Research

JAMA Ophthalmology | Original Investigation

Evaluation of a Myopic Normative Database for Analysis

of Retinal Nerve Fiber Layer Thickness
Sayantan Biswas, MPhil; Chen Lin, BM; Christopher K. S. Leung, MD

Invited Commentary
IMPORTANCE Analysis of retinal nerve fiber layer (RNFL) abnormalities with optical coherence page 1040
tomography in eyes with high myopia has been complicated by high rates of false-positive
errors. An understanding of whether the application of a myopic normative database can
improve the specificity for detection of RNFL abnormalities in eyes with high myopia is
relevant.

OBJECTIVE To evaluate the diagnostic performance of a myopic normative database for

detection of RNFL abnormalities in eyes with high myopia (spherical equivalent, –6.0 diopters
[D] or less).

DESIGN, SETTING, AND PARTICIPANTS In this cross-sectional study, 180 eyes with high myopia
(mean [SD] spherical equivalent, –8.0 [1.8] D) from 180 healthy individuals were included in
the myopic normative database. Another 46 eyes with high myopia from healthy individuals
(mean [SD] spherical equivalent, –8.1 [1.8] D) and 74 eyes from patients with high myopia and
glaucoma (mean [SD] spherical equivalent, –8.3 [1.9] D) were included for evaluation of
specificity and sensitivity. The 95th and 99th percentiles of the mean and clock-hour
circumpapillary RNFL thicknesses and the individual superpixel thicknesses of the RNFL
thickness map measured by spectral-domain optical coherence tomography were calculated
from the 180 eyes with high myopia. Participants were recruited from January 2, 2013, to
December 30, 2015. The following 6 criteria of RNFL abnormalities were examined: (1) mean
circumpapillary RNFL thickness below the lower 95th or (2) the lower 99th percentile;
(3) one clock-hour or more for RNFL thickness below the lower 95th or (4) the lower 99th
percentile; and (5) twenty contiguous superpixels or more of RNFL thickness in the RNFL
thickness map below the lower 95th or (6) the lower 99th percentile.

MAIN OUTCOMES AND MEASURES Specificities and sensitivities for detection of RNFL
abnormalities.

RESULTS Of the 46 healthy eyes and 74 eyes with glaucoma studied (from 39 men and
38 women), the myopic normative database showed a higher specificity (63.0%-100%) than
did the built-in normative database of the optical coherence tomography instrument
(8.7%-87.0%) for detection of RNFL abnormalities across all the criteria examined
(differences in specificities between 13.0% [95% CI, 1.1%-24.9%; P = .01] and 54.3% [95% CI,
37.8%-70.9%; P < .001]) except for the criterion of mean RNFL thickness below the lower
99th percentile, in which both normative databases had the same specificities (100%) but
the myopic normative database exhibited a higher sensitivity (71.6% vs 86.5%; difference in
sensitivities, 14.9% [95% CI, 4.6%-25.1%; P = .002]).

CONCLUSIONS AND RELEVANCE The application of a myopic normative database improved

the specificity without compromising the sensitivity compared with the optical coherence
tomography instrument’s built-in normative database for detection of RNFL abnormalities in Author Affiliations: Department of
eyes with high myopia. Inclusion of myopic normative databases should be considered in Ophthalmology and Visual Sciences,
The Chinese University of Hong Kong,
optical coherence tomography instruments.
Hong Kong, People’s Republic of
China.
Corresponding Author: Christopher
K. S. Leung, MD, Department of
Ophthalmology and Visual Sciences,
The Chinese University of Hong Kong,
JAMA Ophthalmol. 2016;134(9):1032-1039. doi:10.1001/jamaophthalmol.2016.2343 Hong Kong, People’s Republic of
Published online July 21, 2016. Corrected on September 22, 2016. China (tlims00@hotmail.com).

