Mean deviation values for healthy subjects and patients with optic neuritis or optic neuritis and multiple sclerosis (ON/MS) on the same day (A) and on different days (B). Note the large variability in the patients.
Pattern SD values for healthy subjects and patients with optic neuritis or optic neuritis and multiple sclerosis (ON/MS) on the same day (A) and different days (B). Note the large variability in the patients.
Short-term fluctuation values for healthy subjects and patients with optic neuritis or optic neuritis and multiple sclerosis (ON/MS) on the same day (A) and different days (B). Note the large variability in the patients.
Corrected pattern SD for healthy subjects and patients with optic neuritis or optic neurtitis and multiple sclerosis (ON/MS) on the same day (A) and different days (B). Note the large variability in the patients.
Gray-scale and probability results in healthy subjects and patients with optic neuritis. Results are shown for the same day and different days matched for time of day. The 2 fields at the top of each graph are from 8 AM, the next pair down from 10 AM, etc. A, Typical example of a normal subject's consistent results. B, A patient with optic neuritis with consistent results. C, A patient with optic neuritis with variable results. Note the variation from near normal to near complete hemianopia. D, Another patient with variable results.
Results of visual field simulation of 30 visual fields by means of thresholds from a patient with optic neuritis who had variable results of conventional automated perimetry strategy. A, Actual values. B and C, Visual field with the highest (B) and lowest (C) right-upper-quadrant sensitivity. Note how the appearance of a quadrant defect can be simulated.
Wall M, Johnson CA, Kutzko KE, Nguyen R, Brito C, Keltner JL. Long- and Short-term Variability of Automated Perimetry Results in Patients With Optic Neuritis and Healthy Subjects. Arch Ophthalmol. 1998;116(1):53-61. doi:10.1001/archopht.116.1.53
To measure the short- and long-term variability of automated perimetry in patients with optic neuritis and normal subjects.
Prospective case-control design of patients with recovered optic neuritis with intraday and interday repetitions to obtain robust variability measurements. Entry criteria included a corrected pattern SD that was worse than the normal 5% probability level and a mean deviation worse than −3 dB but better than −20 dB. Five Humphrey 30-2 full threshold tests were administered during a 7-hour period (1 test every 2 hours) on the same day and at the same periods on 5 separate days.
Seventeen patients with recovered optic neuritis and 10 healthy subjects of similar age.
Main Outcome Measures
Short-term variability and long-term variability for global visual field data.
Patients with optic neuritis demonstrated variations in visual field sensitivity that were outside the entire range of variability for normal controls. These variations occurred for multiple tests performed on the same day at specific times and for tests performed at specific times on different days. There were no consistent patterns of sensitivity changes that could be attributed to time of day. The most dramatic fluctuations occurred in a patient whose visual fields varied from normal to a hemianopic defect from one week to another and from a partial quadrant loss to a hemianopic defect at different times on the same day. Seven of the patients with optic neuritis also demonstrated intermittent vertical step defects.
Patients with resolved optic neuritis can have large variations in visual field results on different days and at different times on the same day. The variations affect both the severity and the pattern of visual field loss and do not appear to be consistent across patients. These data indicate that care must be taken when automated visual field results in patients with optic neuritis are interpreted. Distinguishing systematic changes in sensitivity from variability requires more than a comparison of the current visual field with the most recent previous visual field.
THE ABILITY to distinguish visual field progression or improvement from one visit to the next is difficult because of the variability that occurs with threshold measurements, especially in areas of visual field damage. This variability of conventional automated perimetry has been investigated extensively in normal subjects and patients with glaucoma1- 6 and is closely related to loss of sensitivity. Beyond approximately 1 log unit (10 dB) of sensitivity loss in patients with glaucoma, variability rises exponentially and encompasses nearly the full measurement range of the instruments used.7 For example, a test location with 15 dB of loss in patients with glaucoma has a 95% prediction interval that ranges from about 5 to 30 dB.
