Donald C. Fletcher
Topographic measurement of vision is essential in all patients with scotomas. Patients trying to search for and identify targets on a screen have to make eye movements. This allows measuring their latency of target recognition. Latencies will thus depend on the quality of vision where the target first appeared. Experiments in low vision patients have shown that their latencies were longer and showed more variation than those of a control group. Target identification correlated only weakly with age and visual acuity, but strongly with best achievable reading speed.
Introduction Characterizing visual impairment needs to include an assessment of the patient’s ability to interact with the visual environment. The most important responses of this kind are eye movements that are elicited by attracting visual attention to an interesting target on the peripheral retina. Patients with low vision can perform feature search, and it has been shown that search performance can be trained . To conduct training, we have to know where to concentrate our efforts. Since patients with maculopathies tend to have retinal areas with diminished visual capacity, we need to find out where these areas are. This article aims to introduce a novel approach towards the assessment of visual impairment by removing some traditional barriers.
It would be desirable to be able to conduct topographic testing without the constraint of strict fixation. This requirement has been a limitation of perimetry for a long time, because steady fixation is hard for patients with damaged foveal vision. A second important issue is the fact that the task in classic perimetry is to simply detect the appearance of a target. In real life, our eyes are always moving, and eye movements can limit the impact of field disruptions on visually guided activities. Thus, we suggest using a task that is similar to a real-life viewing situation, which makes the task more demanding. This can be achieved by requiring to actually recognize a target.
Thus, the “Macular Search Test” (MST) paradigm consists of two tasks: 1. to find the target, and 2. to identify it. This paradigm allows patients to make any eye movements they need in order to solve these tasks, which abolishes the demand for steady fixation. The difference between individual trials is the location where the target first appears. To measure performance, we determine the response latency, i.e. the time from target appearance to the correct response. Methods
We used this paradigm on an experimental group of 135 patients (age 18 to 98 years) with a wide variety of diagnoses from neurological damage after a stroke to medication toxicity, although 71.8 % had Age-Related Maculopathy (ARM). Visual acuity (v.a.) in the better eye varied between 20/20 and 20/800 (median 20/139, i.e. appr. 0.15 visus in metric notation). There was no significant difference in v.a. between ARM patients and those with other diagnoses (Mann-Whitney U test, p = 0.567). The control group consisted of 30 healthy subjects from 19 to 84 years of age (15 women) with best corrected v.a. between 20/20 and 20/40.
The experiments were in compliance with the tenets of the Declaration of Helsinki. Stimuli – Targets were Landolt rings with a gap in one of four orientations (right, left, top and bottom). The target stayed visible until the correct answer had been given on the keyboard. Target contrast in Weber notation [CW = (L max – L min) / L min] was always maximal at 240 %.
Procedure – Subjects (Ss) sat comfortably and viewed a computer screen from a distance of 40 cm wearing their best available optical correction. Viewing was binocular, and a tone signaled the upcoming appearance of a new target. Ss never knew where the target would appear, because the sequence of locations was randomized. There were 32 such possible locations, which were arranged on four circles of 2, 4, 6, and 8 degrees eccentricity (8 each). Each test location was used only once, so that a trial block had 32 individual trials. The Ss used a ring of 12 mm diameter (1.72 deg) with a center hole of 2 mm (0.29 deg) in the middle of the screen as a reference point by initially “centering” their gaze on it casually. The ring always disappeared before a target appeared. Thus, the instructions were: “Start by looking in the middle, find the target, and tell me where the gap is.” Ss were encouraged to make any eye movements they needed to identify the target as quickly as possible.
We always made sure that the Ss could solve the task by determining the threshold. A series of Landolt rings appeared one at a time, beginning with the largest size and declining in 1/10 log steps. For each size, the subject was asked to tell the gap position. The smallest that could be identified was taken as the size threshold. This size was then doubled and used throughout the entire rest of the experiment. This procedure made sure that the subject could solve the task provided their eye movements could put the target on the right spot on the retina. Thresholding typically took less than one minute.
As the program accepted only correct responses, recorded latencies lasted from the target appearance to the correct identification. Ss responded verbally, and the examiner entered the response on the keyboard. This indirect performance measure was deemed acceptable because it prevented contaminating the data by other variables based on inter-individual differences in age, gender and educational status. Thus, each recorded response latency contained a component added by the examiners’ reaction times. Their medians were calculated after 200 trials for each examiner to be 633 ms (DCF) and 603 ms (MM).
