Development of an Age-corrected Normative Database for Saccadic Vector Optokinetic Perimetry (SVOP).

PRECIS
Normal age-corrected threshold sensitivity values were determined for a new eye tracking perimeter and compared with standard automated perimetry (SAP).


PURPOSE
The purpose of this study was to determine threshold visual field sensitivities in normal subjects performing saccadic vector optokinetic perimetry (SVOP), a new eye tracking perimeter.


PATIENTS AND METHODS
A total of 113 healthy participants performed SVOP and SAP in both eyes with the order of testing randomized. The relationship between SAP and SVOP sensitivity was examined using Bland-Altman plots and 95% limits of agreement. The relationship between sensitivity and age was examined by pointwise linear regression and age-corrected normal threshold sensitivities were calculated.


RESULTS
After excluding unreliable tests, 97 participants with a mean age of 65.9±10.1 years were included. Average SAP mean deviation was -0.87±1.56 dB, SAP sensitivity was 29.20±1.68 dB and SVOP sensitivity was 32.18±1.96 dB. SVOP had a longer test duration (431±110 compared with 307±42 seconds for SAP, P<0.001). On average, the mean sensitivity obtained using SVOP was 2.98 dB higher than average SAP sensitivity, with 95% limits of agreement of -0.11 to 6.15 dB. For each decade older, SAP sensitivity decreased by 0.93 dB (95% confidence interval: 1.21 to 0.64) and SVOP sensitivity decreased by 1.15 dB (95% confidence interval: 1.47 to 0.84).


CONCLUSIONS
The results provide age-corrected normative values for threshold sensitivities from SVOP. Overall, SVOP provided a similar shaped hill of vision as SAP however threshold sensitivities were higher, meaning results are not interchangeable.

Perimetry is routinely used to screen for abnormalities of visual function caused by diseases such as glaucoma, with standard automated perimetry (SAP) the prevailing modality. SAP uses a white static stimulus displayed against a white background, with the contrast of the stimulus varied according to a staircase strategy to determine differential light sensitivity (DLS). 1,2 SAP has become the 'gold standard' perimetric test, yet patients often find it difficult to perform. 3 Though patients accept that visual field testing is important, they find it more demanding than other common clinical tests and a qualitative investigation discovered a perception among patients that multiple visual field tests were needed to become comfortable and to gain an accurate representation of their vision. 3,4 A further problem is that SAP is subject to considerable test-retest variability with the result that patients may need repeat testing to establish a diagnosis and to confidently identify change over time. 5 In 1989 Johnston and colleagues described a method of computerized perimetry using a moving fixation target. 6,7 Throughout testing patients were required to use a mouse to hold a circle-shaped cursor over a moving fixation target. Stimuli were only presented when the cursor was held in the correct position and the patient was instructed to press a response button when the stimulus was seen. A similar approach is used by the Melbourne Rapid Fields (MRF) visual field test, an FDA cleared application, to enable testing of 30 degrees of visual field using a tablet computer with a screen size of only 9.7 inches. 8 The MRF also relies on the patient touching the screen, clicking a mouse, or pressing a key to register responses to stimuli. Eye trackers have long been used be used to monitor fixation during perimetry, however, several groups have recently developed software for eye-tracking based perimetry. [9][10][11][12] For example, Jones and colleagues used a 50 Hz eye tracker mounted on a tablet computer using magnets to test visual field using fixed luminance stimuli across 20 degrees of visual field. 12 In a feasibility study, there was strong correlation between SAP Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.
A C C E P T E D mean deviation and measurements obtained from the eye-tracking perimeter, and eyetracking perimetry had good ability to differentiate patients with glaucoma from controls.
We have recently described a new method of automated threshold visual field assessment, known as Saccadic Vector Optokinetic Perimetry (SVOP). [13][14][15][16][17] SVOP uses eye tracking to quantify natural eye movements that occur in response to stimuli presented using a computerized screen. SVOP was initially developed for use in children and so was designed to not require a chin rest or need the patient to press a response button. 13. The software uses the preceding stimulus as the fixation spot for the next stimulus and automatically adjusts the size and position of the stimulus allowing the patient to move their head freely during testing.
A stimulus is registered as seen if there is a saccadic eye movement towards the stimulus within a prespecified time from first presentation. SVOP therefore potentially provides a more intuitive experience for patients, which is likely to be particularly beneficial for those that struggle to maintain fixation or have difficulty holding and pressing a button to register a response.
We have previously shown good correlation between threshold sensitivity values obtained with SVOP and SAP in patients with glaucoma and demonstrated SVOP to have high repeatability. 16,17 Most patients found SVOP comfortable, with 71% of patients preferring SVOP compared to SAP. 17 The aim of this study was to examine threshold sensitivity values of healthy individuals using SVOP and to compare results to SAP. These values could then be used to develop a normative database of differential light sensitivity values for the SVOP device.

