- Email: [email protected]
The Sensitivity and Specificity of Nerve Fiber Layer Measurements in Glaucoma as Determined With Scanning Laser Polarimetry
The Sensitivity and Specificity of Nerve Fiber Layer Measurements in Glaucoma as Determined With Scanning Laser Polarimetry MARTHA J. TJON-FO-SANG, MD, AND HANS G. LEMIJ, MD, PHD
• PURPOSE: To determine the sensitivity and spec ificity for detecting glaucoma by scanning laser polarimetry and to assess the relation between nerve fiber layer (NFL) measurements and visual field indices. • METHODS: The peripapillary NFL was divided into four segments: superior, inferior, temporal, and nasal. The mean polarimetric NFL for each segment was calculated out of six selected areas of 2 5 6 pixels each. Ratios relative to the nasal seg ment were determined for the superior and inferior segments. With the use of previously obtained normograms for polarimetric NFL readings, the sensitivity of scanning laser polarimetry was as sessed in 200 glaucomatous eyes (155 subjects). The specificity was assessed in a normal popula tion of 150 eyes (150 subjects). The relation between hemifield polarimetric NFL and visual field indices was assessed by linear regression analysis. • RESULTS: The sensitivity of scanning laser polarimetry was 96% and the specificity was 93%. The correlation between NFL parameters and visual field indices ranged from —0.18 to + 0 . 2 6 .
Accepted for publication Aug 6, 1996. From the Glaucoma Service, Rotterdam Eye Hospital, Rotterdam, The Netherlands. Reprint requests to Martha J. Tjon-Fo-Sang, MD, Rotterdam Eye Hospital, PO Box 70030, 3000 LM Rotterdam, The Netherlands; fax: 31 10-401 76 55.
62
The amount of variation by the linear regression model ranged from 3% to 6%. • CONCLUSIONS: Although quantitative mea surements of the NFL with scanning laser polar imetry relate poorly to visual field indices, the technique seems to be promising for screening populations for glaucoma. Whether measurements of the NFL with scanning laser polarimetry are also sensitive enough to detect change over time requires further study.
S
CANNING LASER POLARIMETRY (THE NERVE FIBER
Analyzer I, Laser Diagnostic Technologies, Inc, San Diego, California) can be used to quantita tively assess the peripapillary retinal nerve fiber layer (NFL) in vivo. The technique is based on the specific arrangement of the microtubules in the retinal NFL.1,2 In scanning laser polarimetry, the phase shift of the reflected light beam (equal to the retardation) is linearly correlated with the NFL thickness.1,2 Scan ning laser polarimetry has potential value in the detection and follow-up in glaucoma because it pro vides an objective and fast measurement of the NFL with good reproducibility of measurements. The coef ficient of variation for repeated measurements with scanning laser polarimetry has been reported to be 3.6% to 5.2% in normal subjects (references 3 and 4 and Tjon-Fo-Sang MJ, de Vries J, Lemij HG, unpub lished data). In a previous study,5 we presented normograms for NFL retardation in white subjects. In these normograms, we showed a gradual decrease in NFL retardation with age.5 This decrease with age has
© AMERICAN JOURNAL OF OPHTHALMOLOGY i23,-62-69
JANUARY 1997
also been observed by other authors. 4 In addition, significantly lower retardation in patients with glauco ma compared with normal subjects has been recog nized, although there was also considerable overlap.3,4 Weinreb and associates6 reported an only moderate association between NFL retardation and visual field indices; they concluded, however, that scanning laser polarimetry seemed promising for the detection and follow-up of glaucoma, although sensitivity and speci ficity measures were not assessed. Therefore, we have determined the sensitivity and specificity of scanning laser polarimetry as a diagnostic tool in a large population. We have also assessed the relation be tween NFL retardation and visual field indices.
