable patients by using the F-variance ratio test. We also evaluated the association of the severity of the glaucoma and the variance of the mean defect and pattern standard deviation. The severity of glaucoma was determined by the mean defect on the patient's initial visual field examination included in this study. This initial mean defect was categorized into levels of progressive severity (0 to -5, -5 to -10, -10 to -15 and < -15 decibels [dB]).For each level of severity of glaucoma the variances of
the mean defect and pattern standard deviation for unreliable patients were compared to those of reliable patients by using the F-variance ratio test. Because of the lack of independence and the large number of F-tests performed, the Bonferroni correction was used to determine an acceptable level of significance to assure a type I experimentwise error rate for this study of at least 0.05. To evaluate for a learning effect a paired t-test was used to assess the statistical significance between differences in the average mean defect and pattern standard deviation between tests. The association of unreliability upon the average mean defect and pattern standard deviation was assessed for each type of catch trial by using a paired t-test in patients who had one reliable and one unreliable visual field.
Results We included in this study 414 visual fields of 207 glaucoma patients (207 eyes). The average patient age was 62.6 k 10.4 years. One hundred and one right eyes and 106 left eyes were included. Table 1 shows the association of unreliability with threshold variance. A significant difference from the reliable group is noted for patients who had > 30%false positive or negative errors on one examination, and in whom the other examination also was unreliable (> 10% catch trial loss) (p < 0.05). In total, 41 patients (19.8%)had unreliable visual fields associated with statistically increased threshold variation. Table 2 shows the association of unreliability with threshold variance based on the severity of glaucoma. For patients with one unreliable test from false negative errors the variance of the mean defect was significantly greater than that of reliable patients only when the glaucoma was severe enough to cause a mean defect of < -15 dB (p < 0.05). False positive errors were associated with a significant improvement in the variance of the pattern standard deviation 'when little if any glaucomatous damage existed on the visual field (mean defect > -5 dB; p < 0.05). Table 3 shows the effect of unreliability upon the average mean defect and pattern standard deviation. No significant change was observed for unreliable visual fields compared to reliable visual fields except that the pattern standard deviation 667
was increased in patients who were unreliable from false negative errors (p < 0.05). The difference between visual fields for patients in whom both tests were reliable in the mean defect was 0.72 k0.53 dB (p < 0.05) and in the pattern standard deviation - 0.43 k 0.33 dB (p < 0.05). No significant difference existed between tests which helped exclude a learning effect.
Discussion In automated static perimetry a retest variability exists which creates uncertainty when interpreting measured threshold levels. Isolating factors associated with increased or decreased retest variability would help a physician better interpret successive automated static visual field examinations. Retest variability in normal subjects has been
Table 2 Association of the severity of glaucoma and the standard deviation of the differences in the mean defect and pattern standard deviation between tests.
Mean defect Seventy of glaucoma (mean defect)
Reliable fields
Fixation loss
False positives
2.07, n = 7 3.10, n = 9 3.71, n = 4 0.87, n = 5
1.89, n = 19 3.00, n = 4 2.82, n = 7 O+
1.22, n = 5 O+ O+ O+
1.92 , n = 1.77 , n = 2.99 ,n = 7.47*, n =
2.07, n = 7 3.10, n = 9 3.71, n = 4 0.87, n = 5
3.00, n = 32 3.00, n = 22 5.67, n = 7 3.80, n = 10
7.58, n = 4 8.07, n = 2 O+ Of
4.32, n = 8 1.74, n = 10 6.75, n = 10 4.24, n = 16
False negatives
False negatives
One unreliable test Oto- 5 5 to -10 -10 to -15 < -15
-
5 4 8 9
Two unreliable tests oto- 5 - 5 to -10 -10 to -15 < -15 +
Insufficient data to provide variance
* Significant difference from reliable group,p < 0.05 Pattern standard deviation Severity of glaucoma (mean defect)
Reliable fields
Fixation loss
False positives
2.18, n = 7 1.49, n = 9 0.65, n = 4 1.55, n = 5
1.24, n = 19 2.01, n = 4 1.78,n= 7 O+
0.22*, n = 5 O+ O+
2.80, n = 0.96, n = 1.26, n = 1.89, n =
2.18, n = 7 1.49, n = 9 0.65, n = 4 1.55, n = 5
1.95, n = 32 2.25, n = 22 2.00, n = 7 2.05, n = 10
3.42, n = 4 5.47, n = 2 O+ O+
3.12,n= 8 2.61, n = 10 1.81, n = 10 2.50, n = 16
One unreliable test 0 to- 5 5 to -10 -10 to -15 < -15
-
Of
5 4 8 9
Two unreliable tests Oto- 5 5 to -10 -10 to -15 < -15
-
Numbers in decibels +
Insufficient data to provide variance
* Significant difference from reliable group, p < 0.05 668
Table 3 Change in average mean defect and pattern standard deviation associated with unreliable visual fields compared to reliable visual fields.
