Visual Field Evaluation Using Zippy Adaptive Threshold Algorithm (ZATA) Standard and ZATA Fast in Patients With Glaucoma and Healthy Individuals

Abstract

Purpose: To evaluate visual fields obtained with Zippy Adaptive Threshold Algorithm (ZATA) Standard and ZATA Fast from patients with glaucoma and healthy individuals.

Methods: Fifty-five patients with glaucoma (median mean deviation [MD], −7.6 dB; interquartile range [IQR], −15.3 to −2.6 dB) and 22 healthy participants (median MD, −0.6 dB; IQR, −1.7 to 0.2 dB) performed ZATA Standard and ZATA Fast tests on a Henson 9000 perimeter and Swedish Interactive Thresholding Algorithm (SITA) Standard and SITA Fast tests on a Humphrey Field Analyzer. Tests were repeated within 90 days (median, 14 days; range, 7–26 days) to evaluate the test–retest variability.

Results: The mean difference between the MD of the ZATA Standard and SITA Standard tests was 1.7 dB (95% confidence interval [CI], 0.9–2.4). Between ZATA Fast and SITA Fast, it was 0.9 dB (95% CI, 0.2–1.5 dB). Although there were systematic differences between the distributions of sensitivity estimates with ZATA and SITA, they did not affect the overall representation of damage by these tests. ZATA Standard and ZATA Fast were approximately 30% and 6% faster, respectively, than the corresponding SITA tests.

Conclusions: ZATA Standard and ZATA Fast are suitable for clinical practice. However, differences between ZATA and SITA tests suggest that they should not be used interchangeably when patients with glaucoma are followed over time.

Translational Relevance: This study examined the characteristics of ZATA visual field tests in a clinical population, and it supports the adoption of these tests for assessing patients with glaucoma.

Introduction

Glaucoma is an optic neuropathy that leads to progressive and irreversible visual field loss. It is a major cause of blindness in India and worldwide.^1–4 Static automated perimetry is a key measure of visual function in glaucoma⁵; it measures contrast sensitivity across the visual field by testing the patient's ability to detect low-intensity white light stimuli against a uniformly illuminated white background.⁶ Because a large number of visual field locations must be examined, highly efficient test algorithms are needed to keep the overall number of stimuli and the test duration within acceptable limits. 

Swedish Interactive Thresholding Algorithm (SITA) Standard and SITA Fast incorporated in the Humphrey Field Analyzer (Carl Zeiss Meditec, Jena, Germany) have been used for over two decades and have been thoroughly evaluated clinically.^7–10 SITA has become the de facto standard for visual field evaluation in glaucoma and has been used in landmark clinical trials such as the Ocular Hypertension Treatment Study and the United Kingdom Glaucoma Treatment Study.^11,12 

Zippy Adaptive Threshold Algorithm (ZATA) Standard and ZATA Fast, incorporated in the Henson 9000 perimeter (Topcon Healthcare, Oakland, NJ), are more recent innovations, and, to date, no published evidence of their performance is available. In this study, we focused on comparing the test–retest variability of ZATA Standard and ZATA Fast to that of SITA Standard and SITA Fast. 

Methods

Participants

In this prospective cross-sectional study, consecutive patients with glaucoma were recruited from the ophthalmology clinics of Kasturba Hospital, Manipal Academy of Higher Education, Manipal, Karnataka, India, from December 2020 to December 2021. Healthy participants were recruited from people attending their annual health check for employment or self-care. All participants received a complete eye examination. For both groups of participants, inclusion criteria were 35 to 85 years of age and a best-corrected visual acuity (BCVA) of 0.3 logMAR or better. Glaucoma patients had characteristic optic disc changes such as focal or diffuse neuroretinal rim thinning, localized notching, or nerve fiber layer defects. Only patients with primary glaucoma were included (primary open-angle or primary angle-closure glaucoma); patients with secondary glaucoma were not included. We did not include glaucoma patients with a recent history of uncontrolled intraocular pressure, those who were non-compliant with prescribed medications, or those who had previously failed to attend scheduled follow-up visits. All glaucoma patients had previously undergone at least one visual field test, but visual field loss was not an inclusion criterion for this study. 

Healthy participants had no optic disc abnormality or ocular pathology other than mild cataract or a previous history of uncomplicated cataract surgery. On the baseline visit, all participants underwent a demonstration visual field test before the study tests. 

Ethics

The study was registered in the Clinical Trial Registry of India (CTRI/2020/10/028348) and was approved by the Institutional Ethics Committee, Kasturba Medical College and Kasturba Hospital (approval no. 370/2020). The study adhered to the tenets of the Declaration of Helsinki. Written informed consent was obtained from all participants following a detailed explanation of the study in the participants’ preferred language (Kannada or English). 