PATIENTS AND METHODS FROM OUR OUTPATIENT GLAUCOMA SERVICE, 155 PA-
tients with glaucoma (200 eyes) were recruited con secutively. Patients were included only when they had multiple intraocular pressure readings over 21 mm Hg, no secondary glaucoma, no diabetes or other chorioretinal disorders, no intraocular lens implants, a pupil diameter greater than 2 mm, and typical glaucomatous defects on the visual field test. All patients were subjected to a Humphrey Field Analyzer (HFA) central 30-2 threshold test. To qualify patients for further analysis, the HFA central 30-2 test should satisfy the standard reliability criteria for patients with glaucoma7,8: less than 20% fixation losses and less than 33% false positive errors. In addition, the Glaucoma Hemifield Test (GHT) 9 should score pa tients on the visual field test as "outside normal limits." Typical glaucomatous defects were defined as arcuate scotomas or hemifield defects. When a patient scored on the G H T visual field as "within normal limits," "borderline," or "general reduction of sensi tivity," the eyes were not included in this study. All polarimetric recordings were obtained under the following conditions: the subject's head was placed as upright as possible in the chin rest, pupils were left undilated, and ambient lights were left on. During image acquisition and analysis of the data, the investigator had been masked for the corresponding visual fields but not for the patient diagnosis. For data
VOL.123,
No. 1
analysis, we applied the squares calculation method (reference 5 and Tjon-Fo-Sang MJ, de Vries J, Lemij HG, unpublished data). In brief, the retardation map was divided into 16 X 16 squares of 256 pixels each. A circle was placed at the optic disk's margin, and the retardation map was divided in a superior and an inferior segment of 120 degrees each, a nasal segment of 50 degrees, and a temporal segment of 70 degrees. Subsequently, we averaged NFL retardation in six squares for each superior, inferior, and nasal segment. Selected by the investigator, the squares were outside a peripapillary circle of 1.75 disk diameters and contained no blood vessels because blood vessels may adversely affect the reproducibility of the recordings.3 Ratios for the superior to nasal (superior NFL), inferior to nasal (inferior NFL), and superior to inferior (S/I) NFL retardation were then computed. Because the superior and inferior NFL ratios re spectively represent the superior and inferior arcuate bundles, we determined the mean deviation of the HFA central 30-2 test separately for the superior and inferior hemifields. The hemifield mean deviations were calculated for the nonedge points only, except for the most nasal points (x = —27 degrees, y = 3 degrees, and x = — 27 degrees, y = — 3 degrees for a right eye field), which were included. Linear regres sion analysis was then applied to assess the correlation between superior NFL and mean deviation of the lower hemifield. In addition, R2, as explained in the linear regression model, was calculated. The same procedure was repeated for the relation between inferior NFL and mean deviation of the upper hemifield. The GHT was developed to assess hemifield (updown) differences in the Statpac probability maps9'12 to better differentiate localized from diffuse loss caused by, for example, media opacities or miotic therapy. Based on the probability symbols in the pattern deviation map of an HFA central 30-2 test, the G H T score is calculated for each hemifield. To differentiate localized from diffuse loss, the up-down difference of the hemifields is determined. The dis criminative capability of the G H T as a method of analysis has a sensitivity of 92% to 94% and a specificity of 90%.13,14 With linear regression analysis, the correlation between the superior NFL and inferi or G H T score was determined. A similar correlation
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
63
was determined between the inferior NFL and superi or G H T score. Nerve fiber layer measurements determined with scanning laser polarimetry4,5 exhibit great intersubject variability and a gradual decrease of NFL retardation with age. Both these factors possibly affect the corre lation between the visual field indices (mean devia tion and G H T score) and the NFL retardation measurements. Therefore, to account for the age effect, we normalized the NFL ratios by dividing a patient's NFL retardation by the mean normal retar dation for his or her age cohort (per decade), ex pressed as a percentage. Linear regression analysis was thereafter repeated for determination of the relation between normalized NFL retardation and visual field indices. Only when all the retardation data in the glaucomatous and normal populations had been established did we determine the second and fifth percentiles for superior and inferior NFLs for each decade, begin ning at age 20 years, in our normograms for NFL retardation.6 We also assessed the 2.5 and 97.5 percentiles for the S/I ratio. These cutoff points were subsequently used to determine the sensitivity and specificity of scanning laser polarimetry in 155 pa tients with glaucoma (200 eyes) and 150 normal subjects (150 eyes). T h e normal eyes had not been included in the determination of the normograms. All normal subjects had intraocular pressures of 21 mm Hg or less, had no evidence of disorders of the optic disk or retina, and had completed an HFA 120-point full-field suprathreshold screening test, ful filling its reliability and normality criteria.15,16 For determination of mean NFL retardation in patients with glaucoma and normal subjects, only the eye that satisfied our inclusion criteria for each patient in the glaucomatous population was included. When both eyes were eligible, one eye was randomly selected for inclusion. As many patients with glaucoma as possible (100) were randomly matched with normal subjects for race, sex, age, and included eye. For the statistical analysis of NFL retardation values in pa tients with glaucoma and normal subjects, Student's t test was applied. A P value below .05 was considered statistically significant. Only one eye of each subject was included for this part of the study to avoid intereye correlation bias in the Student's t test.
64
frequency (N)
Mean Defect (dB)
Figure 1. Histogram of the distribution of the mean defect (mean deviation in dB) in the visual fields of 200 glaucomatous eyes, determined with the Humphrey Field Analyzer Central 30-2 test.