PATIENTS AND METHODS
Healthy participants were recruited through the Scottish Health Research Register (SHARE), a database of over 206,000 people interested in participating in medical research. 18 Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.
A C C E P T E D All participants were provided written informed consent prior to enrolment and all study methods were prospectively approved by the South-East Scotland Research Ethics Committee (Reference 13/SS/0045). The study adhered to the tenets of the Declaration of Helsinki.
Participants underwent a baseline optometric examination, including best-corrected visual acuity and focimetry of the current spectacles. Healthy participants were required to have no previous history of significant eye disease, no known history of visual field defect and no neurological conditions that might affect the visual field. All participants were required to have a best corrected visual acuity of 20/30 or better in each eye. Participants with refractive error of greater than  7D spherical equivalent or more than 3D cylinder were excluded.
All participants completed SAP and SVOP in both eyes with the order of testing randomized. SAP was performed using a Humphrey Field Analyzer (HFA) 750i (Carl Zeiss Meditec, Dublin, CA) using the 24-2 test pattern and SITA Fast algorithm. SVOP was performed at the same visit using a research device, which has previously been described in detail. 16,17 Visual field tests were reviewed for reliability. SAP tests with ≥15% false positives or ≥20% fixation losses were considered unreliable and excluded. SVOP does not provide information about false positives or fixation losses as it inherently accounts for fixation by only presenting stimuli once fixation is achieved for each test location. SAP and SVOP tests were reviewed for artefact, e.g. lid artefact, and were excluded if artefact was present.
Participants found to have a visual field defect confirmed by repeat testing were excluded as the study was focused on normal individuals.

Saccadic Vector Optokinetic Perimetry (SVOP)
The threshold SVOP device consists of a personal computer with a 24" highresolution LCD screen (Eizo ColorEdge CG243W, Hakusan, Japan) and an eye tracker (X2-Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited. A C C E P T E D 60 model, Tobii Technology, Stockholm, Sweden) ( Figure 1). 16 The eye tracker assesses gaze responses to stimuli presented on the display screen. A stimulus is registered as seen if there is a saccadic eye movement towards the stimulus within a prespecified time from first presentation. The eye tracker also provides 'real time' data on eye location meaning that the size and position of the stimuli can be automatically and continually adjusted to compensate for changes in the patient's position during testing.
During testing, participants were seated in front of the LCD screen with their eyes aligned with the screen's center at an initial distance of 55 cm. Each eye was tested separately using custom made test spectacles, which occluded the non-test eye with an infrared bandpass filter. A best vision sphere lens was then placed in front of the test eye, the power of which was calculated based on focimetry readings taken from the subjects' own glasses and adjusted for the (55 cm) working distance of the test where necessary. The infrared bandpass filter enabled the eye tracker to detect the position of the occluded eye to monitor the position of the patient. The test began with a 20 second demonstration which was followed by a calibration sequence. During testing the patient was instructed to follow their natural reaction to fixate towards any stimulus perceived. The eye tracker evaluated responses to the stimuli and software determined whether the stimulus had been seen based on the direction and amplitude of the gaze response. Whether or not the stimulus was seen was determined based on the direction and amplitude of the change in eye position relative to the stimulus and the point of fixation (preceding stimulus) . The start of a fixation change was defined as the start point of a >50 pixels change in gaze and the end location of a fixation change was defined by the point at which 5 consecutive gaze data samples were separated by a distanced of <50 pixels, occurring after the detection of a fixation change start point.
Stimuli were equivalent to Goldmann size III and each stimulus was presented for 200ms using coordinates equivalent to the SAP 24-2 test pattern. A duration of 200 ms was selected Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.

A C C E P T E D
as it is the same duration used by the HFA and previous work has shown visual processing speed and speed of the saccade response is sufficiently fast to reach a stimulus within this time period. 19,20 Saccades larger than 5 degrees take only approximately 20 to 30 ms, with an additional 2ms for each additional degree. 19 As SVOP uses a flat LCD screen rather than a projection system, the way in which stimuli of different luminance are displayed is inherently different to SAP. In an LCD screen, light is provided by a fluorescent backlight, which passes through layers of polarizing material, attenuating and filtering light to produce different colors. Different colors have different luminance as more or less backlight is allowed to pass and therefore different colors can be used to produce different levels of luminance. Grey-scale level colors were produced by setting red, green and blue (RGB) levels to equal each other and luminance varied by adjusting levels. RGB levels range between 0 and 255, therefore the maximum level of luminance was obtained by setting the RGB level to 255, 255, 255 and the minimum to 0, 0, 0. Screen calibration was performed using a Look-Up Table to  Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.