Fixation loss False positives False negatives
Mean defect
Pattern standard deviation
0.26 f 3.0 1.25 f 1.7 -0.54 f 5.6
0.10 f 1.6 -0.39 f 0.6 0.95 f 2.4*
Numbers in decibels * Signifies significant difference between visual field examinations (p < 0.05)
shown to increase with greater distance from fixation (Heijl et al. 1989). Also, retest variability is greater when testing with a small stimulus size (Goldmann size I and 11) (Gilpin et al. 1990) or a mesopic background intensity (Crosswell et al 1991) instead of stimulus size III or a low photopic background intensity. In glaucoma patients retest variability is greater than in normal individuals (Flammer, Drance & Zulauf 1984) and is larger within a scotoma than in the remaining normal areas of the field (Flammer, Drance & Schulzer 1984). In the present study we examined the association of patient reliability upon threshold reproducibility. The Humphrey Field Analyzer estimates patient reliability by using three different catch trial tests (fixationlosses, as well as false positive and false negative errors). Previous reports studying reliability have shown in healthy subjects that global indices, as measured by the Humphrey Statpac, become significantlyaltered from control fields at a prevalence of 20% false negative errors and 33%false positive errors as well as fixation losses (Cascairoet al. 1991).However, the number of abnormal locations and their distribution become significantlyaltered from the control visual fields at 10%incidence of catch trial losses (Cascairo et al. 1991).Unreliable glaucoma patients have been shown to have decreased threshold levels from false negative errors, and improved threshold levels from false positive errors and fixation losses as compared to reliable patients (Katz & Sommer 1990, Reynolds et al. 1990).Consequently, unreliable patients potentially may falsely alter their measured threshold levels and increase inter-test variation.
This paper showed that when one of two visual fields was unreliable (> 10%incidence of a missed catch trial) no significant differences in variation were noted from reliable patients. However, when both tests were unreliable because of false positive or negative errors, a significant difference in variation of the mean defect from reliable patients was observed. This increase in variation was most strongly associated with patients who incurred > 30%false positive or negative errors or whose glaucoma was severe enough to cause a mean defect of < -15 dB and who were unreliable because of false negative errors. Unexpectedly, in patients with little or no glaucomatous visual field loss who were unreliable from false positive errors, signigicantly less variance in the pattern standard deviation was noted than in reliable patients. The reason for an improvement in variance from false positive errors was not clear from our data. Although not statistically significant after the Bonferroni correction, patients who were unreliable from fixation loss on both examinations showed a trend to increased variance in the mean defect. This was especially evident with > 30%incidence of fixation loss or when a patient’s glaucoma was severe enough to have a mean defect < -10 dB. Also, our data tended to support past studies which indicate that false positivc errors and furation losses increase while false negative errors decrease ,thresholdlevels in glaucoma patients (Katz & Sommer 1990, Reynolds et al. 1990). Knowledge of patient unreliability upon threshold variation may have application in visual field interpretation. Software currently is available on the Humphrey Field Analyzer (San Leandro, California) which analyzes the significance of change in threshold values between two visual fields. However, this software does not account for differences in variation among reliable and unreliable patients (Heijl et al. 1989). Since this paper indicates that unreliability increases threshold variation, a measured change in threshold may more likely indicate progressive disease in a reliable patient than in an unreliable patient. Future software development possibly could isolate groups of patients with less variation not only for reliability, but also for other factors which influence the visual field such as a media opacity, refraction, and learning effect. This study suggests when evaluating the change 669
in threshold between two successive visual field tests, greater caution should be exercised in diagnosing visual field progression in unreliable than reliable glaucoma patients. This study did not evaluate the effect of increasing eccentricity hom fixation on threshold variation and reliability. Also, the influence of unreliability upon the number of abnormal locations in the visual fields was not assessed. Further evaluation may clarify the role of assessing threshold variation differently among reliable and unreliable patients.