RESULTS WE MEASURED NFL RETARDATION IN 95 LEFT AND 105
right eyes of 155 patients with glaucoma (87 men, 68 women; 152 white, three black). The mean age of the glaucomatous patients was 67.1 years (total range, 26 to 92 years). The mean deviation of the HFA 30-2 tests, as noted on the visual fields, was —10.33 dB (range, —31.5 to 0.76 dB). Therefore, patients with early defects as well as patients with severe visual field loss were included in this study. The distribution of the mean deviations in this glaucoma population has been illustrated in Figure 1. With linear regression analysis, the correlation between the superior NFL and mean deviation of the lower hemifield was only 0.26 (R2 = 0.07; see Figure 2, top). The correlation between the inferior NFL and mean deviation of the upper hemifield was identical: 0.26 (R2 = 0.07) (Figure 2, bottom). The correlation between the superior NFL and the G H T score of the lower hemifield was only —0.24 (R2 = 0.06), but between the inferior NFL and the G H T score of the upper hemifield, it was even less: - 0 . 1 8 (R2 = 0.03). Normalization of the NFL retardation parameters did not improve these correlations. Of 200 glaucomatous eyes, 113 (56.5%) had an S/I ratio below the 2.5 percentile or over the 97.5 percentile. Of the remaining eyes, 81 had a superior or inferior NFL below the fifth percentile, of which even 79 (39.5%) had a superior or inferior NFL below the
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
3
superior
NFL R = 0.26
*
2.5 2
-
1 S
.
•
*.• •
: ••
•
•
• ^-^s^^^s^-' • • -S%~T~ -i>' >"\ • . ••» u . .
a*
••
1 0.5
-30
-10
0 Mean Defect (dB)
inferior NFL R=0.26
-40
-30
-20
-10
0
10
Mean Defect (dB)
Figure 2. (Top) The relation between superior nerve fiber layer (NFL) retardation and the mean defect (dB) of the inferior hemifield. Linear regression analysis yielded a correlation of +0.26; R2 = 0.07. (Bottom) The relation between inferior nerve fiber layer retarda tion and the mean defect (dB) of the superior hemifield (correlation +0.26; R2 = 0.07).
tive outcome of the test) is, however, only 12%. By contrast, the negative predictive value in a normal polarimetric reading is 100%. Because of the size of our normal database, only 100 patients (44 men, 56 women; 97 white, three black) of the 155 could be matched with normal subjects. The mean age of this glaucomatous subpopulation was 65 years (total range, 26 to 92 years) and, in the matched normal population, was 64 years (total range, 28 to 88 years). The mean deviation of the HFA 30-2 tests in the glaucomatous subpopulation was -10.27 dB (total range, - 3 1 . 5 to +0.76 dB). The mean superior NFL in patients with glauco ma was 1.4 (95% range, 0.8 to 2.0), whereas the mean superior NFL in matched normal subjects was statistically significantly higher: 2.2 (95% range, 1.1 to 3.3; P = .0004) (Figure 3, top). The mean inferior NFL in patients with glaucoma was identical to the superior NFL (1.4), but with a slightly smaller 95% range of 0.9 to 1.9 (Figure 3, bottom). The mean inferior NFL in normal subjects was again statistically significantly higher: 2.2 (95% range, 1.1 to 3.3; P = .0004). For patients with glaucoma, the mean S/I was 1.05 (95% range, 0.71 to 1.39), whereas for normal subjects, this was 1.01, with, however, a much smaller 95% range of 0.89 to 1.13.
DISCUSSION IN THIS STUDY, WE FOUND A SENSITIVITY OF 96% AND A
second percentile. Six glaucomatous eyes had NFL readings within normal limits. Therefore, when the cutoff point was set at an abnormal S/I or when the superior or inferior NFL was below the second percentile, the sensitivity of scanning laser polarimetry was 96%. Of 150 normal eyes (56 white male and 94 white female subjects; mean age, 49 years [total range, 21 to 88 years]), four (2.7%) had abnormal S/I ratios and 13 had a superior or inferior NFL below the fifth percentile. Seven (4-7%) of these were even below the second percentile. The specificity of scan ning laser polarimetry was 93%. Based on Bayes' theorem of conditional probabilities17 and a preva lence rate of glaucoma of 1% in the mass population more than 40 years old,18'20 the predictive value for glaucoma in an abnormal polarimetric reading (posi
Vot. 123, No. 1
specificity of 93% for the detection of glaucoma using scanning laser polarimetry. These numbers are better than those reported for tonometry, optic disk evalua tion, NFL red-free fundus photographs, the GHT, and even NFL height measurements (Appendix). In previous studies, doubts had been cast on the poten tial of scanning laser polarimetry as a screening instrument, mainly because of the great variability in NFL retardation found between normal subjects.3,5 However, the sensitivity reported here indicates that glaucomatous NFL loss exceeds age-related NFL loss to an extent readily detectable. Although the sensitiv ity and specificity for scanning laser polarimetry were very good, they might have been overstated because these numbers were based on glaucoma service cases rather than on the general population. Moreover, the
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
65
theorem is applied for the G H T with a sensitivity of 94% and specificity of 90%, the positive predictive value is only 9%. Both positive predictive values are poor because of the low prevalence rate of glaucoma in the general population. Therefore, the use of scanning laser polarimetry as a single screening procedure for glaucoma should be reserved for highly selected settings with a higher prevalence rate of glaucoma—for example, patients of older age, black subjects, patients with a family history of glaucoma, and patients with diabetes or other vascular diseases.21 The Public Health Service of the United States has suggested that any medical screening program, to be considered a cost-effective measure, must detect at least a 2% pathology in the general population.22 However, when polarimetry is combined with subse quent static perimetry (GHT), the positive predictive value increases to 53%.