1032 (Reprinted) jamaophthalmology.com

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021

 Retinal Nerve Fiber Layer Imaging in Myopia
D
iscerning glaucomatous changes to the optic nerve
head is challenging in eyes with high myopia.1,2 Para-
papillary atrophy and tilted disc configuration—2
highly prevalent optic disc findings in eyes with high myopia
(spherical equivalent,–6.0 diopters [D] or less)3—often obfus-
cate reliable assessment of the neuroretinal rim configura-
tion and optic disc excavation. Likewise, visualization of the
retinal nerve fiber layer (RNFL) is difficult in eyes with high
myopia. Although optical coherence tomography (OCT) af-
fords reproducible measurement of the RNFL thickness,4-6 de-
tection of RNFL abnormalities in eyes with high myopia is com-
plicated by high rates of false-positive errors,7-12 which is likely
related to the lack of inclusion of individuals with high myo-
pia in the normative databases of many OCT instruments. For
example, the mean (SD) refractive error of the 271 healthy in-
dividuals included in the normative database of the Cirrus high
definition (HD)-OCT (Carl Zeiss Meditec)—a spectral-domain
OCT instrument—was –0.82 [1.96] D.13 The convergence of
the superotemporal and/or inferotemporal RNFL bundles
toward the macula in eyes with high myopia may render the
RNFL measurements at the superior and/or inferior quad-
rants relatively abnormal with reference to the built-in
normative data.14 We hypothesize that incorporating a my-
opic normative database in the OCT instrument for analysis
of RNFL thickness would be important to decrease the
frequency of false-positive errors in eyes with high myopia.
We calculated the normal reference ranges of RNFL thick-
nesses across the circumpapillary RNFL (cRNFL) profile, as well
as in the individual superpixels of the RNFL thickness map
(50 × 50 superpixels), obtained with the Cirrus HD-OCT in-
strument in healthy individuals with high myopia. We then
compared the specificity and sensitivity of the myopic nor-
mative database and the Cirrus HD-OCT built-in normative da-
tabase for detection of RNFL abnormalities in a separate group
of healthy individuals with high myopia and patients with high
myopia and glaucoma.
Methods
Participants
A total of 315 individuals with myopia (spherical equivalent
range, –2.25 to –17.13 D) seeking corneal refractive surgery at
the University Eye Center, the Chinese University of Hong Kong,
were consecutively recruited from January 2, 2013, to April 30,
2015. Among the 315 individuals with myopia, 351 eyes of 196
individuals with a spherical equivalent of –6.0 D or less were
considered for inclusion in the myopic normative database. A
separate group of 46 eyes with high myopia (spherical equiva-
lent, –6.0 D or less) from 27 healthy individuals (the control
group) were consecutively enrolled during another period be-
tween October 1 and December 30, 2015, for evaluation of the
specificities of the myopic normative database. Seventy-four
eyes with high myopia (spherical equivalent, –6.0 D or less)
from 50 patients with glaucoma (the glaucoma group) were en-
rolled from the glaucoma clinic at Hong Kong Eye Hospital for
evaluation of the sensitivities of the myopic normative data-
base. All participants received a complete ophthalmic exami-
jamaophthalmology.com
Copyright 2016 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 05/07/2021
Original Investigation Research
Key Points
Question Does the application of a myopic normative database
improve the diagnostic performance for detection of retinal nerve
fiber layer (RNFL) abnormalities in eyes with high myopia?
Findings In this cross-sectional study, specificities of a myopic
normative database for detection of glaucomatous RNFL
abnormalities in the circumpapillary RNFL profile and in the RNFL
thickness map were higher (63.0%-100%) compared with the
normative database of the optical coherence tomography
instrument (8.7%-87.0%) at a comparable level of sensitivities.
Meaning Myopic normative databases should be included in
optical coherence tomography instruments to improve the
diagnostic specificity for detection of RNFL abnormalities.
nation with intraocular pressure measured with Goldmann ap-
planation tonometry, central corneal thickness measured with
ultrasound pachymetry, and axial length (AL) measured with
partial coherence laser interferometry (IOL master, Carl Zeiss
Meditec). All eyes examined in the study were phakic at the
time of examination, with best-corrected visual acuity of
20/40 or more. No individual had a history of macular dis-
ease, refractive or intraocular surgery, or neurologic disease
that may result in visual field (VF) abnormalities. Color optic
disc stereophotographs were taken with a fundus camera
(TRC-50DX, Topcon), and the RNFL was imaged by the Cirrus
HD-OCT instrument. Standard automated white-on-white
perimetry (Swedish Interactive Threshold Algorithm stan-
dard 24-2; Humphrey Field Analyzer II-I, Carl Zeiss Meditec)
was performed in eyes with evidence of glaucomatous optic
disc changes (described below) among the 351 eyes to be con-
sidered for inclusion in the myopic normative database and in
all eyes in the control group (46 eyes) and the glaucoma group
(74 eyes). After excluding 13 individuals with glaucomatous
optic disc changes (6 had perimetric and 7 had preperimetric
glaucoma) and 3 individuals with nonglaucomatous optic neu-
ropathies (described below) in at least 1 eye, 319 eyes of 180
participants were eligible for inclusion in the myopic norma-
tive database. If both eyes were eligible, only the right eye was
selected. Consequently, 180 eyes with high myopia from 180
participants were included in the myopic normative data-
base. No eyes in the control group displayed glaucomatous op-
tic disc changes or VF abnormalities. All eyes in the glaucoma
group had perimetric glaucoma. The study was conducted in
accordance with the ethical standards stated in the 1964 Dec-
laration of Helsinki and approved by the Kowloon Central/
Kowloon East Research Ethics Committee with written
informed consent obtained from all participants.
Diagnosis of Glaucoma
Perimetric glaucoma was diagnosed based on the presence of
glaucomatous optic disc changes, which included narrowed
neuroretinal rim and optic disc excavation with or without vis-
ible RNFL defects in color optic disc stereophotographs evalu-
ated by a glaucoma specialist (C.L.) together with correspond-
ing VF abnormalities (≥3 nonedge contiguous points significant
at P < .05, including ≥1 point significant at the P < .01 level on
(Reprinted) JAMA Ophthalmology September 2016 Volume 134, Number 9 1033
 Research Original Investigation Retinal Nerve Fiber Layer Imaging in Myopia

Figure 1. Examples of Retinal Nerve Fiber Layer (RNFL) Thickness Analysis

A Healthy eye with high myopia B Eye with high myopia and glaucoma

104 58
94 114 75 54

163 36
63 67

96 30
58 49

133 32
56 54

64 179 62 62
78 54

Mean RNFL thickness 100 Mean RNFL thickness 53

Cirrus HD-OCT normative database RNFL thickness analysis Cirrus HD-OCT normative database RNFL thickness analysis

104 58
94 114 75 54

163 36
63 67

96 30
58 49

133 32
56 54

64 179 62 62
78 54

Mean RNFL thickness 100 Mean RNFL thickness 53

Myopic normative database RNFL thickness analysis Myopic normative database RNFL thickness analysis

