This prospective, non-randomized, cross-sectional analysis of normal, OHT and glaucomatous eyes was carried out in 184 subjects (184 eyes). One eye from each of the 52 healthy (28.26%), 38 OHT (20.65%), 58 early glaucomatous (31.52%) and 36 moderate glaucomatous (19.56%) subjects was analysed. We did not include in the study normal-tension glaucoma patients. All the procedures were conducted in accordance with the principles of the Helsinki Declaration. Detailed consent forms were obtained from each of the patients.

All patients underwent a complete ophthalmic examination, which included systemic, ocular, and family history, best-corrected visual acuity by means of a standard Snellen projector system, slit-lamp biomicroscopy including gonioscopy; diurnal applanation tonometry measurement and stereoscopic optic disc evaluation using a 78-diopter lens (with cup/disc ratios assessed by the same experienced examiner). Achromatic visual field testing with SITA standard strategy were carried out using the Humphrey 750 Visual Field Analyser program 24-2 (Zeiss Humphrey Systems, San Leandro, CA, USA). SLP measurements were obtained with the GDx-VCC, software version 5.1.0 (Laser Diagnostic Technologies, San Diego, California, USA). OCT measurements were carried out using the Stratus-OCT, version 3.0 (Carl Zeiss Meditec Inc, Dublin, CA, USA).

Normal eyes were defined as those with intraocular pressure (IOP) not exceeding 21 mmHg and having no prior history of elevated IOP, reproducible normal visual field as measured with AAP [glaucoma hemifield test, mean deviation (MD) and pattern standard deviation (PSD) within normal limits] and a healthy optic nerve head (ONH) appearance (asymmetry of vertical cup/disc ratio <0.2, no evidence of rim thinning, notching, excavation, haemorrhage, or RNFL defects). OHT subjects had an IOP greater than 21 mmHg but reproducible normal visual fields with healthy ONH appearance. Glaucoma patients were defined as having either specific and repeatable glaucomatous achromatic visual field defects or clinical judgment of glaucomatous optic neuropathy (GON), based on the presence of at least one of the following characteristics that are indicative of glaucoma: cup-to-disc ratio asymmetry between fellow eyes greater than 0.2, cup excavation greater than 0.6, altered neuroretinal rim with focal or general thinning, peripapillary haemorrhages, notches, or localized pallor.

Exclusion criteria for all patients included: best-corrected visual acuity worse than 20/40, refractive error exceeding +/-5.00 dioptres of sphere or 2.00 dioptres of cylinder, evidence of vitreous or retinal pathology not attributed to glaucoma (including diabetic retinopathy, age-related macular degeneration, cystoid macular oedema and peripapillary atrophy extending to 1.7 mm from disk centre), advanced visual field loss (defined by defects close to fixation 5 dB below normal and/or an AGIS score > 11), unreliable AAP or other pathological conditions that could affect the visual field (pituitary lesions, demyelinating diseases), secondary causes of IOP increase (corticosteroid use, iridocyclitis, trauma) and prior incisional surgery or laser treatment.

AAP was performed with the Humphrey Field Analyser (Humphrey Systems Inc, Leandro, CA, USA) using the Swedish Interactive Threshold Algorithm (SITA) standard strategy, program 24-2 test procedure and a size-III stimulus. The analysis of the data was carried out by the program STATPAC 2, included in the software of the perimeter. All patients had undergone automated perimetry before.

Perimetry was performed within 3 months of clinical examination and determination of RNFL thickness. AAP reliability criteria to accept a VF examination as valid included: fixation-loss rates of less than 20% and maximum false-positive and false-negative rates of 25%. The unreliable VF examinations were rejected and repeated.

The criteria used to establish an abnormal visual field consisted of a specific and repeatable glaucomatous achromatic visual field defect that produced three or more significant (P<0.05) non-edge-contiguous points with at least one at P<0.01 on the same side of the horizontal meridian in the pattern deviation plot and classified outside normal limits in the Glaucoma Hemifield Test. Visual field (VF) defects were considered reproducible if they met the criteria for abnormality and were present in exactly the same field locations in at least one consecutive VF confirmatory examination.

