agreement among spectral-domain optical coherence tomography instruments for assessing retinal nerve...
TRANSCRIPT
●
ra(●
●
ciGSRBdom●
tApgSaaeS.iVi●
tccimab
A
URdF
s0
0d
Agreement Among Spectral-Domain Optical CoherenceTomography Instruments for Assessing Retinal Nerve
Fiber Layer Thickness
MAURO T. LEITE, HARSHA L. RAO, ROBERT N. WEINREB, LINDA M. ZANGWILL, CHRISTOPHER BOWD,
PAMELA A. SAMPLE, ALI TAFRESHI, AND FELIPE A. MEDEIROSAtQcoEOaRhrtn
atCEvaicpbilsTR
aatlm
●
s
PURPOSE: To assess the agreement of parapapillaryetinal nerve fiber layer (RNFL) thickness measurementsmong 3 spectral-domain optical coherence tomographySD-OCT) instruments.
DESIGN: Observational, cross-sectional study.METHODS: Three hundred thirty eyes (88 with glau-
oma, 206 glaucoma suspects, 36 healthy) from 208ndividuals enrolled in the Diagnostic Innovations inlaucoma Study (DIGS) were imaged using RTVue,pectralis and Cirrus in a single visit. Agreement amongNFL thickness measurements was assessed usingland-Altman plots. The influence of age, axial length,isc size, race, spherical equivalent, and disease severityn the pairwise agreements between different instru-ents was assessed by regression analysis.RESULTS: Although RNFL thickness measurements be-
ween different instruments were highly correlated, Bland-ltman analyses indicated the presence of fixed androportional biases for most of the pairwise agreements. Ineneral, RTVue measurements tended to be thicker thanpectralis and Cirrus measurements. The agreement inverage RNFL thickness measurements between RTVuend Spectralis was affected by age (P � .001) and sphericalquivalent (P < .001), whereas the agreement betweenpectralis and Cirrus was affected by axial length (P �
004) and spherical equivalent (P < .001). Disease severitynfluenced the agreement between Spectralis and both RT-ue and Cirrus (P � .001). Disc area and race did not
nfluence the agreement among the devices.CONCLUSIONS: RNFL thickness measurements ob-
ained by different SD-OCT instruments were not entirelyompatible and therefore they should not be used inter-hangeably. This may be attributable in part to differencesn RNFL detection algorithms. Comparisons with histologiceasurements could determine which technique is most
ccurate. (Am J Ophthalmol 2011;151:85–92. © 2011y Elsevier Inc. All rights reserved.)
ccepted for publication Jun 30, 2010.From the Hamilton Glaucoma Center, Department of Ophthalmology,niversity of California San Diego, La Jolla, California (M.T.L., H.L.R.,.N.W., L.M.Z., C.B., P.A.S., A.T., F.A.M.); and Universidade Federale São Paulo, Department of Ophthalmology, São Paulo, Brazil(M.T.L.,.A.M.).Inquiries to Felipe A. Medeiros, Hamilton Glaucoma Center, Univer-
tity of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-946; e-mail: [email protected]
© 2011 BY ELSEVIER INC. A002-9394/$36.00oi:10.1016/j.ajo.2010.06.041
LTHOUGH THE ASSESSMENT OF THE PARAPAPIL-
lary retinal nerve fiber layer (RNFL) is essential indiagnosis and management of glaucoma, its objec-
ive evaluation remains a challenge in clinical practice.1
uantitative measurements of RNFL thickness have be-ome possible with the development of imaging technol-gies, such as optical coherence tomography (OCT).arlier versions of this technology, known as time-domainCT (TD-OCT), have demonstrated good reproducibility
nd accuracy for detection of RNFL loss in glaucoma.2–4
ecently, the introduction of spectral-domain optical co-erence technology (SD-OCT) has greatly enhanced theesolution and decreased scan acquisition times comparedo TD-OCT,5,6 potentially improving the ability to diag-ose and follow glaucoma.7–12
Three of the current commercially available SD-OCTsre the RTVue (Optovue Inc, Fremont, California, USA),he Cirrus SD-OCT (Carl Zeiss Meditec, Inc, Dublin,alifornia, USA), and the Spectralis OCT (Heidelbergngineering, Dossenheim, Germany). The principle in-olved in image acquisition is similar for all these devicesnd involves a scan with a diode laser that collectsnformation of RNFL thickness in a 3.4-mm-diameterircle centered on the optic disc. Although the workingrinciples are similar among the SD-OCTs, the agreementetween them has not yet been reported. With an increas-ng number of commercially available SD-OCTs, it isikely that patients examined with 1 machine will haveubsequent examinations performed with another device.herefore, it is important to evaluate the agreement inNFL thickness measurements among those devices.The purpose of the present study was to assess the
greement of parapapillary RNFL thickness measurementsmong the RTVue, Cirrus, and Spectralis, and to evaluatehe influence of age, race, spherical equivalent, axialength, disease severity, and optic disc size on the agree-ent among the devices.