A, Optic disc photographs, RNFL thickness deviation maps, and clock-hour 26.39 mm). B, Optic disc photographs, RNFL thickness deviation maps, and
circumpapillary RNFL profiles analyzed by the Cirrus high-definition optical clock-hour circumpapillary RNFL profiles analyzed by the Cirrus HD-OCT and
coherence tomography (HD-OCT) and the myopic normative databases of a the myopic normative databases of an eye with high myopia and glaucoma
healthy eye with high myopia (spherical equivalent, –8.75 diopters; axial length, (spherical equivalent, –7.00 diopters; axial length, 26.10 mm).

the same side of the horizontal meridian in the pattern devia-
tion plot) in a reliable VF (fixation losses, false-negative and
false-positive errors ≤20%) that were repeatable in at least 2
consecutive VF tests. Eyes with glaucomatous optic disc
changes but without VF abnormalities were diagnosed as pre-
perimetric glaucoma. Eyes exhibiting neuroretinal rim pallor
and RNFL loss but without narrowed neuroretinal rim or op-
tic disc excavation were diagnosed as nonglaucomatous op-
tic neuropathies.15
(50 × 50 superpixels) were reported in the Cirrus HD-OCT RNFL
analysis printout taking reference from the built-in norma-
tive database comprising 271 healthy individuals from 4 racial/
ethnic groups (Chinese, Hispanic, African descent, and Euro-
pean descent).13 Individual clock-hour and superpixel RNFL
thicknesses below the lower 99th and 95th percentiles were
encoded in red and yellow, respectively, in the circumpapil-
lary RNFL profile and in the RNFL thickness deviation map
(Figure 1).

OCT Imaging of the RNFL Development of the Myopic Normative Database

The Cirrus HD-OCT Optic Disc Cube scan imaged and mea-
sured the RNFL thickness in the 6 × 6–mm2 optic disc region
(200 × 200 pixels). Only images with signal strength of 6 or
greater and without motion artifact were included. Retinal
nerve fiber layer abnormalities in the cRNFL profile (scan di-
ameter, 3.46 mm) and in the RNFL thickness deviation map
The individual superpixel RNFL thickness of the RNFL thick-
ness map (50 × 50 superpixels; 6 × 6 mm2) of the 180 eyes with
high myopia were exported from the Cirrus HD-OCT instru-
ment via a research browser (version 6.5.0.772; a software pack-
age for export of raw data) provided by Carl Zeiss Meditec. With
the use of a custom-designed computer program (MATLAB;