The severity of the disease in the case of glaucomatous eyes was graded using the Advanced Glaucoma Intervention Study (AGIS) scores.17 The AGIS algorithm for scoring visual field defects is based on reliability, the number of adjacent test locations with depressed sensitivity, the depth of the depression, and the region of the field affected.18 In the present study, the stratification of glaucomatous field losses were classified according to the AGIS criteria as follows: early visual field damage (AGIS score < 6), moderate visual field damage (AGIS score between 6 and 11) and severe visual field damage (AGIS score > 11). Patients with severe visual field damage were not included in the study, in order to avoid truncation effects and their difficulty to undergo VF examinations.

Measures of VF depression presented on the Humphrey printout include the mean deviation (MD) and the pattern standard deviation (PSD). Therefore, MD will reflect generalised glaucomatous VF loss, while PSD will reflect small localised defects that would appear in early stages of glaucoma. As a result, MD will detect better a generalized glaucomatous VF loss than small localized defects that would appear in early stages of glaucoma; these ones will be better revealed using the PSD. The MD and PSD were used for the statistical analysis, in order to evaluate the correlation between RNFL structural measurements obtained with the RNFL analysers and the visual field global indices measured with AAP.

With the advent of SLP with Variable Corneal Compensation (GDx-VCC, Laser Diagnostic Technology, Inc., San Diego, CA, USA), an objective means of evaluating relative RNFL thickness has been developed, with automated individualized compensation of anterior segment birefringence. This technology uses a scanning laser to obtain images of the RNFL, from which it calculates the relative RNFL thickness. With SLP, linearly polarized light traversing the RNFL, which is presumably a birefringent medium, is elliptically polarized, and bounces back out of the retina. The amount of linear retardation of the light at each retinal location has been shown to be proportional to the corresponding RNFL thickness.7,8 This technology has been described in detail elsewhere.19-21

The GDx-VCC Nerve Fiber Layer Analyser is a commercially available scanning laser polarimeter, designed to estimate RNFL thickness “in vivo”. In our study, at each imaging session, a well trained operator performed all the RNFL thickness measurements. Three consecutive values of the RNFL thickness were obtained in each eye in an undilated state along the visual axis and central cornea. Only high-quality images that had passed the internal software's automated quality control criteria were accepted. Images that were obtained during eye movement, or unfocused and poorly centred images were rejected and taken again. Lastly, the optic disc margin was approximated by a circle or an ellipse placed by an experienced technician around the inner margin of the peripapillary scleral ring. Default quadrant positions (as supplied by the manufacturer) were applied: the peripapillary band was divided into superior and inferior segments of 120° each, a temporal segment of 50°, and a nasal segment of 70°.

The Zeiss Optical Coherence Tomography Model 3000 (Stratus-OCT; Humphrey Systems, Dublin, CA, USA) is a noncontact and noninvasive technology that allows crosssectional imaging of the human retina at histologic levels of resolution (approximately 10 μm), based on the principles of low-coherence interferometry.4,22 It is designed to provide real-time, objective, cross-sectional measurements of various layers of the retina, including the RNFL.2 A scanning interferometer is used to obtain a cross section of the retina based on the reflectivity of its different layers.8 A high reflectance layer located just under the inner surface of the retina that corresponds to the RNFL thickness is measured with a computer algorithm to give RNFL thickness.7,8 Detailed descriptions of the principles of OCT have already been published.23-25

Image acquisition was as follows: after pupillary dilation with 1% tropicamide to achieve a minimum diameter of 5 mm to maximize the likelihood of acquiring a high-quality image, the RNFL was scanned with the Stratus-OCT. Three circular scans of 3.4 mm diameter centred on the optic disc, judged to be of acceptable quality by an experienced observer, were obtained and stored for each eye tested. The scan process was started and the image of the ONH was focused and aligned using the real time video monitor. When the operator was satisfied with the focus and quality of the scan image, it was frozen and saved in a database. RNFL thickness is quantified by an automated computer algorithm that identifies the anterior and posterior borders of the RNFL. The data are presented by clock hours, by quadrants, and overall.

