PATIENTS AND METHODS

Subjects

Sixty two white patients with primary open angle glaucoma were recruited from the outpatient clinics of Manchester Royal Eye Hospital without knowing the severity of visual field defects and optic nerve head damage. Informed consent was obtained from all patients after the examination procedure was fully explained. All patients underwent evaluations including abbreviated medical history, full ocular history, ophthalmological examination, automated visual field test, optic nerve head tomography, and scanning laser polarimetry. The ophthalmological examinations included logMAR visual acuity, slit lamp biomicroscopy, applanation tonometry, gonioscopy, and dilated indirect ophthalmoscopy for optic disc and nerve fibre layer evaluations. Visual field tests were performed with the Humphrey field analyser (HFA) (Model 740, Humphrey-Zeiss, Dublin, CA, USA) using the full threshold 24-2 program. Optic nerve head tomography was performed using HRT (software version 2.01, Heidelberg Engineering GmbH, Heidelberg, Germany). Scanning laser polarimetry was conducted with the GDx (software version 2.0.09, Laser Diagnostic Technologies, Inc, San Diego, CA, USA). Automated visual field testing, optic nerve head tomography, and scanning laser polarimetry were completed on each patient within 6 months of each other. The inclusion criteria included age ⩾40 years, best corrected visual acuity not worse than logMAR 0.4, spherical refractive errors ⩽+5.00 and ⩾−5.00D and cylinder ⩽3.00D, previous experience of full threshold perimetry, reliable HFA visual fields (fixation losses, false positives, and false negatives <25%) and good image quality with the HRT and GDx. The exclusion criteria included neurological disease, history of ocular trauma, intraocular surgery, other intraocular disease except early senile cataract, and generalised depression or an abnormal high sensitivity result with the glaucoma hemifield test (GHT). For the purpose of this study, glaucoma was defined as: (1) glaucomatous optic disc based on either cup/disc asymmetry between two eyes >0.2, rim defect, notching, excavation, or nerve fibre layer defect, and (2) two consecutive reliable abnormal HFA visual fields with corrected pattern standard deviation (CPSD) outside 95% normal limit, or GHT outside 99% normal limit, or three or more adjacent points with p <5% on the pattern deviation probability plot, and one of which must have p <1%.²³ The index MD was not used because of the possibility of a reduced value resulting from cataract. If both eyes were eligible only one eye of each patient was chosen randomly.

Humphrey field analyser

The visual field was divided into superior, inferior, nasal, and temporal sectors based on the anatomical relation established by Garway-Heath et al.²⁴ The number of points that represented p <1% and p <5% on the pattern deviation probability plot were also used as global VF indices. With regard to the sectoral VF index (sectoral mean deviation), this was represented by the mean value of all points for each sector on the total deviation plot of the HFA.

Confocal scanning laser ophthalmoscope

The optic nerve head tomography was performed by well trained technicians using the Heidelberg retina tomograph (HRT). Magnification error was corrected by keratometry values and refraction results for each individual. Imaging was performed with a 10 × 10° field of view through a ⩾3 mm diameter pupil when the subject viewed at a distant fixation target. Each eye had five scan series recorded at one sitting. The quality of images was assessed with the aid of the HRT software and by the experience of the technician. The best three of these five scan series were used to produce an overall mean topographical image for further analysis. The mean topography had a clear image with a clearly definable disc margin and the mean standard deviation for the height position of each pixel element of the three topographic images was under 40 μm.