1034 JAMA Ophthalmology September 2016 Volume 134, Number 9 (Reprinted) jamaophthalmology.com

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021

 Retinal Nerve Fiber Layer Imaging in Myopia Original Investigation Research

MathWorks, Inc), the RNFL thickness maps were centered at

Table 1. Participant Demographics and Biometric Parameters
the optic disc and superimposed. The RNFL thicknesses at the
individual superpixel coordinates were registered. The lower Group, Valuea
95th and 99th percentile RNFL thickness values of the indi- Myopic
Normative Patients With
vidual superpixel coordinates of the RNFL thickness map and Demographic Database Controls Glaucoma
the mean and clock-hour for the cRNFL profile were derived. Eyes/Participants, No. 180/180 46/27 74/50
The lower 95th and 99th percentile values represented the cut- Sex, No.
offs below which RNFL thickness measurements were en- M 49 9 30
coded in yellow and red, respectively, in the RNFL thickness F 131 18 20
deviation map and in the clock-hour for the cRNFL profile Age, y 37.5 (9.0) 36.0 (12.3) 47.9 (8.6)
(Figure 1). Spherical equivalent, −8.0 (1.8) −8.1 (1.8) −8.3 (1.9)
diopter
Definitions of RNFL Abnormalities Axial length, mm 26.7 (1.0) 26.6 (1.0) 26.7 (1.1)
Six criteria including (1) mean cRNFL thickness below the lower Intraocular pressure, 13.3 (2.4) 16.2 (2.4) 17.2 (4.1)
mm Hg
95th percentile and (2) the lower 99th percentile; (3) one clock-
Thickness, μm
hour for cRNFL thickness or more below the lower 95th per-
RNFL 89.6 (7.2) 90.7 (8.3) 66.1 (8.9)
centile and (4) the lower 99th percentile; (5) twenty contigu-
Central corneal 553.4 (31.2) 540.0 (29.0) 545.3 (36.8)
ous superpixels or more of RNFL thickness in the RNFL
thickness map below the lower 95th percentile and (6) the lower Abbreviation: RNFL, retinal nerve fiber layer.
a
99th percentile were evaluated. These criteria were exam- Data are presented as mean (SD) unless otherwise indicated.
ined because they were widely adopted in clinical practice and
clinical studies to determine RNFL abnormalities. Specificities and Sensitivities for Detection
of cRNFL Abnormalities
Statistical Analysis Another 46 eyes with high myopia from 27 healthy individu-
Statistical analyses were performed using Stata, version 14.0 als and 74 eyes with high myopia from 50 patients with glau-
(StataCorp). Comparisons of spherical equivalent, AL, and coma were included for evaluation of the specificities and
age between the control and glaucoma groups were per- sensitivities of the myopic normative database and the Cirrus
formed with linear mixed modeling. The sensitivities and HD-OCT normative database for detection of RNFL abnormali-
specificities and positive and negative predictive values for ties (Table 1). The mean (SD) spherical equivalent and AL did
detection of RNFL abnormalities determined by the Cirrus not differ between the control and glaucoma groups (control
HD-OCT (as reported in the analysis printout) and the group: spherical equivalent, –8.1 [1.8] and AL, 26.6 [1.0] mm;
myopic normative databases were compared with a 2-sided glaucoma group: spherical equivalent, –8.3 [1.9] and AL, 26.7
McNemar test and generalized score statistic, respectively. [1.1] mm; P ≥ .42), although individuals in the glaucoma group
Power calculation revealed the current sample size had a were older (mean [SD] age, control group, 36.0 [12.3] years;
power of 0.835 at an α of 0.05 to detect a difference in the glaucoma group, 47.9 [8.6] years; P < .001).
discordant proportions of 0.014 (1 of 74) and 0.149 (11 of 74) Across all the diagnostic criteria examined, the myopic
between the myopic normative database and the Cirrus normative database outperformed the Cirrus HD-OCT nor-
HD-OCT normative database in the glaucoma group and a mative database for detection of RNFL abnormalities in eyes
power of 0.846 at an α of 0.05 to detect a difference in the with high myopia (Table 2). For example, with use of the cri-
discordant proportions of 0.022 (1 of 46) and 0.239 (11 of 46) terion of 1 clock-hour or more cRNFL thickness below the
between the myopic normative database and the Cirrus lower 99th percentile, higher sensitivity and specificity were
HD-OCT normative database in the control group. observed for the myopic normative database (100% and
87.0%, respectively) compared with the Cirrus HD-OCT nor-
mative database (91.9% and 56.5%, respectively; P = .01 for
sensitivity and P < .001 for specificity). With use of the lower
Results 99th percentile of the mean cRNFL thickness as a cutoff to
A total of 180 eyes with high myopia from 180 healthy indi- define RNFL abnormality, high specificities were seen for
viduals were included in the development of the myopic nor- both the myopic normative and the Cirrus HD-OCT data-
mative database (Table 1). All participants were Chinese. The bases (100%), but the sensitivity was higher for the myopic
mean (SD) spherical equivalent and AL were –8.0 (1.8) D (range, normative database (86.5%) than the Cirrus HD-OCT data-
–17.1 to –6.0 D) and 26.7 (1.0) mm (range, 24.4 to 29.9 mm), re- base (71.6%; P = .002). The myopic normative database was
spectively. The mean cRNFL thickness was 89.6 (7.2) μm (range, more sensitive than the Cirrus HD-OCT normative database
70 to 110 μm), which was positively correlated with spherical for detection of eyes with early glaucoma (mean deviation,
equivalent (β = 1.02; P = .001) and negatively correlated with more than –6 dB) (n = 35) and all eyes with glaucoma (n = 74)
AL (β = –0.92; P = .09). The mean (SD) intraocular pressure and using the criterion of 1 clock-hour or more for cRNFL thick-
central corneal thickness were 13.3 (2.4) mm Hg (range, ness below the lower 99th percentile (early glaucoma:
9-21 mm Hg) and 553.4 (31.2) μm (range, 488-656 μm), myopic normative database, 100%; Cirrus HD-OCT norma-
respectively. tive database, 88.6%; P = .046; all glaucoma: myopic

jamaophthalmology.com (Reprinted) JAMA Ophthalmology September 2016 Volume 134, Number 9 1035

