Peripheral Color Contrast

Transcription

1 Investigative Ophthalmology & Visual Science, Vol. 32, No. 10, September 1991 Copyright Association for Research in Vision and Ophthalmology Peripheral Color Contrast A New Screening Test for Preglaucomarous Visual Loss Tak C. Yu,*t Fernando Falcao-Reis,:J: Werner Spileers, and Geoffrey D. Arden*f A new test of peripheral color contrast is described. A high-definition color monitor driven by a personal computer with a graphics interface card displays an annulus subtending 25 at the eye. The color contrast between the annulus and the background can be varied. Forty-five degrees of the annulus is randomly removed in one of four quadrants. Patients are asked to identify the position of the gap while fixating a central spot. The minimum color contrast between annulus and background at which the identification is possible is between 13-16% for the protan, deuteran, and tritan axis in normal subjects. This threshold value changes little with age, refractive error, or pupillary aperture, and test-retest variability is low. Testing one eye takes only 1-2 min. The test was applied to ocular-hypertensive and glaucomatous patients. All patients with glaucoma had thresholds greater than two standard deviations (SD) above the normal mean. In addition, 97% of glaucoma patients had thresholds greater than four SDs, and 95% had thresholds more than five SDs above the normal mean. Most patients with ocular hypertension and clinical signs indicating a low or medium risk of conversion to glaucoma had thresholds under the upper limit of normal. High-risk patients with ocular hypertension fell into two groups. One approximated to normal; the other had elevated thresholds, which in many cases were more than four SDs above the normal mean. The epidemiologic consequences of this test are discussed. Invest Ophthalmol Vis Sci 32: , 1991 Arden and Jacobson 1 described a method of measuring contrast sensitivity to black and white gratings and showed losses at low spatial frequencies in patients with glaucoma and ocular hypertension. They suggested that this test would permit mass screening. However, this was hopelessly optimistic; the upper limit of normal overlapped values obtained from glaucomatous patients with the best preserved vision. Improved techniques, in which the grating was temporally and spatially modulated improved discrimination, 2 but the incidence of false-positive and false-negative results was unacceptably high. Thus, although many authors agree that losses of contrast sensitivity From the *Department of Clinical Ophthalmology, Institute of Ophthalmology, University of London, and the felectrodiagnostic Clinic, Moorfields Eye Hospital, London, United Kingdom, the ^Department of Ophthalmology, Oporto Medical School, Oporto, Portugal, and the Department of Ophthalmology, KU Leuven, Leuven, Belgium. Supported in part by grants from the Wellcome Trust and the Wolfson Foundation. TCY was supported by a studentship from Fight For Sight (UK), and FF-R had a scholarship from Calouste Gulbenkian Foundation, Lisbon. Submitted for publication: November 19, 1990; accepted April 25, Reprint requests: Professor G. B. Arden, Electrodiagnostic Clinic, Moorfields Eye Hospital, City Road, London EC IV 2PD, UK. can be seen in patients with 20/20 vision or better, testing contrast sensitivity was not valuable for screening. Recently, contrast sensitivity has been measured at various peripheral loci in patients with glaucoma and ocular hypertension. 3 These authors followed the hypothesis that, because the field defects in glaucoma initially are circumscribed and in the peripheral visual field, at a still earlier stage the losses of contrast sensitivity might be limited to the same arcuate zone. They investigated a group of patients with early glaucomatousfielddefects, detectable by automated perimetry, in whom central low spatial frequency contrast sensitivity was within normal limits. At 15 and 20 off axis, contrast sensitivity was grossly elevated from normal means. Moreover, in a group of high-risk glaucoma suspects, there was a bimodal distribution of results. Those in whom contrast sensitivity was abnormal (9 of 20) all had abnormalities by Humphrey automated perimeter visualfieldanalysis (Humphrey Allergan, San Leandro, CA) that were insufficient to diagnose a glaucomatous defect. Only 1 of 11 who had normal contrast sensitivity had any indication of a field defect. These results suggested strongly that it might be possible to use such tests to diagnose the defects occurring in "glaucoma suspects" but that measurements had to be made in all retinal quadrants. This testing took a long time. Therefore, it was