The contour line (disc margin) was drawn using a computer mouse system by a trained observer who was masked to the patient diagnosis and characteristics. A series of optic nerve head topographic parameters (optic disc area, cup area, cup/disc area ratio, cup volume, rim area, rim volume, height variation contour, mean cup depth, maximum cup depth, and cup shape measure) and RNFL (mean RNFL thickness and cross section area along the corrected contour line) for the whole disc and predefined sectors were calculated by the software. The sectors used in this study were defined as temporal: −30°–30°, superior: 30°–135°, nasal: 135°–225°, and inferior: 225°–330°.²⁴ HRT also provides a glaucoma index and a classification of “normal” (positive index) or “glaucoma” (negative index) based on a discriminant function elaborated by Mikelberg et al.²⁵

Scanning laser polarimeter

RNFL thickness measurement was performed with a laser diagnostics glaucoma scanning system (GDx) by well trained technicians using a 15 × 15° field of view through a ⩾3 mm diameter pupil with the untested eye fixating on a fixation light attached to the GDx. Five high quality images per eye were taken in one sitting. The quality of images was assessed with the aid of the GDx software and by the experience of the technician. The three best images were used to generate an overall mean image. The mean image was clear and had a clearly definable disc margin. On the mean image, the disc margin was established with an ellipse whose parameters were adjusted by an experienced observer who was masked to the patient diagnosis and characteristics. A 10 pixel-wide elliptical band was then automatically positioned concentric with the disc margin outline and 1.75 disc diameters from the centre of the optic disc. A series of RNFL parameters for the whole area, peripapillary ellipse, and predefined sectors were calculated by the software. The default sectors were defined as temporal: −25°–25°, superior: 25°–145°, nasal: 145°–215°, and inferior: 215°–335°. The “number,” average thickness, ellipse average, total polar integral, maximum modulation, and ellipse modulation were regarded as global parameters. Superior (inferior) maximum, superior (inferior) ratio, as well as average, integral, and deviation from normal for four sectors were regarded as sectoral parameters.

Analysis

Correlations were evaluated by scatter plot and linear regression for a range of global and sectoral parameters with the HRT, GDx, and HFA. A statistically significant association was taken when the p value was ⩽0.05.

RESULTS

This study enrolled 62 white patients with primary open angle glaucoma. Patients’ age ranged from 40 to 81 (mean 65.76 (SD 8.95)) years of age. The R/L ratio was 33/29 and the male/female ratio was 36/26. The global visual field indices, global topographic parameters by HRT, and global RNFL parameters by GDx are shown in Table 1.

The correlations between global topographic parameters by HRT and global VF indices were only found to be significant for rim area and disc area. Rim area was correlated to MD, CPSD, and the number of points with p <5%. Surprisingly, disc area was related to MD (Table 2). Figure 1 shows the scatter plot of the relation between rim area and mean deviation.

Among the sectoral topographic parameters, correlations were found between the C/D ratio, rim area, rim volume, thickness, and cross sectional area of RNFL and MD for combined superior and inferior sectors and inferior sector (Table 3). There were also correlations between rim area and height variation contour and MD for the superior sector. Although statistics showed associations between some sectoral topographic parameters and MD in nasal and temporal sectors, the scatter plots did not reveal any strong relation. The associations may have been heavily biased by one or two extreme values.

With regard to the RNFL parameters by GDx, the “number,” maximum modulation, and ellipse modulation correlated with CPSD, number of points with p <1% and 5%, and with MD (correlation coefficient ~ 0.4) (Table 4). The scatter plot of the relation between maximum modulation and mean deviation is shown in Figure 2.

All the sectoral RNFL parameters including maximum, average, integral, ratio and deviation from normal showed correlation with MD for combined and inferior sectors. None of these sectoral RNFL parameters demonstrated correlation with MD for other sectors (Table 5).

Unexpectedly, the RNFL parameters, including average thickness, ellipse average, and total polar integral by GDx and RNFL parameters, including thickness and cross section area by HRT, had no correlation. As for the correlation among the topographic parameters of the HRT, many optic nerve head parameters correlated with RNFL parameters. C/D area ratio, rim area, rim volume, and maximum cup depth correlated with both RNFL thickness and cross section area (correlation coefficient: 0.273–0.695). Rim volume had the highest correlation coefficient, nearly 0.7. There were also correlations between cup area and height variation contour and RNFL thickness (correlation coefficient: 0.344, 0.251) as well as disc area and RNFL cross section area (correlation coefficient: 0.299). RNFL thickness and RNFL cross section area were also correlated with a correlation coefficient of 0.965.