Acknowledgments Sponsored in part by an unrestricted grant from Research to Prevent Blindness, Inc. The authors have no commercial or proprietary interest in Allergan-Humphrey, Inc. or the Humphrey Field Analyzer.
Flammer J, Drance S M & Schulzer M (1984): Covariates of the long-term fluctuation of the differential light threshold. Arch Ophthalmol 102: 880-882. Flammer J, Drance S M & Zulauf M (1984): Differential light threshold. Short- and long-term fluctuation in patients with glaucoma, normal controls, and patients with suspected glaucoma. Arch Ophthalmol 102: 704706. Gilpin L B, Stewart W C, Hunt H H &Broom C D (1990): Threshold variability using different Goldmann stimulus sizes. Acta Ophthalmol (Copenh)68: 674-676. Heijl A, Lindgren A & Lindgren G (1989).Test-retest variability in glaucomatous visual fields. Am J Ophthalmol 108: 130-135. Heijl A, Lindgren G & OlssonJ (1987):111.2 A package for the statistical analysis of visual fields. In: Greve E L & Heijl A (eds).Seventh International Visual Field Symposium, pp. 153-168. Martinus Nijhoff/ Dr. W Junk Publishers, Amsterdam. Katz J & Sommer A (1990):Screening for glaucomatous visual field loss. The effect of patient reliability. Ophthalmology 97: 1032-1037. Reynolds M, Stewart W C & Sutherland S (1990):Factors which influence the prevalence of positive catch trials in glaucoma patients. Graefes Arch Clin Exp Ophthalmol 228 338-341.
References Received on April 2nd, 1992. Cascairo M A, Stewart W C & Sutherland S E (1991): Influence of missed catch trials on the visual field in normal subjects. Graefes Arch Clin Exp Ophthalmol229: 437-441 Crosswell H H, Stewart W C, Cascairo M A & Hunt H H (1991): The effect of background intensity upon the components of fluctuation as determined by thres hold-related automated perimetry. Graefes Arch Clin Exp Ophthalmol229: 119-122
670
Author’s address:
Dr. William C. Stewart, Medical University of South Carolina, Storm Eye Institute, 171 Ashley Avenue, Charleston, South Carolina 29425-2236, USA.
70 (1992)665-670
Association of reliability with reproducibility of the glaucomatous visual field Tod A. McMillan’, William C. Stewart’ and Hurshell H. Hunt’ Glaucoma Service at Department of Ophthalmology’ and the Department of Biostatistics, Epidemiology and System Sciences*at the Medical University of South Carolina, Charleston, South Carolina, USA
Abstract. We studied 207 glaucoma patients (207 eyes) who had two successive visual tests on the Humphrey Field Analyzer to determine the association of unreliability (> 10%incidence of a missed catch trial test) with threshold variation. We found that in patients where one visual field showed > 30%false negative or positive errors, and the other visual field also was unreliable, a significant increase in the variance of the mean defect existed from reliable patients (p < 0.05). Of patients who had only one of two visual fields which was unreliable (from false negative errors), only those with marked glaucomatous visual field loss (< -15 dB depth of defect) showed a greater variance in mean defect between examinations over reliable patients (p < 0.05). In total, 41 patients (19.8%)had unreliable visual fields associated with statistically increased threshold variation. This study suggests that patient reliability should be considered when interpreting changes in threshold between automated visual field examinations. Key words: false negative errors - false positive errors - fixation loss - catch trial tests - threshold fluctuation - glaucoma.
