Open Access
Glaucoma  |   February 2025
Detecting and Quantifying Glaucomatous Visual Function Loss With Continuous Visual Stimulus Tracking: A Case-Control Study
Author Affiliations & Notes
  • Anne C. L. Vrijling
    Laboratory of Experimental Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
    Royal Dutch Visio, Centre of Expertise for Blind and Partially Sighted People, Huizen, The Netherlands
  • Minke J. de Boer
    Laboratory of Experimental Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
  • Remco J. Renken
    Cognitive Neuroscience Center, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
    Department of Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
  • Jan-Bernard C. Marsman
    Cognitive Neuroscience Center, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
  • Joost Heutink
    Royal Dutch Visio, Centre of Expertise for Blind and Partially Sighted People, Huizen, The Netherlands
    Department of Clinical and Developmental Neuropsychology, University of Groningen, Groningen, The Netherlands
  • Frans W. Cornelissen
    Laboratory of Experimental Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
  • Nomdo M. Jansonius
    Laboratory of Experimental Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
    Department of Ophthalmology, University Medical Center Groningen, University of Groningen, Groningen, The Netherlands
  • Correspondence: Nomdo M. Jansonius, University Medical Center Groningen, Department of Ophthalmology, HPC BB61, PO BOX 30.001, Groningen 9700 RB, The Netherlands. e-mail: [email protected] 
  • Footnotes
     ACLV and MJB contributed equally to this work and share first authorship.
Translational Vision Science & Technology February 2025, Vol.14, 3. doi:https://doi.org/10.1167/tvst.14.2.3
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Anne C. L. Vrijling, Minke J. de Boer, Remco J. Renken, Jan-Bernard C. Marsman, Joost Heutink, Frans W. Cornelissen, Nomdo M. Jansonius; Detecting and Quantifying Glaucomatous Visual Function Loss With Continuous Visual Stimulus Tracking: A Case-Control Study. Trans. Vis. Sci. Tech. 2025;14(2):3. https://doi.org/10.1167/tvst.14.2.3.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose: Continuous visual stimulus tracking could be used as an easy alternative to standard automated perimetry (SAP) for visual function screening. With continuous visual stimulus tracking, we simplified the perimetric task to following a moving dot on a screen with the eyes. Here, we determined whether tracking performance (the agreement between gaze and stimulus position) enables the detection and quantification of glaucomatous visual function loss (in terms of SAP), and whether it shows a learning effect.

Methods: We evaluated the tracking performance of 36 cases with early, moderate, or severe glaucoma (median with interquartile range [IQR] age = 70 [67–74] years) and 36 controls (median = 70, IQR = 67–72 years). All participants monocularly tracked a moving stimulus (Goldmann size III) at 3 Weber contrast levels: 40, 160, and 640%, while their eye movements were recorded.

Results: Glaucoma decreased the tracking performance, with the most severe reduction in the severe glaucoma cases. A distinction between groups was possible, but depended on the contrast level: tracking performance of early glaucoma cases was significantly different from controls only at 40% contrast. Within the cases, glaucomatous visual function loss (SAP Mean Sensitivity [MS]) was best correlated with tracking performance when using 160% contrast. There was no significant learning effect.

Conclusions: Overall, the data indicate that it is possible to detect and quantify glaucomatous visual function loss with continuous visual stimulus tracking.

Translational Relevance: Continuous visual stimulus tracking is an easy, fast, and intuitive technique that has the potential for diagnostic applications in detection of new glaucoma cases and monitoring of previously diagnosed cases.