DISCUSSION

Previous reports have described the topographic and quantitative relation between the optic nerve head, RNFL, and visual field defects. The introduction of the HRT and GDx is helpful for investigating quantitative associations. Many global or sectoral topographic parameters measured with the HRT have been found to be correlated with global or sectoral visual field indices. They included cup area, cup/disc area ratio, rim area, rim volume, cup shape measure, RNFL thickness, and RNFL cross section area. Different studies revealed different parameters as having the strongest correlation with the visual field. Most of the studies emphasised the importance of cup/disc ratio,²⁶ rim area,²⁷ cup shape measure,^27–30 RNFL thickness,^26,31–33 and RNFL cross section area.^26,30 These correlations were stronger in the group combining healthy and glaucoma subjects than in the glaucoma subjects alone.^26,27,33 In addition, the correlations were better for MD than for CPSD.^27,30 With regard to the correlations for sectors, some studies showed less association for inferior RNFL measurements and the corresponding VF,²⁶ but some demonstrated less association for superior RNFL measurements.^33,34 However, most of the studies agreed that the temporal sector had the poorest association.^30,31 This is explained by the temporal sector usually remaining unaffected until a late stage of the disease.

In our study, correlations between HRT parameters and VF indices were only found between rim area and MD, CPSD and the number of points for p <5% on pattern deviation probability plot. There are several reasons why this study did not find as many significant correlations as reported in previous studies. Firstly, the sample size may not have been large enough. Tole et al³⁰ found all the significant correlations disappeared when analysis was confined to the glaucoma patients with MDs of <−10 dB. They supposed it reflected the statistical effect of reducing the numbers from 106 eyes to 61 eyes. However, relations between topographic parameters and MD were found with smaller sample sizes (47) in other studies.²⁷ Secondly, our study only included glaucoma patients. As mentioned above, the correlations were better for combined groups owing to the larger range of values. Thirdly, the visual field defects in our study were less severe compared with other studies. Mean MD in our study was −5.77 (SD 4.91) dB. Only two (3.2%) patients had an MD worse than −20 dB. In other studies, the mean MD ranged from −6.38 to −10.12 dB^26,27,30,32 with a higher percentage of glaucoma patients with the MD worse than −20 dB (11.5%,²⁶ 7.5%³⁰). The mean CPSD in our study was 6.21 (SD 3.29) dB with 83.9% of them having a CPSD of <10 dB. The mean CPSD for other studies ranged from 5.16 to 9.16 dB^27,32 with a higher percentage of glaucoma patients with CPSD >10 dB (28.3%³⁰). From the scatter plots, we noticed that advanced glaucoma cases had a large influence on the relation between topographic parameters and VF indices. Since our study included less advanced glaucoma cases, it could be expected that it would not reveal such high associations. This finding is supported by the research of Teesalu et al.^29,34 Their scatter plots also demonstrated that the advanced glaucoma patients played an important part in establishing an association. When the advanced glaucoma patients, with an MD worse than −10 dB, were excluded from the data set, the associations were considerably weakened. Fourthly, from our scatter plots, for a specific VF index, the value of the corresponding topographic parameters had a large range. The large intersubject variability weakens any relation between topographic parameters and VF indices. Based on the 95th percentile of the standardised rim/disc area ratio, Bartz-Schmidt et al³⁵ calculated the relative rim area loss and correlated this with the visual field index MD. The scatter plot showed an exponential change with 40% of the neuroretinal rim area lost before the appearance of the first VF defect. Half of the neuroretinal tissue had been lost when a 6 dB decrease in the MD occurred. In the early stage of glaucoma, a small change in MD was accompanied by a large amount of rim loss. While in the later stages of glaucoma, a large change in MD was only accompanied by a small amount of change in rim loss. Their scatter plot also demonstrated great interindividual variation among glaucoma patients. The variation was greater for early glaucoma than for advanced glaucoma. For MD between 0 to −6 dB, the relative rim loss can range from 0% to 70%. On the other hand, for MD between −6 dB and−12 dB, the relative rim loss only ranged from 40% to 80%. Hence, greater interindividual variation in early glaucoma further hampered the relation, especially when most of the subjects had only mild VF loss.

