Reproducibility of Nerve Fiber Layer Thickness Measurements Using 3D Fourier-Domain OCT

Abstract

purpose. Conventional time-domain optical coherence tomography (OCT) has been shown to provide reproducible retinal nerve fiber layer (RNFL) measurements. Recently, high-speed, high-resolution Fourier-domain 3D-OCT has been introduced to improve OCT quality. It can provide 6-mm² high-density scans to provide RNFL thickness measurements. The purpose of this study was to test the reproducibility of 3D-OCT RNFL thickness measurements in healthy volunteers.

methods. Thirty-eight eyes were included in the study. High-density 6-mm² 3D scans were registered by two independent operators. RNFL thickness was calculated for eight areas corresponding to the ETDRS areas and for two ring areas. The ETDRS grid was centered on the optic disc. Intraclass correlation coefficients (ICC) and coefficients of variation (COV) were calculated. Interobserver reproducibility was visualized by using Bland-Altman analysis.

results. Intrasession reproducibility was good with a mean ICC of 0.90. The mean COV for operator 1 and 2 was 4.2% and 4%, respectively (range, 1.9%–6.7%). Highest reproducibility was found for the two ring areas and the superior and inferior quadrants. Mean differences in RNFL thickness measurements for ring 1 and 2 between operator 1 and 2 were 0.9 μm (limits of agreement, −11.4 to +9.6 μm) and 0.1 μm (limits of agreement −4.1 to +3.9 μm), respectively.

conclusions. 3D-OCT RNFL thickness measurements in healthy volunteers showed good intra- and interobserver reproducibility. 3D-OCT provides more RNFL thickness information compared to conventional time-domain OCT measurements and may be useful for the management of glaucoma and other optic neuropathies.

Evaluation of the retinal nerve fiber layer (RNFL) is fundamental for diagnosing and managing glaucoma and other optic neuropathies. In the past, RNFL could be assessed subjectively only by slit lamp examination. This method requires clinical experience and offers only qualitative data. In addition, comparisons over time are almost impossible. Successively, other techniques such as color photographs of the optic disc or red-free photographs of the RNFL have become available and have facilitated comparisons over time. Scanning laser ophthalmoscopy and scanning laser polarimetry were the first instruments to allow objective and quantitative evaluation of the RNFL and the optic disc. ¹  

Optical coherence tomography (OCT) was first introduced in 1995 as an imaging technique for glaucoma diagnosis. ² Studies have been conducted to investigate the reproducibility of OCT RNFL thickness measurements, to assess the value of OCT as a clinical tool for distinguishing between healthy and glaucomatous eyes. ^{3 4 5 6 7 8 9}  

However, in all previous studies, conventional time-domain OCT was used for testing the reproducibility of RNFL thickness measurements. Time-domain OCT uses a scanning interferometer and an 820-nm infrared light source that is split into two separate beams. One beam is scanning a tissue being analyzed, and the other one acts as a reference beam that is reflected by a reference mirror. The distance of the reference mirror can be adjusted, and therefore the time it takes for the reference beam to reach the sensor can be changed. By comparing the two light beams, time-domain OCT measures the optical backscattering of light to generate a cross-sectional image of the tested tissue. 

Recently, improvements in OCT technology have been introduced. ^{10 11} Fourier-domain (FD)-OCT provides increased resolution and scanning speed by recording the interferometric information with a Fourier-domain spectrometric method instead of adjusting the position of a reference mirror. Resolution is up to five times higher, and imaging speed is 60 times faster than in conventional time-domain OCT. ^{12 13}  

In addition to high image quality, it is important to have reliable and reproducible software programs to analyze the data acquired by FD-OCT. Previous versions of OCT (Stratus OCT3; Carl Zeiss Meditec, GmbH, Oberkochen, Germany) mostly used a 3.4-mm diameter circle scan centered on the optic disc to generate 512 A-scans. The RNFL thickness profile showed a characteristic curve with two peaks, one in the superior and one in the inferior quadrant. FD-OCT can perform a high-density raster-scan (512 × 128 axial B-scans in a 6 mm²area). Recently, the peripapillary nerve fiber layer thickness profile was determined with FD-OCT by using high-density scanning. ¹⁴ These raster scans provide considerably more data for RNFL thickness analysis. The purpose of this study was to test the reproducibility of RNFL thickness measurements in healthy subjects by using FD-OCT high-density raster scans (3D OCT1000; Topcon, Tokyo, Japan). 

