Our results showed that pattern discrimination perimetry detected
progression of glaucomatous visual field damage in fewer patients (26
as opposed to 32) than conventional perimetry. Additionally, among
patients that progressed with both techniques, only 6 of 17 were
detected earlier by pattern discrimination perimetry. Analysis of those
patients who progressed with conventional perimetry showed that the
large majority continued to show progression, suggesting that our
criteria were specific enough and that progression was not occurring by
chance. This degree of repeatable confirmed progression was not
observed in pattern discrimination perimetry. Our study suggests that
pattern discrimination perimetry may not be a good indicator of
progressive visual field damage in glaucoma.
In another study, we demonstrated that high-pass resolution perimetry
is significantly more sensitive than conventional perimetry at
identifying progressive visual field damage.
27 We
therefore felt justified that by combining the data from conventional
and high-pass resolution perimetry, we would produce an adequate
standard against which data from pattern discrimination perimetry could
be reviewed. As illustrated in
Table 1 , we evaluated 14 different sets
of criteria (number of locations where the difference in threshold
deviation was outside the 95% confidence limits) for progression with
pattern discrimination perimetry and chose the optimum number of
stimuli that yielded the highest sensitivity and specificity for
progression. Despite this, we obtained poor results with pattern
discrimination perimetry. Also as evident from
Table 1 , when confirmed
progression required only one location to be outside the variability
limits (a criterion that would favor high sensitivity but low
specificity), the sensitivity of pattern discrimination perimetry to
detect progressive damage still remained low, that is, only 85% while
the specificity was 57%.
There are several potential factors that could explain our results.
First, it should be understood that our purpose was to identify
progressive damage in patients with established visual field damage.
Johnson et al.
28 29 in their work with short-wavelength
automated perimetry noted that once a significant amount of
glaucomatous damage had occurred, little or no short-wavelength
sensitivity remained at those locations. Likewise, it is possible then
that by the time glaucoma is detected (using conventional clinical
criteria), substantial damage has already taken place such that not
enough ganglion cell reserves remain for evaluating progressive damage
by pattern discrimination perimetry. Therefore, it is likely that
pattern discrimination perimetry may be sensitive in detecting damage,
but once damage has occurred, it is not sensitive in detecting the
difference as damage progresses.
Second, the coherence threshold scale used in pattern discrimination
perimetry reaches its maximum value of 100% when the defect in
conventional perimetry is moderate or even relatively mild (1–2 log
units).
18 This coherence limit could underestimate the
degree of advanced defects in glaucoma patients and consequently
restrict the ability to detect progression.
Third, our study showed that a high degree of variability exists in
glaucoma patients tested with pattern discrimination perimetry. As a
result of this high variability, the changes in coherence thresholds
were quite large to confirm progression. For baseline coherence
threshold of 70%, the 95% confidence interval was exceeded only if
the subsequent coherence threshold was ≥ 90%. For baseline
coherence threshold of 80%, the corresponding value was ≥ 97%.
For higher initial baseline coherence thresholds, the upper end of the
confidence bounds was truncated due to a limitation in dynamic range.
This effect has also been reported in conventional
perimetry.
26 The increase in variability may be attributed
to factors such as fatigue and difficulty of the task.
Finally, we observed that no correlation between pattern discrimination
and conventional techniques existed in 8 of 17 patients with regard to
the hemifield that showed progression first. This lack of correlation
has been reported previously.
14 19 Another reason for the
difference in sensitivity to detect progression could be attributed to
the fact that the two techniques were detecting different aspects of
glaucomatous damage. If this were the case, different techniques may be
appropriate for different mechanisms of glaucomatous damage, which to
date have not been identified.
Although the units for conventional and pattern discrimination
perimetry are different, we used test–retest variability
characteristics for each technique to derive confidence limits based on
probability levels. Progression was documented to occur at any location
if the measurements exceeded the variability limits, hence we feel the
comparison between the two perimeters was justified.
In assessing whether a new technique can detect glaucomatous
progression earlier than conventional perimetry, it is important to
know the rate of false progression by that technique in a parallel
group of normal controls. However, since our study indicated that
pattern discrimination perimetry did not identify progressive damage in
a greater number of patients than conventional perimetry, the issue of
false cases of progression was not applicable. For this reason, we feel
that, although having a control group would have been useful to study
the characteristics of pattern discrimination perimetry in normals, it
would not have made a meaningful difference to our conclusions.
A sharp distinction lies between detecting early visual field damage
and detecting early progressive visual field damage. The former
requires testing a group of subjects in whom no previous damage has
been noted and who presumably have a higher ganglion cell reserve. The
latter involves evaluating further damage in patients with established
visual field damage, who presumably have a relatively lower ganglion
cell reserve. It is therefore likely that a test may be very sensitive
at detecting early visual field damage but poor in detecting
progressive damage and vice versa. Visual field testing strategies in
the future may need modification such that different strategies are
used to detect these two equally important aspects of clinical glaucoma
management.
Previous studies have shown that pattern discrimination may detect
significant visual field damage in glaucoma
suspects.
16 18 20 24 Although we have shown that pattern
discrimination perimetry is not sensitive for detecting progressive
damage, our study is not necessarily incompatible with reports showing
its efficacy in demonstrating preclinical damage in glaucoma suspects.
However, pattern discrimination perimetry is as time-consuming as
conventional perimetry (using the conventional bracketing thresholding
algorithm) in addition to being a difficult test for most patients. In
contrast several new techniques such as high-pass
resolution
30 31 and frequency doubling
perimetry
32 are much faster (test time 5 to 6 minutes),
primarily because they test fewer locations and have a higher patient
acceptance. Additionally, high-pass resolution perimetry
27 has been shown to detect progressive glaucomatous visual field damage
earlier than conventional perimetry. Furthermore, faster testing
strategies with conventional perimetry such as the Swedish Interactive
Thresholding Algorithm (SITA)
33 34 35 may prove to
be more efficacious for detecting progression than conventional
techniques. Given these facts and that pattern discrimination perimetry
is not effective in evaluating progressive glaucomatous visual field
damage, its role in clinical glaucoma management currently appears
limited.