superior NFL
inferior NFL
Figure 3. The mean superior nerve fiber layer (NFL) retardation in 100 patients with glaucoma (A) and 100 matched normal subjects (O). (Top) Superior nerve fiber layer retardation was statistically significantly higher in normal subjects (2.2; 95% range, 1.1 to 3.3) compared with glaucomatous patients (1.4; 95% range, 0.8 to 2.0). (Bottom) Inferior nerve fiber layer retardation. The mean inferior nerve fiber layer retardation was also significantly higher in normal subjects (2.2; 95% range, 1.1 to 3.3) compared with glaucomatous patients (1.4; 95% range, 0.9 to 1.9).
predictive value of a positive predictive value test was quite disappointing: 12%. However, this was still better than the equivalent for the GHT: when Bayes'
66
What does this process of serial testing mean in practice? Suppose we were to screen a population of 100,000 adults with an estimated 1% glaucoma prevalence. For serial testing, any screening procedure is most efficient when the test with the highest sensitivity and specificity is used first.17 Hence, by starting with polarimetry, 960 of the 100,000 patients (sensitivity, 96%) would be correctly diagnosed. Forty patients would be missed. With a specificity of 93%, 6,930 normal subjects would have abnormal polarimetric readings, leading to wrong referral. Followed by static perimetry (GHT), however, only an addi tional 1,595 of the 100,000 adults would be referred and another 58 glaucomatous patients would be missed. Therefore, the low rate of referral occurs at the expense of a total of 98 (9.8%) missed glaucoma tous cases. If scanning laser polarimetry were not available, the initial screening for glaucoma with visual field testing would result in wrong referrals of 9.9% (3% more than initial screening with polarime try). Moreover, the acquisition of polarimetric read ings is much faster3 (including patient instructions, about 10 minutes) and objective, whereas visual fields require longer testing and are subjective. Moreover, in a study assessing various analytic strategies, almost 30% of the subjects failed to meet the reliability criteria for the HFA central 30-2 test.14 Hence, scanning laser polarimetry might be useful as the first test when the general population is screened for
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
glaucoma; automated perimetry may function as a second, additive test. One limitation of our method of data analysis is that it requires an operator to select the squares to represent each segment. This might introduce great interoperator variability. However, this method was necessary for us to exclude the areas with blood vessels and to correct for the lack of absolute calibra tion in the Nerve Fiber Analyzer I. In its newest version, the Nerve Fiber Analyzer II, the calibration problem has been dealt with; and, with the newest software, it will be possible to assess areas without blood vessels in an automated manner. Scanning laser polarimetry has a higher sensitivity than the G H T does, that is, 96% compared with 94% (Appendix). This may concur with our previous finding that 58% of subjects with ocular hypertension had an abnormal polarimetric reading.5 Our data therefore suggested that high intraocular pressure had caused early NFL loss before visual field defects had become apparent. Nerve fiber layer or optic disk abnormalities preceding visual field defects were in deed reported by Quigley and associates,23 who found substantial loss of axons in patients with ocular hypertension. Whether these findings suggest that current definitions of glaucoma and ocular hyperten sion should be modified and whether ocular hyper tensive patients with abnormal polarimetric readings should be treated as glaucomatous patients are ques tions yet to be answered. As did Weinreb and associates,6 we found a poor correlation between NFL retardation and visual field indices. The amount of variation explained by the linear regression models, however, was somewhat better in their study. Our variation may have been adversely affected by the great variability in NFL retardation at a mean defect of about 0 dB, or a G H T score of zero. Quigley and associates24 assessed the relation between decibel sensitivity loss on perimetry and the percent of normal ganglion cells at precisely measured locations. They concluded that a 5-dB loss required 20% to 50% ganglion cell death. However, in locations with a 0-dB sensitivity, 3 % to 36% of normal density of ganglion cells was still found.24 Although these cells might be physically present, their function in terms of retinal sensitivity on perimetry is absent.24 Therefore, the relation between
VOL.123, No. 1