Less is known about the variability in visual field sensitivity that occurs in optic neuritis and multiple sclerosis. In contrast to glaucomatous damage, optic neuritis is characterized by immune-mediated inflammatory damage to the optic nerve, with the myelin sheath being the primary target of the inflammation. The main difference between glaucomatous visual field damage and optic neuritis–associated visual field loss is that in optic neuritis, the nerve fiber bundles that subserve central vision (the papillomacular bundles) are involved as often as the paracentral superior and inferior arcuate nerve fiber bundles.8 Early visual field loss in glaucoma, on the other hand, is predominantly found in the superior and inferior arcuate nerve fiber bundle regions, and central visual loss typically does not occur until advanced stages of the disease process.
In view of the differences between glaucoma and optic neuritis, it is not clear that the response variability found in glaucomatous visual fields can be generalized to visual field loss derived from optic neuritis or multiple sclerosis. Results from the Optic Neuritis Treatment Trial8 indicate that optic neuritis produces a variety of patterns and degrees of visual field loss, both among different individuals and in the same individual at different times during the disease process. Although this qualitative information is helpful, it is also important to quantitatively document these variability characteristics to provide an empirical basis for evaluating whether visual fields in these patients are improving, getting worse, or simply demonstrating fluctuations. The purpose of the present study was to perform a formal investigation of the short-term and long-term variability of automated perimetry threshold determinations in patients with residual visual field loss caused by optic neuritis and/or multiple sclerosis in comparison with findings in healthy control subjects of similar age.
Seventeen patients with a clinical history of optic neuritis and/or multiple sclerosis and 10 healthy subjects of similar age participated in the study. Before testing, informed consent was obtained from each participant according to the tenets of the Declaration of Helsinki. The study protocol was approved by the institutional review boards of both the University of Iowa, Iowa City, and the University of California, Davis.
The control subjects were paid volunteers who were hospital employees or students. We selected them so that their ages would fall within the expected age range of patients with optic neuritis (20-50 years). Five healthy subjects were enrolled at each institution. These subjects were included if they had no history of eye disease or surgery, had no more than 5 diopters (D) of spherical equivalent and 3 D of astigmatic refractive error, and had normal results of ophthalmologic examination and automated perimetry by means of the Humphrey Field Analyzer (Humphrey Instruments, San Leandro, Calif) program 30-2. Potential controls were excluded if they had visual field indexes (mean deviation [MD], pattern SD [PSD], short-term fluctuation [SF], or corrected pattern SD [CPSD]) that were at the P<.05 probability level or worse, had 3 or more adjacent test locations with a total deviation score at the P<.05 probability level or worse, or had 2 or more adjacent test points with a total deviation score at the P<.01 probability level or worse. One eye was randomly selected for testing in the normal control subjects.
Patients with optic neuritis were recruited from Optic Neuritis Treatment Trial patients (Iowa), by telephone calls after a patient database search (Iowa and Davis), and by referral from local multiple sclerosis societies (Davis). Ten patients with optic neuritis were enrolled at the University of California, Davis, and 7 were enrolled at the University of Iowa. All of the patients with optic neuritis had a clinical history of optic neuritis with residual visual field loss. Nine of the 10 Davis patients and 5 of the 7 Iowa patients also had multiple sclerosis. To be eligible, patients with optic neuritis had to have at least 2 previous visual fields in which the CPSD was outside the P<.05 normal probability level and an MD that was worse than −3 dB but better than −20 dB in one or both eyes. If none of the previous visual fields had been conducted within the past 6 months, an additional eligibility visual field was performed. In the event that both eyes qualified for the study, one was randomly selected for testing. The patients underwent a complete neuro-ophthalmologic examination at theinitial visit and were excluded if they had any ocular or neurologic disorders other than optic neuritis and/or multiple sclerosis or if they had refractive errors greater than 5 D spherical equivalent or 3 D astigmatic error. One patient only underwent the multiple test sessions on the same day, and 1 patient only participated in the multiple test sessions on different days. Therefore, comparisons within the 2 testing intervals were performed on the data from 16 patients, and comparisons between the 2 testing intervals (same day compared with weekly) were conducted on the data from 15 patients.
