Comparison of the areas under the receiver operating characteristic curves (AUCs) for discriminating between healthy and glaucomatous eyes using manufacturer-assumed fixed corneal compensation(FCC)–measured and subject-specific variable corneal compensation (VCC)–measured ellipse average thickness. The AUCs are significantly different from each other (P<.01), with an increased area using VCC.
Comparison of the areas under the receiver operating characteristic curves (AUCs) for discriminating between healthy and glaucomatous eyes using manufacturer-assumed fixed corneal compensation(FCC)–measured and subject-specific variable corneal compensation (VCC)–measured superior average thickness. The AUCs are significantly different from each other (P<.01), with an increased area using VCC.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Weinreb RN, Bowd C, Zangwill LM. Glaucoma Detection Using Scanning Laser Polarimetry With Variable Corneal Polarization Compensation. Arch Ophthalmol. 2003;121(2):218–224. doi:10.1001/archopht.121.2.218
Copyright 2003 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2003
To compare the ability of scanning laser polarimetry (SLP) to discriminate between healthy and glaucomatous eyes with manufacturer-assumed fixed and subject-specific variable corneal polarization magnitude (CPM) and corneal polarization axis (CPA) values.
An SLP was modified to enable the measurement of CPM and CPA values so that compensation for corneal birefringence could be corrected on a subject-specific variable basis. We examined 40 healthy eyes and 54 glaucomatous eyes with repeatable visual field damage (average ± SD mean deviation, −6.5± 4.9 d B) were examined by SLP using the manufacturer-assumed fixed corneal compensation (FCC-SLP) values and subject-specific variable corneal compensation (VCC-SLP) values. Areas under the receiver operating characteristic(ROC) curve for discriminating between healthy and glaucomatous eyes using FCC-SLP and VCC-SLP parameters were compared.
The areas under the ROC curve increased with VCC-SLP compared with FCC-SLP, particularly for all thickness parameters. The parameters with which the area under the ROC curve improved significantly from FCC-SLP to VCC-SLP included average thickness (ROC curve area, 0.62 vs 0.75), superior integral (0.66 vs 0.79), ellipse average (0.65 vs 0.80), inferior average (0.66 vs 0.80), and superior average (0.68 vs 0.83).
Variable corneal compensation to correct for subject-specific CPM and CPA can improve the ability of SLP to discriminate between healthy and glaucomatous eyes.
SCANNING LASER polarimetry (SLP) is an optical imaging technique used clinically as an aid to diagnose and monitor glaucoma. This technique relies on the birefringent property of the retinal nerve fiber layer (RNFL) to provide an indirect measurement of RNFL thickness based on the linear relationship between the retardation of monochromatic reflected light and histologically measured RNFL tissue thickness.1
However, the RNFL is not the only source of birefringence in the eye. The cornea and Henle fiber layer of the macula, and to a lesser extent the lens, also are birefringent. To address anterior chamber (corneal) birefringence, the commercial SLP (GDx Nerve Fiber Analyzer; Laser Diagnostic Technologies, San Diego, Calif) employs a fixed compensator that assumes that all individuals have a slow axis of corneal birefringence (corneal polarization axis [CPA]) 15° nasally downward with a magnitude (corneal polarization magnitude[CPM]) of 60 nm. Recent studies have shown a wide variation in the axis2,3 and magnitude of corneal polarization in healthy and glaucomatous eyes.3,4 Because reliable parapapillary RNFL thickness measurements with SLP require separating RNFL birefringence from corneal birefringence, it is likely that improperly compensated measurements are a significant source of error with this technique.5 Further, it is likely that inaccurate compensation decreases the ability of SLP parameters to discriminate between healthy and glaucoma eyes.6,7
The purpose of this study was to compare the ability of SLP to discriminate between healthy and glaucomatous eyes when using manufacturer-assumed and subject-specific CPM and CPA values. To do so, we used a modified SLP with a variable corneal compensator (VCC). We hypothesized that the discriminating abilities of SLP parameters, described using receiver operating characteristic(ROC) curve analysis, would improve when all subjects were appropriately CPM and CPA compensated using subject-specific VCC.