A variation in repeat threshold measurements exists when testing glaucoma patients with automated static perimetry. This variation is important because it decreases the accuracy of individual measurements for threshold. Furthermore, unreliable patient responses to static threshold testing potentially may increase retest variability. Greater understanding of this effect would help the clini-
cian better interpret changes between successive automated visual field examinations We studied the visual fields of glaucoma and ocular hypertensive patients tested on the Humphrey Field Analyzer to determine the association of reliability with threshold ’reproducibility in automated perimetry.
Materialsand Methods We included in this study patients with ocular hypertension (increased intraocular pressure without optic nerve or visual field damage) or glaucoma (glaucomatous optic nerve damage with increased intraocular pressure with or without visual field loss) who had performed at least two automated visual fields within one year between 1988 and 1991. Visual fields were acquired consecutively in alphabetical order from patient files at the Medical University of South Carolina. We excluded patients who showed progressive glaucomatous optic disc changes or who had an intraocular pressure > 21 mmHg between glaucomatous visual field examinations. Only one randomly selected eye from each patient was evaluated. Ifa patient had completed multiple visual field examinations the last two were evaluated in this study to limit any influence of a learning effect. The patient’s best corrected refraction for a 330 mil665
limeter (mm) distance was used for each test. Each patient was tested on the Humphrey Field Analyzer, which is a single unit projection perimeter with a 330 rmn bowl. Each patient was tested with the program 30-2 which examines 76 locations within the central 30" with a static full-threshold strategy. Locations are spaced 6" apart and offset symmetrically across the vertical and horizontal midlines. The Humphrey uses three catch trial tests to es-
timate reliability. A fixation loss is recorded when a projected stimulus into the expected area of the patient's blind spot is perceived. A false positive is determined if a patient responds to an imagined stimulus when none is presented. A false negative is recorded when a patient fails to perceive a more intense stimulus than seen earlier at the same location. Reliability was evaluated for individual visual fields by determining for each catch trial the per-
Table 1 Association of increasing unreliability and the standard deviation of the differences in the mean defect and pattern standard deviation between tests.
Mean defect Percent Unreliability
I
I
Fixation Loss
False Positives
I
False Negatives
One unreliable test: 10 - 20 20 - 30 >30
2.73, n = 12 3.37, n = 11 2.49, n = 8
1.37, n = 4 O+ O+
1.35, n=13 1.80, n = 8 3.51, n = 6
2.24, n = 10 1.79, n = l l 3.95, n = 50
O+ O+ 8.13*, n = 5
3.28, n = 6 1.95, n = l l 6.15*, n = 27
Two unreliable tests: 10 - 20 20 - 30 >30 Reliable fields
=
2.63, n = 25 Pattern standard deviation ~
Percent Unreliability
Fixation Loss
False Positives
False Negatives
0.94, n = 12 1.70, n = l l 1.87, n = 8
0.25, n = 4 O+ O+
2.71, n = 13 1.11,n= 8 1.83, n = 6
1.83, n = 10 1.18, n = 11 2.15, n = 50
O+ O+ 4.25, n = 5
2.43, n = 6 2.19, n = 11 2.78, n = 27
One unreliable test: 10 - 20 20 - 30 >30 Two unreliable tests: 10 - 20 20 - 30 >30 Reliable fields
=
1.68, n = 25
Numbers in decibels +
Insufficient data to determine variance
* Signifies significant difference between visual field examinations compared to reliable patients (p < 0.05) 666
centage of positive responses to the number of times tested. Patients then were categorized as reliable (all three catch trials with less than 10%positive responses for both tests), unreliable on one exam (one or more catch trial tests with greater than 10%positive responses on one visual field examination, but the other examination was reliable), or unreliable on both examinations (a catch trial test with greater than 10%positive responses on both visual field examinations).Unreliable patients were grouped also by the individual catch trial (fixation losses, false positive or negative errors) with the greatest percentage of missed questions. Ten percent was used as a figure for unreliability because of previous evidence that statistically significant alterations may occur above this level compared to visual fields with fewer than 10% missed catch trial questions (Cascairo et al. 1991). As a measure for threshold several global indices were used as determined