Visual Fields

The four visual field tests performed were ZATA Standard, ZATA Fast, SITA Standard, and SITA Fast. The Henson 9000 perimeter (Topcon Healthcare) was used for the ZATA Standard and ZATA Fast tests, and the Humphrey Field Analyzer (Carl Zeiss Meditec) was used for the SITA Standard and SITA Fast tests. In both perimeters, the background intensity is 10 cd/m² and the nominal maximum stimulus intensity is 3183 cd/m². The four strategies used in the study are explained briefly below. 

ZATA Standard and ZATA Fast are Bayesian strategies constructed according to the principles of Zippy Estimation by Sequential Testing (ZEST).¹³ Bayesian strategies use prior information that is already available at the outset of the test to guide the presentation of subsequent stimuli. The prior information is expressed in the form of a probability density function (PDF). After each presentation, this prior PDF is updated with a likelihood function corresponding to the response (“seen” or “not seen”) to form a posterior PDF. In turn, the posterior PDF becomes the prior of the next presentation—until the width of the posterior PDF (i.e., the uncertainty of the threshold estimate) becomes smaller than a set criterion value, and testing is terminated. In the ZATA strategies, the mathematically rigorous principles of ZEST are modified by a small set of rules. For example, locations at which the first stimulus (mode of the PDF) is seen are not tested further. Also, if the mode of the PDF falls below 10 dB, testing is terminated at that location. Finally, no more than six stimuli (ZATA Standard) or four stimuli (ZATA Fast) are presented at each location. In patients who have completed a previous test, ZATA adjusts the initial PDFs according to the previous test results (personal communication, David B. Henson). 

SITA strategies are a combination of staircase and Bayesian algorithms that use prior knowledge of age-corrected normal sensitivities and intersubject variability of glaucomatous and normal visual fields.¹⁴ Based on this prior information, SITA builds two PDFs during the test, one for patients with glaucoma and another for normal observers.¹⁴ If a stimulus is not seen, the PDF decreases for the normal model and increases for the glaucoma model; the opposite occurs if a stimulus is seen.¹⁴ The sequence of stimulus presentations is similar to a full-threshold staircase.¹⁴ The SITA strategies follow a growth pattern where thresholds are first measured at four seed locations, one in each quadrant at 12.7° from fixation, and these thresholds are subsequently used to set the starting intensities for neighboring, as yet untested locations.¹⁴ The difference between SITA Standard and SITA Fast is that the latter allows a slightly greater tolerance to uncertainty and, therefore, requires fewer stimulus presentations than SITA Standard.¹⁵ 

Testing Protocol

All visual field tests were conducted on a 24-2 test pattern with Goldmann size III stimuli. The order of tests between the ZATA and SITA strategies was chosen randomly by a coin flip. Participants were allowed a break of 5 to 15 minutes between tests to avoid fatigue. If clinically indicated, glaucoma patients also underwent visual field tests on the eye not included in the study. All tests were repeated within 90 days to evaluate the test–retest variability. 

Data Analysis

The visual field data were converted to right-eye format, and blind-spot locations (15°, −3° and 15°, 3°) were excluded from further analysis. The data analysis was conducted using R 4.0.3.¹⁶ Mean deviation (MD) values, pointwise sensitivity, and test duration of ZATA Standard and ZATA Fast were compared to those of SITA Standard and SITA Fast. 

Results

Participants

Visual field data were obtained from 55 patients with glaucoma and 22 healthy participants. Demographic and ocular characteristics of the two groups are shown in Table 1. Figure 1 illustrates the distribution of MDs and pattern standard deviations (PSDs) for all participants. 

Table 1.

 

Details of the Study Participants

 

Figure 1.

 

Distribution of visual field indices in healthy individuals and patients with glaucoma (SITA Standard, baseline visit). For completeness, analogous plots for the three other tests are shown in Supplementary Figure S1.

MD Values of ZATA Standard and ZATA Fast Compared to SITA Standard and SITA Fast

The MD value provides an index of the overall visual field damage,¹⁷ with more negative MD values indicating more severe damage. There were statistically significant practice effects between the test (baseline) session and the retest session with ZATA Fast and SITA Standard, but these were small (<1 dB) and therefore of no clinical significance (Table 2). 

Table 2.

 

Median (25th, 75th Percentile) of MD Values and Their Differences Across the Two Visits for the Four Tests

 

The Bland–Altman limits of agreement plots in Figure 2 show the difference between MD values for the ZATA and SITA tests as a function of the average MD.¹⁸ The shaded region is the agreement interval—that is, the region that encompasses 95% of the data (mean ± 1.96 SD of the differences). 