Scanning laser polarimetry data from 94 consecutive subjects (40 healthy subjects and 54 patients with glaucoma) older than 50 years who met the diagnostic inclusion criteria were evaluated at the Hamilton Glaucoma Center, University of California, San Diego. One eye per subject was included by random selection. Forty-one subjects were male, and 53 were female; 79 subjects were white, 5 were Hispanic, 3 were of African decent, 2 were Asian American, 2 were Indo-European, and 3 self-reported their race/ethnicity as unknown.
Before SLP imaging, all subjects underwent a complete ophthalmologic examination, including slitlamp biomicroscopy, intraocular pressure (IOP) measurement, dilated stereoscopic fundus examination, stereoscopic photography of the optic disc, and Swedish Interactive Test Strategy (SITA) automated perimetry or standard full-threshold automated perimetry (SAP) (Humphrey Field Analyzer; Zeiss-Humphrey Systems, Dublin, Calif). Only eyes with a visual acuity of 20/40 or better were included, and refractive error ranged from −9.00 diopters (D) to 4.75 D (mean [SD], −0.73 D [2.22 D]). Eyes with coexisting retinal disease, uveitis, or nonglaucomatous optic neuropathy were excluded from this investigation. Informed consent was obtained from all participants, and all methods were approved by the University of California, San Diego, Human Subjects Committee and adhered to the Declaration of Helsinki for research involving human subjects.
The inclusion criteria for healthy eyes were no history of ocular disease or increased IOP and normal results on a dilated ophthalmologic examination, including IOPs of 22 mmHg or less (Goldmann applanation tonometry), healthy appearance of the optic disc and RNFL (no diffuse or focal rim thinning, cupping, or RNFL defects indicative of glaucoma or other ocular abnormalities), and normal results on SITA or full-threshold 24-2 SAP. Normal visual field indices were defined as a mean deviation (MD) and corrected-pattern standard deviation within 95% confidence limits and a glaucoma hemifield test result within normal limits. The mean age of the healthy subjects was 64.0 years (SD, 10.4 years; range, 49-86 years). These subjects were chosen so that their age range was similar to the age range of patients with glaucoma (50-89 years).
Eyes were classified as glaucomatous if they had repeatable (2 consecutive) abnormal visual field test results on SITA or full-threshold 24-2 SAP, defined as a corrected-pattern standard deviation outside of the 95% normal limits, or a glaucoma hemifield test outside of the 99% normal limits. Average (SD) mean deviation on the test nearest the imaging date was −6.49 d B (4.94d B) (range, −20.92 to 0.26 d B). Intraocular pressure was not used to classify this group. The mean age of patients with glaucoma was 68.7 (9.2) years (range, 50-89 years). There was a significant difference in mean age between patients with glaucoma and healthy subjects (t test; P = .02).
Polarimetry images were obtained using a commercial SLP (GDx Nerve Fiber Analyzer) modified so that the original fixed corneal compensator (FCC) was replaced with a VCC as described by Zhou and Weinreb8 (see also Weinreb et al3). Briefly, the VCC-SLP is composed of a set of 4 linear retarders in the path of the measurement beam. The first two adjustable retarders are optical lenses that have equal retardance and form a variable cornea and lens compensator. The third retarder is composed of the cornea and lens, and the fourth retarder is the retinal birefringent structure (RNFL or macular Henle fibers).
To determine eye-specific CPM and CPA values, the compensating retarders were adjusted to 0 nm, and the macula was imaged. The resulting retardation profile represented the additive effects of cornea, lens, and macular Henle fiber birefringence. The compensating retarders were then adjusted to minimize the effects of anterior segment birefringence. For each subject, the macula was imaged 3 times, and the mean values of CPM and CPA from the 3 macular scans that resulted in adequate compensation (ie, a macular retardation profile showing uniform low retardation) were recorded. Nasally upward CPA values(in degrees) were recorded as negative; nasally downward CPA values were recorded as positive. The macula was then imaged again using the subject-specific CPM and CPA values to assure that compensation was adequate.
