Download Response Variability in the Visual Field: Comparison of Optic

Response Variability in the Visual Field: Comparison of
Optic Neuritis, Glaucoma, Ocular Hypertension,
and Normal Eyes
David B. Henson,1 Shaila Chaudry,1 Paul H. Artes,1 E. Brian Faragher,2 and Alec Ansons1
PURPOSE. To compare the relationship between sensitivity and response variability in the visual field
of normal eyes and eyes with optic neuritis (ON), glaucoma (POAG), and ocular hypertension
(OHT).
METHODS. Frequency-of-seeing (FOS) data were collected from four visual field locations in one eye
of 71 subjects (12 ON, 25 POAG, 11 OHT, and 23 normal), using a constant stimulus method on
an Henson 4000 perimeter (Tinsley Instruments, Croydon, UK). At each location, at least 20 stimuli
(subtending 0.5°) were presented for 200 ms at six or more intensities above and below the
estimated threshold. The mean and SD of the probit fitted cumulative Normal function were used
to estimate sensitivity and response variability. Cluster regression analysis was carried out to
determine whether there were differences in the sensitivity-log (variability) relationship between
the four groups.
RESULTS. Variability was found to increase with decreased sensitivity for all four groups. The
combined data from the four groups was well represented (R2 ⫽ 0.57) by the function loge(SD) ⫽
A䡠sensitivity (dB) ⫹ B, where the constants A and B were ⫺0.081 (SE, ⫾0.005) and 3.27 (SE,
⫾0.15), respectively. Including other statistically significant covariates (false-negative errors, P ⫽
0.004) and factors (diagnosis, P ⫽ 0.005) into the model increased the proportion of explained
variance to 62% (R2 ⫽ 0.62). Stimulus eccentricity (P ⫽ 0.34), patient age (P ⫽ 0.33), fixation loss
rate (P ⫽ 0.10), and false-positive rate (P ⫽ 0.66) did not reach statistical significance as additional
predictors of response variability.
CONCLUSIONS. The relationship between response variability and sensitivity is similar for ON, POAG,
OHT, and normal eyes. These results provide supporting evidence for the hypothesis that response
variability is dependent on functional ganglion cell density. (Invest Ophthalmol Vis Sci. 2000;41:
417– 421)
O
ptic neuritis is a demyelinating condition which in its
acute form gives rise to a series of symptoms, including abrupt loss of visual acuity, reduced contrast sensitivity, dyschromatopsia, ocular pain, and visual field loss.
Although most visual functions show substantial recovery
within 6 months of the attack, some residual visual field loss is
common.1 There is also evidence of retinal nerve fiber loss in
the majority of patients.2 Increased variability has been established both for the global visual field indices derived from
automated static threshold perimetry3 and at individual test
locations using a frequency-of-seeing (FOS) technique.4
An increase in visual field variability also occurs in primary
open angle glaucoma (POAG).5–11 Using FOS techniques, apFrom the 1Department of Ophthalmology, University of Manchester; and 2Department of Organisational Health Psychology, University
of Manchester Institute of Science & Technology, Manchester, United
Kingdom.
Supported by The Guide Dogs for the Blind Association, The
Wellcome Trust, and The Manchester Royal Eye Hospital Endowment
Funds.
Submitted for publication May 21, 1999; revised September 21,
1999; accepted October 5, 1999.
Commercial relationships policy: N.
Corresponding author: David B. Henson, Department of Ophthalmology, University of Manchester, Royal Eye Hospital, Oxford Road,
Manchester M13 9WH, UK. [email protected]
Investigative Ophthalmology & Visual Science, February 2000, Vol. 41, No. 2
Copyright © Association for Research in Vision and Ophthalmology
plied to single locations within the visual field, it has been
established that variability increases as the sensitivity reduces.12–15 Weber and Rau12 also investigated the relationship
between sensitivity and variability in ocular hypertensive
(OHT) and normal eyes. They found that peripheral locations
in the visual field of OHT and normal eyes, where sensitivity is
lower, demonstrated more variability.