Methods

Thirty-eight eyes of 19 healthy subjects (10 women) with a mean age of 26 ± 3 years were included in the study. Exclusion criteria were history of glaucoma, history of any other ocular disease, intraocular pressure greater than 21 mm Hg, or a refractive error of more than −5 or +5 D. FD-OCT high-density scans were performed with the 3D OCT1000 system. 

The 3D OCT1000 is an FD-OCT device providing OCT images up to 50 times faster than time-domain OCTs with a sweep-scan technique. The device has a field angle of 45° with a color fundus camera included. The scanning range of the device is from 3 to 6 mm². Horizontal resolution is ≤20 μm and depth resolution is up to 5 μm. As a light source, the system uses superluminescent diodes with a wavelength of 840 nm. 

Pupil diameter had to be at least 4 mm for scanning. High-density raster scans (512 × 128 B-scans in a 6 mm² area) were centered on the optic disc by moving the patient’s fixation target on the OCT observer screen. Scans were performed six times in one session by two operators (three scans each in changing order). All subjects gave informed consent to participate in the study, which adhered to the tenets of the Declaration of Helsinki. The FD-OCT software provides a quality (Q)-factor comparable to the scan strength number given in Stratus OCT3 for each examination. Scans with a Q-factor less than 45 were excluded, and measurements were repeated until six scans of good quality were acquired. In addition, scans with blinks during the scanning process were excluded and repeated. Eighteen scans had to be repeated because of low Q-factors or blinks (7.9%). The 3D OCT1000 system contains a high-resolution camera for color fundus pictures. Pictures are automatically taken after each examination. Before data analysis, stored infrared fundus images were registered with the corresponding color fundus image. Scans were automatically aligned to compensate for eye movement artifacts during the scanning process. The FD-OCT system provides a software algorithm for RNFL thickness measurements. Each high-density raster scan was separately analyzed by using the RNFL algorithm to generate RNFL thicknesses in micrometers. Mean RNFL thicknesses can be plotted as an area of 6 mm² containing 36 squares of mean RNFL thicknesses, or alternatively as nine areas corresponding to the nine ETDRS areas also known from the Stratus OCT3. The 3.4-mm circle scan for RNFL measurements known from the Stratus OCT was not available in the software version of the 3D OCT1000. To obtain good centration on the optic disc, it is beneficial to use a circle-shaped target area that can easily be centered on the optic disc. Therefore, for testing RNFL thickness reproducibility, the ETDRS plot (Fig. 1)was chosen, because one can easily center the inner ring of the plot on top of the optic disc. The inner circle of the ETDRS plot has a diameter of 500 μm. The middle circle represents a diameter of 3 mm and the outer circle represents a diameter of 6 mm. Both left and right eyes were analyzed. Therefore, data were adjusted so that all quadrants could be appropriately assessed. Left eyes were treated as mirror images of right eyes. In all tables, area 3 and 7 correspond to temporal quadrants and areas 5 and 9 correspond to nasal quadrants. Figure 1shows an example of an RNFL thickness measurement showing mean RNFL thicknesses for each of the nine ETDRS areas. The most inner ring (area 1) of the ETDRS plot was excluded from analysis as measuring RNFL thickness is not possible directly on the optic disc cup. In previous measurements, we observed that the 3D OCT1000 actually performed RNFL thickness measurements on the optic disc rim. Therefore, we decided to include areas 2 to 5 (inner ring) to test reproducibility of such measurements. Areas 2 to 5 are between the inner and middle rings and were included in the analysis even if parts of the optic disc rim crossed the inner ring. Areas 5 to 9 were unaffected by the optic disc and included in the analysis (Fig. 1) . 

For statistical analysis areas 2 to 9 were analyzed. In addition, mean RNFL thicknesses for the inner ring (ring 1, consisting of areas 2–5) and outer ring (ring 2, consisting of areas 6–9) were calculated. Square root of variance components and 95% confidence intervals (95% CI) were determined for subjects, eyes, operators, and scans using a linear mixed-effect model. In addition, the bias between operators was tested. Commercial software (Stata, ver. 9.2; StataCorp, College Station, TX) was used for analysis. Three different kinds of intraclass correlation coefficients (ICCs) were determined: ICC¹ (for measurements within the same subject, eye and operator), ICC² (interoperator), and ICC³ (intraoperator). 