Quantitative variables were compared using the one-way analysis of variance (ANOVA) to evaluate different measures among the groups. Results of statistical significance were also provided after Bonferroni correction based on the number of comparisons within each analysis, in order to determine which pairs of means differ, as reflected in the correlation tables. Prior to the use of parametric tests, the variables under study (data from AAP, GDx-VCC or OCT) were confirmed to behave as normal variables in any of the examinations performed by means of Kolmogorov-Smirnov test, and homogeneity in group variances were contrasted.

Statistical correlations between parameters were assessed using the bivariate Pearson's correlation coefficient, multivariate linear regression techniques and nonlinear regression models. Different nonlinear regression models for curve estimation were used and compared: logarithmic, inverse, quadratic, cubic, power, compound, S-curve, and logistic. The coefficient of determination (r2) was calculated. A P-value less than or equal to 0.05 was considered to be statistically significant.

One eye was used from each subject for the statistical analysis. For healthy and OHT subjects the eye was randomly selected. For the data to be as unbiased as possible, in glaucoma patients the selected eye was the worst one, since some glaucoma subjects only had one damaged eye. By choosing a random eye from these subjects, there would be a significant proportion of healthy eyes from glaucomatous subjects included in the glaucoma group, which would inevitably affect the results.

Five parameters generated by the Nerve Fiber Analyzer (NFA) available mapping software were selected from each instrument for the statistical analysis. In the SLP GDx-VCC, we evaluated the following: the calculation circle average thickness (TSNIT), the superior 120° average thickness (SAVG), the inferior 120° average thickness (IAVG), the calculation circle standard deviation (TSNIT SD) and the nerve fiber indicator (NFI). The first three are average-based parameters, the fourth is a ratio-based parameter and the fifth is a constructed retardation parameter. These five parameters were selected for the statistical analysis because they have proven to be more sensitive than those presented in the TSNIT parameters table.26 With the Stratus-OCT we evaluated the maximum thickness in the superior quadrant (SMAX), the maximum thickness in the inferior quadrant (IMAX), the average thickness in the superior quadrant (AVGSQ), the average thickness in the inferior quadrant (AVGIQ) and the average thickness for the total circumference (AVGT). These five parameters were considered because they provide a quantitative evaluation of the thickness of the peripapillary retinal nerve fiber layer (which is useful to assess current or progressive glaucoma damage) and since they are probably the most relevant ones to the clinical community.

SLP makes use of the birefringence features of ganglion cell axons, measuring the linear retardation in order to calculate RNFL thickness. Recently, GDx was modified into the GDx- VCC, which includes a variable cornea compensator (VCC) that uses macular-based strategies for neutralization of corneal polarization magnitude. It has been stated that, compared with fixed compensation, VCC improves the relationship between mean-based SLP retardation parameters and visual function, particularly with standard automated perimetry visual field MD and PSD.7,43,44,(Garcia-Sánchez J, et al. IOVS 1998;39(Suppl): S933.Abstract 4298) Moreover, the nerve fiber indicator was the best parameter from the GDx-VCC to discriminate between healthy eyes and eyes with glaucomatous visual field loss using the receiver operating characteristic curves (AUC=0.91).30

In this study, we found that GDX-VCC mean-based retardation parameters were correlated with AAP MD and PSD as outcome variables. These findings are consistent with the results reported by other authors. It has been shown36 that RNFL thickness decreases are associated with an increase of field defects (MD value) both globally and by hemi-field/hemiretina. It has also been found8 that there is a weaker correlation between mean-based retardation parameters and visual function (r2=0.17 to 0.27), as compared with constructed retardation parameters (modulation parameters, ratio parameters, and neural network number; r2=0.36 to 0.51) using SLP with fixed compensation. Some authors,31,43 (Garcia-Sánchez J, et al. IOVS 1998;39(Suppl): S933.Abstract 4298) have reported that constructed SLP parameters (modulation, ratio, neural network number, and linear discriminant function values) have greater discriminating power compared with mean-based parameters. In fact, in this study, a GDx-VCC constructed parameter, the NFI, showed the strongest correlation between the SLP parameters and the VF global indices (tables 4 and 5), which suggests that GDx-VCC could be a well suited technique for the early detection of glaucoma.