METHODS
SUBJECTS: This was an observational cross-sectionaltudy. Subjects included in this study were recruited from
he longitudinal Diagnostic Innovations in GlaucomaLL RIGHTS RESERVED. 85
SC
lcpadSZIcITocggd
vwodw
cishmsanhandNpedrtrfohn
●
ltZnp
ciqqrqiimea
lh0sgadaada
MwAeot23dnna
HlniotiCass3ttf�
●
8
tudy (DIGS) conducted at the Hamilton Glaucomaenter (University of California, San Diego).Each participant underwent a comprehensive ophthalmo-
ogic examination including review of medical history, best-orrected visual acuity, slit-lamp biomicroscopy, intraocularressure measurement, gonioscopy, dilated funduscopic ex-mination with a 78-diopter (D) lens, stereoscopic opticisc photography, and automated perimetry with the 24-2wedish Interactive Threshold Algorithm (SITA; Carleiss Meditec, Inc). Axial length was acquired withOLMaster (Carl Zeiss Meditec). Optic disc area wasalculated using the Heidelberg Retina Tomograph (HRTI; software version 3.1.2.0, Heidelberg Engineering).hree scans centered on the optic disc were automaticallybtained for each test eye, and a mean topography wasreated. Trained technicians outlined the optic disc mar-in while they viewed simultaneous stereoscopic photo-raphs of the optic disc. Only images with a standardeviation of �50 �m were included.To be included, subjects had to have best-corrected
isual acuity of 20/40 or better, spherical refractionithin �5.0 D, cylinder correction within �3.0 D, andpen angles on gonioscopy. Eyes with coexisting retinalisease, uveitis, or nonglaucomatous optic neuropathyere also excluded from the investigation.To study the agreement among the 3 devices in a broad
ohort of patients, we included glaucomatous patients,ndividuals suspected of having the disease, and normalubjects. To be classified as glaucomatous, patients had toave at least 2 consecutive and reliable standard auto-ated perimetry (SAP) examinations with either a pattern
tandard deviation (PSD) outside the 95% normal limits orglaucoma hemifield test (GHT) result outside the 99%
ormal limits. Patients considered suspect for glaucomaad either an IOP greater than 21 mm Hg or suspiciousppearance of the optic nerve head with at least 2 reliableormal visual fields, defined as a PSD within 95% confi-ence limits and a GHT result within normal limits.ormal control subjects were recruited from the general
opulation and had IOP �22 mm Hg with no history oflevated IOP and with at least 2 reliable normal visual fields,efined as a PSD within 95% confidence limits and a GHTesult within normal limits. All visual fields were reviewed byhe VisFACT (Visual Field Assessment CenTer) visual fieldeading center. VisFACT checked for the presence of arti-acts such as lid and rim artifacts, fatigue effects, inattention,r inappropriate fixation. To be considered reliable, all testsad to have false-positive responses, fixation loss, and false-egative responses �33%.
INSTRUMENTATION: Measurements of the parapapil-ary RNFL thickness were obtained in the same visit usinghe RTVue (Optovue Inc), the Cirrus HD-OCT (Carleiss Meditec) and the Spectralis OCT (Heidelberg Engi-eering). Patients were examined without correction. Para-
apillary RNFL thickness parameters that were automatically mAMERICAN JOURNAL OF6
alculated by the machines and investigated in this studyncluded average thickness (360-degree measure), temporaluadrant thickness (316 degrees to 45 degrees), superioruadrant thickness (46 degrees to 135 degrees), nasal quad-ant thickness (136 degrees to 225 degrees), and inferioruadrant thickness (226 degrees to 315 degrees). Allncluded images were checked for motion artifacts bynspection of the continuity of the scanned images (align-ent of blood vessels). In addition, all images that had
rrors on RNFL segmentation were excluded from thenalysis.
The RTVue (software version 4.0.5.39) uses a scanningaser diode with a wavelength of 840 nm to provideigh-resolution images and has an acquisition rate of 2600 A-scans per second. The imaging protocol used in thistudy was ONH (optic nerve head scan). This protocolenerates a polar RNFL thickness map that is measuredlong a circle 3.45 mm in diameter centered on the opticisc. The RNFL thickness parameters are measured byssessing a total of 2325 data points between the anteriornd posterior RNFL borders. Only good-quality images, asefined by a signal strength index of �30, were used fornalysis.