Introduction
Glaucoma is a chronic eye disease, which eventually can lead to irreversible blindness. Visual field (VF) testing is important for detecting, characterizing, and monitoring vision loss in glaucoma. The current gold standard for VF assessment is standard automated perimetry (SAP). During SAP, stimuli at different luminances are systematically presented at various test locations throughout the VF. Meanwhile, the person being assessed has to maintain fixation on a central target and press a button when a stimulus is perceived somewhere in the VF. Performing SAP requires prolonged focused attention, understanding of the test, and multi-tasking. These requirements in part complicate valid and reliable use of SAP in children, the elderly, and persons with cognitive and/or motor impairments, critically limiting the quality of ophthalmic and rehabilitation care. An additional limitation is that SAP requires some learning before a reliable and stable VF can be measured. Studies have shown that there is a significant learning effect between the first and second SAP tests, with smaller effects after the second test.13 All together, easier and more intuitive ways to assess the VF may contribute to improving quality of care. 
Previous studies have shown that the manual responses as used in SAP can be replaced by eye-movement-based responses that can be obtained using an eye tracker.411 This simplifies the task for the person being assessed considerably. Eye-movement-based responses can be used to detect glaucomatous VF defects using the assessment of saccadic reaction times when going from a fixation target to a stimulus presented somewhere in the VF.5,6,9,11,12 Recently, we took a next step by further simplifying the task to tracking a continuously moving stimulus with the eyes – thus additionally eliminating the need to return to a fixation target after each saccade.13 Performance, defined as the agreement between gaze and stimulus position during the tracking of the continuously moving stimulus, was shown to be influenced by the presence of VF defects14 and this tracking might thus serve as the core of a novel perimetric approach. Moreover, this technique is generally preferred by patients over SAP.15 As humans have a natural tendency to look at moving or appearing stimuli, we hypothesize that continuous stimulus tracking will not require significant learning as it is sufficiently intuitive. 
In our previous study, we have shown that, in healthy participants, tracking performance improves with increasing stimulus contrast and that, with aging, a higher stimulus contrast is needed to maintain performance.16 For a Gaussian-shaped stimulus with a size similar to that of a Goldmann size III stimulus, 40% was the lowest contrast usable at all ages. At this contrast, tracking was possible for both older and younger participants. At the same time, the performance improved when using higher contrast levels, indicating that reductions in contrast sensitivity should lead to a decrease in tracking performance, thus allowing for the detection of visual function loss. The next step is to further develop the method to ensure that it is both able to detect small amounts of visual function loss (as measured with SAP) and has a sufficiently large dynamic range to allow quantifying glaucomatous visual function loss over a wide range of disease severities. Moreover, it should be minimally affected by learning. 
Therefore, the aims of our current study were to determine (1) whether tracking performance enables the detection of glaucomatous visual function loss, (2) what stimulus contrast level or levels are best suited to quantify the glaucomatous visual function loss, and (3) whether tracking performance shows a learning effect. For this purpose, we evaluated tracking performance in three groups of glaucoma cases having early, moderate, or severe glaucoma, as well as in a group of age-matched controls. We asked the participants to monocularly track a stimulus that moves continuously on a screen and makes occasional jumps, requiring the participant to make both smooth pursuit and saccadic eye movements. The stimulus jumps were directed toward specific locations on the screen, in order to cover the test locations of the Humphrey Field Analyzer (HFA) 24-2 grid. Gaussian-shaped stimuli with a size similar to that of a Goldmann size III stimulus were presented at 3 contrast levels: 40%, 160%, and 640% Weber contrast. In this study, we focused on the overall agreement between the gaze and stimulus position. Finally, the presence of a learning effect was studied by having the glaucoma cases repeat the task within a week and comparing their performances. 
Methods
Study Population
In this case-control study, we included 36 people with glaucoma (cases) and 36 age-matched participants without glaucoma (controls) between 50 and 80 years of age. Written informed consent was obtained from the participants after explanation of the nature and possible consequences of the study. The study was approved by the Medical Ethical Committee of the University Medical Center Groningen (NL70382.042.20). The study followed the tenets of the Declaration of Helsinki. 
Recruitment
Glaucoma cases were selected from the ophthalmic outpatient department of the University Medical Center Groningen (UMCG), using the VF database of the Groningen Longitudinal Glaucoma Study (GLGS).17 Eight cases of other hospitals in the Netherlands were recruited via an advertisement and provided their medical history and most recent visual function assessment (visual acuity and at least 3 prior reliable VF assessments) for selection. We included cases who had a minimum of three consecutive reliable SAP assessments with the Glaucoma Hemifield Test reproducibly outside normal limits in at least one eye. The test pattern of these previous SAP assessments could be either the HFA 24-2 or 30-2 (depending on the test pattern used clinically in the concerning case. For both grids, the used thresholding strategy was SITA-Fast If the most recent SAP assessment was older than 6 months, the assessment was repeated, using the SAP 24-2 test pattern and SITA-Fast strategy (HFA; Carl Zeiss Meditec AG, Jena, Germany). Reliability criteria were defined as false positive responses ≤ 10% and fixation losses ≤ 20%. In the case of an increased number of fixation losses, fixation was considered acceptable if a well-defined blind spot was present and the technician had observed and reported a stable fixation.17,18 One eye per case was selected based on the SAP mean deviation (MD) in decibels (dB) as a measure of severity and on the location of the VF defect — ensuring the presence of defects in both the superior and inferior hemifield in the study population — resulting in 3 subgroups of 12 cases each with early (MD > −6 dB), moderate (–12 dB < MD ≤ −6 dB), and severe (MD < −12 dB) glaucoma with diverse locations of VF defects within each group. Cases were allowed to have different types of glaucoma and concurrent other eye diseases, like myopia, as long as glaucoma was the presumed sole cause of VF loss and loss of visual acuity. Exclusion criteria were a best corrected visual acuity (BCVA) of worse than 0.3 logMAR and neurological disorders that could affect test performance, as assessed with a questionnaire (asking if the person is currently or was previously under the care of a neurologist, and, if yes, for what reason). 
Controls were recruited via advertisement. Potential controls who responded to the advertisement were asked to fill out a questionnaire to screen for any known eye abnormalities, a positive family history of glaucoma, and neurological disorders that could affect the test performance. Hereafter, a short eye-health check was performed, which included an intraocular pressure (IOP) measurement (Ocular Response Analyzer G3 non-contact tonometer; Reichert Technologies, Inc., Depew, NY, USA), a refraction and BCVA measurement (Nidek ARK-1s. Nidek Co., Ltd., Gamagori, Japan), a frequency doubling technology visual field test (FDT; C20-1 screening mode; Carl Zeiss, Jena, Germany), and a measurement of the peripapillary retinal nerve fiber layer thickness as assessed by spectral domain optical coherence tomography (OCT; Copernicus, Optopol Technologies, Zawierci, Poland). Exclusion criteria were any known eye abnormality, a positive family history of glaucoma, any known neurological disorders that could affect test performance, a BCVA worse than 0.1 logMAR, a cornea-compensated IOP (IOPcc) above 22 millimeters of mercury (mm Hg), any reproducibly abnormal test location on the FDT test result, or any temporally located red clock hour abnormality in the peripapillary retinal nerve fiber layer on the OCT. One eye that met the eligibility criteria was included. During the recruitment process, we selected the cases and controls, per glaucoma severity subgroup (early, moderate, and severe), to make the groups age and sex similar, and we equalized the numbers of included right and left eyes. 
Cognitive Screening