Software and statistics – The experiments were run by software on a PC computer under Windows ME. The program was written in Delphi 6 (Borland Inc.). The data base in the program saved all single trial data, so that they could later be retrieved if necessary.
For statistical analysis, we transferred the data to StatView (Abacus Software, Inc.) and used non-parametric statistics to avoid assumptions regarding a normal distribution of the data. Relationships between variables were expressed as coefficient of determination R2. Since we expected correlation between v.a. and reading speed, we also performed multiple regression analysis to gain a more realistic estimate of the contribution of each variable while adjusting for the others. This procedure followed the strategy described by Legge et al. We used SPSS (Chicago, IL) software to regress median (med) latency on reading speed, age and v.a. simultaneously. Logarithmic transformations were used on all variables to reduce significant positively skewed distributions. Values of R2 given here are all based on the log-transformed data.
Test-Retest Reliability –Twenty patients and 20 control Ss performed the test twice within a few minutes. Calculating the correlations between individual trials of two test runs is not appropriate here, because the sequence of locations of target appearance was randomized. Hence, any trial #M in test run 1 might not be comparable with trial #M in test run 2 in the same patient, because the two trials would most likely have tested different locations. The point is exacerbated by the functional heterogeneity of the retinae of our Ss due to scotomas and other deficits. Instead, we used the sums of latencies for each block of 32 trials and subject. The correlation coefficient R served as a traditional measure of test-retest reliability. We also used the coefficient of determination R2 to indicate what percentage of the variation was accounted for just by repeating the test. The coefficient of repeatability was calculated as 1.96 X the standard deviation of the mean differences between the two sets of data.
RESULTS The MST showed very good test-retest reliability. It was more difficult for the patients than for the control group. This was reflected in its duration (mean of 140.0 s vs. 87.1 s), higher variability (maximum/minimum response latency ratio of 6.3 vs. 3.4) and a longer latency (med = 2.521 s vs. 1.486 s).
Test duration – It took Ss between 55 and 478 seconds to complete all 32 trials of the MST (med = 122 s, IQR = 82.5 s). This duration showed no appreciable correlation with patients’ age (R2 = 0.048). Men were a bit faster (med duration = 105.5 s, 74 s IQR) than women (med duration = 133 s, 77.7 s IQR), which was statistically significant (MWU, p = 0.025).
Reading Speed – Best reading speed for a 60-character paragraph of MNread text showed no gender-related differences. It declined slightly under the influence of age (R2 = 0.0538) and varied between 45 and 2118 characters per minute (CPM; med = 699 CPM, IQR = 714 CPM. Given the average word length of 4.05 characters/word in MNread, the median reading speed was equivalent to 172 words/minute.) As the most frequent diagnosis was ARM, we looked for the same relationships as above considering only the 97 patients with ARM. The results were very similar to those from the whole group, i.e. R2 = 0.0335 for best reading speed vs. age.
Age and v.a. together can make independent contributions to reading speed. This showed in multiple regression analysis, with latencies from the MST paradigm not included. It yielded R2 = 0.4315 (= 43.15 % of the variance of reading speed was accounted for.)
Test performance – The most conspicuous result of the “search and identify” paradigm was how much performance levels varied between patients. Median latencies of correct responses lay between 730 ms and 10,195 ms (a factor of almost 14). The effect was just as strong when using the sum of all 32 latencies per trial block, which varied between 24.4 and 404.0 seconds (a factor of >16).
Gender differences – Men performed slightly better (med. latency = 2410 s, 1640 s IQR) than women (med. latency = 133 s, 1650 s IQR), which was statistically significant (MWU, p = 0.044). Latencies also varied strongly between individuals: The longest could be 1.4 – 31 times longer than the shortest, depending on where the target first appeared. There was a statistically significant performance difference between the subgroups ARM vs. non-ARM, the latter being slightly faster (med. latency 1730 ms vs. 2500 ms (MWU, p = 0.018)).
Correlations – Since age and v.a. varied greatly between patients, we investigated possible correlations with the performance level. For med. latency vs. age, R2 = 0.0967 and for med. latency vs. v.a., R2 = 0.176. This means that only a small part (27.3 %) of the variance of the dependent variable (MST performance) could be accounted for by the two independent variables age and v.a. As both were essentially independent of each other (R2 = 0.003), these two variables could not be interpreted as major influences on search performance.