2 2780 INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE / September 1991 Vol. 32 desirable to improve the speed of testing and also, if possible, its sensitivity. During this period we observed that several groups of workers and ourselves found that color vision defects frequently occur in early glaucoma and ocular hypertension, 4 " 10 and we therefore modified the color vision system previously described 11 to test specifically for glaucoma. Our results suggest that, in large populations, we might have a method of reliably detecting, not only all patients with glaucoma, but a desired proportion of those patients with lesser defects who cannot be found by perimetry. 12 Materials and Methods Equipment The color vision testing system previously described was used. It consisted of a high-definition "multisync" color monitor, driven by a computer that contained a special graphics card and a clinical software program. In a previous report, we described a color vision test using alphabetic letters. 13 In this study, the image on the screen was a central white fixation spot, a uniform background of 16 cd/m 2, and a color-contrasting annulus concentric with the fixation spot. The annulus had the same luminance as the background. The luminance varied somewhat depending on the subject'sflickermatches. The patients sat at a slit-lamp table with a chin rest, and the monitor screen was placed 45 cm from their corneas. The annulus had a radius of 12.5 in the extramacular field and a width of 1 of arc. The borders of the annulus were not sharp because the image was spatially low-pass filtered by the software. The patient fixated on the center of the screen, and a simple television camera was placed under the monitor to give a view of the subject's eyes. Thus it was easy to see if fixation moved. Procedures The test consisted of the following sequence. First, the subject adjusted a flickering red-green patch on the monitor screen for a minimum sensation of flicker. This was repeated for a flickering blue-green patch. Minimum flicker is obtained when the two lights are equally bright. The actual ratio varies for each observer, and such values were used automatically by the program to calculate the number of colors that lay on the color confusion lines for trichromatic observers and that were precisely equiluminous for the particular subject. After this, the test began. The annulus was displayed so it was equiluminous with the background, and the color difference between it and the background adjusted to threshold. During the test, 45 of the annulus was removed in one of four quadrants: upper right, upper left, lower left, and lower right. The patient had to identify the correct quadrant (Fig. 1). This is a "four-way forced choice," and the threshold color contrast for this determination was obtained with a modified binary search technique as already described.' l Upper and lower bounds were defined and placed on computer stacks; the contrast in the display was midway between these bounds. Depending on whether the patient gave the correct or incorrect answer, the displayed contrast value was "pushed" onto the appropriate stack to become the new bound, and the process was repeated. If the correct or incorrect results were obtained more than twice in succession, the program "popped" the lowest stack level and set a new boundary. This system rapidly corrected errors and converged to a stable value. What was measured was a function of the average color contrast threshold in the entire annular zone of the retinal image of the stimulus. The relationship between the measured threshold and the exact location and area of localized defects will be analyzed elsewhere. For the purposes of this report, it is important that no information about the site of a localized lesion CONTRAST = % RESPONSE - UL B CONTRAST = % RESPONSE=LR CONTRAST = % RESPONSE = LL CONTRAST = % RESPONSE=UL CONTRAST = 15.75% RESPONSE = UR CONTRAST = % RESPONSE = LR Fig. 1. A diagram of the stimulus used. A-F are part of a sequence leading to determination of a threshold.

3 No. 10 A NEW TEST FOR GLAUCOMA SCREENING / Yu er ol 2781 was preserved. The time to determine a threshold was, on average, 45 sec. The maximum total time to complete all formalities and determine protan, deutan, and tritan thresholds for both eyes was less than 15 min. The extreme colors used for the protan axis background were X = and Y = and, for the image, X = and Y = For deutan colors, the values of background and image were, respectively, X = and Y = 0^2988 and X = and Y = For tritan colors, the values of background and image were, respectively, X = and Y = and X = and Y = Q O OUl V) D D I / D Subjects Normal subjects were employees of the hospital and patients' spouses. All had corrected visual acuity of 20/20 or better and no known ocular or systemic disease. In eight men, isolated tripling or quadrupling of the normal protan or deutan threshold was seen. These subjects had congenital color defects assessed by foveal color testing, and their results are not included. 15 No case of possible tritanomaly or tritanopia was encountered. We tested eyes from normal persons (age range, yr). The patients were drawn from a cohort in whom clinical trials of electrophysiologic and psychophysic tests already were being done. These were patients whose clinical and visual functions were well documented. They were compliant and not necessarily representative of the population in general. The patients were classified as having either glaucoma or ocular hypertension using the usual criteria (Table 1). Glaucomatous eyes had a visualfielddefect defined as one or more loci of sensitivity loss of 10 db or greater and one or more adjacent spots of at least 5 db loss. This defect had to be demonstrated on two different occasions. Hypertensive eyes were classified further into three groups: 14 high-risk, medium-risk, and low-risk ocular hypertensives. Table 1. Classification criteria Normal Low risk Medium risk High risk Glaucoma LOP. (mmhg) C/D ratio Visual field >26 >26 