After the qualifying examinations, all participants underwent perimetric testing with the Humphrey Field Analyzer 30-2 full threshold program on one eye, according to the schedule shown in Table 1. Ten perimetric sessions were conducted, 5 at intervals of 1 to 2 hours on the same day and 5 with testing at various times of the day and separated by a 1-week interval between test sessions. The standard test conditions for the Humphrey Field Analyzer, 31.5-apostilb background, Goldmann size III target, and 200-millisecond stimulus duration, were used for all perimetric examinations. An appropriate near correction for the test distance was used for the tested eye, and an eye patch was used to occlude the nontested eye. During each examination, rest breaks were given when requested by the participant.
The visual field data from all participants were imported into the SAS (Cary, NC) and Sigmastat (San Rafael, Calif) statistical analysis packages. The primary outcome variables were all normally distributed according to the Kolmogorov-Smirnov test (P>.05). All had similar variances by means of the Levene median test (P>.05). Analysis of variance (ANOVA) with post hoc t tests corrected for multiple comparisons were used for all statistical evaluations except those that failed the above tests for normality and homoscedasticity, in which case we performed an ANOVA on ranks with post hoc tests by the Student Neuman-Keuls method. The P<.05 probability level was used as the criterion for statistical significance.
Differences in thresholds between groups and at different times of the day were tested for statistical significance by means of repeated-measures ANOVA with a 3-factor nested design. The between-subjects factor was group (healthy subjects and patients with optic neuritis) and the within-subjects factors were interval (same day and different days) and time (hour of the day). The dependent measures were mean threshold, foveal threshold, MD, SF, PSD, and CPSD. An ANOVA was also used to test for differences in variabilities for the same dependent measures. Interval was the within-subject variable and group was the between-subject variable.
Differences in variability (SDs of the 5 tests) between intervals (all tests in 1 day vs different days) were tested for statistical significance by means of a general linear models procedure for a 2-factor nested design. The between-subjects factor was group (healthy subjects and patients with optic neuritis) and the within-subjects factor was interval (same day and different days). Subjects were nested within groups.
We developed software with the use of Visual Basic in Microsoft Excel (Microsoft Corp, Redmond, Wash) to perform simulations of automated visual fields. To estimate the slope (SD of the cumulative gaussian function) of the frequency of seeing function for each possible visual sensitivity of the dynamic range of the Humphrey Field Analyzer, we used results from frequency of seeing experiments in patients with glaucoma.9,10 We evaluated the relationship between the slope of the frequency of seeing function for these data by finding the best fit of various simple functions. A power function with an r2 of 0.84 was used. Using this function, we were able to calculate a slope of the frequency of seeing function for the visual sensitivity for each test location. By then specifying the false-positive rate (5%) and false-negative rate (1%) for each test location, we could estimate the probability that a subject would respond to a stimulus at each possible sensitivity of the instrument we were simulating. Then, using (1) a 4/2 staircase procedure and (2) the starting values (25 dB) for the primary or "seed" points, (3) taking the mean of the doubly determined primary point, and (4) using the rules for passing other starting values within each quadrant (once a threshold was found, eccentricity-corrected values were passed to all adjacent locations not yet tested), we were able to simulate a standard algorithm Humphrey visual field examination. We assumed that any variability we found could be attributed mostly to the testing method rather than patient factors.
The mean age of the patients was 41.7±9.9 years; the mean age of the controls was 34.6±7.5 years. The age difference between the 2 groups was not statistically significant. As indicated in Table 2, the patients with optic neuritis had significantly greater elapsed test times, questions asked, false-negative trials, false-negative errors, and false-positive trials than the controls. There was no significant difference in the number of false-positive errors between the 2 groups.
The MD, PSD, SF, and CPSD in the patients with optic neuritis were all significantly different from the results of the controls (Table 3). The mean within-subject SDs of MD for the 5 tests in control subjects was 0.46 dB for the same-day measures and 0.36 dB for weekly visual field measures. In the patients with optic neuritis, the SDs of MD for the 5 same-day measures was 1.35 dB and for the different-day measures it was 3.10 dB. This represents a 3-fold increase in variability for the patients with optic neuritis in comparison with the control subjects for the same-day measures and a 9-fold increase in variability for the different-day measures. These differences were even more pronounced when the visual field quadrants were analyzed in a similar manner (Table 4).