Interobserver reproducibility was visualized by providing limits of agreement in Bland-Altman plots to compare every first measurement of operator 1 and 2 of the inner and outer rings and areas 2 to 9, based on the assumption of equal imprecision between operators. Additional limits of agreement were provided to take into account that eyes were nested within subjects. Bland-Altman plots were created with commercial software (version 9.3.9.0; MedCalc, Mariakerke, Belgium). Coefficients of variation (COV) were determined for each area and both rings for operators 1 and 2. 

Results

RNFL thickness measurements were performed three times by each operator. Mean RNFL thicknesses values were calculated out of the three measurements for areas 2 to 9 and rings 1 and 2 separately for operators 1 and 2 (Table 1) . 

RNFL thicknesses were higher in the superior and inferior quadrants (areas 2, 4, 6, and 8) compared with the temporal and nasal quadrants. Regardless of the observer, mean RNFL thicknesses were very similar. COVs of measurements for each area are shown in Table 2 . Square root of variance components with 95% confidence intervals for patients, eyes, operator, and residuals and the corresponding ICCs are shown in Table 3 . 

Reproducibility was good with a mean ICC¹, ICC², and ICC³ of 0.9, 0.88, and 0.9, respectively. Mean COV for operators 1 and 2 was 4.2% and 4%, respectively. COVs ranged from 1.9% to 6.7% (Table 2) . Best reproducibility was found for rings 1 and 2 with highest ICCs (ring 1: 0.95; ring 2: 0.96), and lowest COVs (operator 1 for ring 1: 2.9%; ring 2: 1.9%; operator 2 for ring 1: 2.3%; ring 2: 2.0%). Lowest reproducibility was found for areas 5, 9 (nasal quadrants), and 7 (outer temporal area) with lowest ICCs (areas 5: 0.82; 7: 0.8; and 9: 0.81) and highest COVs (operator 1 for areas 5: 6.8%; 7: 4.1%; and 9: 6.5%; operator 2 for areas 5: 6.2%; 7: 4.5%; and 9: 6.4%). 

Mean differences in RNFL thickness measurements for rings 1 and 2 between operators 1 and 2 were 0.9 μm (range, −14.0 to +9.8 μm) and 0.13 μm (range, −4 to +4.8 μm), respectively. To assess interobserver reproducibility, limits of agreement were provided as 2× SD with upper and lower limits of the differences between measurements of operators 1 and 2 for RNFL thickness measurements of rings 1 and 2. For ring 1, bias was −0.9 μm with limits of agreement from −11.4 to +9.6 μm. For ring 2 bias was −0.1 μm with limits of agreement from −4.1 to +3.9 μm. Figure 2shows Bland-Altman plots for RNFL measurements of rings 1 and 2, to demonstrate interobserver differences. Bland-Altman plots for all other areas tested looked similar. In addition, adjusted limits of agreement were provided to take into account that eyes were nested within subjects. No significant bias was found between operators (P > 0.09). 

Discussion

FD-OCT represents the latest commercially available generation of OCT. With higher axial resolution and higher scan acquisition speed FD-OCT enables high-density scanning of larger retinal areas compared with conventional time-domain OCT. Before FD-OCT, only circle scans could be registered to calculate RNFL thickness at certain points around the disc. RNFL thickness maps were available with Stratus OCT3, but data were interpolated out of only three circle scans with increasing diameters, centered on the disc. This can cause considerable error. In addition, localized RNFL defects can be missed. With higher scan density of FD-OCT, RNFL thickness information of a 6-mm² area around the disc becomes available. This additional information may be helpful in diagnosing and observing glaucoma or other diseases that might affect the RNFL. Reproducibility of any diagnostic test is important for diagnostic accuracy. Especially in glaucoma, reproducibility of RNFL measurements is critical if the device is used to monitor progression of the disease. Consequently, the goal of this study was to determine the reproducibility of high-density FD-OCT RNFL thickness measurements. Data on the reproducibility and reliability of first-, second-, and third-generation OCT (time-domain OCT) RNFL thickness measurements have been reported before in normal and glaucomatous eyes. To our knowledge, we are reporting the first data on reproducibility of high density FD-OCT RNFL thickness measurements in healthy subjects. 

Schuman et al. ¹⁵ used first-generation OCT, performing a 3.4-mm circle scan on 11 normal subjects and 10 with glaucoma on five separate occasions. He reported ICCs of 0.56 and 0.52, respectively. The authors concluded that measurements are reproducible, particularly when stable fixation could be maintained during measurements. Carpineto et al. ⁷ used first-generation OCT to test reproducibility in 24 patients with glaucoma compared with 24 age-matched control subjects. They reported an ICC of 0.50 for mean RNFL thickness, but poor ICCs for temporal and nasal quadrants (0.36 and 0.31, respectively). 