However, we found that the highest correlation between search performance (med. latency) and any other variable was the one with best achievable reading speed. As these data included several extreme outliers, so that we performed a 10% winsorization. Thus, all values of median latency below the 5th and all above the 95th percentile% were set to the 5th and 95th percentile level. The result yielded R = 0.748 and R2 = 0.560. The results of multiple regression analysis tell us that the contributions of age and acuity are not statistically significant. We conclude that the dominating correlation is the one between search performance and best possible reading speed.
DISCUSSION The most important finding of the reported research is that search test performance varied greatly despite the fact that visual acuity was neutralized by relating the acuity demand to the individual thresholds. Furthermore, it was surprising to see that the major correlations were essentially the same for all patients with a wide range of diagnoses and for the subgroup with ARM. Furthermore, it was interesting that the patient age did not seem to significantly influence performance. All three points support the notion that “search and identify” performance is influenced by factors other than age, visual acuity and diagnosis. We conclude from the good test-retest reliability (see above), that the variations between patients are truly patient characteristics and not an effect of poor test-retest reliability.
It is not surprising that some of the patients could read with acceptable speed even with a dense central scotoma in both eyes, which has also been found by others. Note that the current results do not allow direct conclusions from the results of detailed micro-perimetry by SLO, since the latter can only be performed monocularly. Since reading is a learned behavior, it cannot be ruled out that the found differences might have been influenced by inter-individual differences either before onset of low vision, like educational status, or those including low vision, like current reading habits. This indicates that maintaining some reading practice, albeit with adequate magnification, may still pay off for patients with low vision.
Lott et al. found that good high-contrast v.a. does not assure that elderly Ss (58 – 102 years) can read satisfactorily and that age alone is not a good predictor of reading performance. Thus, one might expect that our results may come out differently because of the wider age range of our patients (18 – 98 years) and of their universally compromised vision. However, our results also showed that age alone is not a good predictor of reading performance. This could be explained by the fact that the presence of low vision in our cohort may have simulated the compromising conditions that led Lott and colleagues to their conclusion, i.e. low contrast vision, motor ability, and attentional field integrity. A recent study has shown that performing well in some tasks puts more emphasis on normal functioning in the cognitive domain. Though we did not test for cognitive status here, psychological profiles taken from all our patients made sure that none with cognitive deficits were included. This makes it unlikely that deficits in cognitive abilities may have emerged as a major factor. The presented findings show that measuring visual acuity alone in patients with low vision could have led to entirely misleading conclusions. They demonstrate that there are other factors that influence performance in a task that bears resemblance with those that have to be faced daily by patients with low vision.
Conclusions Because steady fixation is not required, the Macular Search Test allows functionally relevant vision testing in a relaxed atmosphere and at a wide range of ages, acuities, and reading skill levels. We conclude from our findings that continuous text reading and the “find-and-identify” paradigms share an important behavior that determines performance in both tasks. We hypothesize that the factor enabling patients to perform well in both paradigms is oculomotor control, which has been shown to be a learnable skill. Hence, we propose that oculomotor training should be a focus of activity for low vision therapists.
One could argue that the high correlations found here with reading performance indicate that a conventional reading test can yield the same results. However, the “find-and-identify” paradigm has four distinct advantages relative to testing reading: 1. Independence of reading ability, so that very young and illiterate Ss can also be tested. 2. In patients who can read, differences between levels of reading skill and habits introduce a source of noise into the data that can be avoided in the “search and identify” paradigm. 3. As long as patient and examiner can communicate verbally, this method allows comparisons between patient cohorts who speak different languages. 4. It demonstrates the spatial positions where oculomotor training could intervene, since patients could be trained to direct exploratory and compensatory movements in those directions, in which the MST latencies are found to be longest.
The figure shows a report card with result of an appr. 4 minute run of the Macular Search Test. The disks in the diagram of the visual field (right) show the locations of first appearance of the targets (Landolt rings) out to 8 deg eccentricity. Their gray values indicate the relative duration of the correct responses (white = shortest, black = longest). These findings indicate that the patient has a scotoma in the right and upper right section of the visual field.
Reference For more detail see: MacKeben M & Fletcher DC. (2011) Target search and identification performance in low vision patients. Invest Ophthalmol Vis Sci. 2011, Sep 29;52(10): 7603-9