The Cirrus SD-OCT (software version 4.5; Carl Zeisseditec) uses a superluminescent diode laser with a centeravelength of 840 nm and an acquisition rate of 27 000-scans per second. The protocol used for RNFL thickness
valuation was the optic disc cube. This protocol is basedn a tridimensional scan of a 6 � 6-mm2 area centered onhe optic disc where information from a 1024 (depth) �00 � 200-point parallelepiped is collected. Then, a.46-mm-diameter circular scan is placed around the opticisc and the information about parapapillary RNFL thick-ess is obtained. To be included, images were reviewed foroncentered scans and had to have a signal strength �7nd the absence of movement artifact.
Spectralis OCT (software version 3.1; Model SpectralisRA�OCT) uses a dual-beam SD-OCT and a confocal
aser scanning ophthalmoscope (CSLO) that uses a scan-ing laser diode with a wavelength of 870 nm and an
nfrared reference image simultaneously to provide imagesf ocular microstructures. The instrument has an acquisi-ion rate of 40 000 A-scans per second. Spectralis OCTncorporates a real-time eye tracking system that couplesSLO and SD-OCT scanners to adjust for eye movements
nd to ensure that the same location of the retina iscanned over time. The protocol used was the RNFL circlecan, which consists of 1024 A-scan points from a.45-mm circle centered on the optic disc. All patients hadheir corneal curvature inputted into the machine beforehe examination. To be included, all images were reviewedor noncentered scans and had to have a signal strength
15 dB.
STATISTICAL ANALYSIS: The agreement among RNFL
easurements obtained by the different instruments wasOPHTHALMOLOGY JANUARY 2011
ibatmbrltdvwuArapwewlttt
iei
cd
ebsiap
cCIw
T
gTiabd�f
V
nvestigated using Bland-Altman plots.13 The differencesetween measurements for each parameter were plottedgainst their mean. These plots allowed us to determinehe existence of any systematic differences between theeasurements (ie, a fixed bias). The mean difference
etween RNFL measurements obtained by the instrumentseflects an estimation of the bias. In addition, we calcu-ated the 95% limits of agreement for each comparison;hat is, an estimate of how much the measurementsiffered in most individuals. Because both eyes per indi-idual were used for the analysis, the limits of agreementere corrected for multiple measurements per individual,sing a method described by Bland and Altman.14 Bland-ltman plots were also used to evaluate any possible
elationship of the difference between the measurementsnd their average (ie, a proportional bias). The presence ofroportional bias indicates that the limits of agreementill depend on the actual measurement. To formallyvaluate this relationship, the difference between methodsas regressed on their average. If the slope of the regression
ine was statistically significant, we considered the exis-ence of proportional bias. To assess correlation betweenhe 3 devices, we calculated the coefficients of determina-ion (R2) for pairwise measurements.
The influence of covariates on the agreement amongnstruments was assessed by regression analysis. The differ-nces in RNFL thickness measurements between pairs of
TABLE 1. Clinical and Demographic C
Normals n � 36
Age (years) 58 � 12
Sex (% male) 29.4
Race (% African-American) 50
Axial length (mm) 24.02 � 1.12
Disc area (mm2) 2.03 � 0.57
Mean deviation (dB) �0.2 � 0.96
Spherical equivalent (diopters) �0.24 � 2.07
Pseudophakia (%) 0
aMean � standard deviation values of age, ax
deviation reported by diagnostic group.
TABLE 2. Mean (95% Confidence Interval)Layer Thickness Me
RTVue
Average thickness (�m) 92 (91.2–94.2)
Superior thickness (�m) 110 (108.4–112
Temporal thickness (�m) 69 (67.8–70.6)
Inferior thickness (�m) 119 (116.5–121
Nasal thickness (�m) 72 (70.9–73.8)
nstruments were included as dependent variables and the s
AGREEMENT AMONG SPECTOL. 151, NO. 1
ovariates (age, axial length, spherical equivalent, race,isc size, mean deviation [MD]) as independent variables.Generalized estimating equations with robust standard
rrors were used to adjust for potential correlationsetween both eyes of the same individual. Our sampleize was sufficient to provide a narrow 95% confidencenterval of approximately �1.2 �m for the limits ofgreement for the parameter average thickness in allairwise comparisons.All statistical analyses were performed with commer-
ially available software (Stata version 10; StataCorp,ollege Station, Texas, USA, and SPSS ver 16.0; SPSS
nc, Chicago, Illinois, USA). The alpha level (type I error)as set at 0.05.