Variability in the visual field of POAG eyes decreases with
increasing stimulus size.16 –22 This relationship has led to a
hypothesis linking variability to functioning ganglion cell density.22,23 The hypothesis is based on three assumptions: (1) that
individual ganglion cells give variable responses when repeatedly stimulated, (2) that adjacent ganglion cells do not vary in
synchrony with each other, and (3) that there is pooling of
responses from ganglion cells. The hypothesis predicts that
variability increases when the number of stimulated ganglion
cells is reduced, either by reduction of stimulus size or by a
reduction in the density of ganglion cells.
According to this hypothesis the variability versus sensitivity relationship would be independent of the underlying
cause of any ganglion cell loss. The variability versus sensitivity
relationship, therefore, would be similar in POAG and ON
despite the significant differences in the mechanism of nerve
fiber damage and the nature (depth, location, and permanency) of the visual field defects.
417
418
Henson et al.
IOVS, February 2000, Vol. 41, No. 2
The aim of this study was to establish whether there are
differences in the relationship between sensitivity and response variability in ON, POAG, OHT, and normal eyes and
which other variables could be clinically important predictors
of response variability.
METHODS
Subjects
Data were collected from one eye of 12 patients with a history
of optic neuritis (5 men, 7 women), 25 patients with POAG (12
M, 13 F), 11 patients with OHT (5 M, 6 F), and 23 normal
subjects (13 M, 10 F).
The ON, POAG, and OHT patients were recruited from
the outpatient clinics at the Manchester Royal Eye Hospital.
The ON patients all had defective color vision (Ishihara), reduced visual acuity (VA; ⱖ 6/9, equivalent to 0.18 logMAR) and
a relative afferent pupillary defect. Nine patients had a history
of numbness. Data were collected from these patients once
their VA had returned to 6/18 or better. The interval between
diagnosis and data collection ranged from 1 to 169 weeks
(median, 26 weeks). The POAG patients all had glaucomatous
visual field loss (AGIS score: range, 1 to 19; median, 5) combined with either (or both) glaucomatous changes the optic
nerve head or a raised intraocular pressure (IOP; ⬎21 mm Hg).
OHT patients all had IOPs greater than 21 mm Hg on two
separate occasions, with normal disc appearance and no visual
field loss using program 24-2 and the Glaucoma Hemifield Test
of the Humphrey Visual Field Analyzer (HFA; Humphrey Instruments Inc., San Leandro, CA).
Normal subjects were recruited from hospital staff and
had no history of ophthalmic disease, a normal ophthalmic
examination, no systemic illness, and a visual acuity better than
6/9 (equivalent to 0.18 logMAR). The study was approved by
the Central Manchester Research Ethics Committee and follows the tenets of the Declaration of Helsinki. Informed consent was obtained from each subject. All subjects underwent a
visual field test (HFA 24-2) before the collection of FOS data.
They were only included in the study if they fulfilled the HFA
reliability criteria (⬍20% fixation losses, ⬍33% false-positive
errors, ⬍33% false-negative errors).
Test Locations
FOS data were collected at four visual field locations during a
single experimental session. For the normal and OHT eyes, the
locations were 12.7° from fixation along the 45, 135, 225, and
315 meridians. For the ON and POAG eyes, one location was
chosen to lie in an area where sensitivity was within normal
limits and, if possible, three locations in or adjacent to a
damaged area of the visual field. The damaged locations were
chosen on the basis of the HFA visual field test.
Data Collection
FOS data were collected using a modified program on a Henson
4000 bowl perimeter (Tinsley Instruments, Croydon, UK).
