Abstract
Purpose:
We aimed to develop and evaluate the Computerized Glaucoma Visual Function Test (CoGVFT), among a cohort of glaucoma patients, and identify potential new items to optimize the test.
Method:
A cross-sectional study involving 84 patients with open-angle glaucoma of varying severity and 18 controls without glaucoma were recruited. Better and worse eye visual field parameters, visual acuity, contrast sensitivity, 6-Part Cognitive Impairment Test (6CIT) and Glaucoma Activity Limitation-9 (GAL-9) questionnaire responses were recorded. The CoGVFT was administered to all participants. Rasch analysis was used to assess the psychometric properties of the CoGVFT, which was then evaluated with criterion, convergent, and divergent validity tests. Regression modeling determined factors predictive of CoGVFT performance.
Results:
The 38-item CoGVFT demonstrated convergent validity with statistically significant differences in glaucoma severity groups (P < 0.001, analysis of variance). The correlation coefficient for CoGVFT person measures (logits) with GAL-9 person measures (logits) and better eye (BE) mean deviation was 0.528 (P < 0.001) and 0.762 (P < 0.001), respectively, demonstrating convergent validity. Divergent validity was suboptimal as the 6CIT score demonstrated moderate correlation (r = 0.463, P < 0.001) with CoGVFT person measures (logits). Multivariable analysis revealed that better BE contrast sensitivity, lower age, and better BE visual acuity were associated with better CoGVFT performance (P < 0.001).
Conclusions:
The CoGVFT retains most of the features of its predecessor to estimate vision-based activity limitation related to glaucoma.
Translational Relevance:
The CoGVFT is an easily accessible tool that can potentially be used in the community to help detect undiagnosed glaucoma in the population.
The GAL-9 scores displayed good fit to the Rasch model, with no evidence of multidimensionality, ordered thresholds, no differential item functioning or item misfit. Person separation and person reliability indexes were acceptable on initial analysis (2.04, 0.81 respectively), however, targeting was suboptimal (−2.6).
With the initial 116 items included, a PCA of the residuals was performed and the unexplained variance explained by the first contrast was 25.9 eigenvalue units, with 52 items loading (>0.4) onto the first contrast. The unexplained variance explained by the second contrast was 8.7 eigenvalue units, with 16 items loading (>0.4) onto the second contrast. The unexplained variance explained by the third contrast was 4.9 eigenvalue units, with 4 items loading (>0.4) onto the third contrast. The unexplained variance explained by the fourth contrast was 3.9 eigenvalue units, with two items loading (>0.4) onto the fourth contrast. The unexplained variance explained by the fifth contrast was 3.6 eigenvalue units, with two items loading (>0.4) onto the fifth contrast.
The items within each of the five contrasts formed five distinct domains for further psychometric testing. Rasch analysis was performed for the first domain containing 52 items, and the item fit statistics indicated that four items misfitted. These were removed and on a second iteration five items were found to misfit. These were removed and on a third iteration two further items were found to misfit. On the fourth iteration no misfitting items were detected however differential item functioning was detected for 3 items for gender, which was removed. No further DIF was identified. Person separation and reliability were acceptable with values of 2.98 and 0.9, respectively. Targeting was suboptimal with a difference of 2.6 between the mean patient and item values. Rasch analysis of each of the second to fifth domains indicated that none provided valid measurement. Items were grossly misfitting and the person separation was inadequate (
Table 2).
Table 2. Steps Involved in Reducing the Computerized Glaucoma Visual Function Test From 116 Items to 38 Items
Table 2. Steps Involved in Reducing the Computerized Glaucoma Visual Function Test From 116 Items to 38 Items
A CoGVFT person measure (logit) score was created for each of the 102 subjects based on the 38 items that fit the Rasch model (
Table 3). All 38 items involved the binary Correct/Incorrect test information; details of timing per item did not pass Rasch analysis for any item.
Table 3. The Computerized Glaucoma Visual Function Test Items
Table 3. The Computerized Glaucoma Visual Function Test Items
Factors Predictive of CoGVFT (logit) Score: Univariable and Multivariable Analysis
This study demonstrates that the CoGVFT is a potentially useful test for simulating activity limitation related to glaucoma but will likely benefit from some modifications.
The strength of the CoGVFT is that it is an objective computer based test, which can therefore be widely accessible and utilizable. It is a useful bridge between daily patient function and peripheral visual testing, allowing clinicians, patients, and policy makers to better understand the impact of glaucoma on daily life. It is also safer to administer compared to performance-based assessments that requires individuals to physically perform tasks such as ambulation.
21 The CoGVFT also does not preclude individuals with neurological or musculoskeletal disease affecting mobility or speech from undertaking the test.