Figure 2.

 

Bland–Altman plots of MD for ZATA Standard and SITA Standard (a) and ZATA Fast and SITA Fast (b).

The mean difference in the MD values of the ZATA Standard and SITA Standard was 1.7 dB (95% confidence interval [CI], 0.9–2.4) dB, and the 95% agreement interval extended from −5.2 to 8.5 dB (Fig. 2a). The mean difference in the MD values of ZATA Fast and SITA Fast was 0.9 dB (95% CI, 0.2–1.5), and the agreement interval extended from −4.7 to 6.4 dB. The MDs show that, on average, the ZATA Standard MDs were slightly less negative than those of the SITA Standard (Fig. 2a), and the ZATA Fast MDs were slightly less negative than those of the SITA Fast (Fig. 2b). The limits of agreement in the two figures (Figs. 2a, 2b) show that, in individual cases, the differences in the MDs of ZATA and SITA tests could be substantial. 

Test–Retest Variability of MD Values

Results of Bland–Altman analyses of test–retest variability¹⁹ in the MD values (difference in the MD values across the two visits) for the four tests are shown in Figure 3. The shaded region in the plot is the 95% test–retest interval; it is estimated by ±1.96 SD of the differences between test and retest and encompasses 95% of the data. It represents the overall test–retest variability of the MD values, which appeared broadly similar in the four tests. With all four tests, the data appeared less dispersed at more positive MD values (e.g., better than −5 dB), suggesting a relatively lower variability in the MD with less damaged visual fields.²⁰ 

Figure 3.

 

Bland–Altman plots showing MD values across the two visits for ZATA Standard (a), ZATA Fast (b), SITA Standard (c), and SITA Fast (d).

Frequency Distribution of Pointwise Sensitivity Estimates of ZATA Standard, ZATA Fast, SITA Standard, and SITA Fast

Figure 4 shows the frequency distribution and cumulative distribution of pointwise sensitivity estimates of ZATA Standard, ZATA Fast, SITA Standard, and SITA Fast tests. Frequency and cumulative distributions of ZATA Standard were close to those of ZATA Fast; similarly, the distributions of SITA Standard were close to those of SITA Fast. The frequency distributions of all four tests had a left-most peak at 0 dB (plotted as solid circles in Fig. 4a). For SITA Standard and SITA Fast, 9% and 12% of the sensitivity estimates, respectively, were 0 dB. For ZATA Standard and ZATA Fast, the proportions were 13% and 15%, respectively. The right-most peaks (high sensitivity) of the ZATA distributions were considerably taller and shifted to the right compared to the SITA distributions. 

Figure 4.

 

Frequency distribution (a) and cumulative distribution (b) of the sensitivities ranging from 0 to 35 dB obtained from the four tests across the two visits. The solid circles (a) represent the percentage of sensitivity estimates equal to 0 dB in the four tests (shown separately for better visibility). The ZATA Fast data have been shifted by 0.25 dB for better visibility.

Principal Curve Analysis of the Relationship Between ZATA and SITA Estimates

The conspicuous differences between the frequency distributions of sensitivity estimates of ZATA and SITA motivated an analysis of the relationship between the point-by-point sensitivity estimates of the two techniques. To obtain the best possible precision, we compared the average of all four ZATA tests with that of all four SITA tests, ignoring the minor differences between the Standard and Fast versions of each family. Because both techniques are subject to uncertainty and measurement error, a principal curve²¹ was fitted to describe the relationship (Fig. 5). For test locations in the near-normal range of sensitivity (approximately 25–35 dB with SITA), the slope of the principal curve was shallower than the 1:1 diagonal line, suggesting that ZATA compresses this spectrum into a somewhat narrower range. In addition, there was a tail of points with high SITA but low ZATA sensitivities. These were almost exclusively from a small number of ZATA tests with likely lens rim effects in the periphery. 

Test–Retest Variability of Pointwise Sensitivity Estimates

Figure 6 shows the distributions of the pointwise sensitivities on visit 2 (retest) as a function of those at visit 1 (baseline), stratified into broad categories. In the absence of variability, the test and retest sensitivities should fall into the same interval. Clearly, the retest distributions spread out over a much wider range. Tighter distributions (lower variability) are seen at higher baseline sensitivities and wider distributions (higher variability) at lower baseline sensitivities, such as 0 to 15 dB and 15 to 22 dB. 