As tracking performance is influenced by cognition, and more specifically, the ability to sustain attention to the task, we performed two cognitive screening tests. The first test was the Dutch translated Montreal Cognitive Assessment (MoCA) test.19 This test screens for any mild cognitive impairment and provides a rapid assessment of different cognitive domains. The maximum score on the MoCA is 30 points. Scores of 26 or higher are considered in the normal range. The second test was the Test of Sustained Selective Attention (TOSSA, Kovács, Pyramid productions). This is an auditory vigilance task in which the participant has to react to a target (group of 3 beeps) amidst distractor stimuli (groups of 2 or 4 beeps) during a time period of about 8 to 10 minutes. Three scores were extracted, being (1) the Concentration Strength (CS), (2) the influence of the Length of Administration on the Detection Strength (LADS), which is the true positive hit rate, and (3) the influence of the Length of Administration on the Response Inhibition Strength (LARIS), which is based on the false positive hit rate as a function of test duration. Scores above the fifth percentile were considered normal. This test was selected for the similar duration of SAP and the need to multitask (press a button) and to suppress reacting on a distractor (cf. making an eye movement to the projected stimulus instead of fixating the central target in SAP). Test scores were used for descriptive statistics and as a covariate in the analyses; participants were not excluded based on the results of these two tests. 
Experimental Set-Up Tracking Task
Participants were seated in front of a computer monitor at a viewing distance of 60 cm with their head placed on a chinrest and stabilized with a forehead rest to minimize head movements. All tests were performed monocularly with the selected eye with optimal correction (spherical equivalent) for the viewing distance, and the fellow eye occluded. The experiment was performed in a dimly lit room with the monitor being the only source of illumination. 
Stimuli were presented on a 24.5-inch IPS monitor (OptixMag25R1X, Micro-Star International Co., Ltd., New Taipei City, Taiwan) with a framerate of 240 hertz (Hz). We chose to use an IPS monitor as the screen luminance – and thus stimulus contrast – is more homogeneous across the screen and less dependent on the viewing angle. The monitor had a resolution of 1920 × 1080 pixels (49 × 29 degrees of visual angle at the viewing distance of 60 cm). The luminance of the monitor was calibrated as described in the study by Vrijling and de Boer et al.16 Stimulus display and gaze recording were controlled by the Psychtoolbox and Eyelink Toolbox extensions2023 for MATLAB (The Mathworks, Inc., version 2021a). The PC for the stimulus presentation was an HP Elitedesk 800 G3 with a NVIDIA GeForce GTX 1080 graphics card, running Windows 10. A desktop-mounted Eyelink Portable Duo eye-tracker (SR Research Ltd., Ottawa, Ontario, Canada), running software version 6.10.01, was used to measure the participants’ eye movements. The Eyelink eye-tracker recorded monocular gaze data at a sampling frequency of 1000 Hz. At the start of the tracking task and after breaks or movement of the participant, the eye-tracker was (re)calibrated using the built-in nine-point calibration routine. The calibration accuracy was verified with the built-in validation routine and was accepted when the accuracy was defined “good” by the tracker software (i.e. average error less than 0.5 degrees and maximum error < 1.0 degrees). If for at most, one of the nine points, the error was > 1.0 degrees (usually one of the upper corners) and this could not be solved on repeated calibrations, the measurement was continued. The participant was excluded if a good calibration accuracy could not be achieved. A drift correction was performed prior to presenting each trajectory set (see the Methods section, subsection Data collection). There were no significant differences in drift between cases and controls, or between the different stimulus contrasts (all P > 0.46). 
Stimulus
The stimulus was a Gaussian blob with 0.22-degree standard deviation. The diameter of this stimulus corresponds to the diameter of a Goldmann size III stimulus (0.43 degrees), a commonly used stimulus size in VF testing. The stimulus was presented on a static gray background of 30 cd/m2 (somewhat higher than the commonly used 10 cd/m2 in order to make the contrast sensitivity less sensitive to small changes in luminance due to, for example, inhomogeneities of the monitor or differences/changes in pupil diameter.16 Three different contrasts were used, 40%, 160%, and 640% Weber contrast, where the contrast was calculated with the luminance of the peak of the Gaussian blob. Based on our previous study,16 we selected a stimulus contrast of 40% as the base stimulus contrast. For this contrast, the tracking performance for saccadic pursuit mode (see below) is above detection threshold and has not yet saturated, for both younger and older control participants. The contrasts of 160% and 640% represent a 6 and 12 dB increase in contrast relative to the base stimulus contrast, respectively. 
Stimulus Trajectories
The stimulus moved along a trajectory in saccadic pursuit mode. In this mode, the stimulus moves continuously, whereas the direction and velocity of the motion vary in a pseudo-random manner. Additionally, the stimulus makes a jump to a new VF location at pseudo-random intervals. As the timing, amplitude, and direction of the stimulus jumps are pseudo-randomized, participants cannot predict when and to where the stimulus will jump. Spatially, the jumps were toward one of 56 test locations on a grid. The grid consisted of the 54 locations of the HFA 24-2 grid (a grid with a 6 × 6-degree spatial resolution), modified to be symmetrical by inserting two additional test locations on the temporal side. The order in which these locations were sampled was pseudo-randomized, in such a way that in a set of three 40-second saccadic pursuit trajectories all 56 test locations were covered once. A trajectory always started from the center of the screen. See Vrijling and de Boer et al.16 for a detailed description of the generation of the saccadic pursuit trajectories. 
A total of 15 unique trajectory sets (of three 40-second trajectories each) were generated to limit learning and predictive viewing behavior. Additionally, a separate practice trajectory set was generated. 
Data Collection
Both cases and controls were invited for 2 visits within 1 week, not on the same day. At the first visit, the tracking task and the MoCA were performed, and for the controls also the short eye-health check was performed. If the SAP assessment of a glaucoma case was older than 6 months, the assessment of the SAP test was repeated at the first visit. At the second visit, the tracking task was repeated for the cases, and the TOSSA was performed after the tracking task. Controls performed the TOSSA, but no repeat of the tracking task (their second visit was mainly devoted to another study). For that reason, the existence of a learning effect was studied in the glaucoma cases only. 
For all participants, the tracking task consisted of a single block of four trajectory sets (with each set consisting of three 40-second saccadic pursuit trajectories): a practice trajectory set at 160% contrast followed by testing trajectory sets at 40%, 160%, and 640% contrast. The order of contrast levels was pseudo-randomized such that the number of times each contrast level order was presented, it was balanced both within and between groups. The instruction given to the participant was to follow the stimulus with their gaze to the best of their abilities, without making any head movements. Participants were encouraged to take breaks between the trajectories if needed to maintain concentration and avoid fatigue. The task was self-paced; the participant could start each presentation of a trajectory via a mouse click. Overall, the tracking task consisted of 12 trajectories, together lasting about 15 minutes, including breaks and eye-tracker calibration. 
Data Analyses
Pre-Processing of Eye Movement Data
Eye gaze positions were recorded as horizontal and vertical screen coordinates and these were analyzed separately. Before calculation of the tracking performance (see below), the raw gaze data were filtered for unrealistic peaks in velocity (> 1000 deg/s, not likely to be caused by actual eye movements)24,25 and plateaus in position (zero velocity for at least 2 consecutive samples). For the current analyses, unrealistic peaks and plateaus in gaze data were ignored and the corresponding parts of the stimulus trajectories were also set to Not a Number (NaN) in order to ensure a correct calculation of the tracking performance (see below). 
Definition of Tracking Performance
Tracking performance was defined as the cosine similarity (normalized inner product) between the stimulus and gaze vector (Equation 1):  
\begin{eqnarray} Cosine\ Similarity\ \left( {A,B} \right) = \frac{{A \cdot B}}{{\left| {\left| A \right|} \right| \left| {\left| B \right|} \right|}}\quad \end{eqnarray}
(1)
where A is the vector of stimulus positions and B the vector of gaze positions, ||A|| the Euclidean norm of vector A, and ||B|| the Euclidean norm of vector B. Theoretically, the cosine similarity is bound between −1 and 1. However, in practice, it ranges between values around zero — denoting poor tracking performance — and values close to 1 where 1 indicates perfect tracking performance. Before calculating the cosine similarity, vectors A and B were both mean centered. Obviously, there is a (physiological) delay between stimulus and response. Therefore, we calculated all “time-shifted” cosine similarities with a time shift in the range of −5 and +5 seconds in steps of 0.0041 seconds (1 frame). Next, we identified the first positive peak in the cosine similarity curve (Fig. 1) after 0 seconds using the Matlab function findpeaks. Tracking performance was defined as the cosine similarity corresponding to this peak. If no clear peak could be identified, the maximum cosine similarity between 0 and +3 seconds was used. For each set of saccadic pursuit trajectories, the three cosine similarity peaks were averaged to obtain the mean tracking performance for the concerning set. 
Figure 1.
 
