Glaucoma

GLAUCOMA DIAGNOSIS USING TENDENCY ORIENTED PERIMETRY

MANUEL GONZÁLEZ DE LA ROSA, VICTOR ARTEAGA, GUSTAVO FERNÁNDEZ-BACA and MARTA GONZÁLEZ-HERNÁNDEZ

Hospital Universitario de Canarias, Universidad de La Laguna, Spain

Abstract

Purpose: Looking over the data from a previous study, we noticed that the mean defect (MD) had similar diagnostic ability with TOP as it did with a bracketing strategy, but that the loss variance (LV) was better than MD with TOP and worse with a bracketing strategy. In this study, we evaluated several perimetric indices using TOP. Methods: a. One hundred and thirty-nine visual fields from patients with glaucoma (mean MD = 7.5 dB; SD = 6.7 dB) and 89 non-glaucomatous subjects were examined using TOP-32; and b. sixty-five glaucomatous (mean MD = 6.1 dB; SD = 2.7 dB) and 62 non-glaucomatous subjects were examined with G1-TOP. ROC analysis and differences between the two strategies were evaluated, as follows: number of points deviating by more than 5 dB from age-matched normals (NPP); MD; sLV (square root LV); and an empirical criteria (3/7 criteria) consisting on the presence of at least three of seven of the following criteria NPP > 2, sLV > 3 dB, MD > 6.7 dB, and sLV in areas S3, S2, I2 and I3 higher than 2.55 dB, as well as other cut-off levels close to this last value. Results: In both cases, the best accuracy was achieved with the 3/7 criteria, followed by sLV, NPP, and MD in descending order. The accuracy with the 3/7 criteria was between 91 and 95%, while it was lower than 90% for almost all the other indices. Conclusions: Although LV values are lower with TOP than with a bracketing strategy, its diagnostic ability is higher than that of MD. The best results were obtained with the 3/7 criteria, which takes into account the regional LV at four nasal areas, corresponding to specific ganglion cell axon bundles.

Introduction

Mean defect (MD) is a robust perimetric index that has been widely used to classify the degree of glaucomatous loss.1 The loss variance (LV)2 and the distribution of the local deviations3 have also been used as diagnostic criteria. The number of points with significantly reduced sensitivity, their grouping forming clusters,4 and their asymmetry in the upper and lower visual fields,5 have also been used as diagnostic criteria, using conventional statistical techniques, logistic regression or neuronal networks.

Looking over the data from a previous study,6 in a reduced group of 15 normal subjects and 20 glaucomas, we noted that MD had similar diagnostic ability with TOP (sensitivity = 90.0%, specificity = 90.9%) as with bracketing (sensitivity = 90.0%,

Address for correspondence: Manuel González de la Rosa, C/. 25 de Julio, 34, 38004. Santa Cruz de Tenerife, Spain. Email: mgdelarosa@jet.es

Perimetry Update 2002/2003, pp. 157–163

Proceedings of the XVth International Perimetric Society Meeting, Stratford-upon-Avon, England, June 26–29, 2002

edited by David B. Henson and Michael Wall

© 2004 Kugler Publications, The Hague, The Netherlands

specificity = 88.6%), but that LV was better than MD with TOP (sensitivity = 96.7%, specificity = 95.5%), and worse with a bracketing strategy (sensitivity = 86.7%, specificity = 86.4%). In order to confirm this difference, we also studied the diagnostic precision of several TOP perimetric indices.

Material and methods

Two groups of patients were used in this study, one consisted of control subjects and the other of patients with glaucomatous neuropathy and perimetric experience. Two studies were carried out. The first with the normal and glaucoma patients being examined with TOP-32, and the second with TOP-G1. One eye per subject was examined. The diagnosis of glaucoma was made by experts in the disease, taking into consideration several clinical aspects, such as intraocular pressure, family history, and other risks factors, the state of the optic discs, and, in some cases, previous perimetries with characteristic and persistent glaucomatous defects, significant evolution from previous exams or marked asymmetry between both eyes. The patients were classified as being normal or glaucomatous before the visual field examination used for this paper. No subject/patient suffered from any other pathology or was on any medication that could affect the visual field. However, corrected spherical ametropia of less than 6 diopters, extreme pupil diameters, and incipient cataracts were allowed as long as visual acuity was better or equal to 0.5.