Figure 7 shows the test–retest sensitivity range for the baseline sensitivity on the x-axis. For baseline sensitivities between 1 and 5 dB, the test–retest (10% to 90% percentile) ranged from 0 to 28 dB for ZATA tests and from 0 to 26 dB for SITA tests. At high baseline sensitivity (e.g., 25–35 dB), ZATA Standard and ZATA Fast test–retest ranges appeared slightly tighter than those of SITA Standard and SITA Fast. At locations with sensitivities below 25 dB, the test–retest intervals of ZATA appeared somewhat wider than those of the SITA strategies, particularly so at the margins of the distributions (10th and 90th percentiles, 5th and 95th percentiles). 

Test Duration

All four tests took slightly longer with damaged compared to normal visual fields (Fig. 8). With ZATA Standard and ZATA Fast, test duration in highly damaged visual fields declined. 

Visual Field Representations of ZATA Tests Compared to SITA Tests

Figure 9 shows examples of visual field damage with the four tests. ZATA and SITA tests represented deep visual field damage in a similar way (Figs. 9a, 9b), but shallow losses appeared somewhat less clearly with the ZATA tests in comparison to SITA (Figs. 9c, 9d). For reasons of space, only data from selected glaucoma patients are shown here; data from all other participants are given in the Supplementary Material. 

Discussion

The ZATA Standard and ZATA Fast tests of the Henson 9000 perimeter have not previously been studied in a clinical population. We compared these visual field tests to the widely used SITA Standard and SITA Fast tests of the Humphrey Field Analyzer. Our results suggest that, in most cases, ZATA and SITA tests characterized patients’ visual fields similarly; overall, the results of the tests compared closely to each other. Due to the tighter distribution of near-normal sensitivity estimates with ZATA, however, shallow or diffuse visual field damage was less clearly apparent with these tests compared to SITA. 

The MD values obtained from the ZATA and SITA tests agreed to within approximately ±5 dB in almost all patients (see limits of agreement in Fig. 2). Although this level of agreement can hardly be described as “close,” it is similar to the test–retest intervals of all four tests (Fig. 3). In other words, the differences among the tests appeared similar to the random differences seen when the tests were repeated. This would suggest that the tests are broadly equivalent to each other when patients are assessed for visual field damage. But, when patients are followed over time to monitor progression of visual field damage, the systematic differences between the SITA and ZATA strategies mean that the tests should not be interchanged. The pointwise analysis highlighted conspicuous differences in the shape of the distributions of sensitivity estimates between the ZATA and SITA tests. We believe that these differences reflect the prior PDFs as well as any rules used by the two algorithms.¹⁴ Because this information is not in the public domain, our explanation must remain speculative. However, the comparison of such distributions may provide highly informative clues on the properties of visual field test strategies. With the ZATA strategies, for example, the spectrum of near-normal visual field sensitivities is compressed into a narrower range in comparison to the SITA strategies. This is also apparent in the principal curve analysis, which revealed a systematic departure from the ideal 1:1 relationship between the estimates in the high-sensitivity region. As a consequence, shallow visual field damage may be less conspicuous with ZATA compared to the SITA strategies. 

The distributions of the retest sensitivities (Figs. 5–7) are useful for examining in greater detail the properties of the four tests. With both SITA and ZATA tests, the test–retest intervals increased with visual field damage (Fig. 7), as has been reported by others.^20,23 These data suggest that ZATA tests have somewhat lower test–retest variability than SITA tests in the high-sensitivity region, whereas the variability of ZATA tests is relatively higher in areas of low sensitivity. Both ZATA and SITA are Bayesian algorithms that make an explicit trade-off between accuracy (lack of systematic error or bias) and precision (test–retest variability).²⁴ By permitting a somewhat larger bias, the efficiency of threshold estimation can be increased to reduce the number of stimulus presentations and/or to reduce test–retest variability. It appears obvious that the sensitivity distributions of the ZATA tests have more prominent peaks (Figs. 4, 6) than those of SITA. In other words, ZATA has much fewer sensitivity estimates in the mid-range region (15–25 dB) compared to SITA. This will not impact the measurement of visual field areas that are normal or those with absolute or very deep losses, but it may well affect the appreciation of shallow losses or diffuse damage. 

Figure 5.

 

Sensitivity estimates of ZATA plotted against those of SITA. The principal curve shows the relationship between estimates of both techniques (red line). The 500 bootstrap replications were performed to demonstrate the stability of this fit (pink lines). The tail of green circles with low ZATA and high SITA estimates is largely caused by a small number of healthy participants with peripheral losses that were likely artifactual (participants #1, #3, #16, and #20; see Supplementary Material).

Figure 6.