by Statpac. The mean defect is a weighted average deviation for the normal reference field and estimates the uniform part of the deviation (Heijl et al. 1987).The pattern standard deviation is a weighted standard deviation of the pointwise differences between the measured and normal reference fields and estimates the nonuniform part of the deviation (Heijl et al. 1987). We determined the association of unreliability upon threshold fluctuation by comparing the variances of the mean defect and pattern standard deviation from reliable to unreliable visual fields. Unreliable visual fields were categorized by the catch trial test (fixation losses, false positive and false negative errors) which had the greatest percentage of missed questions, then subcategorized by the percentage (10-20%,20-30%,and > 30%)of missed questions within each category. The variances of the mean defect and pattern standard deviation for each level of unreliability for each catch trial were compared to those of reliable patients by using the F-variance ratio test. We also evaluated the association of the severity of the glaucoma and the variance of the mean defect and pattern standard deviation. The severity of glaucoma was determined by the mean defect on the patient's initial visual field examination included in this study. This initial mean defect was categorized into levels of progressive severity (0 to -5, -5 to -10, -10 to -15 and < -15 decibels [dB]).For each level of severity of glaucoma the variances of
the mean defect and pattern standard deviation for unreliable patients were compared to those of reliable patients by using the F-variance ratio test. Because of the lack of independence and the large number of F-tests performed, the Bonferroni correction was used to determine an acceptable level of significance to assure a type I experimentwise error rate for this study of at least 0.05. To evaluate for a learning effect a paired t-test was used to assess the statistical significance between differences in the average mean defect and pattern standard deviation between tests. The association of unreliability upon the average mean defect and pattern standard deviation was assessed for each type of catch trial by using a paired t-test in patients who had one reliable and one unreliable visual field.
Results We included in this study 414 visual fields of 207 glaucoma patients (207 eyes). The average patient age was 62.6 k 10.4 years. One hundred and one right eyes and 106 left eyes were included. Table 1 shows the association of unreliability with threshold variance. A significant difference from the reliable group is noted for patients who had > 30%false positive or negative errors on one examination, and in whom the other examination also was unreliable (> 10% catch trial loss) (p < 0.05). In total, 41 patients (19.8%)had unreliable visual fields associated with statistically increased threshold variation. Table 2 shows the association of unreliability with threshold variance based on the severity of glaucoma. For patients with one unreliable test from false negative errors the variance of the mean defect was significantly greater than that of reliable patients only when the glaucoma was severe enough to cause a mean defect of < -15 dB (p < 0.05). False positive errors were associated with a significant improvement in the variance of the pattern standard deviation 'when little if any glaucomatous damage existed on the visual field (mean defect > -5 dB; p < 0.05). Table 3 shows the effect of unreliability upon the average mean defect and pattern standard deviation. No significant change was observed for unreliable visual fields compared to reliable visual fields except that the pattern standard deviation 667
was increased in patients who were unreliable from false negative errors (p < 0.05). The difference between visual fields for patients in whom both tests were reliable in the mean defect was 0.72 k0.53 dB (p < 0.05) and in the pattern standard deviation - 0.43 k 0.33 dB (p < 0.05). No significant difference existed between tests which helped exclude a learning effect.
Discussion In automated static perimetry a retest variability exists which creates uncertainty when interpreting measured threshold levels. Isolating factors associated with increased or decreased retest variability would help a physician better interpret successive automated static visual field examinations. Retest variability in normal subjects has been
Table 2 Association of the severity of glaucoma and the standard deviation of the differences in the mean defect and pattern standard deviation between tests.