The first study was carried out on 139 visual fields from patients with glaucoma (mean MD = 7.5 dB; SD = 6.7 dB) and 89 control subjects examined using TOP-32 (mean MD = 1.8 dB; SD = 2.2 dB). The second study was carried out on 65 visual fields from glaucoma patients (mean MD = 6.1 dB; SD = 2.7 dB) and 62 from control subjects examined with G1-TOP (mean MD = 1.0 dB; SD = 1.7 dB).

The resulting visual fields were analyzed in order to calculate the mean defect (MD), the square root of the loss variance (sLV), the number of points with a deviation higher than 5 dB from the normal value estimated for a patient of the same age (NPP), and the sLV value in four (S2M S3, I2, I3) of the seven visual field areas created on the basis of the paths of the retinal ganglion cells (Figs. 1 and 2).7,8 Our purpose was to establish the diagnostic value of regional MD and LV values. Although several tests were performed, no regional MD values or LVs from areas I1, S2, or T were found to enhance diagnostic value. Therefore, these areas will not be discussed any further in this paper.

The distributions of probability scores for control subjects and glaucomatous patients were compared by means of receiver operating characteristic (ROC) curves and four different criteria: 1. number of pathological points (NPP); 2. MD; 3. sLV; and 4. an empirical criterion consisting of the analysis of the seven indices specified in the previous paragraph and the observation of how many of them gave values above a certain level. Therefore, ROC analysis of the first three indices (global NPP, MD, and sLV) was carried out using a conventional automatic procedure. For experiment 4, the ‘cut-off’ of the seven indices was set to give the best diagnostic discrimination between glaucoma and controls. When the index was higher than the cut-off level, it was given a value of 1, and if not, a value of 0. The sum of these seven values was used in the ROC analysis, where the areas, accuracy (percentage of correctly classified

Fig. 1. Delimited areas in the glaucomatous visual field (program 32) for the right eye.

Fig. 2. Delimited areas in the glaucomatous visual field (program G1) for the left eye.

cases), sensitivity, specificity, and positive and negative predictive values were calculated.

Since this type of procedure may overestimate the diagnostic accuracy, due to the criteria being sample-specific, two different samples were used, using two different examination programs (TOP-32 and TOP-G1), and the results analyzed using specific or equal ‘cut-off’ levels for both samples.

Results

The results of the study carried out with the 32 program are shown in Table 1, and those of program G1 in Table 2.

For the first three criteria, the optimum cut-off values obtained by ROC analysis were as follows:

Program 32: NPP = 9.5 (9 = normal; 10 = abnormal), MD = 2.3 dB and sLV = 2.45 dB Program G1: NPP = 9.5 (9 = normal; 10 = abnormal), MD = 2.91dB and sLV = 2.70 dB For the ‘3/7 criteria’, the final cut-off level was 2.5 in all cases (number of indices above the cut-off level: 2 ≤ normal; 3 ≥ abnormal). The best result was achieved when the diagnosis was established for three or more of the seven criteria, with values higher than

the cut-off level being established (‘3/7 criteria’).

Table 1. ROC analysis of the cases examined with TOP-32, using NPP, MD, sLV and ‘3/7 criteria’ as diagnostic arguments

In both cases, the best diagnostic accuracy was achieved with the 3/7 criteria, followed by sLV, NPP, and MD, in descending order. The diagnostic accuracy with the 3/7 criteria was between 91 and 95%, while it was lower than 90% for almost all the other indices.

The best results for program 32 in relation to the ROC area were given by the 3/7 criteria and by the global sLV, while for program G1, they were given by the 3/7 criteria and MD.