RESULTS
HE STUDY INCLUDED 330 EYES (88 WITH GLAUCOMA, 206
laucoma suspects, 36 normals) from 208 individuals.able 1 shows clinical and demographic characteristics of
ncluded subjects. Mean age was significantly differentmong groups; glaucomatous patients were older, followedy glaucoma suspects and normals (P � .001). Meanisease severity as measured by the visual field MD was0.2 dB for normals, �0.66 dB for suspects, and �5.07 dB
or glaucomatous (P � .001). Mean axial length was
cteristics of the Studied Populationa
uspects n � 206 Glaucomatous n � 88 P Value
64 � 12 69 � 10 �.001
39.8 53.4 .026
74.2 57.9 .002
23.99 � 1.1 24.07 � 1.27 .986
2.03 � 0.46 2.09 � 0.51 .802
�0.66 � 1.45 �5.07 � 5.43 �.001
�0.53 � 2.21 �0.445 � 1.77 .738
16.5 32.9 �.001
gth, disc area, spherical equivalent, and mean
ge and Quadrant-wise Retinal Nerve Fiberd by the 3 Devices
Cirrus Spectralis
83 (81.5–85.3) 85 (83.4–86.7)
101 (98.7–102.9) 99 (97–101.7)
58 (56.4–59) 66 (64.2–67.2)
104 (102–106.9) 111 (107.8–113.2)
68 (67.2–69.6) 65 (62.9–66.3)
hara
S
ial len
Averaasure
.3)
.2)
imilar for all groups; normals had a mean of 24.02 mm,
RAL-DOMAIN OCTS 87
s(sam.a
RTat
Smmcwtqqwqd
FRtbt
FRtbt
Blan
8
uspects had 23.99 mm, and glaucomatous had 24.07 mmP � .986). Disc area, as measured by the HRT, did notignificantly differ among groups: normals had a mean discrea of 2.03 mm2, glaucoma suspects had a mean of 2.03m2, and glaucomatous had a mean of 2.09 mm2 (P �
802). No difference was found in spherical equivalentmong groups (P � .738).
Table 2 shows mean values of average and quadrantNFL thickness parameters obtained by each instrument.able 3 shows the agreement between instruments. Forverage thickness, RTVue measurements were significantly
IGURE 1. Bland-Altman plot for the agreement betweenTVue and Cirrus for the average retinal nerve fiber layer
hickness parameter. The regression line of the differenceetween the measurements on their average is represented byhe solid line. Dashed lines show the 95% limits of agreement.
TABLE 3. Bland-Altman Regression-based 95% Limits of ARetinal Nerve Fi
Parameter Agreement
Mean
Difference P Va
Average thickness (�m) RTVue-Cirrus 9.77 �.0
RTVue-Spectralis 7.65 �.0
Cirrus-Spectralis �2.12 �.0
Superior thickness (�m) RTVue-Cirrus 9.61 �.0
RTVue-Spectralis 11.01 �.0
Cirrus-Spectralis 1.4 .0
Temporal thickness (�m) RTVue-Cirrus 11.52 �.0
RTVue-Spectralis 3.51 �.0
Cirrus-Spectralis �8.01 �.0
Inferior thickness (�m) RTVue-Cirrus 14.33 �.0
RTVue-Spectralis 8.34 �.0
Cirrus-Spectralis �5.99 �.0
Nasal thickness (�m) RTVue-Cirrus 3.96 �.0
RTVue-Spectralis 7.77 �.0
Cirrus-Spectralis 3.81 �.0
aCorrected for multiple measurements per individual according to
hicker than Cirrus (difference � 9.77 �m; P � .001) and C
AMERICAN JOURNAL OF8
pectralis (difference � 7.65 �m; P � .001) measure-ents. Statistically significantly thicker RTVue measure-ents were also found for all quadrants (P � .001)
ompared to Cirrus and Spectralis. Cirrus measurementsere significantly thinner than Spectralis for average
hickness (difference � 2.12 �m; P � .001), temporaluadrant (difference � 8.01 �m; P � .001), and inferioruadrant (difference � 5.99 �m; P � .001). Measurementsith Spectralis were thinner than Cirrus for the nasaluadrant (difference � 3.81 �m; P � .001). No significantifference was found for the superior quadrant between
IGURE 2. Bland-Altman plot for the agreement betweenTVue and Spectralis for the average retinal nerve fiber layer
hickness parameter. The regression line of the differenceetween the measurements on their average is represented byhe solid line. Dashed lines show the 95% limits of agreement.
ent for RTVue, Cirrus, and Spectralis for the Parapapillaryayer Thickness
Fixed Bias R2 P Value
Proportional
Bias
95% Limits of
Agreementa
YES 0.029 .002 YES �0.02 to 19.57
YES 0.04 .005 YES �3.68 to 18.98
YES 0.129 �.001 YES �13.13 to 8.89
YES 0.014 .06 NO �7.32 to 26.53
YES 0.096 �.001 YES �10.71 to 32.73
NO 0.063 �.001 YES �17.23 to 20.03
YES 0.013 .291 NO �4.56 to 27.60
YES 0.01 .296 NO �14.77 to 21.79
YES 0.056 �.001 YES �22.77 to 6.74
YES 0.018 .048 YES �3.47 to 32.12
YES 0.107 �.001 YES �12.5 to 29.17
YES 0.057 �.001 YES �24.37 to 12.39
YES 0.056 .003 YES �17.66 to 25.57
YES 0.042 .001 YES �14.73 to 30.27
YES 0.142 �.001 YES �22.38 to 29.99
d and Altman.
greember L
lue
01
01
01
01
01
73
01
01
01
01
01
01
01
01
01
irrus and Spectralis (P � .073).