Example output of the lagged cosine similarity pipeline for a single trial at 160% contrast for a control (left) and glaucoma case (right). The x-axis depicts the delay in seconds, the y-axis depicts the cosine similarity. The dashed line (reference) shows the lagged cosine similarity for the stimulus with itself, and therefore has a value of 1 at delay 0. The control's cosine similarity curve approximates the curve of the reference, but with a small delay (± 300 ms) and near perfect tracking performance (peak of curve at ± 0.95). The curve of the glaucoma case is wider, with a longer delay (± 450 ms) and lower tracking performance (± 0.72).
Figure 1.
 
Example output of the lagged cosine similarity pipeline for a single trial at 160% contrast for a control (left) and glaucoma case (right). The x-axis depicts the delay in seconds, the y-axis depicts the cosine similarity. The dashed line (reference) shows the lagged cosine similarity for the stimulus with itself, and therefore has a value of 1 at delay 0. The control's cosine similarity curve approximates the curve of the reference, but with a small delay (± 300 ms) and near perfect tracking performance (peak of curve at ± 0.95). The curve of the glaucoma case is wider, with a longer delay (± 450 ms) and lower tracking performance (± 0.72).
Mean Sensitivity
To evaluate the agreement between the severity of visual function loss and the tracking performance, we used the Mean Sensitivity (MS) of the HFA 24-2 grid as an absolute rather than age-corrected measure of glaucomatous visual function loss. Mean sensitivity was defined as the sum of the local contrast threshold values (in dB) for each test location divided by 52 (the number of test locations outside the blind spot).26,27 
Statistical Analyses
All statistical analyses were conducted in R (version 4.2.2) using R Studio (version 1.3.1093). The study population was described using nonparametric descriptive statistics (median with interquartile range [IQR]). Univariable comparisons of the continuous variables (age, spherical equivalent, and visual acuity) between cases and controls were made with a Mann-Whitney U test. Spearman correlation tests were performed to test whether age affected the MoCA, CS, LADS, or LARIS scores and to test whether the tracking performances at the different stimulus contrast levels were correlated, with the cor.test function (stats package, version 4.2.2).28,29 Linear regression models were performed with the lm function (stats package).29 Linear mixed-effects regression models were performed with the lmer function (lmerTest package, version 3.1-3.30 To find the best model structure for the regression analyses, model comparison using the anova function (stats package, version 4.2.2)31 was done. For evaluating the models, we calculated the Akaike information criterion (AIC) as well as the Bayesian information criterion (BIC). We considered the best model to be the model with the lowest BIC (the more conservative criterion; in case of discrepancy between AIC and BIC, this heads toward a simpler model with fewer variables). For the best model, significance of main effects and interactions across factor levels was tested with the Anova function (car package, version 3.0-10),32 with the used test set to type III Wald chi-square. Anova post hoc Tukey Tests were performed with the emmeans function (eponymous package package, version 1.8.3).33 The assumptions of normality and homoscedasticity of the residuals were tested with the Shapiro-Wilk normality test. 
Detecting Visual Function Loss Through Tracking Performance
To test whether visual function loss could be detected through tracking performance, we performed a linear mixed-effects regression with the tracking performance as the dependent variable, the variables stage, stimulus contrast, BCVA, refractive error in spherical equivalent (SE), scores from cognition tests (MoCA, CS, LADS, and LARIS), age, and sex as fixed effects, and subject as a random effect. Contrast (40%, 160%, and 640%), stage (control, early, moderate, and severe), and sex were entered as factors, BCVA, SE, age, and scores from cognition tests as numerical values that were centered around the mean. We averaged the horizontal and vertical tracking performance prior to the regression because the effects of contrast, age, and stage on the horizontal and vertical tracking performance were very similar (see Supplementary Fig. A.1). We included the gaze data of the first visit of the glaucoma cases in this analysis (for the controls, we collected the data only once during their first visit; see the Methods section, subsection Data collection). The best model (see Supplementary Table A.1) violated the assumptions of normality and homoscedasticity of the residuals. Therefore, we used bootstrapping with 1000 replications to estimate the confidence intervals (using the function confint from the stats package version 4.2.2).29 Estimates for which the 95th percentile confidence intervals did not include zero were considered significant. Variables that did not affect tracking performance were omitted from the regression analysis that was used to determine the agreement between MS and tracking performance in glaucoma (see the Methods section, subsection Predicting the Mean Sensitivity From Tracking Performance in Glaucoma). 
Predicting the Mean Sensitivity From Tracking Performance in Glaucoma
To determine the agreement between the MS and the tracking performance at different stimulus contrasts and age in the glaucoma cases, we performed a linear regression with the MS as the dependent variable, and built separate models that always included only one stimulus contrast per model. We included only one stimulus contrast per model because the tracking performances at the different stimulus contrasts were all strongly correlated (rho = 0.91 for 40% and 160% contrast; rho = 0.80 for 40 and 640% contrast; rho = 0.90 for 160% and 640% contrast; all P < 0.001). We tested if age contributed to the model by adding age as a numerical value that was centered around the mean. In accordance with the Methods section, subsection Detecting Visual Function Loss Through Tracking Performance, we averaged the horizontal and vertical tracking performance prior to the regression analyses, used the gaze data of the first visit and omitted variables that did not affect tracking performance (BCVA, SE, sex, and scores from the cognition tests [MoCA, CS, LADS, and LARIS]). The best model (see Supplementary Table A.2) met the assumptions of normality (w = 0.98, P = 0.66) and homoscedasticity of the residuals. 
Presence of Learning Effect on Tracking Performance
To test whether there was a significant learning effect on tracking performance, we performed a secondary analysis with visit (first and second) added as a main effect to the best model, as found in the Methods section, subsection Detecting Visual Function Loss Through Tracking Performance. This analysis was performed in the glaucoma cases only (as we had only repeat data for this group; see the Methods section, subsection Data Collection). 
Results
Table 1 shows the general characteristics of the study population. Cases and controls did not differ significantly regarding age and gender. Cases had a slightly reduced visual acuity (higher logMAR value) compared with the controls, and were, on average, more myopic. The median VF MD (HFA 24-2) of the included eye of the glaucoma cases was approximately −9 dB. 
Table 1.
 