Any and and or and <0.6 <0.6 >0.6 >0.6 >0.6 Normal Normal or suspect Normal or suspect Normal or suspect Abnormal Suspect: threshold elevated at 1 or more site ^5 db and < 10 db; Abnormal: threshold elevated at 1 or more site ^ 10 db plus oneflankingspot ^5 db and<10db. 1.._ L FIRST VISIT THRESHOLD ( % ) Fig. 2. Test-retest variability of threshold determination. Units are depth of color-contrast modulation along a color confusion line for dichromatic observers, given in percent of the maximum possible modulation, using the phosphors available. Control Experiments Results The limitations and sensitivity of color vision tests using the video system were described in previous publications, 1113 but because the image we used in this investigation was so different from that in previous articles, a series of check experiments was done. Acceptance: No patient or control did not complete the test. We had little interference with eye movements; the stimulating annulus appeared to the subject to be far from fixation, and the temptation to glance at it seemed to be absent. It was obvious to the patients that the best view was obtained by fixating centrally. All patients preferred this test to automated perimetry. Validation of method: We instructed the patients to "guess where the break is" even when they stated they could not see the ring, and the proportion of correct responses was then above the chance level. Therefore the thresholds obtained were below the level of certainty. An exception occurred when the recorded threshold was 100%. In this case, the stimulus always was invisible. In check experiments, the subjects were instructed always to give the same response or to reply at random. When stable thresholds were obtained, the contrasts were all approximately 100%. Test-retest variability: Twenty-five eyes were tested on two different occasions to measure test-retest variability. These results are shown in Figure 2. The persons tested included normal subjects; low-, medium-, and high-risk patients with ocular hypertension; and patients with glaucoma. Most of the data points in 60

4 2782 INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE / September 1991 Vol. 32 w o _J o X iu 40 cc X H 20 I- a s ill 1 1 I o SUBJECT A PROTAN TRITAN H 1 A,, SUBJECT B PROTAN A TRITAN A H a REFRACTIVE ERROR ( DIOPTERS ) i B i +8 Fig. 3. Effect of refractive error on tritan threshold. Figure 2 clustered around the diagonal line. The mean deviation between test and retest thresholds was only 2.97%. Linear-regression analysis showed that the test and retest thresholds were highly correlated (r = 0.9). This result can be used to calculate the variability of persons whose results are near the "normal" threshold. Thus, if we take the mean plus five standard deviations (SDs) as the upper limit of normal (this corresponds to a tritan threshold of 29.1 % modulation depth), then a patient who had precisely this threshold would, in 95 of 100 trials, give a threshold between %. The former value is considerably greater than mean plus four SDs. Thus the testretest error contributes little to the variability of measurement, and it is unimportant in deciding whether a patient who is suspected of having glaucoma should be considered normal or abnormal. Effect of refractive error: One 60-yr-old normal subject was tested with various corrective lenses, the power ranging from 8-+8 D. These results are shown in Figure 3 as rectangles. There was no significant increase of the protan or tritan threshold for the range tested. The subject's spectacle correction was D, with no astigmatism, and thus the monitor face was in focus with no added correction. The accommodation of this subject was < 0.5 D, and therefore, considerable errors of refraction did not affect the test results (as expected) because the object to be detected occupied about 5 cm on the screen, viewed at a distance of 45 cm. A second young subject was tested after instillation of 1% cyclopentolate to eliminate accommodation. These results are shown in Figure 3 as triangles. Again, there is no significant change of the thresholds with departure from ametropia. Effect of illumination level: Thresholds were determined in four subjects after pupillary dilatation, with and without a 0.6 or 1.2 log unit neutral-density filter placed before the eye. These results are shown in Table 2. The change in retinal illumination affected the tritan thresholds of subjects A (age, 26 yr) and B (age, 32 yr) to a minor extent. Subjects C (age, 60 yr) and D (age, 36 yr) showed that there was an inverse relationship between illumination and threshold. Although the effect of illumination depended on the individual, the threshold approximately doubles for a tenfold change in illumination. In our normal subjects and patients, variation in retinal illumination caused by pupil size and average retinal sensitivity was less than twofold (none of the patients was receiving miotic drugs, and none had extensive or profound scotomas). The maximum threshold in Table 2 is within the upper limit of normal as defined subsequently. The test was done under reduced illumination, and the average pupil diameter was 5 mm. Table 2 shows that the test results were not affected significantly when the pupils were dilated maximally. Therefore, this test could be incorporated into a clinical examination before or after the use of mydriatics. Table 2. Effect of change in retinal illumination on tritan threshold Dilated pupil 0.6 Log unit filter 1.2 Log unit filter Normal pupil Subject A For explanation of units, see Figure 2. Subject B Tritan axis Subject C Subject D