Next, 3 corneal birefringence–compensated parapapillary SLP images from each eye were obtained using both the FCC CPM and CPA values (60 nm and 15°, respectively) and the appropriate eye-specific VCC CPM and CPA values. The 3 images obtained using each technique were automatically combined using standard software to create composite mean images used for RNFL thickness analysis (one mean image using FCC CPM and CPA values and one mean image using eye-specific VCC CPM and CPA values for each eye). The optic disc margin was outlined on each mean image by a trained technician for calculation of ellipse parameters. In 6 eyes (5 glaucomatous and 1 healthy), the measurement ellipse, in its standard position 1.75 disc diameters from the disc center, fell on areas of parapapillary atrophy. In these cases, the size of the measurement ellipse was increased by up to 20% (horizontally or vertically) so that the measurement ellipse was clear of areas of parapapillary atrophy.
We examined 25 parameters automatically provided by SLP software (GDx version 2.0.01). Parameters investigated were grouped as thickness parameters and ratio/modulation parameters. Thickness parameters were superior maximum thickness, temporal maximum thickness, inferior maximum thickness, nasal maximum thickness, temporal median thickness, nasal median thickness, average thickness, total polar integral thickness, superior integral thickness, temporal integral thickness, inferior integral thickness, nasal integral thickness, ellipse average thickness, superior average thickness, temporal average thickness, inferior average thickness, and nasal average thickness. Ratio/modulation parameters were symmetry, superior-nasal thickness ratio, inferior-maximum–nasal-median thickness ratio, temporal–nasal-median thickness ratio, superior ratio, inferior ratio, ellipse modulation, and maximum modulation. Ratio/modulation parameters differ from thickness parameters because they were designed to be independent of absolute values of RNFL retardation and dependent on the amplitude and relative thickness of the RNFL thickness profile. All parameters investigated are shown in Table 1 and have been described in detail elsewhere.9
Measurements taken using the fixed CPA and CPM values are referred to as FCC-SLP measurements, and measurements taken using eye-specific CPA and CPM values are referred to as VCC-SLP measurements.
For each FCC-SLP and VCC-SLP parameter, the area under the ROC curve for discriminating between glaucomatous and healthy eyes was calculated. The area under the ROC curve is a continuous plot of the true-positive rate (sensitivity) by the false-positive rate (1.0 − specificity) that gives a global indication of the overall diagnostic accuracy of the test. In the current study, for example, an ROC curve area of 0.80 indicates that a randomly selected individual from the glaucoma group has a lower SLP RNFL thickness measurement than a randomly selected individual from the healthy group 80% of the time. Therefore, ROC curve areas of 0.5 and 1.0 represent chance and perfect discrimination, respectively. Significant differences in ROC curve area for classifying healthy and glaucomatous eyes using FCC-SLP parameter values compared with VCC-SLP values were determined using the method of Hanley and McNeil.10 For each parameter, we also reported sensitivity at 90% specificity for correctly classifying eyes as glaucomatous based on the "gold" standard of repeatable abnormal SAP and/or SITA results.
The FCC-SLP and VCC-SLP parameter comparisons between glaucomatous and healthy eyes were performed using independent t tests, and comparisons between FCC and VCC techniques were performed using paired t tests. We also investigated the relationship between age and SLP parameters using FCC and VCC with linear regression.
The mean (SD) CPM value was 38.1 nm (17.1 nm) (range, 7-91 nm) in glaucomatous eyes and 39.4 nm (12.4 nm) (range, 17-79 nm) in healthy eyes. The mean (SD) CPA value was 21.1° (18.3°) (range, −13° to 74°) in glaucomatous eyes and 20.9° (16.7°) (range, −8° to 60°) in healthy eyes. There were no significant differences between glaucomatous and healthy eyes for CPM (t test; P = .67) and CPA (t test; P = .90).
The mean values of all thickness parameters measured using VCC-SLP were significantly lower than the same parameters measured using FCC-SLP (all P<.001; paired t tests). There were no significant differences between ratio/modulation parameters using VCC-SLP compared with FCC-SLP (all P>.10). In addition, the variability (SD) of all thickness parameters measured using VCC-SLP was significantly smaller than the variability using FCC-SLP (all P<.001; paired t tests). Mean values and SDs for FCC and VCC parameters are shown in Table 1.