Characteristics of the Study Population
Table 1.
 
Characteristics of the Study Population
Table 2 shows the characteristics of the early, moderate, and severe glaucoma cases. They did not differ regarding age, sex, included eye, and BCVA. They did not differ either regarding the cognition scores (MoCA, Tossa CS, Tossa LADS, and Tossa LARIS). 
Table 2.
 
Characteristics of the Glaucoma Cases
Table 2.
 
Characteristics of the Glaucoma Cases
Detecting Visual Function Loss Through Tracking Performance
Figure 2 shows the tracking performance as a function of stimulus contrast for the cases with early, moderate, and severe glaucoma and the controls. As can be seen in this figure, glaucomatous visual function loss decreased the tracking performance. Moreover, for both cases and controls, the tracking performance improved with increasing stimulus contrast. 
Figure 2.
 
Tracking performance as a function of stimulus contrast for the cases with early, moderate, and severe glaucoma and the controls.
Figure 2.
 
Tracking performance as a function of stimulus contrast for the cases with early, moderate, and severe glaucoma and the controls.
To evaluate these results, we built a series of mixed-effects regression models and compared their performance in terms of the associated BIC value. For an overview of these models, see Supplementary Table A.1 in the Supplementary Material. The mixed-effects regression model that best (lowest BIC) described the data in Figure 2 included the factors age, contrast, and stage as main effects, participant as random intercept, and included an interaction between stage and contrast
The best model showed significant main effects of contrast (P < 0.001), age (P = 0.02), and stage (P < 0.001), and a significant interaction between contrast and stage (P < 0.001). Model estimates (Table 3) indicated that the tracking performance increased for higher stimulus contrast levels and reduced with age and glaucoma severity. The effect of contrast differed per stage, with the strongest influence of contrast observed in severe glaucoma. 
Table 3.
 
Model Estimates With 95% Confidence Intervals for the Best Model With the Intercept Representing the Estimate for the Controls at a Stimulus Contrast of 40%
Table 3.
 
Model Estimates With 95% Confidence Intervals for the Best Model With the Intercept Representing the Estimate for the Controls at a Stimulus Contrast of 40%
Bootstrapping (see Table 3) showed that all estimates of the best model for the tracking performance were significant. Tables 4 and 5 show the post hoc Tukey tests corresponding to the model as presented in Table 3. As can be deduced from these tables, the highest contrast level (640%) only adds information in severe glaucoma (see Table 4), and all stages can be well discriminated at especially 40% contrast (in particular control versus early glaucoma), except for early versus moderate glaucoma, which cannot be discriminated at any of the tested contrasts (see Table 5). 
Table 4.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stage
Table 4.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stage
Table 5.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stimulus Contrast
Table 5.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stimulus Contrast
Predicting the Mean Sensitivity From Tracking Performance in Glaucoma
Figure 3 shows, for the glaucoma cases, the relationship between tracking performance at the three different stimulus contrasts and the SAP MS. The correlations were 0.74 at 40% contrast, 0.77 at 160% contrast, and 0.71 at 640% contrast (all P < 0.001). 
Figure 3.
 
Scatterplot showing the tracking performance as a function of the SAP Mean Sensitivity (MS) in decibels (dB). The lines show the regression lines (solid) for 40% contrast (red), 160% contrast (green), and 640% contrast (blue) with corresponding 95% confidence intervals.
Figure 3.
 