OPHTHALMOLOGY JANUARY 2011
pespnRtrar(
a0(s
btb.b(badbpmt
sr(R(aeb
T
tAsbm
atv(aionratmbwtr
FCtbt
V
Table 3 also reports the evaluation of the presence ofroportional bias. Proportional bias was present if the differ-nce and average of measurements by 2 instruments wereignificantly correlated. We found proportional biases for allairwise measurements except for the superior RNFL thick-ess agreement between RTVue and Cirrus, the temporalNFL thickness agreement between RTVue and Cirrus, and
he temporal RNFL agreement between RTVue and Spect-alis. The agreement among the devices and the 95% limits ofgreement for the average RNFL thickness parameter areepresented graphically in Figures 1 (RTvue vs Cirrus), 2
IGURE 3. Bland-Altman plot for the agreement betweenirrus and Spectralis for the average retinal nerve fiber layer
hickness parameter. The regression line of the differenceetween the measurements on their average is represented byhe solid line. Dashed lines show the 95% limits of agreement.
TABLE 4. Correlation Among RTVue, Cirrus, andSpectralis for Average and Quadrant-wise Parapapillary
Retinal Nerve Fiber Layer Thickness
Parameter Agreement
Coefficient of
Determination (R2) P Value
Average thickness
(�m)
RTVue-Cirrus 0.87 �.001
RTVue-Spectralis 0.85 �.001
Cirrus-Spectralis 0.86 �.001
Superior thickness
(�m)
RTVue-Cirrus 0.80 �.001
RTVue-Spectralis 0.73 �.001
Cirrus-Spectralis 0.81 �.001
Temporal
thickness (�m)
RTVue-Cirrus 0.62 �.001
RTVue-Spectralis 0.58 �.001
Cirrus-Spectralis 0.71 �.001
Inferior thickness
(�m)
RTVue-Cirrus 0.84 �.001
RTVue-Spectralis 0.82 �.001
Cirrus-Spectralis 0.86 �.001
Nasal thickness
(�m)
RTVue-Cirrus 0.37 �.001
RTVue-Spectralis 0.49 �.001
Cirrus-Spectralis 0.30 �.001
RTVue vs Spectralis), and 3 (Cirrus vs Spectralis). i
AGREEMENT AMONG SPECTOL. 151, NO. 1
Table 4 shows the coefficients of determination (R2) forll pairwise comparisons. Correlations ranged from R2 �.30 (nasal quadrant, Cirrus-Spectralis) to R2 � 0.87average thickness, RTVue-Cirrus). All correlations weretatistically significant (P � .001).
Table 5 shows the effects of covariates on the agreementetween the devices for the parameter average RNFLhickness in univariable regression analysis. Agreementetween RTVue and Spectralis was affected by age (P �001) and spherical equivalent (P � .001); agreementetween Spectralis and Cirrus was affected by axial lengthP � .004) and spherical equivalent (P � .001); agreementetween Spectralis and both RTVue and Cirrus wasffected by disease severity (P � .001). For more severeisease, the difference between RTVue and Spectralis andetween Cirrus and Spectralis increased, with Spectralisroviding thinner RNFL thickness measurements withore advanced disease. Disc area and race did not affect
he agreement among the devices.Incorporating all covariates (age, MD, axial length, race,
pherical equivalent, and disc size) in multivariable modelsesulted in a significant influence of spherical equivalentP � .001) and MD (P � .001) on the agreement betweenTVue and Spectralis and a significant influence of MD
P � .001) and spherical equivalent (P � .04) on thegreement between Cirrus and Spectralis. No significantffect of the covariates was found on the agreementetween RTVue and Cirrus.