structure and function of the retinal nerve fibers— that is, between optic disk or NFL properties and visual field indices—is not yet clear-cut. Scanning laser polarimetry would be of great clinical potential for the follow-up of patients if it were capable of detecting change over time more readily than any other technique. We imagine that functional loss precedes anatomic loss of retinal ganglion cells. However, it was estimated in histologic monkey studies that a 0-dB sensitivity on static perimetry may occur only when the overlapping receptive field centers of at least six ganglion cells showed atrophy.24 Therefore, a number of axons will be lost before we will be able to detect the loss psychophysically, so psychophysical tests as perimetry are probably not sensitive enough to detect early loss of ganglion cells. If, in the longitudinal follow-up of patients with ocular hypertension, scanning laser polarimetry were proved to be highly sensitive to early changes in the NFL, then it might become the primary diagnostic tool in the clinical management of glaucoma. Finally, we found a significant difference in mean NFL retardation measurements between patients with glaucoma and matched control subjects. The mean superior NFL or inferior NFL in patients with glauco ma was about 30% lower compared with those in normal subjects. Weinreb and associates3 reported a mean retardation difference of about 13% between glaucomatous and normal eyes. Their glaucomatous population, however, exhibited an average mean defect in their visual fields of 3.1 dB, whereas in our population, the mean defect was greater: 10.3 dB. In addition, Weinreb and associates3 included the highly variable areas with blood vessels in their measure ments, whereas we excluded them. Their inclusion of highly variable areas might have reduced the contrast between glaucomatous and normal subjects. Even though there appeared to be considerable overlap between the retardation values of the glaucomatous and normal populations, the sensitivity and specifici ty of scanning laser polarimetry were quite satisfacto ry. The large overlap could be largely attributed to the better ratio of the nerve fiber bundles, that is, either superior or inferior, in the patients with glaucoma, whereas the worse of the two still allowed the patients to be correctly classified.
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
67
Appendix Reported Sensitivities and Specificities of Various Diagnostic Methods for Screening of Populations for Glaucoma Diagnostic Method
Sensitivity (%) Specificity (%)
Scanning laser polarimetry* Glaucoma Hemifield Test14 Optic disk evaluation25 Red-free nerve fiber layer photographs26 Nerve fiber layer height measurements27 Tonometry28
96
93
94
90
85
90
64
84
83
88
70
30
* Present study.
With a sensitivity of 96% and a specificity of 93% in a clinical population, scanning laser polarimetry may prove to be a valuable first test in screening programs for glaucoma. Whether scanning laser po larimetry is also suitable for the longitudinal followup of patients with glaucoma and subjects with ocular hypertension requires further study.
REFERENCES 1. Dreher AW, Reiter K. Retinal laser ellipsometry: a new method for measuring the retinal nerve fiber layer thickness distribution? Clin Vision Sci 1992;7:481-488. 2. Weinreb RN, Dreher AW, Coleman A, Quigley H, Shaw B, Reiter K. Histopathologic validation of Fourier-ellipsometry measurements of retinal nerve fiber layer thickness. Arch Ophthalmol 1990;108:557-560. 3. Weinreb RN, Shakiba S, Zangwill L. Scanning laser polar imetry to measure the nerve fiber layer of normal and glaucomatous eyes. Am J Ophthalmol 1995;119:627-636. 4. Chi Q, Tomita G, Inazumi K, Hayakawa T, Ido T, Kitazawa Y. Evaluation of the effect of aging on the retinal nerve fiber
68
layer thickness using scanning laser polarimetry. ] Glaucoma 1995;4:1-8. Tjon-Fo-Sang MJ, de Vries J, Lemij HG. Measurement by nerve fiber analyzer of retinal nerve fiber layer thickness in normal subjects and patients with ocular hypertension. Am J Ophthalmol 1996;122:220-227. 6. Weinreb RN, Shakiba S, Sample P, et al. Association between quantitative nerve fiber layer measurement and visual field loss in glaucoma. Am ] Ophthalmol 1996; 120:732-738. 7. Heijl A, Lindgren G, Olsson J. A package for the statistical analysis of visual fields. In: Greve EL, Heijl A, editors. Documenta Ophthalmologica Proceedings, Series 49. Sev enth International Visual Field Symposium: 1986. Dordrecht, The Netherlands: Martinus, Nijhoff, Junk. 1987:153-168. Katz J, Sommer A, Witt K. Reliability of visual field results over repeated testing. Ophthalmology 1991;98:70-75. Asman P, Heijl A. Glaucoma hemifield test: automated visual field examination. Arch Ophthalmol 1992;110:812-819. 10 Heijl A, Lindgren G, Olsson J, Asman P. Visual field interpretation with empiric probability maps. Arch Ophthal mol 1989;107:204-208. 