Although previous studies have demonstrated similar discriminating power for the detection of early to moderate glaucoma,6,40,43 mean-based retardation parameters obtained with SLP (integral and average thickness values) have been reported to have a relatively weak correlation with RNFL thickness yielded by OCT.8 However, in this study, we found a better correlation in the presence of a significant visual field defect (moderate glaucoma group) and a worse correlation when the visual field was close to normal (early glaucoma patients), which is in agreement with other authors’45 findings. Unfortunately, the moderate glaucoma group is a population segment that presents to clinicians fewer diagnostic complications than early glaucoma patients, since a significant visual field loss is easy to detect by perimetry alone. The worse correlation between early glaucoma patients compared to moderate glaucoma patients could it be due to the so-called functional reserve. On the other hand, it has been stated that the lack of correlation between VF and RNFL parameters in early glaucoma patients could be due to the moderate discriminating power of the commercially available instruments, combined with the great inter-individual variability in RNFL thickness, which causes a significant overlap between normal, OHT and early glaucoma patients.27 Moreover, histological studies showed that 25% to 40% of retinal ganglion cells axons are lost before abnormality changes are detected through automated VF testing,46 probably because of the logarithmic scale used in VF.

In addition, the only significant correlation for the early glaucoma group between the GDx-VCC parameters and VF global indices was between TSNIT SD and AAP PSD. We believe that TSNIT SD could be the best GDx-VCC parameter to differentiate early cases of glaucoma (Table 5).

OCT is based on the measurement of the reflectance of the posterior segment structures and incorporates a mathematical algorithm capable of localizing the anterior and the posterior limits of the RNFL. It has proved to be useful as a glaucoma-screening tool in the general population,39 because RNFL measurements by OCT have revealed NFL thinning in areas corresponding to visual field defects.47 Moreover, regional OCT, which measured RNFL thinning in glaucoma patients, was reported to be associated with decreases in VF zone sensitivity,22 and it was stated48 that OCT is a promising tool to provide quantitative data about the location and the extent of retinal nerve fiber layer injury in glaucoma.

In this investigation, OCT provided the strongest correlation between RNFL thickness and MD or PSD for the entire cohort group (Tables 5 and 6), because the mean Pearson's correlation coefficient for OCT was significantly higher than for GDx-VCC. These results agree with those obtained by other authors3,11 and the higher association with visual function in Stratus-OCT RNFL measurements, compared with that carried out with GDx-VCC, suggest that optical coherence tomography might be a better approach to the evaluation of structure/function relationships. Moreover, OCT could attain a higher resolution than GDx

In the current study, the resulting correlation values between mean global RNFL thickness on OCT and each of the two AAP ratios were similar and did not exhibit large differences (PSD=-0.3686; MD=0.3008). Others studies had reported contradictory results. El Beltagi et al.22 reported that localized RNFL thinning, measured by OCT, is topographically related to decreased localized AAP sensitivity in glaucoma patients.

On the other hand, Hoh et al.8 showed a stronger correlation between RNFL measurements with AAP MD rather than with AAP corrected PSD. This finding suggests a week relationship between structural and functional measurements in glaucoma and, therefore, the results of these ancillary tests should be interpreted jointly to increase diagnostic accuracy.26