Stimuli subtended 0.5° and were presented for 200 msec. After
input of the test locations, the program estimated the sensitivity at each location using the full threshold (4-2) strategy.24 For
each location the program then selected five intensities that
straddled the estimated threshold in steps of 2 dB. At three
occasions during each session the experimenter, who received
FIGURE 1. Typical frequency-of-seeing curves from a normal control
eye (A) and an eye with POAG (B). The results from the POAG eye
have a shallower slope (increased variability). The line drawn through
the data points is the probit fitted function along with its 95% confidence limits (dotted lines).
continuous feedback on the selected intensities and current
responses, would interrupt the collection of data and adjust the
intensities (minimum step size, 1 dB) and number of presentations to ensure that (1) the response range approached 0 and
100% seen; (2) data were collected for at least six intensities;
and (3) there were a minimum of 20 presentations at each
intensity. While the adjustments were being made, the subject
was allowed a short rest. The presentation of stimuli was
randomized with respect to intensity and location. A typical
session lasted for approximately 30 minutes.
Data Analysis
The FOS data from each test location were imported into the
statistical package for probit regression analysis (SPSS, Chicago, IL). The mean and SD parameters of the fitted cumulative
normal function were used as estimates of sensitivity and
response variability.
Cluster regression analysis25 was used as observations
were independent between patients but not within patients.
Suitable adjusted (robust) standard errors were computed using STATA V5 (STATA Corporation, College STN, TX) to determine the following:
1. Whether there were differences in the relationship between sensitivity and variability for normal eyes and eyes
with ON, POAG, and OHT.
2. Whether the prediction of variability could be enhanced
by the variables: diagnosis, eccentricity, patient age, fixation losses, and false-positive and -negative response
rates. The fixation loss and false-positive and -negative
response rates were estimated from the catch trial data
of the 24-2 HFA visual field test.
A backward elimination procedure was used in which insignificant variables were removed successively from the model in
order of significance to identify those factors independently contributing to the variance explained by the model.
RESULTS
Figure 1 gives typical FOS data along with the probit fit for a
normal and a POAG eye.
Visual Field Variability
IOVS, February 2000, Vol. 41, No. 2
419
FIGURE 2. Variability versus sensitivity for all four groups (ON, POAG,
OHT, and normal). The line drawn
through the data points is the best
fitting (least-squares) function.
Figure 2 gives a scatter plot of log response variability
versus sensitivity for all four groups (ON, POAG, OHT, and
normal). As sensitivity decreases there is a dramatic increase in
response variability. The data of all four groups were well
described by the simple model loge(SD) ⫽ Aⴱsensitivity ⫹ B,
where parameters A and B are ⫺0.081 (robust SE, ⫾0.005) and
3.27 (SE, ⫾0.15), respectively. This model explains 57% of the
variance observed in our data (R2 ⫽ 0.57, P ⬍ 0.001). Separate
regression analyses for each group produced the parameters A
(slope) and B (intercept) shown in Table 1.
Multivariate regression analysis with backward elimination was used to establish whether the data from the different
diagnostic groups could be represented by a single model and
to establish the contribution of covariates (eccentricity, age,
false-positive and -negative rates, and fixation losses) to the
variance explained by that model. The slopes of the sensitivity–
variability relationship did not differ significantly between the
four groups (P ⱖ 0.24). Table 2 shows the contributions of the
different covariates and the diagnosis factor to the variance
explained by the model.
Inclusion of all covariates and the diagnosis factor increased the coefficient of determination by only a modest
amount (R2 increased from 0.57 to 0.63). The significance of
the diagnosis factor (P ⫽ 0.005) is due to a small difference
TABLE 1. Results of Cluster Regression by Diagnosis
Group
N
A (Robust 95% CI)
B (Robust 95% CI)
Combined
Normal
OHT
POAG
ON
71
23
11
25
12
⫺0.081 (⫺0.091, ⫺0.071)
⫺0.066 (⫺0.101, ⫺0.031)
⫺0.078 (⫺0.109, ⫺0.047)
⫺0.098 (⫺0.112, ⫺0.085)
⫺0.077 (⫺0.105, ⫺0.048)
3.27 (2.98, 3.56)
2.81 (1.64, 3.97)
3.22 (2.33, 4.11)
3.62 (3.24, 4.00)
3.28 (2.49, 4.07)
between the intercept of the POAG group and that of the other
three diagnostic groups. Accounting for the rate of false-negative responses increased the proportion of explained variance
by 2%. Neither stimulus eccentricity (P ⫽ 0.34), age (P ⫽
0.33), fixation loss rate (P ⫽ 0.10), nor false-positive response
rate (P ⫽ 0.66) had a significant effect on variability.