11 Heijl A, Asman A. A clinical study of perimetric probability maps. Arch Ophthalmol 1989;107:199-203. 12. Asman P. Glaucoma hemifield test: physiological variability. Acta Ophthalmol 1992;70(suppl):21-26. 13 Asman P. Glaucoma hemifield test: condensation of data. Acta Ophthalmol 1992;70(suppl):27-38. 14 Katz J, Sommer A, Gaasterland DE, Anderson DR. Compar ison of analytic algorithms for detecting glaucomatous visual field loss. Arch Ophthalmol 1991;109:1684-1689. 15. Katz J, Tielsch JM, Quigley HA, Javitt J, Witt K, Sommer A. Automated suprathreshold screening for glaucoma: the Balti more Eye Survey. Invest Ophthalmol Vis Sci 1993;34: 3271-3277. 16. Kosoko O, Sommer A, Auer C. Screening with automated perimetry using a threshold related three-level algorithm. Ophthalmology 1986;93:882-886. 17. Fletcher RH, Fletcher SW, Wagner EH. Clinical epidemiolo gy: the essentials. Baltimore: Williams and Wilkins, 1996: 43-74. 18. Hollows FC, Graham PA. Intraocular pressure, glaucoma, and glaucoma suspects in a defined population. Br J Ophthal mol 1966;50:570-586. 19. Bengtsson B. The prevalence of glaucoma. Br J Ophthalmol 1981;65:46-49. 20. Leske MC. The epidemiology of open-angle glaucoma: a review. Am J Epidemiol 1983;118:116-191. 21 Levi L, Schwartz B. Glaucoma screening in the health care setting. Surv Ophthalmol 1983;28:164-174. 22. Keltner JL, Johnson CA. Screening for visualfieldabnormali ties with automated perimetry. Surv Ophthalmol 1983;28:175-183. 23 Quigley HA, Addicks EM, Green WR. Optic nerve damage in human glaucoma, III: quantitative correlation of nerve fiber loss and visual field defect in glaucoma, ischemic neuropathy, papilledema, and toxic neuropathy. Arch Oph thalmol 1982;100:135-146. 24 Quigley HA, Dunkelberger GR, Green WR. Retinal gan glion cell atrophy correlated with automated perimetry in human eyes with glaucoma. Am J Ophthalmol 1989;107: 453-464. 25 Quigley HA, Schwartz B. Open-angle Glaucoma. In:
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
Shingleton BJ, Berson FG, Cantor L, Hodapp EA, Lee DA, editors. Basic and Clinical Science course. Section 10. San Francisco: American Academy of Ophthalmology, 1994: 66-69. 26. Wang F, Quigley HA, Tielsch JM. Screening for glaucoma in a medical clinic with photographs of the nerve fiber layer. Arch Ophthalmol 1994;112:796-800.
27. Caprioli J, Miller J. Measurement of relative nerve fiber layer surface height in glaucoma. Ophthalmology 1989;96:633-641. 28. Packer H, Deutsch AR, Deweese MW, Kashgarian M, Lewis PM. Efficiency of screening tests for glaucoma. JAMA 1965;192:693-696.
Authors Interactive™ Comments and questions of 100 words or less regarding this article may be addressed to authors via THE JOURNAL'S Web Site on the Internet. Selected questions and comments along with the authors' statements and responses will be posted on the Authors Interactive™ Bulletin Board, a feature of THE JOURNAL'S home page. Readers may access THE JOURNAL'S home page and the Authors Interactive™ Bulletin Board at http://www.ajo.com/
VOL.123, No. I
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
69
• PURPOSE: To determine the sensitivity and spec ificity for detecting glaucoma by scanning laser polarimetry and to assess the relation between nerve fiber layer (NFL) measurements and visual field indices. • METHODS: The peripapillary NFL was divided into four segments: superior, inferior, temporal, and nasal. The mean polarimetric NFL for each segment was calculated out of six selected areas of 2 5 6 pixels each. Ratios relative to the nasal seg ment were determined for the superior and inferior segments. With the use of previously obtained normograms for polarimetric NFL readings, the sensitivity of scanning laser polarimetry was as sessed in 200 glaucomatous eyes (155 subjects). The specificity was assessed in a normal popula tion of 150 eyes (150 subjects). The relation between hemifield polarimetric NFL and visual field indices was assessed by linear regression analysis. • RESULTS: The sensitivity of scanning laser polarimetry was 96% and the specificity was 93%. The correlation between NFL parameters and visual field indices ranged from —0.18 to + 0 . 2 6 .
Accepted for publication Aug 6, 1996. From the Glaucoma Service, Rotterdam Eye Hospital, Rotterdam, The Netherlands. Reprint requests to Martha J. Tjon-Fo-Sang, MD, Rotterdam Eye Hospital, PO Box 70030, 3000 LM Rotterdam, The Netherlands; fax: 31 10-401 76 55.
62
The amount of variation by the linear regression model ranged from 3% to 6%. • CONCLUSIONS: Although quantitative mea surements of the NFL with scanning laser polar imetry relate poorly to visual field indices, the technique seems to be promising for screening populations for glaucoma. Whether measurements of the NFL with scanning laser polarimetry are also sensitive enough to detect change over time requires further study.