DISCUSSION
HE PRESENT STUDY EVALUATED THE AGREEMENT IN RNFL
hickness measurements obtained by 3 different SD-OCTs.lthough the technology used by these instruments is
imilar, important differences in measurements obtainedy these devices were found, indicating that the measure-ents should not be used interchangeably.The narrowest 95% confidence interval of limits of
greement was obtained for the parameter average RNFLhickness measurements in all pairwise comparisons. Pre-ious studies assessing the agreement between TD-OCTStratus OCT, Carl Zeiss Meditec) and RTVue also foundbetter agreement for average thickness.7,12 Although all
mages were checked for centering of the scans around theptic disc, the better agreement for the average thick-ess parameter compared to quadrant measures mayepresent the effect of small misplacements of the scanround the optic disc. It has been shown that noncen-ered scans induce a greater error in quadrant-wiseeasurements obtained with OCT.15 Despite showing
etter agreement, the RNFL average thickness measuresere still significantly different among the devices;
herefore, it is unlikely that this parameter would beobust enough to be used in the follow-up of patients
maged with different instruments.RAL-DOMAIN OCTS 89
swiiaaLupGuptteaeebFlbelad
mR
rtmstStbwnTPOaaRm
ptompSpmpmg
9
When evaluating whether the agreement between in-truments is acceptable or not, one must consider theithin-subject repeatability of the measurements. If an
nstrument has low repeatability, it is likely that compar-sons with other instruments are going to result in poorgreement. Previous studies have investigated the repeat-bility of measurements using the Cirrus and the RTVue.eung and associates10 reported an intra-visit repeatabilitysing the Cirrus of 5.12 �m for the average thicknessarameter in normal and glaucomatous patients. Gonzalez-arcia and associates7 found a repeatability coefficient
sing the RTVue of approximately 4.34 �m for normalarticipants and 4.68 �m for glaucomatous patients usinghe average thickness parameter. Considering the intra-est variability of approximately 5 �m for each instrument,ven if the instruments had perfect agreement, an error ofpproximately 10 �m (� 5�m) in the agreement could bexpected because of the variability of measurements. How-ver, if this difference is larger than 10 �m, other factorsesides repeatability are probably affecting the agreement.or the average thickness parameter, we obtained 95%imits of agreement of around 20 �m for the differencesetween instrument pairwise comparisons, nearly twice thexpected error from within-instrument variability. It isikely that other factors, such as differences in hardwarend software, may be affecting the agreement among theseevices.A consistent difference (fixed bias) between measure-ents was observed in almost all pairwise comparisons.
TABLE 5. Effect of Covariates on the Agreethe Average Retinal Nerve Fib
Agreement
Age (years) RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
Disc area (mm2) RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
Axial length (mm) RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
Mean deviation (dB) RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
Spherical equivalent (D) RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
Racea RTVue-Cirrus
RTVue-Spectra
Cirrus-Spectra
D � diopters; dB � decibels; RSE � robustaAfrican-Americans � 0; Caucasians � 1.
TVue measurements were consistently thicker than Cir- t
AMERICAN JOURNAL OF0
us and Spectralis for average and quadrant-wise RNFLhickness. Interestingly, the differences between measure-ents using the Cirrus and the Spectralis were not in the
ame direction for all parameters. For average thickness,emporal quadrant, and inferior quadrant thicknesses, thepectralis provided thicker readings, while the Cirrus gavehicker readings for nasal quadrant. This discrepancy maye explained by the angle of incidence of the laser beam,hich can affect RNFL measurements,16 especially in theasal quadrant where the scan light is generally dimmer.17
hese results are in concordance with the literature.revious studies evaluating the agreement between StratusCT and SD-OCTs showed that average RNFL thickness
s measured by the Stratus was thicker than the Cirrus10
nd thinner than the RTVue.7 Therefore, it seems that theTVue tends to report thicker RNFL thickness measure-ents compared to the other devices.In addition to fixed bias, we investigated the presence of
roportional bias. The existence of proportional bias indicateshat the difference between RNFL thickness measurementsbtained by the devices varies according to the actualeasurement. A study by Vizzeri and associates12 showed a
roportional bias for the agreement between Stratus and theD-OCTs. We found proportional biases for almost allairwise comparisons. Interestingly, proportional bias wasore pronounced for comparisons with Spectralis. For exam-
le, the RTVue obtained systematically thicker measure-ents than the Spectralis; however, this difference was
reater for thinner RNFL. A similar effect was observed for
t Among RTVue, Cirrus, and Spectralis foryer Thickness Measurements
Coefficient RSE P Value
�0.041 0.027 .120
�0.082 0.026 .001
�0.06 0.028 .124
0.50 0.56 .379
0.58 0.76 .441
0.10 0.70 .888
�0.415 0.278 .135
0.678 0.39 .083
1.084 0.377 .004
�0.009 0.064 .862
�0.401 0.113 .001
�0.388 0.118 .001
�0.085 0.143 .553
�0.707 0.15 �.001
�0.627 0.157 �.001
�0.335 0.699 .633
�0.542 0.752 .471
�0.122 0.721 .866
ard error.
mener La
lis
lis
lis
lis
lis
lis
lis
lis
lis
lis
lis
lis
stand
he agreement between Cirrus and Spectralis. The obvious
OPHTHALMOLOGY JANUARY 2011
iiarpmttdiedd
ftsaPanmilmcec
dis
tvrRfibftttialCsahepmatscu
sanm
PNirsfiE((oLD
V
mportance of these results is that the agreement amongnstruments is not the same in all ranges of RNFL thicknessesnd, therefore, it is not likely to be the same throughout theange of glaucoma severity. Although all instruments wereeriodically calibrated according to the manufactures’ recom-endations and only good-quality images were included in
his study, differences in laser output power may have affectedhe agreement in RNFL thickness measurements amongevices. In addition, it is possible that a floor effect is present,e, the ability to detect differences in thin RNFL, and is notqual for all instruments. However, no conclusion can berawn regarding accuracy of the measurements without airect comparison to histologic measurements.
