In this article we have given the rationale for the
following proposal for an international definition of (primary) OAG in
epidemiologic research: POAG is a disorder characterized by a GVFD, in
combination with probable or possible GON based on cutoff points
approximating the 99.5th and 97.5th percentiles, respectively, in that
population in at least one eye of a subject with open chamber angle,
and no history or sign of angle closure or secondary glaucoma. Thus, an
algorithm may be created leading to the diagnosis of definite,
probable, or possible POAG (see addendum). If other research groups use
a similar approach, one is free to pool definite and probable OAG or
not.
When looking at
Table 1 and reference 15, it seems that we have
not made much progress in defining glaucoma since Donders coined the
term glaucoma simplex in 1861.
23 It, thus, seems like a
risky enterprise to start defining criteria for POAG nearly 150 years
later. On the other hand, it is clear also from
Figure 2 that there is
a need for valid comparisons between studies. Current variations
in definition allow wide variations in prevalence data, as well as
justification for treatment. In this study we define cutoff values for
OAG determinants based on statistical grounds. This means that on
arbitrary statistical grounds a division is made between normal and
abnormal discs. We realize that this might be artificial and that some
subjects may falsely be defined as healthy or abnormal. However,
because of the large variation in OAG definitions in epidemiologic
and/or clinical research, we think it is for the time being a good
starting point to use such a definition for better comparison and
pooling of study results. In due time further refinement of the
definition may become possible when more incidence data become
available. Using 97.5th or 99.5th percentiles to define abnormality
does not mean that we used these criteria to define someone as being
diseased. Only in combination with other signs we propose the term
definite OAG. We felt that our database allowed for the definition of
OAG on statistical grounds and that such an approach may become a
starting point for future diagnostic fine-tuning. This may not be too
far away because some large population-based studies are working on
incidence data on POAG.
We did not fully exclude pseudoexfoliation as a cause of OAG during
baseline examination and, thus, refer in this article to OAG instead of
POAG. As no case of OAG in this cohort had pseudoexfoliation on
follow-up clinical examination we feel that in practice we may assume
that our data are valid as POAG data as did two studies that included
pseudoexfoliation in the diagnosis of POAG.
11 13
In this study we present prevalence figures for OAG combining GON
data obtained by Imagenet and ophthalmoscopy. We think the Imagenet
data are the more reliable data, especially for follow-up and
risk-factor analyses. However, because the Imagenet module for the
optic disc is not available any more, and neither is the simultaneous
stereoscopic Topcon TRC-SS2 camera, also essential for this module, we
also present prevalence figures based on only ophthalmoscopic optic
disc data for comparison with other studies. We found in a substudy
that Imagenet and the Heidelberg Retina Tomograph had much higher
correlations for the estimation of the VCDR than ophthalmoscopy,
showing that ophthalmoscopy is less reliable than these semiautomated
apparatuses, even carried out by trained examiners. Still we felt that
in daily practice ophthalmoscopy will be the method of choice during
the coming years. Therefore, in choosing cutoff points for the VCDR and
other disc measures we looked primarily at feasibility and tried to
choose cutoff points that were also ophthalmoscopically assessable.
Thus, to create as simple as possible a definition for OAG, based on
glaucomatous VF loss and GON, we propose as a cutoff point for a
statistically abnormal, and thus arbitrarily pathologic, possible GON a
VCDR ≥ 0.7, asymmetry in VCDR between both eyes ≥ 0.2 or
neuroretinal rim width < 0.1 for data obtained by ophthalmoscopy.
The latter was not assessed in this study by ophthalmoscopy but would
probably be necessary in other studies to detect discs with local
notching of the rim. From
Tables 5 and 6 one may see that for the
largest discs the cutoff for pathologic VCDR might have been chosen
as ≥ 0.8, and the same holds for those 75 years of age and older.
All these subdivisions make the definition more and more complicated
and that is why we propose to keep as the cutoff point for a possible
GON a VCDR ≥ 0.7. It should be borne in mind that this definition
exists for the cohort studied and that for other cohorts and especially
different races this type of definition might have different values.
Hitherto, some studies did not specify whether one used information on
IOP or disc measures to grade VF defects as
glaucomatous.
2 8 13 Two were masked in this
respect
9 11 and one was not.
14 Similarly,
before deciding whether a subject had OAG, all studies mentioned in
Table 1 looked at the combined data of a case while we tried to do so
by combining strictly defined determinants without subjective overall
evaluation at the end. We believed that this would lead to less
assessment bias. On the other hand, this resulted in small differences
in prevalence of OAG when the Imagenet or ophthalmoscopy data were
used.