5 No. 10 A NEW TEST FOR GLAUCOMA SCREENING / Yu er ol 2783 In another experiment, thresholds were determined in subject C before during and after repeated miotic applications of 0.25% and 1% pilocarpine. The effect of reduced pupil size on threshold is shown in Figure 4. As the pupil size was reduced, there was an increase in threshold, in keeping with the results in Table 2. Even with a pupil diameter reduced to 1 mm, the threshold was still within the upper limit of normal. Disturbance from equiluminance: The thresholds reported subsequently were obtained with equiluminous color contrast, but because the settings depended on heterochromatic flicker photometry and were therefore to some extent subjective, we investigated the effect of deliberately altering the relative luminosity of the colors on the screen. This was done in two ways. Three subjects deliberately made incorrect settings in the flicker test, and this destroyed equiluminosity. A larger variation could be obtained by altering the look-up tables in the computer software. Finally, thresholds were determined with the ring using a luminance-contrast palette. The results obtained are shown in Table 3. The "errors" deliberately made on the flicker test were judged to be larger than those made by most observers, who have a SD for the setting of only 2-3%. Nevertheless, only trivial changes in threshold occurred. Changing the computer lookup tables altered color rendering to an unacceptable degree and introduced gross luminance changes. These reduced the threshold. If luminance contrast of the annulus was measured with a black-and-white palette under identical conditions, the threshold was still lower. We concluded that any departure from equiluminance encountered in practice causes trivial reductions in threshold and therefore cannot explain the increases seen in disease. Threshold as a function of eccentricity: As expected, color contrast thresholds rose progressively with eccentricity (Fig. 5). The values for foveal viewing were obtained with images of alphabet letters 13 and therefore are not strictly comparable. The change with ecui ECCENTRICITY ( degree ) Fig. 5. Tritan threshold as a function of retinal eccentricity. centricity was not great, and small motions of the patient's head did not cause great changes in threshold. Normal Subjects The relationship between age and color contrast threshold for protan, deuteran, and tritan vision is shown in Figure 6. There was a slight, but statistically insignificant regression of threshold with age in normal eyes. This can be seen more clearly for tritan thresholds in Figure 7. We previously found small changes in color threshold with age for central color vision, using large letters as test objects. 13 Note that the ring thresholds were two to three times higher than those for foveal viewing, 4 but the protan-tritan ratio previously reported was preserved. Using these results, we determined age-related upper limits of normal. The distribution of normal thresholds was approximately Gaussian (Fig. 8). It was possible to give upper limits for the 95, 99.7, or percentiles (Table 4). Patients Glaucomatous patients' tritan values: The results in persons with glaucomatous field defects are shown in Figure 8. The threshold is expressed in logarithmic units. There was considerable spread in the data points, but none of the results overlapped those of the normal population. Many patients had color thresholds approaching 100%. Our equipment was not de- Table 3. Effect of disturbances of retinal image from equiluminance Tritan axis threshold Subject A Subject B Subject C PUPIL SIZE ( mm ) Fig. 4. The effect of reducing pupil size by miotics on tritan threshold. Equiluminance False flicker test B. palette changed G. palette changed Luminance contrast For explanation of units, see Figure

6 27 INVESTIGATIVE OPHTHALMOLOGY & VISUAL SCIENCE / September 1991 Vol O 20 O cc z 10 < u E 5 REGRESSION ANALYSIS P D T Const < 40 YEARS YEARS > 59 YEARS Fig. 6. The mean color contrast thresholds in normal subjects for three dif- PROTAN ferent age groups (<40 years, years, and >59 DEUTAN years), and for three color axes (protan, deutan, and I TRITAN tritan axes). The lines indicate one standard deviation above the mean. Regression analysis was performed for each of the three color axes. Threshold = (coefft.) x (age in years) + constant. These results are given in the inset to the Figure. For all three groups the Pearson correlation coefficient (r) between threshold and age is very low (between 0.17 and 0.22), and the slope of the regression line is not significant. signed to measure such high thresholds, and it is possible that, for these persons, there were some luminance clues in the target. The reason for the high thresholds was obvious. By definition, these patients had losses of sensitivity to luminance of at least 1 logarithmic unit (10 db on the Humphrey scale) in at least part of the region where we tested color contrast; this was a more difficult judgment to make than that posed by the Humphrey perimeter. If the large break in the ring lay on a normal region of retina, and in the