As expected, a high correlation between the 3 devicesor most of the studied sectors was found. The averagehickness, inferior, and superior quadrant thicknesses, weretrongly correlated for all pairwise comparisons. However,
poor correlation was found for the nasal quadrant.revious studies comparing Stratus OCT and SD-OCTlso showed a weaker correlation and repeatability for theasal quadrant.7,9,12 One explanation is that measure-ents of the nasal quadrant may be influenced by the
ncidence angle of the laser beam, leading to a dimmeright in that quadrant, influencing the RNFL thicknesseasurements. It is important to note that the overall good
orrelation between measurements obtained by the differ-nt devices does not represent good agreement, as indi-ated by our study and discussed in the literature.13
The effect of covariates on the agreement between theevices was also investigated. Disc area and race had nonfluence on the agreement among the devices. Disease
everity influenced the agreement between RTVue and Spec- afiber layer, optic nerve head, and macular thickness
AGREEMENT AMONG SPECTOL. 151, NO. 1
ralis and between Cirrus and Spectralis. For more advancedisual field loss, the Spectralis gave proportionally thinnereadings than RTVue and Cirrus. Because visual loss andNFL thickness are closely related, the influence of visualeld loss may be explained by the existence of proportionalias related to RNFL thickness, as discussed above. We alsoound that spherical equivalent affected the agreement be-ween RTVue and Spectralis and between Cirrus and Spec-ralis. As the spherical equivalent became more negative (ie,oward myopia), the Spectralis gave relatively thinner read-ngs than RTVue and Cirrus. Axial length influenced thegreement between Cirrus and Spectralis; for greater axialengths, Spectralis gave thinner measurements compared toirrus, but this effect was not observed when we adjusted for
pherical equivalent. A study by Nagai-Kusuhara and associ-tes18 reported that eyes with greater axial length tend toave thinner RNFL, indicating that our findings may bexplained by thinner RNFL thickness. Alternately, it isossible that magnification errors might be affecting theeasurements obtained from each device and, thus, their
greement. It is difficult to separate the effect of actual RNFLhickness from spherical equivalent, axial length, and diseaseeverity on the agreement among the devices. However, it islear that these covariates should be considered when eval-ating patients with different devices.
In conclusion, RNFL thickness measurements betweeneveral SD-OCT instruments are not entirely compatible,nd therefore measurements from these instruments shouldot be used interchangeably. Comparisons with histologiceasurements could determine which technique is most
ccurate.
UBLICATION OF THIS ARTICLE WAS SUPPORTED IN PART BY NATIONAL EYE INSTITUTE, BETHESDA, MARYLAND (GRANTSEI EY08208 [P.A.S.] and NEI EY11008 [L.M.Z.]); CAPES Ministry of Education of Brazil (Grant BEX1327/09-7 [M.T.L.] and participant retention
ncentive grants in the form of glaucoma medication at no cost (Alcon Laboratories Inc, Allergan, Pfizer Inc, SANTEN Inc). Robert N. Weinreb haseceived financial support from, Optovue, Topcon, Heidelberg Engineering, and Carl Zeiss Meditec, Inc. Linda M. Zangwill has received financialupport from, Carl Zeiss Meditec, Inc, Heidelberg Engineering, Optovue, Inc, and Topcon Medical Systems, Inc. Pamela A. Sample has receivednancial support from, Carl Zeiss, Meditec, Inc, and Felipe A. Medeiros has received financial support from Carl Zeiss Meditec, Inc, Heidelbergngineering, Reichert, Inc. Involved in design of the study (M.T.L., F.A.M.); analysis and interpretation (M.T.L., H.L.R., F.A.M.); writing the articleM.T.L., H.L.R., F.A.M.); critical revision of the article and final approval (M.T.L., H.L.R., R.N.W., L.M.Z., C.B., P.A.S., A.T., F.A.M.); data collectionM.T.L., H.L.R., A.T.); provision of materials, patients, or resources (R.N.W., L.M.Z., P.A.S., C.B., F.A.M.); statistical expertise (M.T.L., F.A.M.);btaining funding (R.N.W., P.A.S., L.M.Z., F.A.M.); literature search (M.T.L., H.L.R.); and administrative, technical, or logistical support (R.N.W.,.M.Z., C.B., P.A.S., F.A.M.). This study was approved by the University of California San Diego Human Subjects Committee and adhered to theeclaration of Helsinki. Informed consent was obtained from all participants.