Figure 3 gives the data from each group on separate axes
along with the individual regression lines and the regression
line fitted to the combined data. This figure highlights the
agreement between the fit of the combined data and that of the
individual groups.
DISCUSSION
The results from this study agree with earlier work that reported
an increase in the variability of visual field measures, where
sensitivity had been reduced by either POAG or ON.3–15,26
The relationship between sensitivity and response variability was similar between the four groups of patients. Statistical
analysis did not detect differences in the slopes of the sensitivity ⫺log(variability) relationship between the four groups.
TABLE 2. Variables and Proportion of Explained Variance after
Inclusion into the Model
Variable
R2
P (F test)
Sensitivity
Diagnosis
False-negative rate
Fixation loss rate
Age
Eccentricity
False-positive rate
0.57
0.60
0.62
0.63
0.63
0.63
0.63
⬍0.001
0.005
0.004
0.10
0.33
0.34
0.67
420
Henson et al.
FIGURE 3. Variability versus sensitivity for all four groups (ON, POAG,
OHT, and normal) separated on the basis of their diagnosis. The solid
lines drawn through the data points are the best fitting (least-squares)
function to the combined data set; the dashed lines are the best fitting
(least-squares) functions to the individual groups.
There was a statistically significant difference in the intercept
between the normal, OHT, and ON groups and the POAG
group, which showed less variability (18%, P ⫽ 0.005). This
difference might be explained by the greater perimetric experience of the POAG group.
The pathophysiology of ON is very different from that of
POAG. In acute ON there is swelling in the area of demyelination
that affects the transmission of impulses along the ganglion cell
axons. This can take the form of a total block, an attenuation, or
an extended refractory period. In certain fibers there is a breakdown of the myelin sheath and a destruction of the ganglion cell
axons, with subsequent proximal and distal degeneration. The
process can occur at any location within the optic nerve and
frequently involves the fibers that supply the fovea. In comparison, damage to ganglion cell axons in POAG is a slow chronic
process that occurs at the optic nerve head and more frequently
involves the fibers at the superior and inferior poles. The similarity
between the data from all groups suggests that there is a common
underlying process linking variability with sensitivity, which is
independent of the pathophysiology of the two diseases (POAG
and ON). A common feature of these two pathologies is the loss
of functional ganglion cell axons.
The results from this study support the hypothesis that a
reduction in the number of stimulated functional ganglion cells
is likely to lead to decreased sensitivity and a concurrent
increase in response variability.22,23 This reduction could come
about via a loss of ganglion cells, such as occurs in ON and
POAG, or transfer of the stimulus to the peripheral visual
field.12 This hypothesis also predicts the reported reduction in
variability with an increase in the stimulus size.22
In short wavelength perimetry, blue stimuli are presented
on a yellow background. This type of perimetry was designed
to isolate a sparse population of ganglion cells and, as a result
of this, identify loss at an earlier stage.27 Short wavelength
perimetry is associated with an increase in response variabili-
IOVS, February 2000, Vol. 41, No. 2
ty.28 –31 FOS curves for motion stimuli, using a line displacement test, show an increase in response variability with increasing motion threshold.32 An increase in variability with
loss in sensitivity also has been reported for frequency-doubling perimetry.23 All these findings are in agreement with the
hypothesis relating variability to the density of functioning
ganglion cells. Some of the benefits resulting from targeting
sparse, vulnerable populations may be lost due to the increased
response variability associated with sparse populations.