Figure 1 presents the MD for all 5 visual field determinations for control subjects and patients with optic neuritis for same-day and weekly determinations. Note that the normal MD values are all tightly clustered, whereas the results for the patients with optic neuritis are more dispersed. In addition, it appears that there are considerable individual differences in variability of MD values among patients with optic neuritis. Similar results for PSD, SF, and CPSD are presented in Figure 2, Figure 3, and Figure 4 with a similar format.
To evaluate the effects of group and test interval on threshold, the repeated-measures ANOVA with the 3-factor nested design showed that each of the dependent measures was significantly different between the control and patient groups (mean threshold, P<.001; foveal threshold, P<.001; MD, P<.001; PSD, P<.001; SF, P<.001; CPSD, P<.001). For MD, there was an interaction between group and test interval; there was no significant difference between same-day and different-day measures in control subjects, whereas there was a significant difference between same-day and different-day measures for the patients with optic neuritis. For the total average score, there was a significantly (P=.05) higher score (better sensitivity) of 1.36 dB for the different-day measures as compared with the same-day measures (includes both patients and normal subjects). There was no significant effect of time of day.
The ANOVA for the SDs of the 5 visual field examinations by time interval disclosed significant differences among the dependent measures (mean threshold, P=.02; foveal threshold, P=.04; MD, P=.02; PSD, P=.01; SF, P=.009; CPSD, P=.02). There were significant differences between the groups but not between the testing intervals (control subjects and patients combined). However, when threshold was entered as a cofactor for the average total score, there were no significant differences between the groups of subjects for these dependent measures. Note in Figures 1 through 4 that there appears to be more variability in the visual field measurements from week to week than from within a day.
Individual examples of visual fields obtained on the same day and weekly are presented in Figure 5. Figure 5, A, shows the results of a typical set of visual field determinations for a control subject. Note that there is minimal variation in the pattern of visual field sensitivity for the 10 test procedures. Figure 5, B, shows visual field loss of a patient with optic neuritis in the inferior nasal quadrant of the right eye. Both the severity and pattern of visual field loss appear to be consistent, both for visual field measures obtained on the same day and for those obtained on different days separated by 1-week intervals. About one fifth of the patients with optic neuritis demonstrated this degree of consistency among multiple visual field tests; about 30% showed moderate amounts of variability. However, approximately half of the patients with optic neuritis exhibited large variations in visual field results from one test to the next.
Two dramatic examples of this high variability are shown in Figure 5, C and D. Figure 5, C, shows that for the same-day measures, this patient had a small superonasal paracentral visual field deficit for the first visual field performed in the morning. The results became progressively worse as the day progressed, resulting in a hemianopic defect by the late afternoon. The measures obtained on different days, separated by a week, range from a nearly normal visual field in the first week to a hemianopic defect in the third week to a small superonasal hemianopic deficit in the fifth week. A similar type of variation in both the pattern and severity of visual field loss is shown in Figure 5, D, for another patient.
Variability of visual field measurements in recovered patients with optic neuritis with −3 to −20 dB of residual visual field sensitivity loss was significantly higher than in healthy subjects of similar age, both for short-term (same day) and long-term (weekly) comparison intervals. We observed 3 patterns of variability among the patients with optic neuritis. Approximately 20% of the patients had good reproducibility for multiple visual field examinations, being equivalent to or only slightly worse than the variability demonstrated by control subjects. Moderate variability of visual field results was observed in approximately 30% of the patients with optic neuritis. About half of these patients demonstrated large variability in visual field measurements, in some instances ranging from normal or minimally abnormal visual fields to dense hemianopic deficits over a week or two or even within the same day. Similar patterns of variability were observed for PSD, CPSD, SF, and reliability indexes (with the exception of false-positive responses). These findings indicate that the response properties of the majority of patients with optic neuritis undergoing visual field testing exhibit significant fluctuations.