The cut-off levels for the different indices analyzed by the 3/7 criteria were always the same for NPP = 2, sLV = 3 dB, MD = 6.7dB, and sLV (S3) = 2.55 dB. However, there were differences between the programs (32 and G1) for the sLV of the remaining three areas (S2, I2 and I3). Diagnostic accuracy was close to 92% when a constant cut-off level of 2.55 dB was selected for all local sLV indexes. The results were slightly modified when the cut-off level was increased to 3 dB in area S2. By choosing values specially adjusted for each of the two studies, accuracy levels of 95% were achieved.

Exclusion of the statistic analysis of any of the seven indices made the results worse in both samples. However, the inclusion of the MD value of the four areas (S2, S3, I2, I3) or the MD and LV of areas I1, S1, and T did not enhance the results.

Discussion

The mean MD values of the non-glaucomatous population were higher than zero, probably as a consequence of including subjects with early loss. As we have already mentioned, our purpose was to establish diagnostic accuracy, in the absence of chorioretinal or visual pathway pathology. We chose a criterion that was relatively insensitive to miosis and media opacities.

The optimum cut-off level for NPP, MD, and sLV turned out to be slightly higher than that found in other papers. The cut-off level for NPP was the same in both

studies, even though program G1 examines less points than program 32. The cut-off levels for MD and sLV were not identical. This is probably due to the specific characteristics of the samples used in each study.

The diagnostic ability of sLV was higher than that of NPP and MD for both studies, but this difference was greater when using the 32 program. NPP diagnostic accuracy was similar in both cases, and higher than MD in the study carried out with TOP-32. Each of the indices that make up the ‘3/7 criteria’ results in an increase in diagnostic value. This is why one index has a lower value when combined with others than when it is used alone (NPP), one has a higher value (MD), and the others have an equivalent one (global sLV and sLV of each area).

The new diagnostic criteria classify as glaucomatous some cases with few pathological points, when these are grouped in a sector or when they produce high variance values in that sector (Fig. 3, top). Similarly, the new diagnostic criteria classify as ‘non-glaucomatous’ diffuse sensitivity losses due to, for example, cataracts or intense miosis, which may give very high MD and NPP values (Fig. 3, bottom).

It is known that TOP smoothes scotoma borders as a consequence of its interpolation system,9 giving LV values lower than those given by the bracketing strate- gy.10-12 However, in a previous paper, we reported that LV was better correlated with MD with TOP than with a bracketing strategy.13 Several factors could explain this fact. On the one hand, the maximum possible threshold value in TOP is 18/16 of the normal value corrected for the patient’s age, so that ‘white scotomas’ are filtered, although the influence of false positives cannot be completely avoided. On the other hand, it should be clarified that we do not use an auditory cue with TOP, as this has been found to increase the number of false positives.14 Other factors that could account for the better information given by sLV with TOP could be the better stability of its results,15 the avoidance of perimetric fatigue and of loss of attention by the patient, with the shorter test.

The reason for analyzing the results using a cut-off of sLV (S2) = 3.0 dB for the

Fig 3. Local deviations of two cases classified as glaucomatous by the 3/7 criteria (top) and two cases with diffuse defects considered as ‘non-glaucomatous’ by the same criteria (bottom). Values above the normality limit are shown with a black background and white letters.

3/7 criteria and of sLV = 2.55 dB for the rest of the areas was as follows: after concluding this paper, we had the opportunity to carry out a new analysis using the 3/7 criteria on data recently published by another research group.16 The original paper reported on the diagnostic accuracy of TOP, SITA, and FDT. With the 3/7 criteria, the best results were obtained using the same limits described in this paper for NPP, sLV, and MD, constant values of sLV = 2.55 dB in three areas, and sLV (S2) = 3.0 dB. The results of 3/7 criteria are better than those given in this paper.

This study shows that LV levels have a high diagnostic potential in TOP, probably because they give average values that reflect a loss of initial regularity in the visual field, while in the case of the bracketing strategy, its value may be overestimated as a consequence of local errors, loss of attention, ‘white scotomas’, and the influence of fatigue.