Reliability parameters (fixation loss rate and false-positive
and -negative response rates), which were extracted from the
prior HFA 24-2 test, did not substantially increase the variance
explained by the model. Although the false-negative response
rate (P ⫽ 0.004) was a significant additional predictor of
variability, when included in the model, the explained variance
rose by only 2% (R2 increased from 0.60 to 0.62). The study’s
inclusion criteria of good patient reliability and the poor precision of estimates of patient reliability33 may account for why
these covariates did not have a larger effect on the total variance explained by the model.
In summary, the relationship between visual field sensitivity and response variability is similar in ON, POAG, OHT, and
normal subjects. This finding lends support to the hypothesis
that variability is dependent on functional ganglion cell density. A similar relationship between sensitivity and response
variability may exist for other types of perimetric stimuli. The
increased variability makes it difficult to differentiate genuine
changes in the visual field from noise and, therefore, has
important clinical implications. Targeting sparse populations
will only be beneficial if it leads to an increase in the “signalto-noise” ratio between defect and variability.
References
1. Beck RW, Cleary PA, Anderson MM, et al. A randomized, controlled trial of corticosteroids in the treatment of acute optic
neuritis. N Engl J Med. 1992;326:581–588.
2. Steel DHW, Waldock A. Measurement of the retinal nerve fibre
layer with scanning laser polarimetry in patients with previous
demyelinating optic neuritis. J Neurol Neurosurg Psychiatry.
1998;64:505–509.
3. Wall M, Johnson CA, Kutzko KE, Nguyen R, Brito C, Keltner JL.
Long- and short-term variability of automated perimetry results in
patients with optic neuritis and healthy subjects. Arch Ophthalmol. 1998;116:53– 61.
4. Patterson VH, Foster DH, Heron JR. Variability of visual threshold
in multiple sclerosis. Effect of background luminance on frequency of seeing. Brain. 1980;103:139 –147.
5. Holmin C, Krakau CET. Variability of glaucomatous visual field
defects in computerised perimetry. Graefes Archiv Clin Exp Ophthalmol. 1979;210:235–250.
6. Werner E, Saheb N, Thomas D. Variability of static threshold
responses in patients with elevated IOPs. Arch Ophthalmol. 1982;
100:1627–1631.
7. Flammer J, Drance S, Zulauf M. Differential light threshold; short
and long term fluctuation in patients with glaucoma, normal controls, and patients with suspected glaucoma. Arch Ophthalmol.
1984;102:704 –706.
8. Wilensky JT, Joondeph BC. Variation in visual field measurements
with an automated perimeter. Am J Ophthalmol. 1984;97:328 –331.
9. Katz J, Sommer A. Asymmetry and variation in the normal hill of
vision. Arch Ophthalmol. 1986;1986:104:65– 68.
10. Lewis RA, Johnson CA, Keltner JL, Labermeier PK. Variability of
quantitative automated perimetry in normal observers. Ophthalmology. 1986;93:878 – 881.
11. Heijl A, Lindgren A, Lindgren G. Test-retest variability in glaucomatous visual field. Am J Ophthalmol. 1989;108:130 –135.
IOVS, February 2000, Vol. 41, No. 2
12. Weber J, Rau S. The properties of perimetric thresholds in normal
and glaucomatous eyes. Germ J Ophthalmol. 1992;1:79 – 85.
13. Olsson J, Heijl A, Bengtsson B, Rootzen H. Frequency-of-seeing in
computerised perimetry. In: Mills RP, ed. Perimetry Update 1992/
1993. Amsterdam: Kugler; 1993:551–556.
14. Chauhan B, Tompkins J, LeBlanc R, McCormick T. Characteristics
of frequency-of-seeing curves in glaucoma in normal subjects,
patients with suspected glaucoma, and patients with glaucoma.