Blumenthal et al. ³ tested reproducibility of RNFL measurements of a second-generation OCT in 10 normal and 10 glaucomatous eyes and reported COVs for mean RNFL thickness of 7% in normal eyes and 13% in glaucomatous eyes. Measurements in the nasal quadrant were more variable (COV of 28%) In addition, Jones et al. ¹⁶ studied reproducibility of the second-generation OCT and found a COV for mean RNFL thickness of 5% in normal subjects. Measurements in the nasal quadrant were more variable with a COV of 20%. 

One would expect an improvement in reproducibility of RNFL measurements with further developments in OCT technology. In particular, Increasing scan resolution and improvements in OCT software regarding the RNFL thickness algorithm should improve reproducibility. 

Paunescu et al. ⁴ reported on reproducibility of RNFL thickness measurements of the third-generation Stratus OCT in 10 normal subjects. Subjects were scanned six times per day on three different days over a 5-month period. ICC for mean RNFL thickness was 0.79 by using the fast scan algorithm. ICCs for the superior, inferior, nasal, and temporal quadrant were 0.75, 0.71, 0.75, and 0.84, respectively. When using the standard circle scan with higher resolution compared with the fast scan, ICCs for mean, superior, inferior, nasal, and temporal measurements were 0.79, 0.73, 0.65, 0.68, and 0.79, respectively. In addition, Budenz et al. ⁵ tested reproducibility of RNFL thickness measurements with Stratus OCT3. Intrasession variability of measurements between 88 normal subjects and 59 with glaucoma was tested. ICCs ranged from 0.84 to 0.97 in normal eyes with a range of COVs from 1.7% (mean RNFL) to 8.2% (nasal quadrant). In glaucomatous eyes the ICC ranged from 0.79 to 0.98 with a range of COVs of 3.7% (mean RNFL) to 11.9% (nasal quadrant). In a second study in 59 subjects with glaucoma ICCs, COVs and test–retest variabilities were virtually identical, despite the use of a different operator and different subjects. The study tested variations of measurements between different days. ⁹  

Our results indicate that RNFL measurements with the 3D OCT1000 showed good reproducibility with a mean ICC¹, ICC², and ICC³ of 0.9, 0.88, and 0.9, respectively. Mean COV for operator 1 was 4.7% (range, 2.4%–6.7%) for areas 2 to 9. Mean COV for rings 1 and 2 for operator 1 was 2.9% and 1.9%, respectively. Mean COV for operator 2 was 4.4% (range, 2.9%–6.4%) for areas 2 to 9. Mean COV for rings 1 and 2 for operator 2 was 2.3% and 2.0%, respectively. As shown in Table 3 , some confidence intervals for operator variance components were fairly wide. This may be due mainly to the relatively small sample size and the general difficulty in estimating variance components. 

All data were acquired by using the automated RNFL thickness algorithm provided by the 3D OCT1000 software. Although changes can be made manually to the algorithm if borders of the RNFL are not correctly recognized, in this study no corrections have been made. Therefore, the reproducibility data of this study can be applied only to automated RNFL measurements. Additional studies are needed to test whether the manually corrected algorithm shows even more reliable and reproducible results for RNFL measurements compared with the fully automated measurement. 

Mean RNFL thicknesses measured with the 3D OCT1000 seemed to be slightly higher than values acquired with the Stratus OCT. For ring 1, mean RNFL thickness was 107 μm for both operators. When measured with Stratus OCT 3.4-mm circle scans RNFL thicknesses are expected to range around 100 μm, depending on the age of the study subject. Direct comparison of RNFL values between the different OCT machines may be difficult due to different technical specifications, imaging protocols, and different thickness measurement algorithms. For example, retinal (macular) thickness measurements are performed differently with the 3D OCT1000 compared with the Stratus OCT. Both instruments outline the inner limiting membrane as the inner retinal border. The outer retinal border is defined in the Stratus OCT on top of a signal believed to correspond to the junction between inner- and outer segments of the photoreceptors. However, the 3D OCT1000 software defines the outer retinal border as being directly above the RPE but underneath the photoreceptor signal which results in larger retinal thickness measurements. Algorithm definitions of the outer border for RNFL thickness measurements are not so obvious, but such differences may account for larger RNFL thickness measurements as well. In addition, measurements closer than 3.4-mm diameter around the disc were included in the analysis of 3D OCT100 RNFL thickness measurements. It is well known that RNFL thickness increases with increasing proximity to the optic disc. That may also explain higher RNFL thicknesses values measured with the 3D OCT1000. 