Mean defect Seventy of glaucoma (mean defect)
Reliable fields
Fixation loss
False positives
2.07, n = 7 3.10, n = 9 3.71, n = 4 0.87, n = 5
1.89, n = 19 3.00, n = 4 2.82, n = 7 O+
1.22, n = 5 O+ O+ O+
1.92 , n = 1.77 , n = 2.99 ,n = 7.47*, n =
2.07, n = 7 3.10, n = 9 3.71, n = 4 0.87, n = 5
3.00, n = 32 3.00, n = 22 5.67, n = 7 3.80, n = 10
7.58, n = 4 8.07, n = 2 O+ Of
4.32, n = 8 1.74, n = 10 6.75, n = 10 4.24, n = 16
False negatives
False negatives
One unreliable test Oto- 5 5 to -10 -10 to -15 < -15
-
5 4 8 9
Two unreliable tests oto- 5 - 5 to -10 -10 to -15 < -15 +
Insufficient data to provide variance
* Significant difference from reliable group,p < 0.05 Pattern standard deviation Severity of glaucoma (mean defect)
Reliable fields
Fixation loss
False positives
2.18, n = 7 1.49, n = 9 0.65, n = 4 1.55, n = 5
1.24, n = 19 2.01, n = 4 1.78,n= 7 O+
0.22*, n = 5 O+ O+
2.80, n = 0.96, n = 1.26, n = 1.89, n =
2.18, n = 7 1.49, n = 9 0.65, n = 4 1.55, n = 5
1.95, n = 32 2.25, n = 22 2.00, n = 7 2.05, n = 10
3.42, n = 4 5.47, n = 2 O+ O+
3.12,n= 8 2.61, n = 10 1.81, n = 10 2.50, n = 16
One unreliable test 0 to- 5 5 to -10 -10 to -15 < -15
-
Of
5 4 8 9
Two unreliable tests Oto- 5 5 to -10 -10 to -15 < -15
-
Numbers in decibels +
Insufficient data to provide variance
* Significant difference from reliable group, p < 0.05 668
Table 3 Change in average mean defect and pattern standard deviation associated with unreliable visual fields compared to reliable visual fields.
Fixation loss False positives False negatives
Mean defect
Pattern standard deviation
0.26 f 3.0 1.25 f 1.7 -0.54 f 5.6
0.10 f 1.6 -0.39 f 0.6 0.95 f 2.4*
Numbers in decibels * Signifies significant difference between visual field examinations (p < 0.05)
shown to increase with greater distance from fixation (Heijl et al. 1989). Also, retest variability is greater when testing with a small stimulus size (Goldmann size I and 11) (Gilpin et al. 1990) or a mesopic background intensity (Crosswell et al 1991) instead of stimulus size III or a low photopic background intensity. In glaucoma patients retest variability is greater than in normal individuals (Flammer, Drance & Zulauf 1984) and is larger within a scotoma than in the remaining normal areas of the field (Flammer, Drance & Schulzer 1984). In the present study we examined the association of patient reliability upon threshold reproducibility. The Humphrey Field Analyzer estimates patient reliability by using three different catch trial tests (fixationlosses, as well as false positive and false negative errors). Previous reports studying reliability have shown in healthy subjects that global indices, as measured by the Humphrey Statpac, become significantlyaltered from control fields at a prevalence of 20% false negative errors and 33%false positive errors as well as fixation losses (Cascairoet al. 1991).However, the number of abnormal locations and their distribution become significantlyaltered from the control visual fields at 10%incidence of catch trial losses (Cascairo et al. 1991).Unreliable glaucoma patients have been shown to have decreased threshold levels from false negative errors, and improved threshold levels from false positive errors and fixation losses as compared to reliable patients (Katz & Sommer 1990, Reynolds et al. 1990).Consequently, unreliable patients potentially may falsely alter their measured threshold levels and increase inter-test variation.