This is an original submission and has not been considered elsewhere.
REFERENCES
1. Herrmann J, Funk J. [Diagnostic value of nerve fibre layerphotography in glaucoma]. Ophthalmologe 2005;102(8):778–782.
2. Jaffe GJ, Caprioli J. Optical coherence tomography to detectand manage retinal disease and glaucoma. Am J Ophthalmol2004;137(1):156–169.
3. Medeiros FA, Zangwill LM, Bowd C, Vessani RM,Susanna R Jr, Weinreb RN. Evaluation of retinal nerve
measurements for glaucoma detection using opticalcoherence tomography. Am J Ophthalmol 2005;139(1):44 –55.
4. Wollstein G, Ishikawa H, Wang J, Beaton SA, SchumanJS. Comparison of three optical coherence tomographyscanning areas for detection of glaucomatous damage.Am J Ophthalmol 2005;139(1):39 – 43.
5. Nassif N, Cense B, Park B, et al. In vivo high-resolutionvideo-rate spectral-domain optical coherence tomography ofthe human retina and optic nerve. Opt Express 2004;
12(3):367–376.RAL-DOMAIN OCTS 91
1
1
1
1
1
1
1
1
1
9
6. Wojtkowski M, Srinivasan V, Fujimoto JG, et al. Three-dimensional retinal imaging with high-speed ultrahigh-reso-lution optical coherence tomography. Ophthalmology 2005;112(10):1734–1746.
7. Gonzalez-Garcia AO, Vizzeri G, Bowd C, Medeiros FA, Zang-will LM, Weinreb RN. Reproducibility of RTVue retinal nervefiber layer thickness and optic disc measurements and agree-ment with Stratus optical coherence tomography measure-ments. Am J Ophthalmol 2009;147(6):1067–1074.e1.
8. Johnson DE, El-Defrawy SR, Almeida DR, Campbell RJ.Comparison of retinal nerve fibre layer measurements from timedomain and spectral domain optical coherence tomographysystems. Can J Ophthalmol 2009;44(5):562–566.
9. Knight OJ, Chang RT, Feuer WJ, Budenz DL. Comparison ofretinal nerve fiber layer measurements using time domainand spectral domain optical coherent tomography. Ophthal-mology 2009;116(7):1271–1277.
0. Leung CK, Cheung CY, Weinreb RN, et al. Retinal nervefiber layer imaging with spectral-domain optical coherencetomography: a variability and diagnostic performance study.Ophthalmology 2009;116(7):1257–1263, 1263.e1–2.
1. Sung KR, Kim DY, Park SB, Kook MS. Comparison ofretinal nerve fiber layer thickness measured by Cirrus HDand Stratus optical coherence tomography. Ophthalmology
2009;116(7):1264–1270.e1.AMERICAN JOURNAL OF2
2. Vizzeri G, Weinreb RN, Gonzalez-Garcia AO, et al. Agreementbetween spectral-domain and time-domain OCT for measuringRNFL thickness. Br J Ophthalmol 2009;93(6):775–781.
3. Bland JM, Altman DG. Statistical methods for assessingagreement between two methods of clinical measurement.Lancet 1986;1(8476):307–310.
4. Bland JM, Altman DG. Agreement between methods ofmeasurement with multiple observations per individual.J Biopharm Stat 2007;17(4):571–582.
5. Vizzeri G, Bowd C, Medeiros FA, Weinreb RN, ZangwillLM. Effect of improper scan alignment on retinal nerve fiberlayer thickness measurements using Stratus optical coher-ence tomograph. J Glaucoma 2008;17(5):341–349.
6. Hong S, Kim CY, Seong GJ. Adjusted peripapillary retinalnerve fiber layer thickness measurements based on the opticnerve head scan angle. Invest Ophthalmol Vis Sci. 2010;51(8):4067–4074.
7. Knighton RW, Qian C. An optical model of the humanretinal nerve fiber layer: implications of directional reflec-tance for variability of clinical measurements. J Glaucoma2000;9(1):56–62.
8. Nagai-Kusuhara A, Nakamura M, Fujioka M, Tatsumi Y,Negi A. Association of retinal nerve fibre layer thicknessmeasured by confocal scanning laser ophthalmoscopy andoptical coherence tomography with disc size and axial
length. Br J Ophthalmol 2008;92(2):186–190.OPHTHALMOLOGY JANUARY 2011
Moa
V
Biosketch
auro T. Leite, MD, is a glaucoma specialist and a post-doctoral fellow at the Hamilton Glaucoma Center, Universityf California, La Jolla, California. Dr. Leite has received his medical degree and completed his residency in ophthalmologyt the Federal University of São Paulo, Brazil.
AGREEMENT AMONG SPECTRAL-DOMAIN OCTSOL. 151, NO. 1 92.e1