Scatterplot showing the tracking performance as a function of the SAP Mean Sensitivity (MS) in decibels (dB). The lines show the regression lines (solid) for 40% contrast (red), 160% contrast (green), and 640% contrast (blue) with corresponding 95% confidence intervals.
The best regression model to estimate the MS from the tracking performance (see Supplementary Table A.2) included tracking performance at 160% as the main effect. The model showed a MS of 19.49 dB (SE = 0.64, P < 0.001) at the mean tracking performance of 0.78 and for every 0.1 point increase in tracking performance (from the mean), MS increases with 3.24 dB (SE = 0.39, P < 0.001). The variance explained was 0.67, indicating that the tracking performance at 160% is a good predictor of the MS as measured by SAP. 
Presence of Learning Effect on Tracking Performance
We used a regression model including the factors age, contrast, stage, and visit as main effects, participant as random intercept, and an interaction between stage and contrast, to study the effect of learning. The mean (SD) tracking performance at the first visit was 0.71 (0.23) and at the second visit 0.73 (0.24). The model showed no significant effect of visit (estimate [SD] = 0.02 [0.01], P = 0.053), indicating that the tracking performance did not improve significantly over 1 week. 
Discussion
The main finding of this study is that continuous visual stimulus tracking can be used to detect and quantify glaucomatous visual function loss. Specifically, we conclude that (1) glaucomatous visual function loss can be best detected at a stimulus contrast of 40%, (2) glaucomatous visual function loss (in terms of SAP MS) is best correlated with tracking performance when using a stimulus contrast of 160%, and (3) tracking performance shows no learning effect for the 2 repeat tests presented in this work. Below, we discuss our results and their implications in more detail. 
Detecting Visual Function Loss Through Tracking Performance
The reduction in tracking performance in glaucoma cases, as compared with controls, is in agreement with studies examining eye-movement based responses to peripheral targets. These studies demonstrated a reduced hit rate to peripheral targets4 or slower saccadic eye-movement responses5,6,9,12,34 in glaucoma cases when compared with controls. Two of the studies investigated saccadic reaction times (SRTs) across varying degrees of glaucoma severity.5,6 The study by Mazumdar et al.6 observed a trend toward increasing SRTs with increasing disease severity using a Goldmann size III stimulus at 16% contrast (estimated from the reported background luminance and stimulus brightness) in a group of 25 glaucoma cases and 54 controls. Similarly, Kadavath Meethal et al.5 found that more severe glaucomatous visual function loss resulted in higher SRTs using a Goldmann size III stimulus at 41% contrast in a group of 50 glaucoma cases and 54 controls. These results are in agreement with our findings, where we were able to separate early glaucoma cases from controls based on tracking performance with a Gaussian-shaped stimulus with a size similar to that of a Goldmann size III stimulus with a contrast of 40%. Unfortunately, neither Kadavath Meethal et al.5 nor Mazumdar et al.6 reported an analysis of SRTs at other contrast levels, nor did they quantify the severity of the glaucomatous visual function loss in their cases. 
Although, we found no significant difference in tracking performance between early and moderate glaucoma cases, we could distinguish severe glaucoma cases from all other subgroups. We found a linear relationship between tracking performance and MS. However, the dynamic range of the HFA exceeds 30 dB, and the stages of the simplified Hodapp classification35 are not evenly distributed across the HFA's dynamic range. Our groups, based on this classification, were relatively close to each other in terms of MD (median HFA MD of −4.4 dB for the early glaucoma cases and −8.8 dB for moderate glaucoma). This might explain our inability to distinguish the early and moderate glaucoma cases based on their tracking performance. Using a stimulus contrast of 160%, we could not distinguish between the tracking performance of controls from that of the early glaucoma cases. This may be due to the stimulus being well above threshold for both groups and, consequently, their performance reaching ceiling levels.16 
Predicting the Mean Sensitivity From Tracking Performance in Glaucoma
We found that, for the glaucoma cases, visual function loss (in terms of SAP MS) could be estimated best from the tracking performance for a stimulus contrast of 160%. This is in agreement with other studies that compared suprathreshold perimetry to SAP.4,3638 Stoutenbeek37 used a 52-point suprathreshold screening test of the HFA with stimuli 6 dB brighter than the predicted sensitivity at a particular test point.39 He found a clear, linear association between suprathreshold perimetry (assessed by the number of missed test points) and SAP MD in a group of 131 glaucoma cases that had glaucomatous VF loss on SAP (HFA 24-2 SITA standard). In agreement with our study, a 6 dB suprathreshold stimulus could not distinguish early glaucoma cases from controls.36,37 Jones et al.,4 aiming to develop an effective glaucoma case-finding approach, investigated eye-movement based responses to a peripherally presented static suprathreshold Goldmann size III stimulus with a contrast level of 868%. They assessed the hit-rate (seen targets) in a cohort of 12 cases (comprising 24 eyes) with varying degrees of glaucoma severity, as well as 6 controls (12 eyes). In agreement with our study (see Fig. 3) as well as Stoutenbeek,37 Jones et al.4 observed a linear correlation between the mean hit-rate and the MD of SAP (HFA 24-2). 
Presence of Learning Effect on Tracking Performance
In general, some learning occurs with psychophysical testing. Learning in SAP results in higher sensitivity values and a reduction in intra-test variability.13 A learning effect is undesirable, as it requires the patient to undergo multiple tests before their VF test result can be considered reliable. We found no evidence for a learning effect for continuous stimulus tracking for the two repeat tests presented in this work. If generalizable to other test intervals, this will save time when using such tests in a clinical setting. It is possible that the effect would become significant after more repeat tests. However, the effect size of 0.02, based on the model estimate, was very small, indicating that any further improvement in tracking performance would also be very small, since the largest effect would be expected between the first two tests. 
Clinical Implications
A functional test for glaucoma should be able to perform two roles, the detection of new glaucoma cases and the monitoring of previously diagnosed cases. Our results suggest that the technique of continuous stimulus tracking can potentially fulfill both roles. For the detection of new cases, a contrast of 40% (in combination with stimulus size III) seems best, as it was the only contrast (of the three contrasts we evaluated) at which we could discriminate between healthy and early glaucoma. Following the proof of principle presented here, the logical next step is a screening study with a large number of glaucoma cases and controls in order to determine its screening performance in terms of sensitivity and specificity, using a receiver operating characteristic (ROC) analysis. Furthermore, the user friendliness should be determined formally and compared with that of other techniques, including SAP. For the monitoring of previously diagnosed glaucoma cases, a contrast of 160% seems best. For this role, our finding of a clear relationship between tracking performance and SAP MS is encouraging. To determine the actual value of continuous stimulus tracking for monitoring, a longitudinal study is pivotal. 
Strengths and Limitations
This study utilized continuous visual stimulus tracking which offers an intuitive method that utilizes eye tracking instead of manual button presses, making it more user-friendly, inclusive, and potentially faster than traditional, trial-based approaches. 
The current tracking task consisted of 12 trajectories per eye (4 contrasts and 3 trajectories per contrast), together lasting about 15 minutes, including 3 practice trajectories, breaks, and eye-tracker calibration. Obviously, this is too long for an efficient visual function test that can be used in a busy clinical environment. However, most likely, test time can be reduced to one quarter of this, because we have shown that for the purpose of either screening or staging, a single contrast level stimulus is sufficient (40% for screening and 160% for staging). Further gains in testing efficiency may come from switching to wearable eye-trackers, such as the Neon (Pupil Labs, GmbH, Berlin) with artificial intelligence (AI)-based automatic calibration. This would eliminate the need for an extensive calibration procedure, potentially further reducing the test time per eye to approximately 2 minutes (3 trials of 40 seconds). 