The areas under the ROC curve for FCC-SLP parameters ranged from 0.33(for temporal integral) to 0.78 (for superior-nasal thickness). The areas under the ROC curve for VCC-SLP parameters ranged from 0.32 (for temporal–nasal-median thickness) to 0.83 (for superior average thickness).
In general, increases in the area under the ROC curve were greater for thickness parameters compared with ratio/modulation parameters (which generally did not improve), using VCC compared with FCC. Parameters with areas under the ROC curve greater than 0.50 for which the area under the ROC curve improved significantly from FCC-SLP to VCC-SLP included average thickness (0.62 vs 0.75), superior integral (0.66 vs 0.79), ellipse average thickness (0.65 vs 0.80), and superior average thickness (0.68 vs 0.83), respectively. Of the 6 ratio parameters reported, the area under the ROC curve area for only one parameter (inferior-maximum–nasal-median thickness) decreased by more than 0.10 using VCC compared with FCC. Although not significant, the magnitude of this decrease (0.12) was similar to the magnitude of the increase observed for all but 2 of 11 thickness parameters.
Table 2 shows areas under the ROC curve for all parameters with areas significantly greater than 0.50, or chance discrimination (determined by the 95% confidence interval of the ROC curve excluding 0.50), using FCC-SLP or VCC-SLP. Figure 1 and Figure 2 show comparisons of areas under the ROC curve between FCC-SLP and VCC-SLP for the 2 parameters from Table 2 with the largest significant area difference between instruments (ellipse and superior average thickness). Two parameters with areas under the ROC curve areas less than 0.50 (chance discrimination) using FCC-SLP had areas significantly greater than 0.50 using VCC-SLP (nasal maximum thickness, 0.48 vs 0.63; nasal average thickness, 0.49 vs 0.63).
Table 2 also includes sensitivities at 90% specificity. Sensitivities at this relatively high specificity increased for some parameters and decreased for other parameters when using VCC-SLP compared with FCC-SLP. The most notable increases in sensitivity using VCC-SLP were for ellipse average thickness (FCC-SLP sensitivity, 44%; VCC-SLP sensitivity, 67%), superior integral (43%; 65%), ellipse modulation (26%; 46%), and inferior integral (37%; 57%). The most notable decreases in sensitivity at 90% specificity were for inferior-maximum–nasal-median thickness (56%; 20%) and superior-nasal thickness (50%; 33%).
There was a significant difference in mean RNFL thickness measurements between healthy and glaucomatous eyes using both FCC-SLP (8 of 25 parameters) and VCC-SLP (15 of 25 parameters) (Table 1; α = .002 after adjustment for 25 comparisons). Seven more parameters, all of which were measures of RNFL thickness, showed significant differences between glaucomatous and healthy eyes using VCC but not FCC (average thickness, total polar integral, superior integral, inferior integral, ellipse average thickness, superior average thickness, and inferior average thickness). In general, differences in SLP measurement between healthy and glaucomatous eyes were greater using VCC-SLP compared with FCC-SLP.
Finally, we examined the effect of subject age on FCC-SLP and VCC-SLP parameters in healthy eyes using linear regression. A significant association(P≤.05) between age and SLP measures was found for 3 FCC-SLP parameters (R2 range, 0.10-0.11) and 9 VCC-SLP parameters (R2 range, 0.10-0.26), suggesting that age is associated more strongly with VCC parameters compared with FCC parameters. The associations between age and SLP parameters are shown in Table 3. We did not examine the relationship between age and SLP parameters in glaucomatous eyes because it is likely that the effect of age is confounded by glaucoma severity. To determine if age affected the overall discrimination ability of VCC-SLP and FCC-SLP parameters, we recalculated the areas under the ROC for all parameters, adding age to the model. The area under the ROC curve did not increase by more than 0.05 in any case (all P>.05 by the method of Hanley and McNeil10; data not shown).