other quadrants contrast sensitivity was reduced, the patient will perceive two, or more, breaks in the ring and will not be able to respond correctly. Age effects: There was no correlation between threshold and age (data not shown). Severity of glaucoma and tritan scores: Following the suggestion of the referee, we graded the severity of glaucoma by using the mean deviation scores in the Humphrey statistics package. We plotted these results 40 SO AGE(YEARS) Fig. 7. Tritan thresholds for all control subjects (N = ) as a function of age. Again, it is shown that there is no clear relation with age. against tritan thresholds. For administrative reasons, we were only able to identify 48 patients' records, but they varied from the minimum scotomas to severe field losses. These results are shown in Figure 9. There was a tendency to have increasing threshold with more severe field defects. The minimum of the ordinate was two standard deviations above the normal mean. The dashed line was the least-squares linear regression. The correlation coefficient, r, was The slope of the line was significant. Inspection of the field test results showed that the position and the degree of localization of the glaucomatousfieldloss varied widely between persons with equal mean deviation scores. This could explain why some patients with small mean deviations cannot see the ring at all, and why the correlation coefficient was not higher. Nine of 48 cases had five points or fewer with an elevation of threshold of 10 db. The mean tritan threshold for these was 57.5% (SD, 23.5%). Thus even minimum scotomas are associated with elevated tritan thresholds. Patients with ocular hypertension: These patients were separated into three groups, ie, those judged to be at high, medium, or low risk of developing a scotoma in the next 5 years as described. 14 There were 77 at high risk, 50 at medium risk, and 30 at low risk. Their tritan threshold results are illustrated in Figure 8. Although low- and medium-risk patients included a few who had high values, for high-risk patients, over 50% of the thresholds were higher than 24%, ie, more than three SDs above the normal mean. Moreover, the results from medium- and high-risk patients were skewed. It appears that there are two populations, and the one with the lower values approximated to normal. If the criterion of abnormality used for glaucoma

7 No. 10 A NEW TEST FOR GLAUCOMA SCREENING / Yu er ol 2785 mean. Averaging these normalized values reduced interindividual variability. The patients' results then were divided by the same normal means, and the transformed data were analyzed. Because some patients had not been tested in all three color axes, the data base was reduced. The relationship between normal subjects, glaucomatous patients, and those with ocular hypertension are shown in Figures 10 and 11. There was an increase in the tritan-protan ratio in glaucomatous eyes, but this was not as marked as it was for foveal viewing. 13 There was little difference between the results for tritan measurements only and the results obtained by averaging results in all three color axes (Fig. 11 compared with Fig. 8). Discussion Measurement of equiluminous color contrast along color confusion axes is a sensitive test of visual func- Table 4. Upper limits of normal corresponding to various criteria Thresholds P D T log threshold modulation Fig. 8. Tritan contrast thresholds for the five different groups: control group; low-, medium-, and high-risk OHT groups, and the glaucoma group. There is no overlap between the values obtained for the control group and those of the glaucoma population. (threshold > 26%) is used for the high-risk group, about 20% would be presumed abnormal. The comparison between controls and different groups of patients including glaucomatous patients is summarized in Table 5. Other Analyses Protan-tritan ratios: Tritan threshold elevation occurs earliest in many acquired conditions, and therefore these ratios were used as an index. These are illustrated for normal subjects, glaucomatous patients, and patients with ocular hypertension (Fig. 10). Combining results for color testing in three axes: Some patients with glaucoma and ocular hypertension had obvious defects in all color axes, and therefore the results were analyzed to determine whether information was lost by confining the test to tritan colors. The mean normal threshold values for protan, deuteran, and tritan colors were calculated, and each individual result was expressed as a fraction of the Age below 40 years Mean Criterion: Mean + (#ofsds) Age Mean Criterion: Mean + (#ofsds) Age 60 or above Mean Criterion: Mean + (#ofsds) P, protan; D, deutan; T, tritan