A C C E P T E D
Thresholds were obtained using a 4-2 bracketing strategy and began by testing four 'seed' locations (one in each quadrant), which were then used to set the starting stimulus luminance levels for neighboring locations which in-turn were used to calculate the remaining starting luminance levels. The SVOP stimulus intensity and background intensity values were matched in luminance to those of SAP to allow direct comparison.

Statistical Analysis
Testing was performed for both eyes however the primary analysis was conducted on right eyes only. As differential light sensitivity is measured using logarithmic units, all sensitivity values were transformed to linear values to calculate average SAP and SVOP sensitivities for each eye. The relationship between SAP and SVOP visual field sensitivity was examined using scatter plots and pointwise linear regression. Bland-Altman plots were used to compare results from SAP and SVOP and determine 95% limits of agreement.
Normality was tested by inspection of histograms and by Shapiro-Wilk test.
Parametric variables were compared using student t-test and non-parametric variables compared using Wilcoxon rank sum test. The relationship between sensitivity values and age was examined by pointwise linear regression analysis and expected sensitivity values were estimated for each test point for a 50 year old individual. All statistical analyses were performed with commercially available software (STATA version 14; StataCorp LP, College Station, TX). The α level (type I error) was set at 0.05.   Agreement between SAP and SVOP was also evaluated for the 40 participants randomized to SAP first and compared to the 57 randomized to SVOP first. For those tested with SAP first, the mean difference in average sensitivity was 2.89 dB (95% CI 2.32 to 3.45 dB), with 95% limits of agreement of -0.54 to 6.32 dB. For participants tested with SVOP first, the mean difference in average sensitivity was 3.12 dB (95% CI 2.71 to 3.52), with 95% limits of agreement of 0.08 to 6.15 dB.

Both
The shapes of the predicted hills of vision for a 50-year-old subject obtained by SVOP and SAP were also compared graphically using a three-dimensional surface plot (Figure 7).

DISCUSSION
The results of this study provide age-corrected normative values for threshold sensitivities using SVOP, a new eye tracking perimeter. Overall, SVOP provided a similar shape 'hill of vision' for normal subjects as SAP using the SITA Fast strategy (Figure 7), however, thresholds with SVOP were on average higher than with SAP. The overall average threshold sensitivity with SVOP was 32.18  1.96 dB compared to 29.20  1.68 dB with SAP. Mean sensitivity from SVOP was therefore, on average, almost 3 dB higher than from SAP, with relatively wide 95% limits of agreement of -0.11 to 6.15 dB. These findings indicate that SVOP and SAP cannot be used interchangeably.
Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.

A C C E P T E D
A possible reason for the higher threshold detected with SVOP is the longer test duration, which may have caused participants to become fatigued. However, we have previously found patients prefer SVOP compared to SAP despite the longer test duration. 17 This is likely due to the lack of a need to maintain position throughout the test or for the patient to place their chin on a rest. Another possible reason for the higher thresholds with SVOP is lack of precision, however this is unlikely to be the case as SVOP has previously been shown to have good repeatability. 14 It is also possible that differences were due to calibration of the LCD screen, and though we tested luminance across the screen using a photometer prior to enrolling the first participant, this was not retested during the study.
Several previous studies have examined normal threshold values for perimetry using various SAP testing strategies, with many showing differences in thresholds between tests. 1,2,21-23 Bengtsson and colleagues found average age-corrected normal sensitivity was higher for SITA Standard and SITA Fast compared to Full Threshold, with mean sensitivities of 29.5 dB, 29.9 dB and 28.3 dB respectively. 2 The differences between strategies were largest in the more peripheral test points. It was concluded that the normal hill of vision was likely to be somewhat higher and smoother with SITA Fast and SITA Standard compared to Full Threshold testing. We found, differences between SVOP and SAP using SITA Fast were similar across test points which led to similar shaped hills of vision (Figure 7). Differential light sensitivity is known to decrease with age. We found a similar relationship between age and threshold sensitivity using SVOP and SAP, though the rate of age-related decline was slightly faster with SVOP. On average, SAP sensitivity decreased by 0.93 dB per decade (95% CI 1.21 to 0.64). Bengtsson and colleagues found an age-related reduction of 0.62 dB per year with SITA Fast, however their study used the 30-2 rather than 24-2 test pattern used in the present study. 2 We found SVOP sensitivity decreased by 1.15 dB per decade (95% CI 1.47 to 0.84). Bengtsson and colleagues also observed differences in age-Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.
A C C E P T E D related decreases in differential light sensitivity between testing strategies with 25% and 23% smaller changes compared to Full Threshold testing for SITA Fast and SITA Standard respectively. 2  Heijl and colleagues have previously reported steeper age-related declines in visual field sensitivity in more peripheral test points, when using a Full Threshold test. 1 Point wise age-related slopes ranged from 0.36 to 1.18 dB per decade leading to a depression and steepening of the normal hill of vision with age. Our analysis of SAP tests found point wise age-related slopes of 0.05 to 0.28 dB per decade ( Figure 5). Though we found slower predicted age-related decline in sensitivty than Heijl, likely due to use of the SITA Fast 24-2 rather than 30-2 Full Threshold strategies, there was agreement in finding faster rates of agerelated loss in midperipheral compared to paracentral test locations, leading to depression and steepening of the hill of vision. The relationship between age and visual field sensitivity was similar for SVOP ( Figure 4), with similar point-wise age-related slopes ranging from 0.07 to 0.22 dB per decade. The predicted rates of age-related decline were however, more similar across the visual field. For example, Figure 4 shows paracentral SVOP test points had rates of change (coefficients) of between 0.14 and 0.22 dB per decade, whereas peripheral test ponts ranged between 0.07 and 0.21 dB per decade. Therefore, the age-related decline in SVOP sensitvities appears to be more uniform, with a depression in the hill of vision, similar to that observed with SAP, but without the steepening. SVOP has potential advantages over conventional automated perimetry, including the lack of the need for the patient to concentrate on maintaining fixation on a single point, and the lack of a need to press a response button. And it is perhaps for these reasons that SVOP seems to be preferred by patients. 17 SVOP also provides information on characteristics of saccades such as accuracy and latency, which have been shown to be impaired in glaucoma. 24 Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.