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021

 Research Original Investigation Retinal Nerve Fiber Layer Imaging in Myopia

Table 2. Sensitivities and Specificities of the Cirrus HD-OCT and the Myopic Normative Databases
for Detection of RNFL Abnormalities
Sensitivity, Specificity,
% (95% CI) % (95% CI)
Patients Patients
All Patients With Glaucoma With Glaucoma
Criteria of RNFL With Glaucoma With MD ≤–6 dB With MD >–6 dB Controls
Abnormalities (n = 74) (n = 39) (n = 35) (n = 46)
Circumpapillary RNFL Thickness Profile
≥1 Clock-hour of RNFL
thickness below the lower
99th percentile
Cirrus HD-OCT normative 91.9 (83.2 to 97.0) 94.9 (82.7 to 99.4) 88.6 (73.3 to 96.8) 56.5 (41.1 to 71.1)
database
Myopic normative 100 (95.1 to 100) 100 (91.0 to 100) 100 (90.0 to 100) 87.0 (73.7 to 95.1)
database
Difference in sensitivity 8.1 (5.4 to 15.7) 5.1 (4.4 to 14.6) 11.4 (2.0 to 24.8) 30.4 (13.7 to 47.2)
and specificity
P value .01 .16 .046 <.001
≥1 Clock-hour of RNFL
thickness below the lower
95th percentile
Cirrus HD-OCT normative 100 (95.1 to 100) 100 (91.0 to 100) 100 (90.0 to 100) 30.4 (17.7 to 45.8)
database
Myopic normative 100 (95.1 to 100) 100 (91.0 to 100) 100 (90.0 to 100) 63.0 (47.6 to 45.8)
database
Difference in sensitivity 0 (−1.4 to 1.4) 0 (−2.6 to 2.6) 0 (−2.9 to 2.9) 32.6 (13.3 to 51.9)
and specificity
P value >.99 >.99 >.99 .001
Mean RNFL thickness below
the lower 99th percentile
Cirrus HD-OCT normative 71.6 (59.9 to 81.5) 87.2 (72.6 to 95.7) 54.3 (36.7 to 71.2) 100 (92.3 to 100)
database
Myopic normative 86.5 (76.6 to 93.3) 94.9 (82.7 to 99.4) 77.1 (59.9 to 89.6) 100 (92.3 to 100)
database
Difference in sensitivity 14.9 (4.6 to 25.1) 7.7 (3.2 to 18.6) 22.9 (6.1 to 39.6) 0 (−2.2 to 2.2)
and specificity
P value .002 .08 .01 >.99
Mean RNFL thickness below
the lower 95th percentile
Cirrus HD-OCT normative 87.8 (78.2 to 94.3) 89.7 (75.8 to 97.1) 85.7 (69.7 to 95.2) 87.0 (73.7 to 95.1)
database
Myopic normative 91.9 (83.2 to 97.0) 94.9 (82.7 to 99.4) 88.6 (73.3 to 96.8) 100 (92.3 to 100)
database
Difference in sensitivity 4.1 (1.8 to 9.9) 5.1 (4.4 to 14.6) 2.8 (−5.5 to 11.2) 13.0 (1.1 to 24.9)
and specificity
P value .08 .16 .32 .01
RNFL Thickness Deviation Map
≥20 Superpixels RNFL
thickness below the lower
99th percentile
Cirrus HD-OCT normative 98.7 (92.7 to 100) 100 (91.0 to 100) 97.1 (85.1 to 99.9) 37.0 (23.2 to 52.5)
database
Myopic normative 94.6 (86.7 to 98.5) 100 (91.0 to 100) 88.6 (73.3 to 96.8) 89.1 (76.4 to 96.4)
database
Difference in sensitivity 4.1 (3.1 to 11.3) 0 (−2.6 to 2.6) 8.6 (6.5 to 23.6) 52.2 (35.6 to 68.8)
and specificity
P value .18 >.99 .18 <.001
≥20 Superpixels RNFL
thickness below the lower
95th percentile
Cirrus HD-OCT normative 100 (95.1 to 100) 100 (91.0 to 100) 100 (90.0 to 100) 8.7 (2.4 to 20.8)
database
Myopic normative 100 (95.1 to 100) 100 (91.0 to 100) 100 (90.0 to 100) 63.0 (47.6 to 76.8)
database
Abbreviations: dB, decibel; MD, mean
Difference in sensitivity 0 (−1.4 to 1.4) 0 (−2.6 to 2.6) 0 (−2.9 to 2.9) 54.3 (37.8 to 70.9)
and specificity deviation; HD-OCT, high-definition
optical coherence tomography;
P value >.99 >.99 >.99 <.001
RNFL, retinal nerve fiber layer.

1036 JAMA Ophthalmology September 2016 Volume 134, Number 9 (Reprinted) jamaophthalmology.com

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021

 Retinal Nerve Fiber Layer Imaging in Myopia
Figure 2. Overlay of Retinal Nerve Fiber Layer (RNFL) Thickness Deviation Maps
A Cirrus HD-OCT normative database
25
20
15
10
5 5
0 0
A, Frequency distribution map of RNFL abnormalities (with reference to the
lower 99th percentiles) constructed by overlaying the RNFL thickness deviation
maps analyzed by the Cirrus high-definition optical coherence tomography
(HD-OCT) normative database in the control group (n = 46). B, Frequency
normative database, 100%; Cirrus HD-OCT database, 91.9%;
P = .01) and the criterion of the mean cRNFL thickness
below the lower 99th percentile (early glaucoma: myopic
normative database, 77.1%; Cirrus HD-OCT database, 54.3%;
P = .01; all glaucoma: myopic normative database, 86.5%;
Cirrus HD-OCT database, 71.6%; P = .002). When the clock-
hour and mean cRNFL abnormality criteria were defined
with reference to the lower 95th percentile values, the
myopic normative database showed higher specificities
(63.0% and 100%, respectively) than did the Cirrus HD-OCT
normative database (30.4% [P = .001] and 87.0% [P = .01],
respectively) at comparable sensitivities (P ≥ .08).
Specificities and Sensitivities for Detection of RNFL
Abnormalities in the RNFL Thickness Map
When RNFL abnormalities were defined with reference to 20
superpixels or more of RNFL thickness below the lower 99th
percentiles in the RNFL thickness deviation map, the myopic
normative database showed a higher specificity (89.1%) than
did the Cirrus HD-OCT normative database (37.0%) (P < .001)
at similar sensitivities (94.6% and 98.7%, respectively; P = .18).
Similar findings were observed when the criterion of the 95th
percentile was applied (Table 2).
Figure 2 shows the distribution maps for frequency of
RNFL abnormalities of the control group (ie, false-positive
detection) constructed by overlaying their RNFL thickness
deviation maps with reference to the lower 99th percen-
tiles. Retinal nerve fiber layer abnormalities were located
largely at the inferior and the nasal quadrants followed by
the superior quadrant when the Cirrus HD-OCT normative
database was applied (Figure 2A). By contrast, superpixels
of RNFL abnormalities appeared to be randomly distributed
w h e n t h e myo p i c n o r m at ive d at a b a s e w a s a p p l i e d
(Figure 2B).
jamaophthalmology.com
Copyright 2016 American Medical Association. All rights reserved.
Downloaded From: https://jamanetwork.com/ on 05/07/2021
Original Investigation Research
B Myopic normative database
25
20
15
10
distribution map of RNFL abnormalities (with reference to the lower 99th
percentiles) constructed by overlaying the RNFL thickness deviation maps
analyzed by the myopic normative database in the control group (n = 46).
Positive and Negative Predictive Values
The positive predictive values of the myopic normative data-
base were higher (81.3%-100%) than those of the Cirrus
HD-OCT normative database (63.8%-91.6%) except for the cri-
terion of mean cRNFL thickness below the lower 99th percen-
tile, in which both normative databases had a positive predic-
tive value of 100% (Table 3). The negative predictive values
were largely comparable between the normative databases, al-
though the myopic normative database exhibited higher nega-
tive predictive values for the criteria based on the mean cRNFL
thickness below the lower 95th and the lower 99th percentile
(myopic normative database, 88.5%; Cirrus HD-OCT norma-
tive database, 81.6%; P = .03 for the lower 95th percentile; my-
opic normative database, 82.1%; Cirrus HD-OCT normative
database, 68.7%; P = .001 for the lower 99th percentile).
Discussion
We demonstrated that integrating a myopic normative data-
base for analysis of RNFL thickness is relevant for detection
of RNFL abnormalities in eyes with high myopia. Using dif-
ferent criteria to define RNFL abnormalities in the cRNFL
profile and in the RNFL thickness map, the myopic norma-
tive database always showed a higher specificity (63.0%-
100%) than did the Cirrus HD-OCT built-in normative data-
base (8.7%-87.0%) at similar sensitivities for detection of
RNFL abnormalities in eyes with high myopia, except for the
criterion of mean RNFL thickness below the lower 99th per-
centile, in which both normative databases showed a speci-
ficity of 100%. Our finding underscores the importance of
incorporating a myopic normative database in OCT instru-
ments for evaluation of RNFL measurements in individuals
with high myopia.
(Reprinted) JAMA Ophthalmology September 2016 Volume 134, Number 9 1037
 Research Original Investigation Retinal Nerve Fiber Layer Imaging in Myopia