The confirmation of these results in new studies should guide us to including this analysis in those programs that use TOP.

References

1.Caprioli J, Zulauf M: The relationship of optic nerve fiber layer parameters with visual field loss in glaucoma. In: Mills P, Heijl A (eds) Perimetry Update 1990/1991, pp 9-14. Amsterdam/Milano/New York: Kugler & Ghedini Publ 1991

2.Damms T, Dannheim F, Ahlers S: A visual field index for nerve fiber bundle defects. In: Mills P (eds) Perimetry Update 1992/1993, pp 9-13. Amsterdam: Kugler Publ 1993

3.Bebie H, Flammer J, Bebie TH: The cumulative defect curve: separation of local and diffuse components of visual field damage. Graefe’s Arch Clin Exp Ophthalmol 227:9-12,1989

4.Fankhauser F1st, Fankhauser F2nd, Larsen SE, Cruz-Orive LM: Fluctuations of visual field interpretation related to cluster and scotoma analysis by one interpreter. German J Ophthalmol 3:43-47,1994

5.Asman P, Heijl A: Glaucoma hemifield test: automated visual field evaluation. Arch Ophthalmol 110:812-819,1992

6.Morales J, Weitzman M, Gonzalez de la Rosa M: A preliminary comparison between tendency oriented perimetry (TOP) and traditional threshold perimetry. Ophthalmology 107:134-142, 2000

7.González de la Rosa M, González Hernández M, Abraldes M, Azuara-Blanco A: Quantification of inter-point topographic correlations of threshold values in glaucomatous visual fields. J Glaucoma 11:30-34, 2002

8.González de la Rosa M, González Hernández M, Aguilar Estévez J, Abreu Reyes A, Pareja Ríos A: Clasificación topográfica del campo visual glaucomatoso. Arch Soc Esp Oftalmol 77:87-94, 2002

9.González de la Rosa M, Martinez A, Sanchez M, Mesa C, Cordovés L, Losada MJ: Accuracy of tendency oriented perimetry (TOP) in the Octopus 1-2-3 perimeter. In: Wall M, Wild J (eds) Perimetry Update 1996/1997, pp 119-123. The Hague: Kugler Publ 1997

10.Lachkar Y, Barrault O, Lefrancois A, Demailly P: Rapid tendency oriented perimeter (TOP) with the Octopus visual field analyzer. J Fr Ophtalmol 21:180-184, 1998

11.González de la Rosa M, Losada MJ, Serrano M, Morales J: G1-tendency oriented perimetry (TOP): introduction and comparison with G1-standard bracketing. In: Wall M, Wild J (eds) Perimetry Update 1998/1999, pp 43-49. The Hague: Kugler Publ 1999

12.Horikoshi N, Osako M, Goto H, Tamura Y, Okano T: Clinical evaluation of tendency oriented perimetry in Octopus perimeter. Jpn J Clin Ophthalmol 53:889-893, 1999

13.González Hernandez M, Martinez Piñero A, Fariña FJ, González de la Rosa M: Relation between the mean defect and the loss variance in TOP and bracketing perimetric strategies. (ARVO Abstract No. 361) Invest Ophthalmol Vis Sci 41:S68, 2000

14.Pareja Rios A, Arteaga Hernandez V, Gonzalez Hernandez M, Gonzalez de la Rosa M: Efecto del uso de estímulos acústicos en perimetría. Arch Soc Can Oftalmol 11:83-86, 2000

15.González de la Rosa M, Martínez Piñero A, González Hernández M: Reproducibility of the TOP algorithm results versus the ones obtained with the bracketing procedure. In: Wall M, Wild J (eds) Perimetry Update 1998/1999, pp 51-58. The Hague: Kugler Publ 1999

16.Wadood AC, Azuara-Blanco A, Aspinall P, Taguri A, King AJ: Sensitivity and specificity of fre- quency-doubling technology, tendency-oriented perimetry, and Humphrey Swedish interactive threshold algorithm-fast perimetry in a glaucoma practice. Am J Ophthalmol 133:327-332, 2002