A C C E P T E D
Though not directly examined in the present study, it would be interesting to examine whether characteristics of saccades may be of additional diagnostic value to differential light sensitivity alone. The present study has however highlighted potential disadvantages of SVOP compared to SAP, including the significantly longer test time, with SVOP on average taking over 7 minutes compared to only 5 minutes for SITA Fast. At present SVOP uses a full threshold algorithm so it is likely test time could be reduced with further modificiations to the algorithm.
It is also important to emphasize limitations of the study, including the relatively small number of participants, and that testing was conducted using a single SVOP device at a single site. In addition, patients performed only one SAP and SVOP test in each eye meaning that we were not able to assess variablity in thresholds in this particular group. Variability in visual fields is likely to be greater in those without previous experience, which may affect the accuracy of test results. Previous studies assessing normal threshold sensitivity values have tended to include results from second or third tests. 2 We did however test right and left eyes and found similar results from both eyes, with no sign of learning effect affecting results. A further potential limitation is the choice of SITA Fast tests as the reference standard. SITA Fast was selected as this is the test used in routine practice in our department, however it is important to appreciate that threshold sensitivity values tend to be higher with shorter test, perhaps as visual fatigue tends to decrease threshold values with increasing test time. 25 A comparison between SITA Standard and SVOP would have likely shown a smaller difference in test time and threshold sensitivity values. It is also important to emphasize that in some circumstances eye tracking may be problematic, for example in patients with nystagmus, ptosis, or corneal or pupil abnormalities, including anisocoria and abnormal pupil shape, eye trackers may not function correctly. In preliminary work developing SVOP we also noted a failure of eye tracking in a patient with a large iridectomy. In this case the tracker jumped Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.
A C C E P T E D between the pupil and iridectomy and the test could not be completed. A further limitation is that we did not examine missing or invalid eye movement data, which may have been useful a measure of gaze tracking reliability. This data is not currently provided as an SVOP output measure, but it may be useful to develop as an additional reliability index for SVOP testing.
Despite limitations, eye-tracking perimetry is an attractive proposition which overcomes some of the disadvantages of SAP. Several groups have developed devices for eye-tracking perimetry proving feasability and widely reporting that it is preferred by patients to SAP. 7-12,16 17 Eye tracking perimeters have also shown good ability to determine threshold visual field sensitivity and produce maps of visual field defects with patterns exhibiting close agreement to SAP. 17 However, to the best of our knowledge, no attempt has been made to generate a normative database for an eye-tracking perimetry device. The present study provides age-corrected normative data for the threshold SVOP test, which may be useful in determining cut off values for detection of abnormality. The normative values could be used to generate age-corrected deviation maps, similar to those generated by other automated perimeters.
Copyright © 2020 Wolters Kluwer Health, Inc. Unauthorized reproduction of the article is prohibited.