Burde and Gallin11 reported elevated thresholds in patients with resolved optic neuritis by means of static perimetry, even though kinetic perimetry results yielded normal findings. This variation in sensitivity between static and kinetic perimetry results in optic neuritis, sometimes referred to as "statokinetic dissociation," has subsequently been reported by other investigators.12,13 Harms14 in 1976 described the generally high variability among repeated visual field tests in patients with optic neuritis, and the first systematic study of the variability of visual thresholds in patients with multiple sclerosis was conducted by Patterson and colleagues in 1980.15
A number of factors, including temperature, fatigue, attention, time of day, axonal damage, "cross talk" among nerve fibers, extent of visual damage, and other related determinants, have been proposed to account for this high variation in visual field sensitivity. In the present study, we found no evidence that fatigue or time of day was strongly related to the variability of visual field measurements in patients with optic neuritis. As reported in glaucoma,1- 6 we observed a strong direct relationship of threshold and variability.
The magnitude of our long- and short-term variability is somewhat higher than that found in glaucoma (Table 5). Variability might be inherently higher in optic neuritis because of myelin loss surrounding surviving axons and related neuronal impulse slowing and instability. Alternatively, the variability of optic neuritis in this study may appear higher because our sample size is smaller than those in the published glaucoma studies, and our selection criteria excluded patients with mild defects (<3 dB of MD loss). Variability measures might also be different because the studies cited in Table 5 were performed on the Octopus (Interzeag, Schlieren, Switzerland) rather than the Humphrey perimeter. As discussed below, some of the visual field variability is likely caused by an interaction between the shallow slopes of frequency of seeing curves at damaged test locations and the method of passing starting values within each quadrant as a test progresses. Early Octopus perimeters used as starting values the first visual field of normal values corrected for age. Subsequent fields used values from a "master field" that was a composite of previous thresholds from that patient. Later Octopus perimeters have adopted the 4-quadrant primary point to adjacent test location strategy (see below).
About 40% of the patients with optic neuritis demonstrated visual field deficits that appeared to have a vertical step component (usually in the form of a quadrant defect). None of the control subjects had a step defect. None of the patients with vertical step defects had a homonomous defect in the fellow eye, and the defects respected the horizontal meridian in all of the patients but 2. Many of these vertical step deficits were present intermittently over repeated test sessions, and their presence was more frequently observed in those patients with greater amounts of variability. Seven of the weekly determinations and 3 of the same-day sessions exhibited vertical step deficits that were intermittent. Although some of the vertical steps may be related to anatomical or pathophysiological factors in these patients, we believe that many of these vertical step patterns in this patient population may be caused by the testing algorithm used by the Humphrey Field Analyzer (and the Octopus perimeter).
With the full-threshold strategy of the Humphrey Field Analyzer, "primary points" (9° vertical and 9° horizontal displacement from fixation) are the first to be determined in each visual field quadrant. These primary point thresholds are doubly determined and the mean is calculated. The primary point threshold is then used as a basis (with a correction for eccentricity) for the starting value for each of the 8 neighboring points surrounding the primary point. In turn, the thresholds of these secondary points are used as the basis for starting values of adjacent test locations within the quadrant. That is, the passing of starting values does not cross the horizontal or vertical meridians. Because the results of staircase procedures can be influenced by the starting position of the staircase (especially when variability and response errors are high), a low threshold for a primary point could make it more likely that the entire quadrant might produce lower sensitivity values and the appearance of a vertical step or quadrant defect.