8 2786 INVESTIGATIVE OPHTHALMOLOGY 6 VISUAL SCIENCE / September 1991 Vol r Fig. 9. Tritan threshold contrast plotted against the Humphrey mean deviation score for 48 glaucoma patients' eyes, for which statistics were available. The equation of the dotted line, which is the least squares regression, is Tritan threshold% = (mean deviation) The standard error of the slope is MEAN DEVIATION IN LUMINANCE THRESHOLD (db) tion, and it should not be confused with the use of colored targets in perimetry Such tests measure luminance increment threshold for selected wave bands, and at best, identify particular retinal mechanisms such as Stiles mechanisms Although the use of color contrast was pioneered early, 20 clinical studies using color contrast had to wait for the development of video technology, but they are beginning to be valuable. In this report, we additionally modified the image in our system to determine peripheral Table 5. Peripheral color contrast thresholds Control group Number Mean ± SD Maximum* Glaucoma Number Mean ± SD Minimum* Pi Low-risk ocular hypertensives Number Mean ± SD Pi Medium-risk ocular hypertensives Number Mean ± SD Pi High-risk ocular hypertensives Number Mean ± SD P* Age 36 ± ± ± ± 11 Protan ± ± <? ±3.74 < ± ± 9.94 Deutan ± ± <« ±3.15 < rb ± Tritan ± ± « ±4.24 < ± * Maximum values for-normals to be compared to minimum for glaucoma. * t-test of normal against each class of patient: figure is the probability that the result could be obtained by chance. retinal color contrast sensitivity, not previously investigated in ocular disease to our knowledge. A novel feature of the test was that, although the function of a perimetric region of interest was investigated, the method did not specify where the loss of sensitivity might be. The result depended on the development of local regions of malfunction, but several regions of similar eccentricity were tested at the same time, and the final result "throws away" the information about the precise retinal locus or loci of malfunction. For this reason, the speed of determining threshold was increased. Additionally, the specificity and sensitivity were enhanced relative to a color contrast test that requires foveal viewing. 21 The relationship between tritan thresholds and other measures of the severity of glaucomatous defects (such as the area of the scotoma or degree of cupping or nervefiberlayer loss) requires further investigation. Epidemiologic Consequences The test we described was capable of discriminating between glaucomatous patients and normal subjects with great reliability and sensitivity. The numbers of patients tested was sufficient to extrapolate how the test might be used in screening. We know several general retinal conditions, apart from ocular hypertension, in which color contrast sensitivity, measured with large letters or rings, is reduced significantly; these include diabetic and age-related maculopathy. However, minimum lens opacities do not appear to raise the thresholds greatly. It is probable that patients with other serious retinal diseases would also be detected by a color contrast test in any glaucoma screening program, but because we have no precise data on this, this consideration was omitted from the following discussion.

9 No. 10 A NEW TEST FOR GLAUCOMA SCREENING / Yu er ol 2787 Fig. 10. Ratio of tritan:protan thresholds in normals, and in patients with ocular hypertension and glaucoma. control low risk medium risk high risk glaucoma The prevalence of glaucoma and ocular hypertension generally are considered to be 1% and 10%, respectively, in persons over 40 yr of age. 22 " 25 Among patients with ocular hypertension, the overall conversion rate to glaucoma is 4.3%. This figure was calculated as the mean of the conversion rates in nine different surveys 26 " 34 in which the mean follow-up was 8 yr. Therefore, most of these people will not have field defects, and it is important to identify those who will. In our sample, patients with ocular hypertension were classified into three groups. In a 5-yr follow-up, the conversion rate for high-risk patients was 100%; for medium-risk patients, it was 42%; and for low-risk patients, it was 2%. 14 The proportion of high to medium to low-risk patients with ocular hypertension in the general population was taken as 5:15:80 to harmonize the overall conversion rate with the conversion rates in these different subgroups of patients. If we assume that the total population older than 40 yr of age to be screened is 1,000,000, the number of patients with glaucoma will be 10,000, and the number of patients with ocular hypertension will be 100,000. Thus the population contains 89% normal subjects, 1% glaucomatous patients, and 0.5% high-risk, 1.5% medium-risk, and 8% low-risk patients with ocular hypertension. The actual numbers are given in Table 6. We can define the upper limit of normal by reference to the mean and the SD and inquire how many of normal subjects, glaucomatous patients, and patients with ocular hypertension of each class will fail the test. These data are shown in Table 6 for the tritan axis only, using a probability table 35 of normal distribution. The numbers for each patient group were extrapolated from our results. For example, if the mean plus two SDs (5% confidence interval) is considered as the upper limit of normal, 5% of all the normal eyes (890,000 X 0.05 = 42,500) will fail the test. All glaucomatous eyes (10,000) will be detected. In addition, 3450 (69%) of high-risk, 9300 (62%) of medium-risk, and 18,400 (23%) of low-risk patients with ocular hypertension will be diagnosed. Such a result is valuable because all eyes with field defects can be detected easily, but large numbers of low-risk patients without field defects would need to undergo further medical examination, and the incidence of subsequent field «'o c <D 3 O* <D l I control glaucoma threshold / normal mean (log scale ) Fig. 11. Normalized threshold values, including results for all color axes, in normals, and in patients with ocular hypertension and glaucoma.