At this point, no specific scan protocol has been implemented in the 3D OCT1000 software to measure the RNFL thickness in a specific area around the disc. In this study the ETDRS grid (normally used for macular thickness measurements) was used to divide RNFL thickness in four quadrants around the disc. However, the central ring (area 1) of the ETDRS grid has a diameter of only 500 μm. The average optic disc has a larger diameter than area 1. RNFL thickness measurements cannot be performed directly on the optic disc cup. Therefore, area 1 had to be excluded from analysis. Frequently, the optic disc rim reached into areas 2 to 5. The 3D OCT1000 RNFL software algorithm was able to identify RNFL boundaries on the optic disc rim as seen in the corresponding OCT B-scans. To our knowledge, RNFL measurements on the optic disc rim have not been tried previously with OCT because such measurements where not possible with the standard 3.4 mm circle scans provided by previous OCT versions. Despite the problem that reliability of RNFL thickness measurements on the optic disc rim is not known yet, areas 2 to 5 and ring 1 were included in the analysis. Surprisingly, reproducibility was good for areas 2 and 4 and ring 1 and acceptable for areas 3 and 5. 

As found in previous studies, reproducibility was higher in the superior and inferior quadrants and lowest in the nasal quadrants. ^{3 5 7 9} The reason for that is not clear. Knighton and Qian ¹⁷ have suggested that the angle of incidence of the illuminating beam makes the RNFL image on the nasal side dimmer and therefore harder to be identified by the measurement algorithm. In addition, ICCs may be reduced mathematically because of a smaller population variance nasally. ⁹ Our data showed better reproducibility in the larger sample areas (rings 1 and 2) than in the smaller areas 2 to 9. These findings are consistent with previous studies using time-domain OCT. ^{5 9} Larger sampled areas contain more individual measurements that add into the mean of that area. This signal averaging leads to more reliable measurements. Gurses-Ozden et al. ¹⁸ actually showed that increasing the sampling density or the number of A-scans can increase the reproducibility of OCT measurements. 

In addition to intraobserver reproducibility, we tested interobserver reproducibility. Differences between the two operators were small. Mean difference between RNFL thickness measurements of operators 1 and 2 for rings 1 and 2 were only 0.9 and 0.1 μm, respectively. Limits of agreement were calculated as 2 × SD of the mean difference between multiple measurements of the two operator for ring 1 (2 × SD = 9.6 μm) and ring 2 (2 × SD = 4.4 μm). Our data suggest that reproducibility of RNFL thickness with FD-OCT is good, regardless of the operator. However, an analysis based on Bland-Altman plots depends on the assumption of equal imprecision. If this assumption is unreasonable, the conclusion from the Bland-Altman plot could be incorrect. 

Scan quality is important to facilitate the recognition of the RNFL by the measurement algorithm. Eighteen (7.9%) scans with poor quality factors (Q-factor) or because of blinks during scanning process had to be excluded. In clinical routine, there might be a greater difference in RNFL measurements between experienced operators and unexperienced operators that are expected to produce scans with lower Q-factors. In our study, the experience level of both operators was the same. Therefore, we assume that the imprecision for both operators was about the same, allowing the use of the linear mixed effect model for analysis and ICC calculation. 

Multiple prior studies with previous generations of OCT have tested reproducibility. ^{3 4 5 7 9 15 16} However, ICCs cannot be directly compared since measurements were performed in different groups of subjects and with different OCT models. The ICC is calculated as the ratio of variability due to differences between subjects to the variability from all sources such as noise and/or fluctuations within subjects. The best way to compare study results would probably be a comparison of the residual error variance components as an absolute measure of imprecision of measurements. Residual error variance components of this study are given in Table 3 . In our case, ICCs were calculated from measurements of young healthy volunteers. One would expect only little between-subject variance in such group. Caution should be used when applying these data in older subjects or patients with glaucoma, in whom a greater a greater between-subject variance is expected. 

Additional studies testing the reproducibility of RNFL thickness measurements in patients with glaucoma are needed before the 3D OCT1000 can be safely used as a tool for diagnosing and monitoring glaucoma and other optic neuropathies. 