malities were based on 1 clock-hour or more of cRNFL thick-

Table 3. Positive and Negative Predictive Values of the Cirrus HD-OCT
and the Myopic Normative Databases for Detection of RNFL nesses below the lower 99th and 95th percentiles, respec-
Abnormalitiesa tively (Table 2). False-positive errors were more obvious in the
analysis of the RNFL thickness map, in which 63.0% and 91.3%
Predictive Value, % (95% CI)
of eyes with high myopia in healthy individuals had at least
Criteria of RNFL Abnormalities Positive Negative
20 contiguous superpixels of RNFL thickness below the lower
Circumpapillary RNFL Thickness Profile
99th and 95th percentiles, respectively. As shown in the dis-
≥1 Clock-hour of RNFL thickness
below the lower 99th percentile tribution map of the frequency of RNFL abnormalities
Cirrus HD-OCT normative 77.3 (67.1-85.5) 81.3 (63.6-92.8) (Figure 2A), the inferior and nasal RNFL measurements are
database more likely to be rendered abnormal by the Cirrus HD-OCT nor-
Myopic normative database 92.5 (84.4-97.2) 100 (91.2-100)
mative database. Although the nasal RNFL measurements are
P value <.001 .09 not typically affected in eyes with early and moderate glau-
≥1 Clock-hour of RNFL thickness coma, the high frequency of false-positive errors at the infe-
below the lower 95th percentile
Cirrus HD-OCT normative 69.8 (60.1-78.4) 100 (76.8-100)
rior quadrant has weakened the diagnostic performance of OCT
database to detect RNFL abnormalities in eyes with high myopia.
Myopic normative database 81.3 (71.8-88.7) 100 (88.4-100) The myopic normative database (n = 180) differs from
P value <.001 >.99 the Cirrus HD-OCT built-in normative database (n = 271)
Mean RNFL thickness below the because it has eyes with a longer mean (SD) AL (26.7 [1.0]
lower 99th percentile
vs 23.9 [1.0] mm, respectively) and a more negative mean
Cirrus HD-OCT normative 100 (93.3-100) 68.7 (56.2-79.4)
database (SD) spherical equivalent (–8.0 [1.8] vs –0.8 [2.0] D,
Myopic normative database 100 (94.4-100) 82.1 (69.6-91.1) respectively).13 We were not able to compare the measure-
P value >.99 .001 ments of RNFL thickness between the normative databases
Mean RNFL thickness below the because the Cirrus HD-OCT normative database was propri-
lower 95th percentile etary. The effect of applying the myopic normative database
Cirrus HD-OCT normative 91.6 (82.5-96.8) 81.6 (68.00-91.2) for analysis of RNFL thickness in eyes with high myopia is
database
signified by the decreases in false-positive errors in analysis
Myopic normative database 100 (94.7-100) 88.5 (76.6-95.7)
of both the cRNFL profile and the RNFL thickness map
P value .01 .03
(Table 2). The improvement in specificities did not come at
RNFL Thickness Deviation Map
the cost of a reduction in sensitivity for detection of glau-
≥20 Superpixels RNFL thickness
below the lower 99th percentile coma. In fact, an improvement in sensitivity was observed
Cirrus HD-OCT normative 71.6 (61.8-80.1) 94.4 (72.7-99.9) when RNFL abnormalities were defined using the criterion
database of 1 clock-hour or more of cRNFL thickness below the lower
Myopic normative database 93.3 (85.1-97.8) 91.1 (78.8-97.5) 99th percentile and the criterion of the mean cRNFL
P value <.001 .62 thickness below the lower 99th percentile.
≥20 Superpixels RNFL thickness Although the myopic normative database has consider-
below the lower 95th percentile
ably improved the specificity for detection of RNFL abnor-
Cirrus HD-OCT normative 63.8 (54.4-72.5) 100 (39.8-100)
database malities in eyes with high myopia, 13.0% and 10.9% of
Myopic normative database 81.3 (74.8-91.0) 100 (88.1-100) healthy high myopic eyes remained false positives for the
P value <.001 >.99 criteria of the circumpapillary clock hour (≥1 abnormal clock
hour of RNFL thickness) and the RNFL thickness deviation
Abbreviations: HD-OCT, high-definition optical coherence tomography;
RNFL, retinal nerve fiber layer. map (≥20 superpixels) using the lower 99th percentile cut-
a
Total of 74 eyes with glaucoma and 46 healthy eyes. offs, respectively. The fact that the criteria for detection of
RNFL abnormalities in eyes with high myopia have not been
optimized and that the influence of the orientation of the
Optical coherence tomography RNFL measurements are fovea relative to the Bruch membrane opening has not been
generally smaller in eyes with high myopia compared with the considered might have negatively influenced the diagnostic
healthy population. For instance, the mean (SD) cRNFL thick- performance of the myopic normative database.17,18
ness of the myopic normative database was 89.6 (7.2) μm, There are other limitations of the study. First, the applica-
whereas it was 97.2 (10.5) μm in a population-based study in tion of the myopic normative data was not age-matched because
healthy Chinese individuals with a mean spherical equiva- the current sample size may not be sufficient to derive reliable
lent of –0.39 D using the same OCT instrument and scanning RNFL thickness cutoff estimates at specific age ranges. Patients
protocol.16 The smaller RNFL measurements in eyes with high in the glaucoma group were older (mean [SD] age, 47.9 [8.6] years)
myopia may in part contribute to an increase in false-positive than individuals included in the myopic normative database (37.5
errors when a normative database without data on eyes with [9.0] years) and those in the control group (36.0 [12.3] years). Had
high myopia is applied for analysis of RNFL thickness. In our age matching been performed in the analysis of RNFL thickness
study, 43.5% and 69.6% of eyes with high myopia in healthy using by the myopic normative database, the sensitivity for de-
individuals had false-positive errors detected by the Cirrus tection of RNFL abnormalities in the glaucoma group could have
HD-OCT normative database when the criteria of RNFL abnor- been enhanced (older individuals have thinner RNFL19). Second,