The quantitative relation between RNFL and established VF defects can also be analysed with the GDx. Global RNFL parameters, including the “number,” average thickness, ellipse average, total polar integral, maximum modulation, and ellipse modulation have all been reported to correlate with the MD for populations including normal and glaucoma subjects. The sectoral parameters, such as superior and inferior average thickness, superior/temporal and inferior/temporal ratios, and the deviation from normal for superior and inferior sectors, have also been found to correlate with MD.³⁶ The “number,” maximum modulation, and ellipse modulation have also been found to be correlated with CPSD. But the correlation coefficients were not as high as that for MD. However, when analysing a population of glaucoma patients, correlations were only found between the “number” and mean deviation and between maximum modulation and mean deviation.³⁷ Generally speaking, the “number,” maximum modulation, and ellipse modulation had a better relation with MD and CPSD than other RNFL parameters.

In this study, we found that the “number,” maximum modulation, and ellipse modulation correlated well with the global visual field indices, including MD, CPSD, and the number of points missed on pattern deviation probability plot at the 5% and 1% levels. The importance of modulation parameters was emphasised by the study in which modulation parameters correlated more strongly with MD than summed or ratio based parameters. However, no strong correlation was found between any parameter and PSD.³⁸ Our scatter plots demonstrated that the advanced cases have greater influence on the relation between RNFL parameters and visual field indices. Moreover, for a specific value of VF index, there was a larger degree of variability in the RNFL parameters. Thus it is very difficult to accurately predict RNFL measurements or VF from each other. The study of Kwon et al³⁹ also mentioned this problem. The VF mean sensitivity and deviation showed a bilinear correlation with peripapillary nerve fibre layer thickness. This bilinear correlation made the accurate predictions of one from the other difficult and made it difficult to establish a critical RNFL thickness. A normal VF can be associated with a wide range of RNFL thicknesses. When the RNFL thickness was greater than 70 μm, the VF mean sensitivity was nearly normal and changed little. When RNFL thickness was below this level, it was associated with a rapid decrease in the VF sensitivity. The precise correlation between VF and GDx was further hampered by a large individual variability around the bilinear fit.

For both HRT and GDx, the relations were better and more parameters were found to have correlation with VF indices for sectors than for the whole area. A certain amount of ONH or RNFL damage must occur before the global parameters showed significant changes. Thus, interpretation of the global parameters may overlook focal damage of ONH or RNFL. This may be the reason why sectoral parameters and sectoral VF indices had a better association than the global parameters and global VF indices.

It is hard to explain why the relations between the RNFL parameters, by either HRT or GDx, and VF were better in the inferior hemiretina than in the superior hemiretina. One explanation for this may be related to different amounts of damage in the inferior and superior hemiretinas. The inferior sector is usually the first to be damaged,⁴⁰ and in our study there were more cases with advanced VF damage in the superior field. Nevertheless, we did not find significant differences between the MD of superior and inferior hemifields although there were significant differences for the RNFL parameters. Surprisingly, different results were obtained from these two instruments. HRT revealed that the RNFL thickness was greater in the superior sector while the GDx found that the RNFL thickness was greater in the inferior sector. Furthermore, no correlation was found between GDx and HRT in RNFL parameters. The discrepancy and lack of any relation between these two instruments may derive from different principles of RNFL measurements. The HRT does not measure the RNFL directly, rather it measures the distance between a defined reference plane and the vitreal surface of the retina. Thus, the RNFL thickness reflects not only the RNFL, but also the supporting glial tissue including astrocytes and Müller cell axons. In contrast, the GDx is independent of a reference plane or magnification.