21
The CoGVFT may have many potential applications once further refined. It may allow a form of glaucoma detection and monitoring by individuals in their own home. Currently up to 50% of glaucoma remains undiagnosed in developed countries,
22 largely because undetected cases have not attended an optometrist for glaucoma screening. Tests that can be performed without attending the optometrist (i.e., on a personal computer) may have a role in increasing detection rates. In addition, computer-based tests like the CoGVFT may be a suitable alternative for use as a visual function assessment in low resource areas that lack accessibility to Ganzfield-bowl perimetry. Given that the test can be quickly completed on an average of 8 minutes and 34 seconds, it can also be widely used in busy clinical environments to assess individuals and identify any visual deficits while they wait for their appointments.
On Rasch analysis the CoGVFT displayed good person separation and reliability, and no DIF. However, targeting was suboptimal, indicating the cohort overall were too able for the test. This is similar to the GAL-9 and other glaucoma-specific tools with good Rasch metrics and reflects that glaucoma tends to not greatly impact activity limitation until more advanced stages.
23 Such a finding could be due to the test being too easy. Alternatively, it may be because of binocular administration of the test, allowing the better eye to compensate for the worse eye until later stages of disease. We feel the test could benefit from monocular administration, as well as the inclusion of more challenging tasks to improve interperson discrimination.
Multivariate analysis revealed that BE contrast sensitivity, lower age, and BE VA were the best combination of predictors of CoGVFT ability. This finding is logical, because many of the tasks require VA and contrast sensitivity. It is possible that older age may correlate with poorer performance because those with more advanced glaucoma were older in our cohort. This is generally unavoidable because glaucomatous damage tends to accumulate with age. Another possible explanation for this observation is unfamiliarity with the technology being used to conduct the CoGVFT among older individuals. During administration of the test, it was observed that the older participants had more difficulty with using the mouse, and it was this inexperience that in some cases resulted in failing to complete a task. Simplifying the tasks, using a touch screen instead of a mouse, and perhaps a trial learning (nonscored) task at the beginning of the test might help improve usability and consistency of measurement.
Although the CoGVFT was assessed successfully using criterion and convergent validity testing, divergent validity was suboptimal as 6CIT score demonstrated moderate correlation (
r = 0.463,
P < 0.001) with CoGVFT score; there was a weaker correlation between 6CIT and BE MD of 0.221 (
P = 0.004). This finding suggests that increasing cognitive impairment is associated with poorer performance on the CoGVFT. This is consistent with findings that cognitive impairment influences visual field test performance
24; however, the CoGVFT may have higher cognitive requirements than HFA, because participants are asked to read, understand, and follow different instructions listed for each task.
25 Future versions of the CoGVFT might benefit from reducing the cognitive requirements of the tasks, so that it can test visual ability more and cognition less. The test also requires participants to be competent with the English language, as the tasks are accompanied with English instructions. Future versions may have potential to translate the tasks into other languages or onto different platforms such as touchscreen tablets or mobile phones to help increase accessibility.
The test displayed suboptimal diagnostic ability as seen in
Figure 1, with sensitivity and specificity levels that did not satisfy the Prevent Blindness America's criteria for minimum performance of a screening test.
26 The test was best at differentiating severe versus moderate cases of glaucoma and not as good at moderate versus mild, as seen in
Figure 2. However, there are many potential avenues available for improving the test discrimination. The test was administered binocularly, but monocular occlusion (testing one eye at a time) will likely result in increased ability to distinguish a wider range of glaucoma severity levels. Furthermore, it may be of benefit to test individual loci on the computer screen methodically; doing so will help distinguish smaller, focal scotoma.
This study itself has potential drawbacks. The sample population was recruited from glaucoma subspecialty clinics at a multisite private clinic and therefore may not be representative of the general population. The study can benefit from having larger sample sizes, especially in the control group (which ideally would be age-matched). However, the current study was powered a priori and was required as a pilot study to help refine the computerized test before larger studies could be undertaken.
Additionally, the study validates the CoGVFT for use on a specific monitor. When the test is disseminated and administered on different monitors, the differing resolution, brightness, and contrast settings of the monitor will impact on the difficulty of the test. It is therefore vital to be conscious of this potential impact and attempt to control for these influencing factors. In addition, it is unknown whether the tasks included in the CoGVFT are a true representation of the real-life tasks that patients experience on a day to day basis. It is at best an estimation of the potential visual difficulties that patients encounter.
In conclusion the CoGVFT retains many of the functions of the original validated CGVFT despite being administered on a smaller computer screen. There are many potential avenues to improve the test's ability to evaluate visual function related to glaucoma.