To test this possibility, we performed 2 analyses. The first consisted of comparing the mean of the primary point sensitivities in a particular quadrant (eg, 9° up and 9° right from fixation) to thresholds from 6 equidistant test locations within the same quadrant displaced vertically (eg, 3° up/21° right, 9° up/21° right, and 15° up/21° right) and horizontally (eg, 21° up/3° right, 21°up/9° right, and 21° up/21° right). A correlation coefficient was then computed for the primary point sensitivity and the mean of these 6 within-quadrant points. This was compared with 6 other locations that were outside of the quadrant (thereby influenced by a different primary point) but were the same distance from the primary point. These test sites, outside the quadrant, were the mirror image of the 6 test locations within the same quadrant. Our hypothesis was that if the starting value of the staircase was a significant factor in the sensitivity measures of the quadrant, because of a combination of the primary point sensitivity and high response variability, then the within-quadrant points (directly influenced by the primary point) would be more highly correlated with the primary point sensitivity than the equidistant outside-quadrant points (predominantly influenced by a different primary point). If this relationship is true, it may account for the presence of some vertical step (quadrant) defects. We suspect this mechanism also applies to step defects in other patients with visual field damage, including patients with glaucoma. Future observations should clarify the generality of this mechanism.
Results of this analysis are presented in Table 6. They show that the within-quadrant sensitivity values are more highly correlated with the primary point sensitivity than those equidistant test locations outside the quadrant. For the within-quadrant points, the r2 was 0.55 for same-day measurements and 0.50 for weekly measurements. In comparison, the outside-quadrant points had an r2 of 0.19 for the same-day measures and 0.16 for the weekly measurements. These differences were statistically significant at the P=.009 level (paired t test).
The second analysis consisted of simulating visual field tests with conditions of the primary point having a moderately low sensitivity value. We chose visual field data from one perimetric examination from a patient with high variability and used these results as the "true" visual sensitivities used to calculate frequency of seeing curve slopes for each test location (see Figure 5, D, fourth field of different days; and Figure 6, A). We then simulated 30 visual fields by means of the Humphrey perimetry methods of program 30-2. Figure 6, B, shows the field with the lowest right-upper-quadrant thresholds; Figure 6, A, shows the field with the highest right-upper-quadrant thresholds. All of the simulated fields showed defects that were less dense (higher sensitivities) than the actual values. This is probably because of the right-upper-quadrant primary point of 18 dB and values being passed to the surrounding points that are more than 1 log unit higher than the "true" thresholds. In Figure 6, it can be seen that these simulated conditions can produce the appearance of intermittent quadrant and vertical steplike deficits. In view of both of these results, we believe that at least some of the transient vertical steps that were present in the visual fields of patients with optic neuritis in our investigation were the result of an interaction between the Humphrey Field Analyzer's testing algorithm and the high response variability (related to the shallow slope of the frequency of seeing curve in damaged test locations).
In summary, we found large variations in visual field measurements obtained in many patients with recovered optic neuritis who had residual visual field loss of −3 to −20 dB. Our study suggests that some of these large changes in the visual field are likely related to the testing method rather than to new disease activity. Therefore, treatment decisions that are based solely on a comparison of the current visual field examination with the most recent previous visual field should be avoided. We hope that the use of larger test stimuli,10 more robust test strategies, and other refinements of perimetry in the future will help to reduce the difficulties associated with high variability in the visual fields of patients with optic neuritis. Until that time, caution in the interpretation of test results, multiple retests of the visual field, and examination of long-term trends over many visual field examinations should be exercised in any treatment decisions concerning patients with optic neuritis that are based on perimetric test results.
Accepted for publication August 29, 1997.
This study was supported in part by a Veterans Affairs Merit Review Grant (Dr Wall); research grants EY-03424 (Dr Johnson) and EY-09435 (Dr Keltner) from the National Eye Institute, Bethesda, Md; unrestricted research grants from Research to Prevent Blindness Inc, New York, NY (to the Departments of Ophthalmology at University of Iowa and University of California, Davis); and cooperative agreement EY09435 (Roy W. Beck Jaeb Center for Health Research, Tampa, Fla) from the National Eye Institute.
We thank our patients for their cooperation and for tolerating the rigors of the automated perimetry testing protocol used in this study. We also thank Kimberly Cello and Jacqueline Nelson-Quigg for their assistance in collecting visual field data.
Reprints: Michael Wall, MD, Department of Neurology, University of Iowa College of Medicine, 200 Hawkins Dr, #2007 RCP, Iowa City, IA 52242-1053 (e-mail: email@example.com).