The correlation between structural damage and functional damage depends on two aspects. From a statistical point of view, the sample size and the characteristics of subjects have a great influence on the result. Larger sample sizes and a larger range of values all contribute to a better correlation. Thus, different studies may get different results because of different sample sizes and different inclusion criteria. From a clinical aspect, the large interindividual variation is the main reason for the poor correlation. It prevents us from predicting accurately the structural damage and functional damage. In an unselected clinical population, the interindividual variation will be even greater than the study results. The larger variation in early glaucoma than advanced glaucoma makes the prediction even more difficult during the early stages of the disease. Moreover, the temporal lag between changes in structure and function may also hamper the prediction. However, a reliable structural or functional estimate is very important clinically, especially for early glaucoma, because it will influence our decision about diagnosis and treatment. In this study, the associations were better between the GDx results and the visual field than between the HRT results and the visual field. GDx has modulation parameters that correct for individual variation by using the patient’s own temporal and nasal RNFL thickness as a baseline, which may account for the better correlation. Another reason may be that the RNFL is responsible for the visual function, while the optic nerve head changes such as cup enlargement and deepening are the consequence of RNFL loss.

There are two explanations for the large individual variation in the relations between topographic and RNFL parameters and visual field indices. One is the large variation of these parameters within the normal population. The other is individual differences in the amount of RNFL damage necessary for visual field loss to occur. The study of Bartz-Schmidt et al³⁵ revealed that the amount of rim area loss for certain degree of visual field loss is different for each individual. Evaluating the correlation between ONH or RNFL changes and visual field changes by longitudinal study, instead of looking at the cross sectional data, could, therefore, be of value. Longitudinal analysis is also a good method of removing the influence of the individual variations. Poinoosawmy et al⁴¹ investigated longitudinal RNFL thickness change in patients with normal tension glaucoma. Patients with initially better visual fields had a greater reduction in RNFL thickness than did those with initially more advanced visual field defect. Significant correlations between change in RNFL thickness and change in MD over the same period were found.

The HRT and GDx instruments provide quantitative measurements of ONH and RNFL suitable for accurate comparison and the monitoring of change.⁴² The examination procedures are simple and fast with fewer restrictions with regard to pupil size and cataract status than standard fundus imaging. However, global or sectoral measurements might overlook focal damage. Since focal damage of ONH or RNFL is a very important and specific sign of glaucoma, we need better parameters to reveal focal damage. The ranked segment distribution (RSD) curve technique⁴³ in HRT was designed for this purpose. Although the GDx does not have this kind of analysis, it does compare the patient’s data with an ethnic specific normal database. The ratio and modulation parameters of the GDx corrected for individual variation have been found to be more suitable for discriminating normal from abnormal.^21,38

Some important information about the configuration of optic nerve head, such as rim pallor, ISNT rule of rim width, vertical elongation of cup, disc haemorrhage, as well as alterations of lamina cribrosa and vessels, are difficult to obtain from the HRT measurements. Therefore, professional judgments, considering different aspects of information, are essential. Nevertheless, the quantitative measurements from HRT and GDx can assist the clinical judgment for early diagnosis and detection of progression. It is especially helpful for detecting diffuse RNFL damage.

The visual field test depends upon the patient’s responses and is more time consuming than HRT and GDx examinations. It would, therefore, be convenient if we could predict visual function from the ONH and RNFL morphology. Unfortunately, the correlation between structure and visual function is not good enough. More appropriate parameters are required to overcome the problem of individual variation. Furthermore, longitudinal studies that establish the value of HRT and GDx in detecting progression, and the correlation between structural changes and functional changes are essential for early diagnosis. At present, visual field examination can’t be replaced by imaging of the optic nerve head. Structural information provided by HRT and GDx and functional information provided by VF are both important and are complementary to each other. To obtain a full characterisation of glaucomatous damage, both structural and functional measurements are necessary and should be considered for clinical judgment and treatment.