The current version of the test is able to detect and stage visual function loss in glaucoma by analyzing tracking performance. For future clinical use, it is essential to derive a VF map. In principle, reconstructing a VF from continuous stimulus tracking data is possible.40 Our method already anticipates this need by using a stimulus grid with the same locations as the HFA 24-2 grid. However, to move from “proof-of-principle” to a robust visual field map further research is needed. We will address this in a follow-up study. 
Conclusions
Glaucomatous VF defects reduce continuous visual stimulus tracking performance and this reduction becomes larger with more severe glaucomatous visual function loss. This finding paves the way for diagnostic applications of continuous visual stimulus tracking in glaucoma. The logical next steps are a large screening study, the reconstruction of a VF map, and a longitudinal data collection. 
Acknowledgments
The authors thank Kim Westra, Suzanne Veneboer-Metzlar, and Sandra Boekhorst for helping us with the HFA and continuous visual stimulus tracking measurements. The authors also thank Eline Will for her help with the selection of the cognitive screening test (TOSSA). 
Supported by the Visio Foundation, The Netherlands, and ZonMw, programme Expertisefunctie Zintuiglijk Gehandicapten, grant number 637005001, and by Uitzicht grant number UZ2019-20 (via funds provided by the ANVVB, Oogfonds, Stichting blindenpenning, LSBS). The funding organizations had no role in designing, conducting, analyzing, or publishing this research. 
Open Practices Statement: The data are available at the DataverseNL repository via https://doi.org/10.34894/QMIJDH, while materials (experimental code) are available at request from the corresponding author. The experiments were not pre-registered. 
Disclosure: A.C.L. Vrijling, None; M.J. de Boer, None; R.J. Renken, is listed as inventor on the patent application WO2021096361A1 (Grillini, A., Hernández-García, A., Renken, R.J. 2019; Method, system, and computer program product for mapping a visual field) on which the method used in this manuscript is partially based; J.-B.C. Marsman, None; J. Heutink, None; F.W. Cornelissen, None; N.M. Jansonius, None 
References
Heijl A. The effect of perimetric experience in normal subjects. Arch Ophthalmol. 1989; 107(1): 81. [CrossRef]
Heijl A. The effect of perimetric experience in patients with glaucoma. Arch Ophthalmol. 1996; 114(1): 19. [CrossRef]
Wild JM, Dengler-Harles M, Searle AET, O'Neill EC, Crews SJ. The influence of the learning effect on automated perimetry in patients with suspected glaucoma. Acta Ophthalmologica. 1989; 67(5): 537–545. [CrossRef]
Jones PR, Smith ND, Bi W, Crabb DP. Portable perimetry using eye-tracking on a tablet computer—a feasibility assessment. Trans Vis Sci Tech. 2019; 8(1): 17. [CrossRef]
Kadavath Meethal NS, Mazumdar D, Asokan R, et al. Development of a test grid using eye movement perimetry for screening glaucomatous visual field defects. Graefes Arch Clin Exp Ophthalmol. 2018; 256(2): 371–379. [CrossRef]
Mazumdar D, Pel J, Panday M, et al. Comparison of saccadic reaction time between normal and glaucoma using an eye movement perimeter. Indian J Ophthalmol. 2014; 62(1): 55.
Mazumdar D, Meethal NSK, Panday M, et al. Effect of age, sex, stimulus intensity, and eccentricity on saccadic reaction time in eye movement perimetry. Trans Vis Sci Tech. 2019; 8(4): 13. [CrossRef]
Mazumdar D, Pel JJM, Kadavath Meethal NS, et al. Visual field plots: a comparison study between standard automated perimetry and eye movement perimetry. J Glaucoma. 2020; 29(5): 351–361. [CrossRef]
McTrusty AD, Cameron LA, Perperidis A, et al. Comparison of threshold saccadic vector optokinetic perimetry (SVOP) and standard automated perimetry (SAP) in glaucoma. Part II: patterns of visual field loss and acceptability. Trans Vis Sci Tech. 2017; 6(5): 4. [CrossRef]
Murray IC, Fleck BW, Brash HM, MacRae ME, Tan LL, Minns RA. Feasibility of saccadic vector optokinetic perimetry: a method of automated static perimetry for children using eye tracking. Ophthalmology. 2009; 116(10): 2017–2026. [CrossRef]
Pel JJM, van Beijsterveld MCM, Thepass G, van der Steen J. Validity and repeatability of saccadic response times across the visual field in eye movement perimetry. Trans Vis Sci Tech. 2013; 2(7): 3. [CrossRef]
Thepass G, Lemij HG, Vermeer KA, van der Steen J, Pel JJM. Slowed saccadic reaction times in seemingly normal parts of glaucomatous visual fields. Front Med. 2021; 8: 679297. [CrossRef]
Grillini A, Ombelet D, Soans RS, Cornelissen FW. Towards using the spatio-temporal properties of eye movements to classify visual field defects. In: Proceedings of the 2018 ACM Symposium on Eye Tracking Research & Applications. ACM; 2018: 1–5.
Soans RS, Grillini A, Saxena R, Renken RJ, Gandhi TK, Cornelissen FW. Eye-movement–based assessment of the perceptual consequences of glaucomatous and neuro-ophthalmological visual field defects. Trans Vis Sci Tech. 2021; 10(2): 1. [CrossRef]
Soans RS, Renken RJ, John J, et al. Patients prefer a virtual reality approach over a similarly performing screen-based approach for continuous oculomotor-based screening of glaucomatous and neuro-ophthalmological visual field defects. Front Neurosci. 2021; 15: 745355. [CrossRef]
Vrijling ACL, de Boer MJ, Renken RJ, et al. Stimulus contrast, pursuit mode, and age strongly influence tracking performance on a continuous visual tracking task. Vision Res. 2023; 205: 108188. [CrossRef]
Heeg GP, Blanksma LJ, Hardus PLLJ, Jansonius NM. The Groningen Longitudinal Glaucoma Study. I. Baseline sensitivity and specificity of the frequency doubling perimeter and the GDx nerve fibre analyser. Acta Ophthalmol Scand. 2005; 83(1): 46–52. [CrossRef]
Garway-Heath DF, Lascaratos G, Bunce C, et al. The United Kingdom Glaucoma Treatment Study: a multicenter, randomized, placebo-controlled clinical trial: design and methodology. Ophthalmology. 2013; 120(1): 68–76. [CrossRef]
Nasreddine ZS, Phillips NA, Bédirian V, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment: MOCA: a brief screening tool for MCI. J Am Geriatr Soc. 2005; 53(4): 695–699. [CrossRef]
Brainard DH. The Psychophysics Toolbox. Spat Vis. 1997; 10(4): 433–436. [CrossRef]
Cornelissen FW, Peters EM, Palmer J. The Eyelink Toolbox: eye tracking with MATLAB and the Psychophysics Toolbox. Behav Res Methods, Instrum Comput. 2002; 34(4): 613–617. [CrossRef]
Kleiner M, Brainard D, Pelli D. What's new in Psychtoolbox-3? Published online 2007. Available at: http://psychtoolbox.org/.
Pelli DG. The VideoToolbox software for visual psychophysics: transforming numbers into movies. Spat Vis. 1997; 10(4): 437–442. [CrossRef]
Boghen D, Troost BT, Daroff RB, Dell'Osso LF, Birkett JE. Velocity characteristics of normal human saccades. Invest Ophthalmol. 1974; 13(8): 619–623.
Van Opstal AJ, Van Gisbergen JA. Skewness of saccadic velocity profiles: a unifying parameter for normal and slow saccades. Vision Res. 1987; 27(5): 731–745. [CrossRef]
Anderson DR, Patella VM. Automated Static Perimetry. 2nd ed. St. Louis, MO: Mosby; 1999.
Schiefer U, Pätzold J, Wabbels B, Dannheim F. [Conventional techniques of visual field examination: part 4 Static perimetry: interpretation – perimetric indices – follow-up – perimetry in childhood] [in German]. Ophthalmologe. 2006; 103(3): 235–256. [CrossRef]
Best DJ, Roberts DE. Algorithm AS 89: the upper tail probabilities of Spearman's Rho. J Royal Statist Soc Series C (Applied Statistics). 1975; 24(3): 377–379.
R: a language and environment for statistical computing. Published online 2022. Available at: http://www.r-project.org.
Bates D, Mächler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Soft. 2015; 67(1): 1–48. [CrossRef]
Chambers JM , ed. Statistical Models in S. Reprint. London, England, UK: Chapman & Hall; 1997.
Fox J, Weisberg S. An R Companion to Applied Regression, Third Edition. Available at: https://www.john-fox.ca/Companion/index.html.
Lenth R . emmeans: estimated marginal means, aka least-squares means. Published online 2022. Available at: https://cran.r-project.org/web/packages/emmeans/index.html.
Murray IC, Perperidis A, Cameron LA, et al. Comparison of saccadic vector optokinetic perimetry and standard automated perimetry in glaucoma. Part I: threshold values and repeatability. Trans Vis Sci Tech. 2017; 6(5): 3. [CrossRef]
Hodapp E, Parrish RK, Anderson DR. Clinical Decisions in Glaucoma. New York, NY: Mosby; 1993.
Mills RP, Barnebey HS, Migliazzo CV, Li Y. Does saving time using FASTPAC or suprathreshold testing reduce quality of visual fields? Ophthalmology. 1994; 101(9): 1596–1603. [CrossRef]
Stoutenbeek R. Population Based Glaucoma Screening. Groningen, Germany: University Library Groningen; 2010.
Topouzis F, Coleman AL, Yu F, et al. Sensitivity and specificity of the 76-suprathreshold visual field test to detect eyes with visual field defect by Humphrey threshold testing in a population-based setting: the Thessaloniki eye study. Am J Ophthalmol. 2004; 137(3): 420–425. [CrossRef]
Dielemans I, Vingerling JR, Wolfs RCW, Hofman A, Grobbee DE, de Jong PTVM. The prevalence of primary open-angle glaucoma in a population-based study in The Netherlands. Ophthalmology. 1994; 101(11): 1851–1855. [CrossRef]
Grillini A, Hernández-García A, Renken RJ, Demaria G, Cornelissen FW. Computational methods for continuous eye-tracking perimetry based on spatio-temporal integration and a deep recurrent neural network. Front Neurosci. 2021; 15: 650540. [CrossRef]
Figure 1.
 