1038 JAMA Ophthalmology September 2016 Volume 134, Number 9 (Reprinted) jamaophthalmology.com

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021

 Retinal Nerve Fiber Layer Imaging in Myopia Original Investigation Research

the Cirrus HD-OCT database did not match race/ethnicity in the
analysis of RNFL thickness. We contributed 63 healthy Chinese
individuals to the development of the Cirrus HD-OCT normative
database but the database also contained 208 healthy individu-
als from 3 other racial/ethnic groups (51 of African descent, 63
of European descent, and 35 of Hispanic descent), and the analy-
sis of RNFL thickness in the Cirrus HD-OCT did not take racial/
ethnic differences in RNFL thickness into consideration.13 Third,
the Cirrus HD-OCT did not correct for ocular magnification for
RNFL measurements. For example, the actual circumpapillary
scan diameter would be greater than 3.46 mm for eyes with an
AL longer than 24.46 mm and the relative increase in the scan
diameter can underestimate the cRNFL thicknesses. Fourth, our
analysis included only eyes with high myopia, as false-positive
errors are more frequently encountered in eyes with high myo-
pia than in those with mild to moderate myopia.7-12 Detecting
RNFL abnormalities is more difficult in eyes with high myopia
than in those with mild or moderate myopia. Finally, central cor-
neal thickness of the myopic normative database was not small
(mean [SD], 550.0 [37.0] μm), as what would have been expected
in eyes with high myopia. Nevertheless, among eyes of Chinese
individuals with myopia, central corneal thickness has been
shown to distribute across a wide range and did not correlate with
the degree of myopia.20
Conclusions
Our study provides data supporting the application of a my-
opic normative database to improve the specificity for detec-
tion of glaucomatous RNFL abnormalities in eyes with high
myopia. To our knowledge, none of the currently available OCT
instruments incorporates a specific normative database col-
lected from individuals with high myopia for analysis of RNFL
thickness, although the retina scan–OCT (Nidek Co, Ltd) has
a normative database for macula map analysis including eyes
of Asian individuals with AL between 26 and 29 mm.21 With a
high prevalence of myopia in Asia, there is an unmet need for
implementation of myopic normative databases in OCT
instruments for analysis of glaucoma.