10 2788 INVESTIGATIVE OPHTHALMOLOGY b VISUAL SCIENCE / September 1991 Vol. 32 Table 6. Number of people showing abnormal values for color test at various discrimination levels from a population of 1,000,000 aged over 40 No. ofpeople failing test Criterion value Tritan threshold Normals (890,000)* Glaucoma (10,000) H/RISK (5000) M/RISK (15,000) L/RISK (80,000) Mean + 2 SDs ,482 11, ,000 10,000 10,000 10,000 10, , Predictions refer to a general population of 1,000,000 above 40 years old, assuming 1% prevalence for glaucoma and 10% prevalance for OH, and a conversion rate of 4.3%, as explained in the text. * Numbers in parentheses are population numbers. defects would be small. Therefore such a criterion is not practicable in most centers. However, if the mean plus 3.4 SDs is taken as the upper limit of normal, the results are different. Only 562 false positives would be detected (from a population of 1 million), and all 10,000 glaucomatous patients will fail the test. In addition, 2250 high-risk, 4200 medium-risk, and no low-risk patients will fail. Table 6 shows how these results vary with the criterion of abnormality used. The consequences for our hospital can be examined. There are approximately 6000 new referrals of suspected glaucoma seen yearly. Of these, 600 have a field defect, and 1080 have moderate- or high-risk ocular hypertension. Currently, all 5400 suspects require repeated field testing and clinical examination. If this were replaced by a single color vision test, the savings in time would be great and the cost, reduced. The outcome, indicated by our results, would be that, if a 3.4 SDs cutoff were used, all glaucomatous patients and 387 patients with ocular hypertension, all from high- and medium-risk groups, would be detected, with no low-risk or normal subjects. These indications suggest that the selected patients with ocular hypertension would include almost all those with retinal damage who require additional examination. Until a longitudinal study is done, we cannot tell if patients with ocular hypertension with the highest thresholds have more severe retinal damage than those with less elevated thresholds, but because there is a correlation between losses of color contrast sensitivity and pattern electroretinography (manuscript in preparation), this is, at least, likely. Furthermore, the results in Figure 9 suggest that the worse the glaucomatous defect, the higher the color threshold. By extrapolation, in patients with ocular hypertension and no field defects, the presence of a high threshold may indicate glaucomatous damage. Our results obtained to date cannot provide information on two important additional points. We cannot tell the discrimination obtainable in other patient populations, including a random population of persons with additional eye diseases and systemic disease. Our patients, all of whom know they are in a clinical trial, cannot be taken as representative of patient populations as a whole. Thus, our results must be supplemented with a larger randomized trial. However it would appear that this new method of screening is superior to perimetry in the following respects: the equipment is less expensive, the test is faster, it is easier to administer and can be done by relatively untrained persons, damage can be diagnosed reliably at an earlier stage in the natural history of the disease, the test result is a single number, complex data analysis is not required, and patients prefer it. For these reasons, it may be possible to provide repeated examinations for large populations on a cost-effective basis, and to screen reliably for glaucomatous damage of a lesser degree than that which produces a minimum field defect. Key words: glaucoma, ocular hypertension, peripheral, color contrast sensitivity, glaucoma screening Acknowledgments The authors thank their colleagues, particularly Mr. R. A. Hitchings and Miss F. O'Sullivan, who supplied information about and sent their patients to the Electrodiagnostic Department for testing and the staff of the Electrodiagnostic Department for their collaboration and help during this study. References 1. Arden GB and Jacobson J: A simple grating test for contrast sensitivity: Preliminary results indicate value for screening glaucoma. Invest Ophthalmol Vis Sci 17:23, Atkin A, Bodis-Wollner I, Wolkstein M, Moss A, and Podos SM: Abnormalities of central contrast sensitivities in glaucoma. Am J Ophthalmol 88:205, Falcao-Reis F, O'Donoghue E, Buceti R, Hitchings R, and Ar-