 

Histograms showing frequency distributions of retest sensitivities for baseline sensitivities of 0 to 15 dB, 16 to 21 dB, 22 to 25 dB, and 30 to 33 dB. Retest distributions show high peaks at 30 dB in the top row (30–33 dB baseline sensitivity) and at 0 dB in the bottom row (0–15 dB baseline sensitivity). These peaks have been truncated to facilitate comparison. The bold number beside the peak is the true size of the peak; n = total number of retest points. Note the differences in the retest distributions between the ZATA and SITA strategies at the near-normal part of the sensitivity spectrum (top two rows).

Figure 7.

 

The 5th and 95th, 10th and 90th, and 25th and 75th percentiles of retest sensitivity at stratified baseline sensitivities. Percentiles of the retest sensitivity range are marked with different shade intensities: light (5th–95th), medium (10th–90th), and dark (25th–75th). The median and mean are indicated by dotted and continuous black lines, respectively.

Figure 8.

 

Duration taken to perform the tests as a function of visual field damage (MD). A Loess curve²² was fitted to the data (red line). For simplicity, the values of test duration and the MD were averaged across the two visits.

In many settings, visual field tests are performed primarily to evaluate patterns of visual field damage. Although there are fundamental differences between ZATA and SITA that are reflected in the sensitivity distributions, it is noteworthy that the patterns of damage reflected in the probability maps of the two tests were similar in a large majority of patients (Fig. 9; Supplementary Material). Because our study focused on test–retest variability, we did not carry out a more formal analysis of how the different tests reflect spatial patterns of visual field damage. Nevertheless, we feel that this is an important aspect of perimetric test performance that deserves further study, either through subjective assessment by trained readers or through more objective tools.²⁵ 

Figure 9.

 

(a) Example of deep damage in the superior hemifield. The damage appears practically identical for ZATA and SITA. (b) Deep damage involving both hemifields. Overall, the four tests highlight a similar pattern of damage. (c) Example with mixed diffuse and focal visual field damage. The ZATA tests showed scattered damaged locations mainly in the superior hemifield, but the total and pattern deviation probability maps of SITA suggest a greater degree of widespread or diffuse damage. (d) Example of a visual field with mainly diffuse visual field damage. For comparisons between the total and pattern deviation probability maps, the SITA tests showed a greater degree of diffuse loss compared to ZATA.

Our study was not designed, nor intended, to assess the diagnostic accuracy (sensitivity and specificity) of the four visual field tests. Such a study will require a different design (comparison of outcomes against an unbiased reference, such as from imaging), different participants (many more with suspected or early disease), and different analyses (e.g., analysis of total and pattern deviation probability maps rather than individual test locations). Our results suggest that attention may have to be focused on how visual field tests perform in patients who have predominantly shallow or diffuse loss, although it is not clear how frequently such losses occur in early glaucoma. 

In our study, SITA Standard showed 45% larger test–retest variability than reported previously,²⁶ and SITA Fast showed approximately 60% larger test–retest variability compared to a study by Heijl et al.²⁷ Our participants and the setting of our study were typical of a tertiary-care facility in India. In clinical studies and randomized trials, participants are typically more highly selected. For example, many participants of our study were relatively inexperienced with visual field tests. They were likely to have had some level of cataract (only 20% were pseudophakes). We deliberately did not exclude test results based on reliability indices, as false-positive or false-negative responses and fixation losses are not dependable indicators of test reliability.^28,29 Furthermore, we did not exclude patients with advanced visual field damage because such patients are an integral part of the case mix in our unit. Selection of patients with a narrower spectrum of visual field damage would have made it more difficult to assess the differences between ZATA and SITA at the extremes of the distribution. All these factors will have contributed to the larger test–retest variability seen in our study compared to others,^27,30 but they should not have affected comparisons among the four tests. 

In conclusion, visual fields obtained with the ZATA tests are broadly similar to those obtained with SITA; therefore, these tests can be used in clinical practice. However, ZATA and SITA tests should not be used interchangeably when patients are followed over time to monitor progression. 

Acknowledgments

The authors thank Elektron Eye Technology for donating Henson 9000 perimeters to the Department of Optometry, Manipal College of Health Professions, Manipal Academy of Higher Education, Manipal India. Elektron Eye Technology was not involved in the design or evaluation of this study. A part of this study was presented as a poster at ARVO 2022. 

Supported by the TMA PAI Endowment (PN) and an intramural grant (RSV and SBG) from the Manipal Academy of Higher Education. 

Disclosure: P. Nasim, None; R.S. Ve, None; N.I.R. Kuzhuppilly, None; P. Naik, None; S. Ballae Ganeshrao, None; P.H. Artes, None