Invest Ophthalmol Vis Sci. 1993;34:3534 –3541.
15. Henson DB, Chaudry SJ, Artes PH. The relationship between sensitivity and variability in normal and glaucomatous visual fields. In:
Wall M, ed. Perimetry Update 1997/1999. Amsterdam: Kugler; In
press.
16. Wilson ME. Spatial and temporal summation in impaired regions of
the visual field. J Physiol. 1967;189:189 –208.
17. Wilensky JT, Mermelstein JR, Siegel HG. The use of different sized
stimuli in automated perimetry. Am J Ophthalmol. 1986;101:710 –
713.
18. Choplin NT, Sherwood MB Spaeth GL. The effect of stimulus size
on the measured threshold values in automated perimetry. Ophthalmology. 1990;97:371–374.
19. Fellman RL, Lynn JR, Starita RJ, Swanson WH. Clinical importance
of spatial summation in glaucoma. In: Heijl A, ed. Perimetric
Update 1988/1989. Amsterdam: Kugler; 1989:313–324.
20. Zalta AH. Use of a central 10 degree field and size V stimulus to
evaluate and monitor small central islands of vision in end stage
glaucoma. Br J Ophthalmol. 1991;75:151–154.
21. Takashima M, Nagata S, Kani K. Examination of receptive fields
using automated perimeter. In Mills RP, ed. Perimetric Update
1992/1993. Amsterdam: Kugler; 1993:537–541.
22. Wall M, Kutzo KE, Chauhan BC. Variability in patients with glaucomatous visual field damage is reduced using size V stimuli.
Invest Ophthalmol Vis Sci. 1997;38:426 – 435.
Visual Field Variability
421
23. Chauhan BC, Johnson CA. Test-retest variability of frequency-doubling perimetry and conventional perimetry in glaucoma patients
and normal subjects. Invest Ophthalmol Vis Sci. 1999;40:648 –
656.
24. Bebie H, Fankhauser F, Spahr J. Static perimetry: strategies. Acta
Ophthalmol. 1976;54:325–328.
25. Rogers W. Regression standard errors in clustered samples. In:
Stata Technical Bulletin Reprints Volume 3. Stata Corporation,
Texas; 1994:88 –94.
26. Harms H. Role of perimetry in assessing of optic nerve dysfunction. Trans Ophthalmol Soc UK. 1976;96:363–376.
27. Johnson CA. Selective versus nonselective losses in glaucoma. J
Glaucoma. 1994;3:S32–S44.
28. Moss ID, Wild JM, Whitaker DJ. The influence of age-related cataract in blue-on-yellow perimetry. Invest Ophthalmol Vis Sci. 1995;
36:764 –773.
29. Wild JM, Moss ID, Whitaker D, O’Neal EC. The statistical interpretation of blue-on-yellow visual field loss. Invest Ophthalmol Vis
Sci. 1995;36:1398 –1410.
30. Wild JM, Cubbidge RP. Pacey IE, Robinson R. Statistical aspects of
the normal visual field in short-wavelength automated perimetry.
Invest Ophthalmol Vis Sci. 1998;39:54 – 63.
31. Nelson–Quigg JM, Johnson CA, Casson EJ, Adams AJ. Long and
short term variability for perimetry of short wavelength sensitive
(SWS) mechanisms [ARVO Abstract]. Invest Ophthalmol Vis Sci.
1990;31(4):S190. Abstract nr 937.
32. Westcott MC, Fitzke FW, Crabb DP, Hitchins RA. Characteristics of
frequency-of-seeing curves for motion stimulus in glaucoma eyes,
glaucoma suspect eyes and normal eyes. Vis Res. 1999;39:631–
639.
33. Vingrys AJ, Demirel S. False-response monitoring during automated perimetry. Optom Vis Sci. 1998;75:513–517.