ARTICLE INFORMATION
Correction: This article was corrected on
September 22, 2016, to fix the Conflict
of Interest Disclosures section.
Accepted for Publication: May 27, 2016.
Published Online: July 21, 2016.
doi:10.1001/jamaophthalmol.2016.2343.
Author Contributions: Dr Leung and Mr Biswas
had full access to all the data in the study and take
responsibility for the integrity of the data and the
accuracy of the data analysis. Mrs Biswas and Chen
contributed equally to the article.
Study concept and design: Leung.
Acquisition, analysis, or interpretation of data: All
authors.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important
intellectual content: Biswas, Leung.
Statistical analysis: All authors.
Administrative, technical, or material support: Lin,
Leung.
Study supervision: Leung.
Conflict of Interest Disclosures: The authors have
completed and submitted the ICMJE Form for
Disclosure of Potential Conflicts of Interest.
Dr Leung reported receiving speaker honorarium
and research support from Carl Zeiss Meditec. No
other disclosures were reported.
REFERENCES
1. Chang RT, Singh K. Myopia and glaucoma:
diagnostic and therapeutic challenges. Curr Opin
Ophthalmol. 2013;24(2):96-101.
diagnostic performance study. Ophthalmology.
2009;116(7):1257-1263, 1263.e1-1263.e2.
5. Kim JS, Ishikawa H, Sung KR, et al. Retinal nerve
fibre layer thickness measurement reproducibility
improved with spectral domain optical coherence
tomography. Br J Ophthalmol. 2009;93(8):1057-
1063.
6. Mwanza JC, Gendy MG, Feuer WJ, Shi W, Budenz
DL. Effects of changing operators and instruments
on time-domain and spectral-domain OCT
measurements of retinal nerve fiber layer thickness.
Ophthalmic Surg Lasers Imaging. 2011;42(4):328-337.
7. Leung CK, Mohamed S, Leung KS, et al. Retinal
nerve fiber layer measurements in myopia: an
optical coherence tomography study. Invest
Ophthalmol Vis Sci. 2006;47(12):5171-5176.
8. Vernon SA, Rotchford AP, Negi A, Ryatt S,
Tattersal C. Peripapillary retinal nerve fibre layer
thickness in highly myopic Caucasians as measured
by Stratus optical coherence tomography. Br J
Ophthalmol. 2008;92(8):1076-1080.
9. Aref AA, Sayyad FE, Mwanza JC, Feuer WJ,
Budenz DL. Diagnostic specificities of retinal nerve
fiber layer, optic nerve head, and macular ganglion
cell-inner plexiform layer measurements in myopic
eyes. J Glaucoma. 2014;23(8):487-493.
10. Leal-Fonseca M, Rebolleda G, Oblanca N,
Moreno-Montañes J, Muñoz-Negrete FJ.
A comparison of false positives in retinal nerve fiber
layer, optic nerve head and macular ganglion
cell-inner plexiform layer from two spectral-domain
optical coherence tomography devices. Graefes
Arch Clin Exp Ophthalmol. 2014;252(2):321-330.
13. Knight OJ, Girkin CA, Budenz DL, Durbin MK,
Feuer WJ; Cirrus OCT Normative Database Study
Group. Effect of race, age, and axial length on optic
nerve head parameters and retinal nerve fiber layer
thickness measured by Cirrus HD-OCT. Arch
Ophthalmol. 2012;130(3):312-318.
14. Leung CK, Yu M, Weinreb RN, et al. Retinal
nerve fiber layer imaging with spectral-domain
optical coherence tomography: interpreting the
RNFL maps in healthy myopic eyes. Invest
Ophthalmol Vis Sci. 2012;53(11):7194-7200.
15. Greenfield DS, Siatkowski RM, Glaser JS, Schatz
NJ, Parrish RK II. The cupped disc: who needs
neuroimaging? Ophthalmology. 1998;105(10):1866-
1874.
16. Tun TA, Sun CH, Baskaran M, et al.
Determinants of optical coherence
tomography-derived minimum neuroretinal rim
width in a normal Chinese population. Invest
Ophthalmol Vis Sci. 2015;56(5):3337-3344.
17. Chauhan BC, Burgoyne CF. From clinical
examination of the optic disc to clinical assessment
of the optic nerve head: a paradigm change. Am J
Ophthalmol. 2013;156(2):218-227.e2.
18. He L, Ren R, Yang H, et al. Anatomic vs.
acquired image frame discordance in spectral
domain optical coherence tomography minimum
rim measurements. PLoS One. 2014;9(3):e92225.
19. Leung CK, Yu M, Weinreb RN, et al. Retinal
nerve fiber layer imaging with spectral-domain
optical coherence tomography: a prospective
analysis of age-related loss. Ophthalmology. 2012;
119(4):731-737.

2. Hsu CH, Chen RI, Lin SC. Myopia and glaucoma:
sorting out the difference. Curr Opin Ophthalmol.
2015;26(2):90-95.
3. Chang L, Pan CW, Ohno-Matsui K, et al.
Myopia-related fundus changes in Singapore adults
with high myopia. Am J Ophthalmol. 2013;155(6):
991-999.e1.
4. Leung CK, Cheung CY, Weinreb RN, et al. Retinal
nerve fiber layer imaging with spectral-domain
optical coherence tomography: a variability and
11. Yamashita T, Kii Y, Tanaka M, et al. Relationship
between supernormal sectors of retinal nerve fibre
layer and axial length in normal eyes. Acta
Ophthalmol. 2014;92(6):e481-e487.
12. Kim KE, Jeoung JW, Park KH, Kim DM, Kim SH.
Diagnostic classification of macular ganglion cell
and retinal nerve fiber layer analysis: differentiation
of false-positives from glaucoma. Ophthalmology.
2015;122(3):502-510.
20. Fam HB, How AC, Baskaran M, Lim KL, Chan
YH, Aung T. Central corneal thickness and its
relationship to myopia in Chinese adults. Br J
Ophthalmol. 2006;90(12):1451-1453.
21. Nakanishi H, Akagi T, Hangai M, et al. Sensitivity
and specificity for detecting early glaucoma in eyes
with high myopia from normative database of
macular ganglion cell complex thickness obtained
from normal non-myopic or highly myopic Asian
eyes. Graefes Arch Clin Exp Ophthalmol. 2015;253
(7):1143-1152.

jamaophthalmology.com (Reprinted) JAMA Ophthalmology September 2016 Volume 134, Number 9 1039

Copyright 2016 American Medical Association. All rights reserved.

Downloaded From: https://jamanetwork.com/ on 05/07/2021