11 No. 10 A NEW TEST FOR GLAUCOMA SCREENING / Yu er ol 2789 den GB: Peripheral contrast sensitivity in glaucoma and ocular hypertension. Br J Ophthalmol 74:712, Gunduz K, Arden GB, Perry S, Weinstein G, and Hitchings R: Color vision defects in ocular hypertension and glaucoma: Quantification with a computer-driven color television system. Arch Ophthalmol 108:929, Lakowski R and Drance S: Acquired dyschromatopsia: The earliest functional loss in glaucoma. Doc Ophthalmol Proc Ser 19:159, Logan N and Anderson DR: Detecting early glaucomatous vivsual field changes with a blue stimulus. Am J Ophthalmol 95:432, Hamil TR, Post RB, Johnson CA, Keltner J: Correlation of color vision deficits and observable changes in the optic disc in a population of ocular hypertensives. Arch Ophthalmol 102:1637, Flammer J and Drance S: Correlation between color vision scores and quantitative perimetry in suspected glaucoma. Arch Ophthalmol 102:38, Hart WM: Acquired dyschromatopsia. Surv Ophthalmol 32:10, Drance SM, Lakowski R, Schulzer M, and Douglas GR: Acquired color vision changes in glaucoma. Arch Ophthalmol 99:829, Arden GB, Gunduz K, and Perry S: Color vision testing with a computer graphics system. Clin Vision Sci 2:303, Dyster-Aas K, Heijl A, and Lvnoqvist L: Computerised visual field screening in the management of patients with ocular hypertension. Acta Ophthalmol (Copenh) 58:918, Arden GB, Berainger TA, Hogg CR, Perry S, and Lund E: A survey of color discrimination in German ophthalmologists: Changes associated with the use of lasers and operating microscope. In Proceedings of the Advisory Group for Aerospace Research and Development of the North Atlantic Treaty Organization. NASA Scientific and Technical Information Division, Vol 70, Yablonsky M, Zimmerman T, Kass M, and Becker B: Prognostic significance of optic disk cupping in ocular hypertensive patients. Am J Ophthalmol 89:585, McKusick VA: Mendelian Inheritance in Man: Catalogs of Autosomal Dominant, Autosomal Recessive and X-linked Phenotypes. Baltimore, The John Hopkins University Press, 1986, pp Abe H, Sakai T, and Yamazaky T: Selective impairment of the color mechanism isolated by new color campimeter. Acta Soc Ophthalmol 87:950, Hard WM and Ronald MB: Color contrast perimetry: The spatial distribution of color defects in optic nerve and retinal diseases. Ophthalmology 92:768, Stiles WS: Increment thresholds and the mechanism of color vision. DocOpthalmol 3:138, Stiles WS: Color vision: The approach through increment threshold sensitivity. Proc Natl Acad Sci U S A 45:100, MacAdam DL: Specification of small chromaticity differences. J Opt Soc Am [A] 33:18, Falcao-Reis F, O'Sullivan F, Spileers W, Hogg C, and Arden GB: Macular colour-contrast sensitivity in ocular hypertension and glaucoma: Evidence for 2 types of defect. Br J Ophthalmol (in press). 22. Graham PA and Hollows FC: A critical review of methods of detecting glaucoma. In Glaucoma: Epidemiology, Early Diagnosis and Some Aspects of Treatment, Hunt LB, editor. Edinburgh, Churchill Livingstone, Bankes JLK, Perkins ES, Tsolakis S, and Wright JE: Bedford glaucoma survey. Br Med J 1:791, Stromberg U: Ocular hypertension. Acta Ophthalmol (Copenh) 37(Suppl): 191, Armaly MF: On the distribution of applanation pressure and arcuate scotoma. In Drug Mechanisms in Glaucoma, Patterson G, Miller SJH, and Patterson GD, editors. Boston, Little, Brown, Sorensen PN, Nielsen NV, and Norskov K: Ocular hypertension: A 15-year follow-up. Acta Ophthalmol (Copenh) 56:363, Kitazawa Y, Aoki S, Suzuki M, and Nishioka K: Untreated ocular hypertension: A long-term prospective study. Arch Ophthalmol 95:1180, Linner E and Stromberg U: Ocular hypertension: Afiveyear study of the total population in a Swedish town, Skovde. In Glaucoma Tutzing Symposium, Leydhecker W, editor. Basel, S. Karger, Graham PA: The definition of pre-glaucoma: A prospective study. Trans Ophthalmol Soc UK 88:153, Armaly MF: Ocular pressure and visualfields:a ten-year follow-up study. Arch Ophthalmol 81:25, Wilensky JT, Podos SM, and Becker B: Prognostic indicators in ocular hypertension. Arch Ophthalmol 91:200, Perkins ES: The Bedford glaucoma survey: I. Long-term follow-up of borderline cases. Br J Ophthalmol 57:179, Norskov K: Routine tonometry in ophthalmic practice: II. Five-year follow-up. Acta Ophthalmol (Copenh) 48:873, Walker WM: Ocular hypertension: Follow-up of 109 cases from 1963 to Trans Ophthalmol Soc UK 94:525, Hodgman CD: Handbook of Chemistry and Physics, 32nd ed. Cleveland, OH, Chemical Rubber Company, 1955.