S
CANNING LASER POLARIMETRY (THE NERVE FIBER
Analyzer I, Laser Diagnostic Technologies, Inc, San Diego, California) can be used to quantita tively assess the peripapillary retinal nerve fiber layer (NFL) in vivo. The technique is based on the specific arrangement of the microtubules in the retinal NFL.1,2 In scanning laser polarimetry, the phase shift of the reflected light beam (equal to the retardation) is linearly correlated with the NFL thickness.1,2 Scan ning laser polarimetry has potential value in the detection and follow-up in glaucoma because it pro vides an objective and fast measurement of the NFL with good reproducibility of measurements. The coef ficient of variation for repeated measurements with scanning laser polarimetry has been reported to be 3.6% to 5.2% in normal subjects (references 3 and 4 and Tjon-Fo-Sang MJ, de Vries J, Lemij HG, unpub lished data). In a previous study,5 we presented normograms for NFL retardation in white subjects. In these normograms, we showed a gradual decrease in NFL retardation with age.5 This decrease with age has
© AMERICAN JOURNAL OF OPHTHALMOLOGY i23,-62-69
JANUARY 1997
also been observed by other authors. 4 In addition, significantly lower retardation in patients with glauco ma compared with normal subjects has been recog nized, although there was also considerable overlap.3,4 Weinreb and associates6 reported an only moderate association between NFL retardation and visual field indices; they concluded, however, that scanning laser polarimetry seemed promising for the detection and follow-up of glaucoma, although sensitivity and speci ficity measures were not assessed. Therefore, we have determined the sensitivity and specificity of scanning laser polarimetry as a diagnostic tool in a large population. We have also assessed the relation be tween NFL retardation and visual field indices.
PATIENTS AND METHODS FROM OUR OUTPATIENT GLAUCOMA SERVICE, 155 PA-
tients with glaucoma (200 eyes) were recruited con secutively. Patients were included only when they had multiple intraocular pressure readings over 21 mm Hg, no secondary glaucoma, no diabetes or other chorioretinal disorders, no intraocular lens implants, a pupil diameter greater than 2 mm, and typical glaucomatous defects on the visual field test. All patients were subjected to a Humphrey Field Analyzer (HFA) central 30-2 threshold test. To qualify patients for further analysis, the HFA central 30-2 test should satisfy the standard reliability criteria for patients with glaucoma7,8: less than 20% fixation losses and less than 33% false positive errors. In addition, the Glaucoma Hemifield Test (GHT) 9 should score pa tients on the visual field test as "outside normal limits." Typical glaucomatous defects were defined as arcuate scotomas or hemifield defects. When a patient scored on the G H T visual field as "within normal limits," "borderline," or "general reduction of sensi tivity," the eyes were not included in this study. All polarimetric recordings were obtained under the following conditions: the subject's head was placed as upright as possible in the chin rest, pupils were left undilated, and ambient lights were left on. During image acquisition and analysis of the data, the investigator had been masked for the corresponding visual fields but not for the patient diagnosis. For data
VOL.123,
No. 1
analysis, we applied the squares calculation method (reference 5 and Tjon-Fo-Sang MJ, de Vries J, Lemij HG, unpublished data). In brief, the retardation map was divided into 16 X 16 squares of 256 pixels each. A circle was placed at the optic disk's margin, and the retardation map was divided in a superior and an inferior segment of 120 degrees each, a nasal segment of 50 degrees, and a temporal segment of 70 degrees. Subsequently, we averaged NFL retardation in six squares for each superior, inferior, and nasal segment. Selected by the investigator, the squares were outside a peripapillary circle of 1.75 disk diameters and contained no blood vessels because blood vessels may adversely affect the reproducibility of the recordings.3 Ratios for the superior to nasal (superior NFL), inferior to nasal (inferior NFL), and superior to inferior (S/I) NFL retardation were then computed. Because the superior and inferior NFL ratios re spectively represent the superior and inferior arcuate bundles, we determined the mean deviation of the HFA central 30-2 test separately for the superior and inferior hemifields. The hemifield mean deviations were calculated for the nonedge points only, except for the most nasal points (x = —27 degrees, y = 3 degrees, and x = — 27 degrees, y = — 3 degrees for a right eye field), which were included. Linear regres sion analysis was then applied to assess the correlation between superior NFL and mean deviation of the lower hemifield. In addition, R2, as explained in the linear regression model, was calculated. The same procedure was repeated for the relation between inferior NFL and mean deviation of the upper hemifield. The GHT was developed to assess hemifield (updown) differences in the Statpac probability maps9'12 to better differentiate localized from diffuse loss caused by, for example, media opacities or miotic therapy. Based on the probability symbols in the pattern deviation map of an HFA central 30-2 test, the G H T score is calculated for each hemifield. To differentiate localized from diffuse loss, the up-down difference of the hemifields is determined. The dis criminative capability of the G H T as a method of analysis has a sensitivity of 92% to 94% and a specificity of 90%.13,14 With linear regression analysis, the correlation between the superior NFL and inferi or G H T score was determined. A similar correlation