 Submitted for publication November 7, 2007; revised March 4, May 21, and July 21, 2008; accepted October 20, 2008.

 Disclosure: M.N. Menke, None; P. Knecht, None; V. Sturm, None; S. Dabov, None; J. Funk, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

(A) ETDRS-scheme applied for RNFL thickness measurements. The central black area (area 1) was excluded for analysis. Mean RNFL thickness was calculated for rings 1 (white) and 2 (gray). (B) Fundus photograph taken during an RNFL measurement. The ETDRS grid was centered on the disc. Arrow: scan location and direction of the B-scan shown in (C). (C) One of 128 B-scans taken during a high-density 3D-OCT scan. White lines: RNFL thickness algorithm. One can see that measurements directly on the optic disc cup cannot be performed correctly. Therefore, area 1 was excluded from the analysis. (D) A 6 mm² RNFL thickness map. RNFL thickness is color-coded from blue (0 to 60 μm) over green (80–140 μm), red (160–180 μm), to white (>200 μm).

Table 1.

 

Mean RNFL Thicknesses for Areas 2 to 9 and Rings 1 and 2

	Area 2	Area 3	Area 4	Area 5	Area 6	Area 7	Area 8	Area 9	Ring 1	Ring 2
Operator 1	132	82	140	106	92	55	91	56	107	74
Operator 2	133	82	142	107	94	55	91	57	107	74

 Data are expressed as the mean micrometers of RNFL thickness. n = 38.

Table 2.

 

COVs for All Areas Measured by Operators 1 and 2

	Area 2	Area 3	Area 4	Area 5	Area 6	Area 7	Area 8	Area 9	Ring 1	Ring 2
Operator 1	5.1	5.8	3.5	6.7	2.4	4.1	3.1	6.6	2.9	1.9
Operator 2	3.5	4.6	4.2	6.2	2.9	4.5	2.9	6.4	2.3	2.0

 Both operators performed three measurements in each eye. Data are COV (%). n = 38.

Table 3.

 

Square Root of Variance Components with 95% CIs for Patients, Eyes, Operators, and Residuals in Each Measured Area.

Areas Tested	SqRVar. Patient	95% CI	SqRVar. Eye	95% CI	SqRVar. Rater	95% CI	SqRVar. Residual	95% CI	ICC 1	ICC 2	ICC 3
Ring 1	14.1	10.1–19.7	4.1	2.8–6	1.4	0.6–2.8	3.5	3.2–3.9	0.95	0.94	0.95
Ring 2	8.3	6–11.6	2.3	1.4–3.2	1	0.6–1.5	1.7	1.4–2	0.96	0.95	0.96
Area 2	18.7	13.2–26.4	7.2	5–10.4	1	0.05–20.3	6.3	5.7–7.1	0.91	0.91	0.91
Area 3	9.8	5.7–17.1	10.7	7.7–15	1	0.14–9.5	5.4	4.8–6.1	0.88	0.88	0.88
Area 4	14.8	10–21.7	8.6	6.1–12.4	1.4	0.17–11.1	6.3	5.9–7.4	0.87	0.86	0.87
Area 5	15.5	10.3–23.5	10.7	7.5–15.4	2.3	0.4–9.9	8.5	7.6–9.5	0.83	0.82	0.83
Area 6	10.3	6.1–15.8	7.6	5.5–10.8	2.4	1.7–3.5	2.8	2.4–3.2	0.96	0.92	0.95
Area 7	4.6	2.8–7.1	3.7	2.8–5.3	0.8	0.2–2.8	2.8	2.4–3.2	0.82	0.8	0.81
Area 8	12.3	8.7–17.4	5	3.5–7.2	1.7	1–3	3.3	2.8–3.6	0.94	0.93	0.96
Area 9	10	6.9–14.6	4.4	2.8–6.5	1.7	1.4–2	5	4.5–5.6	0.83	0.81	0.83

 The linear mixed model was used to calculate ICCs. All units are micrometers. SqRVar, square root of variance; ICC1, measurements within the same subject, eye, and operator; ICC2, interoperator; ICC3, intraoperator. P for operator bias was always >0.09.

Figure 2.

 

Bland-Altman plots of differences in RNFL thickness in measurement 1 by operator 1 and 2 for rings 1 and 2 versus the means of the two operator’s measurements in a set of 38 eyes. Units for both axes are micrometers. adj.: adjusted limits of agreement, taking into account that data of eyes were nested within subjects.