In the present study, dependent and independent variables were quantitative, and categorization of paired (X,Y) data is generally not quite linear. They are generally more S-shaped, though very shallow curves can be almost linear. There are already several publications11,44,49 indicating the lack of a linear relationship between functional and structural losses. The top candidates are cubic smoothing spline and logistic functions, because they have this particular shape. Garway- Heath et al.49 demonstrated, in a group of 74 healthy and glaucomatous eyes, that there was significant linear and quadratic relationships between temporal neuroretinal rim area (in square millimeters) measured with retinal tomography, and visual sensitivity (dB) measured by means of standard automated perimetry. Coefficient of determination (r2) for the linear fit was 0.30 and r2 for the quadratic fit was 0.38. Schlottmann et al.44 showed small but significant improvements in structure–function associations with logarithmic regression using analysis of model residuals. Leung et al.11 used linear and four nonlinear models to assess the association between global visual sensitivity and mean RNFL thickness in healthy, suspect, and glaucomatous eyes, as measured with the Stratus OCT and GDx VCC, and found secondand third-order polynomial curves, respectively, to be better fits than lines when visual sensitivity was expressed in decibels. In the present study, linear regression models showed that the independent variable MD was correlated with the dependent variable NFI and the independent variable PSD was correlated with the dependent variables NFI and SMAX. After that, we calculated the curvilinear regression models adjusting the model to the data points and, according to our results, the cubic curve estimation turned out to be the best fit (in terms of r2), as well as the best theoretical match to the expected shape of our data, when compared to the various offered curves. Anyway, the differences of the determination coefficients were very small, so it cannot really be stated that the cubic model is much better than the linear one.

The results from this study show that, with the use of both technologies, significant structural differences can be observed between moderate glaucoma patients and nonglaucomatous eyes. Moreover, some parameters in glaucomatous eyes showed significant differences compared with normal eyes (Tables 5 and 6). Furthermore, the magnitude of RNFL thinning is related to the decrease of AAP sensitivity. These findings provide validity for both nerve fiber analysers as diagnostic tools, because the results agree with previous knowledge regarding tissue morphology in glaucoma.

As was previously exposed, most parameters in early and moderate glaucomatous eyes showed significant differences with normal eyes. However, there were few significant differences between normal and OHT eyes in terms of the parameters under consideration in this study. Although mean RNFL thickness (Stratus-OCT) and ellipse average (GDx-VCC) measurements were smaller in OHT than in normal eyes, these differences were not statistically significant. These two groups exhibited similar results because, from the functional and anatomic point of view, they have similar characteristics. Our results were in agreement with previous reports3,8,16,35 and disagree with others that found structural differences between these groups11,50 or even between healthy eyes and eyes with localized RNFL defects without VF loss (preperimetric glaucoma).51

It should be noted that it is difficult to compare associations between structure and function across studies because the strength of the correlation coefficient is dependent on many features of the study design. The diagnostic groups included, the severity of the disease for the glaucoma patients, and the range of the measurements will have an impact on the resulting correlation values. For example, weak correlations between RNFL measures and visual field MD and PSD would be expected in studies of healthy subjects and OHT patients without visual field loss, because the visual field indices would all have a limited range. Also, the sample size could probably play a role when the correlations between RNFL thickness and VF sensitivity are determined.

In conclusion, quantitative measurements of RNFL thickness using both GDx-VCC and Stratus-OCT correlated moderately with VF indices for moderate glaucoma patients. In a few cases early glaucomatous eyes showed only borderline correlations and in most cases were not correlated with neither MD or PSD. Cross-sectional studies have limitations when examining the relationship between structural parameters and function because they can assist in determining which parameters are associated to the visual function but they cannot address the temporal relationship between structure and function. Therefore, longitudinal studies must be carried out in order to obtain information regarding how the visual field reflects the anatomic damages that occur in the course of the glaucomatous disease.

Optic-nerve imaging technologies continue to improve, both in terms of their technological ability to measure, and in terms of software to evaluate the optic nerve. These SLP and OCT versions had similar performance to separate glaucomatous and non-glaucomatous populations, but they were unable to differentiate normal from OHT populations in this cohort (Tables 5 and 6). A constructed retardation parameter, the NFI, was the one that was able to discriminate best between different stages of glaucoma in the present study. But, at the moment, none of these instruments is sufficient in and of itself to replace the traditional methods of examination. Either imaging technologies or visual field assessment may show the first evidence of glaucomatous damage; therefore, the combination of optic nerve head parameters and visual field results could improve glaucoma diagnosis and follow-up52. Physician judgment remains crucial in evaluating the output of these devices and in incorporating that into the diagnosis and management of individual patients. Future improvement in imaging technology, which provides higher resolution and better reproducibility, may enhance their sensitivity and specificity, increasing their usefulness for population screening.