This paper showed that when one of two visual fields was unreliable (> 10%incidence of a missed catch trial) no significant differences in variation were noted from reliable patients. However, when both tests were unreliable because of false positive or negative errors, a significant difference in variation of the mean defect from reliable patients was observed. This increase in variation was most strongly associated with patients who incurred > 30%false positive or negative errors or whose glaucoma was severe enough to cause a mean defect of < -15 dB and who were unreliable because of false negative errors. Unexpectedly, in patients with little or no glaucomatous visual field loss who were unreliable from false positive errors, signigicantly less variance in the pattern standard deviation was noted than in reliable patients. The reason for an improvement in variance from false positive errors was not clear from our data. Although not statistically significant after the Bonferroni correction, patients who were unreliable from fixation loss on both examinations showed a trend to increased variance in the mean defect. This was especially evident with > 30%incidence of fixation loss or when a patient’s glaucoma was severe enough to have a mean defect < -10 dB. Also, our data tended to support past studies which indicate that false positivc errors and furation losses increase while false negative errors decrease ,thresholdlevels in glaucoma patients (Katz & Sommer 1990, Reynolds et al. 1990). Knowledge of patient unreliability upon threshold variation may have application in visual field interpretation. Software currently is available on the Humphrey Field Analyzer (San Leandro, California) which analyzes the significance of change in threshold values between two visual fields. However, this software does not account for differences in variation among reliable and unreliable patients (Heijl et al. 1989). Since this paper indicates that unreliability increases threshold variation, a measured change in threshold may more likely indicate progressive disease in a reliable patient than in an unreliable patient. Future software development possibly could isolate groups of patients with less variation not only for reliability, but also for other factors which influence the visual field such as a media opacity, refraction, and learning effect. This study suggests when evaluating the change 669
in threshold between two successive visual field tests, greater caution should be exercised in diagnosing visual field progression in unreliable than reliable glaucoma patients. This study did not evaluate the effect of increasing eccentricity hom fixation on threshold variation and reliability. Also, the influence of unreliability upon the number of abnormal locations in the visual fields was not assessed. Further evaluation may clarify the role of assessing threshold variation differently among reliable and unreliable patients.
Acknowledgments Sponsored in part by an unrestricted grant from Research to Prevent Blindness, Inc. The authors have no commercial or proprietary interest in Allergan-Humphrey, Inc. or the Humphrey Field Analyzer.
Flammer J, Drance S M & Schulzer M (1984): Covariates of the long-term fluctuation of the differential light threshold. Arch Ophthalmol 102: 880-882. Flammer J, Drance S M & Zulauf M (1984): Differential light threshold. Short- and long-term fluctuation in patients with glaucoma, normal controls, and patients with suspected glaucoma. Arch Ophthalmol 102: 704706. Gilpin L B, Stewart W C, Hunt H H &Broom C D (1990): Threshold variability using different Goldmann stimulus sizes. Acta Ophthalmol (Copenh)68: 674-676. Heijl A, Lindgren A & Lindgren G (1989).Test-retest variability in glaucomatous visual fields. Am J Ophthalmol 108: 130-135. Heijl A, Lindgren G & OlssonJ (1987):111.2 A package for the statistical analysis of visual fields. In: Greve E L & Heijl A (eds).Seventh International Visual Field Symposium, pp. 153-168. Martinus Nijhoff/ Dr. W Junk Publishers, Amsterdam. Katz J & Sommer A (1990):Screening for glaucomatous visual field loss. The effect of patient reliability. Ophthalmology 97: 1032-1037. Reynolds M, Stewart W C & Sutherland S (1990):Factors which influence the prevalence of positive catch trials in glaucoma patients. Graefes Arch Clin Exp Ophthalmol 228 338-341.
References Received on April 2nd, 1992. Cascairo M A, Stewart W C & Sutherland S E (1991): Influence of missed catch trials on the visual field in normal subjects. Graefes Arch Clin Exp Ophthalmol229: 437-441 Crosswell H H, Stewart W C, Cascairo M A & Hunt H H (1991): The effect of background intensity upon the components of fluctuation as determined by thres hold-related automated perimetry. Graefes Arch Clin Exp Ophthalmol229: 119-122
670
Author’s address:
Dr. William C. Stewart, Medical University of South Carolina, Storm Eye Institute, 171 Ashley Avenue, Charleston, South Carolina 29425-2236, USA.