Our overall prevalence of definite OAG of 0.8% (with combined
use of Imagenet and ophthalmoscopic data; 0.7% when only using
ophthalmoscopic data) and its rise with age are comparable with
prevalence figures of the Framingham Study
24 (1.2%), of
The Baltimore Eye Survey
7 (1.1%), and among the white
subjects of the Barbados Eye Study
9 (0.8%). The Beaver
Dam Eye Study and the Blue Mountains Eye Study, on the other hand,
found a higher overall prevalence, 2.1%
8 and
3.0%,
11 respectively. The prevalence of definite plus
probable OAG in the Rotterdam Study was 3.2% and this may explain the
gap. Several more reasons for these differences exist. All studies
mentioned in
Table 1 but the Egna-Neumarkt and the Rotterdam Study
used for final assignment to glaucoma diagnosis a review of
all data by one or more principal investigators, glaucoma specialists,
or ophthalmologists. In this study we combined the VF and optic disc
data, and this led straightforward to one of three diagnostic
categories (apart from normals) without additional influence on the
final results. The discrepancy with the Beaver Dam Study could be
explained by their wider criteria for POAG. Other sources for
differences between studies include sampling and perimetry techniques,
screening methods for glaucoma, subjective interpretation of
examination data, diagnostic criteria, age distributions, and real
geographic contrasts in prevalence due to differences in lifestyle or
genetic drift.
The VF screening and grading procedure in our study resulted in a
prevalence of 1.5% of GVFDs compatible with OAG. This is comparable
with the findings of the Framingham Study (1.4%, screening in a subset
only, enlargement of blind spot excluded)
25 but lower than
that found in Australia (3.1%).
11 The Blue Mountains Eye
Study used, after screening, Humphrey full threshold perimetry (C30-2),
which is more sensitive than kinetic Goldmann perimetry,
26 especially in glaucoma where it might detect up to 21% more
defects.
27 Full threshold automated perimetry is nowadays
considered to be the gold standard for VF examination, but at baseline
in 1990 we felt that especially in older subjects it may create more
false-positive errors compared with Goldmann perimetry. This might be
because of poor fixation that accounted for 9% of inadequate Humphrey
fields versus 2% at the Goldmann perimeter.
27 Between
threshold Humphrey perimetry and kinetic Goldmann perimetry there is
88% concordance when both tests appeared reliable.
27 Because the Humphrey algorithms also have changed in the meantime and
because we now perform both Humphrey 30-2 and Goldmann perimetry in the
follow-up study, a more valid comparison between both methods will be
possible within a year from now. It also has been shown that supra
threshold perimetry identifies about two thirds of all cases identified
by full-threshold perimetry.
28 Using this latter test our
prevalence of definite OAG might have risen to approximately 1.4%.
Even then there still would have been a twofold difference in
prevalence by comparison with the Blue Mountains Eye Study. Given the
variation in techniques and differences between various studies we
believe that conclusions on geographic differences are for the time
being not justifiable.
Our study differed from other large population-based studies with
regard to the use of Imagenet to assess the optic nerve. Imagenet used
strict criteria for defining the cup margins, based only on topographic
data, thus reducing variation due to different observers. This makes it
also particularly interesting for follow-up studies.
21 We
found a higher mean VCDR on Imagenet measurements (0.49) compared with
studies using other methods for examining the optic disc (mean VCDR
0.28,
5 0.3
10 using ophthalmoscopy by several
examiners, 0.36
8 and 0.43
11 by grading of
photographs). As a result, the prevalence of an enlarged VCDR was also
higher in our study than in other studies (VCDR ≥ 0.4: 76.7%
compared with 27.1%
5 or 37.0%
8 ). However,
our prevalence of a VCDR ≥ 0.7 (5.1%) was only slightly
different from the findings of the Blue Mountains Eye Study
(5.0%), which examined stereo transparencies with a
viewer.
11 Also, asymmetry in VCDR between both eyes
was more prevalent in our study, compared with findings of other studies (4.6% asymmetry ≥ 0.2,
25 0.7%
asymmetry ≥ 0.3
11 ).
The relation between OAG and gender is still controversial. In
Framingham
5 and Barbados
9 a higher prevalence
of POAG was found in men, which matched our finding. However, in the
Blue Mountains Eye Study a (borderline significantly) higher OR of 1.55
for POAG was found for women,
11 and in
Baltimore
7 and Beaver Dam
8 no difference was
found. It might be that in younger subjects the association between OAG
and gender is not yet present. It would seem possible that if the study
cohort had a greater proportion of younger subjects the gender risk
would disappear.
Our study did not show any correlation between age, gender, or
IOP. This is in contrast to previously published
results,
1 4 29 but is in agreement with
others.
1 5 8 Our findings do agree with prevalence data on
IOP and VCDR in nursing home inhabitants.
30 However, the
response in the nursing homes was low, especially in the older
subjects, increasing the risk of selection bias. This could explain our
lower OAG prevalences compared with that study.
30 One
could adjust the prevalence rates for probable and possible OAGs in the
nursing homes by raising them by 25% similar to the lower
response rates in these homes than in the independently living
subjects.
In conclusion, the overall prevalence of definite OAG in the
Rotterdam Study was 0.8%, which is comparable to findings of other
population-based studies on whites. The OR for men to
have OAG was higher than for women. There was a significant increase in
prevalence of OAG with increasing age. The overall prevalence of OAG
varied 12-fold with different criteria and screening algorithms. We
hope that standardizing diagnostic procedures and our proposed
definitions will improve future (epidemiologic) glaucoma research.