Example output of the lagged cosine similarity pipeline for a single trial at 160% contrast for a control (left) and glaucoma case (right). The x-axis depicts the delay in seconds, the y-axis depicts the cosine similarity. The dashed line (reference) shows the lagged cosine similarity for the stimulus with itself, and therefore has a value of 1 at delay 0. The control's cosine similarity curve approximates the curve of the reference, but with a small delay (± 300 ms) and near perfect tracking performance (peak of curve at ± 0.95). The curve of the glaucoma case is wider, with a longer delay (± 450 ms) and lower tracking performance (± 0.72).
Figure 1.
 
Example output of the lagged cosine similarity pipeline for a single trial at 160% contrast for a control (left) and glaucoma case (right). The x-axis depicts the delay in seconds, the y-axis depicts the cosine similarity. The dashed line (reference) shows the lagged cosine similarity for the stimulus with itself, and therefore has a value of 1 at delay 0. The control's cosine similarity curve approximates the curve of the reference, but with a small delay (± 300 ms) and near perfect tracking performance (peak of curve at ± 0.95). The curve of the glaucoma case is wider, with a longer delay (± 450 ms) and lower tracking performance (± 0.72).
Figure 2.
 
Tracking performance as a function of stimulus contrast for the cases with early, moderate, and severe glaucoma and the controls.
Figure 2.
 
Tracking performance as a function of stimulus contrast for the cases with early, moderate, and severe glaucoma and the controls.
Figure 3.
 
Scatterplot showing the tracking performance as a function of the SAP Mean Sensitivity (MS) in decibels (dB). The lines show the regression lines (solid) for 40% contrast (red), 160% contrast (green), and 640% contrast (blue) with corresponding 95% confidence intervals.
Figure 3.
 
Scatterplot showing the tracking performance as a function of the SAP Mean Sensitivity (MS) in decibels (dB). The lines show the regression lines (solid) for 40% contrast (red), 160% contrast (green), and 640% contrast (blue) with corresponding 95% confidence intervals.
Table 1.
 
Characteristics of the Study Population
Table 1.
 
Characteristics of the Study Population
Table 2.
 
Characteristics of the Glaucoma Cases
Table 2.
 
Characteristics of the Glaucoma Cases
Table 3.
 
Model Estimates With 95% Confidence Intervals for the Best Model With the Intercept Representing the Estimate for the Controls at a Stimulus Contrast of 40%
Table 3.
 
Model Estimates With 95% Confidence Intervals for the Best Model With the Intercept Representing the Estimate for the Controls at a Stimulus Contrast of 40%
Table 4.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stage
Table 4.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stage
Table 5.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stimulus Contrast
Table 5.
 
Post Hoc Tukey Test Results of the Best Model for Tracking Performance Per Stimulus Contrast
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×