The current study shows that the success of several GDx parameters at classifying eyes as glaucomatous or healthy is improved considerably if the effects of CPM and CPA are properly compensated. We observed increases in sensitivities and the areas under the ROC curve using VCC-SLP thickness parameters compared with FCC-SLP thickness parameters.
Evidence from earlier studies using FCC-SLP to discriminate glaucomatous eyes from healthy eyes suggested that ratio parameters and other parameters that describe relative thickness were superior to parameters that measure thickness alone.6,7,11-17 Ratio parameters consider the RNFL thickness profile in one region relative to another region (usually the temporal or nasal region), not the absolute value of RNFL retardation. In retrospect, it now appears that ratio parameters may have demonstrated better diagnostic precision because they controlled for the overall full-field increase in retardance that was often observed when corneal polarization was ineffectively compensated.2 Results from the current study showing increases in sensitivities and the areas under the ROC curve using VCC RNFL thickness values compared with FCC RNFL thickness values support this idea. Similarly, the relatively larger decrease in the variability of RNFL thickness parameters than in ratio parameters with VCC compared with FCC suggests that previously reported poor diagnostic precision using FCC-SLP thickness parameters likely is attributable to inflated intersubject variability resulting from ineffective corneal polarization compensation. The observed decrease in variability for several parameters using VCC suggests that a new normative database with lower variability among healthy eyes could improve the sensitivity for glaucoma detection of the summary parameters available to the clinician on the SLP examination printouts.
In the current study, we neither examined the diagnostic precision of the SLP neural network number nor the SLP linear discriminant function.18 These parameters have previously been shown to best discriminate healthy from glaucomatous eyes.6,7,14,18,19 They were excluded in the current study because they were both developed as a means to best interpret FCC-SLP data. It is likely that the development of new linear discriminant function or neural-network–based summary parameters based on VCC-SLP data will improve SLP diagnostic precision compared with the single best parameter reported in the current study.
Recently, in a study related to ours, Greenfield et al5 showed that adding CPA data to a statistical model with each of several FCC-SLP (GDx Nerve Fiber Analyzer) parameters increased the ability of these parameters to discriminate healthy from glaucomatous eyes. In their study, only the area under the ROC curve of thickness parameters, and not ratio parameters, was improved significantly, similar to our study. For instance, the parameter with the largest increase in discrimination ability in their study was inferior mean thickness, which showed an area under the ROC curve increase of 0.14(from 0.70 without CPA in the model to 0.84 with CPA in the model). In our study, the use of VCC increased the discrimination ability (area under the ROC curve) of the same parameter also by 0.14 (from 0.66 to 0.80).
Regarding the association between age and SLP parameters, we found stronger effects of age on several VCC-SLP parameters than on FCC-SLP parameters, despite the limited age range of healthy subjects enrolled in our study (range, 35 years). However, these effects were not very strong, even using VCC-SLP (R2 range, 0-0.26). This stronger effect of age on VCC-SLP likely is attributable to the decreased variability in parameter measurements using this technique.
In the current study, we used macular imaging to determine and correct for CPM and CPA values. A possible limitation of this technique is that some macular abnormalities may disrupt the Henle layer and/or the macular birefringence, resulting in erroneous corneal polarization measurements. Because both glaucoma and macular degeneration are age-associated diseases, some patients might not provide stable macular images for correct CPM and CPA value determination cross-sectionally and over time. This issue requires additional study, and alternate techniques to effectively compensate corneal birefringence may be necessary in these patients.
In summary, our results indicate that SLP RNFL thickness measurements decrease and their variability is reduced using VCC-SLP compared with FCC-SLP. This observed reduction of variability results in the superior ability of VCC-SLP to discriminate between healthy eyes and eyes with glaucomatous visual field damage.
Corresponding author: Robert N. Weinreb, MD, Hamilton Glaucoma Center, University of California, San Diego, 9500 Gilman Dr, La Jolla, CA 92093-0946.
Submitted for publication June 26, 2002; final revision received September 19, 2002; accepted October 1, 2002.
This study was supported in part by grant NEI EY11008 (Dr Zangwill) from the National Eye Institute, National Institutes of Health, Bethesda, Md, and the Joseph Drown Foundation, Los Angeles, Calif (Dr Weinreb).