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
63
was determined between the inferior NFL and superi or G H T score. Nerve fiber layer measurements determined with scanning laser polarimetry4,5 exhibit great intersubject variability and a gradual decrease of NFL retardation with age. Both these factors possibly affect the corre lation between the visual field indices (mean devia tion and G H T score) and the NFL retardation measurements. Therefore, to account for the age effect, we normalized the NFL ratios by dividing a patient's NFL retardation by the mean normal retar dation for his or her age cohort (per decade), ex pressed as a percentage. Linear regression analysis was thereafter repeated for determination of the relation between normalized NFL retardation and visual field indices. Only when all the retardation data in the glaucomatous and normal populations had been established did we determine the second and fifth percentiles for superior and inferior NFLs for each decade, begin ning at age 20 years, in our normograms for NFL retardation.6 We also assessed the 2.5 and 97.5 percentiles for the S/I ratio. These cutoff points were subsequently used to determine the sensitivity and specificity of scanning laser polarimetry in 155 pa tients with glaucoma (200 eyes) and 150 normal subjects (150 eyes). T h e normal eyes had not been included in the determination of the normograms. All normal subjects had intraocular pressures of 21 mm Hg or less, had no evidence of disorders of the optic disk or retina, and had completed an HFA 120-point full-field suprathreshold screening test, ful filling its reliability and normality criteria.15,16 For determination of mean NFL retardation in patients with glaucoma and normal subjects, only the eye that satisfied our inclusion criteria for each patient in the glaucomatous population was included. When both eyes were eligible, one eye was randomly selected for inclusion. As many patients with glaucoma as possible (100) were randomly matched with normal subjects for race, sex, age, and included eye. For the statistical analysis of NFL retardation values in pa tients with glaucoma and normal subjects, Student's t test was applied. A P value below .05 was considered statistically significant. Only one eye of each subject was included for this part of the study to avoid intereye correlation bias in the Student's t test.
64
frequency (N)
Mean Defect (dB)
Figure 1. Histogram of the distribution of the mean defect (mean deviation in dB) in the visual fields of 200 glaucomatous eyes, determined with the Humphrey Field Analyzer Central 30-2 test.
RESULTS WE MEASURED NFL RETARDATION IN 95 LEFT AND 105
right eyes of 155 patients with glaucoma (87 men, 68 women; 152 white, three black). The mean age of the glaucomatous patients was 67.1 years (total range, 26 to 92 years). The mean deviation of the HFA 30-2 tests, as noted on the visual fields, was —10.33 dB (range, —31.5 to 0.76 dB). Therefore, patients with early defects as well as patients with severe visual field loss were included in this study. The distribution of the mean deviations in this glaucoma population has been illustrated in Figure 1. With linear regression analysis, the correlation between the superior NFL and mean deviation of the lower hemifield was only 0.26 (R2 = 0.07; see Figure 2, top). The correlation between the inferior NFL and mean deviation of the upper hemifield was identical: 0.26 (R2 = 0.07) (Figure 2, bottom). The correlation between the superior NFL and the G H T score of the lower hemifield was only —0.24 (R2 = 0.06), but between the inferior NFL and the G H T score of the upper hemifield, it was even less: - 0 . 1 8 (R2 = 0.03). Normalization of the NFL retardation parameters did not improve these correlations. Of 200 glaucomatous eyes, 113 (56.5%) had an S/I ratio below the 2.5 percentile or over the 97.5 percentile. Of the remaining eyes, 81 had a superior or inferior NFL below the fifth percentile, of which even 79 (39.5%) had a superior or inferior NFL below the
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
3
superior
NFL R = 0.26
*
2.5 2
-
1 S
.
•
*.• •
: ••
•
•
• ^-^s^^^s^-' • • -S%~T~ -i>' >"\ • . ••» u . .
a*
••
1 0.5
-30
-10
0 Mean Defect (dB)
inferior NFL R=0.26
-40
-30
-20
-10
0
10
Mean Defect (dB)
Figure 2. (Top) The relation between superior nerve fiber layer (NFL) retardation and the mean defect (dB) of the inferior hemifield. Linear regression analysis yielded a correlation of +0.26; R2 = 0.07. (Bottom) The relation between inferior nerve fiber layer retarda tion and the mean defect (dB) of the superior hemifield (correlation +0.26; R2 = 0.07).
tive outcome of the test) is, however, only 12%. By contrast, the negative predictive value in a normal polarimetric reading is 100%. Because of the size of our normal database, only 100 patients (44 men, 56 women; 97 white, three black) of the 155 could be matched with normal subjects. The mean age of this glaucomatous subpopulation was 65 years (total range, 26 to 92 years) and, in the matched normal population, was 64 years (total range, 28 to 88 years). The mean deviation of the HFA 30-2 tests in the glaucomatous subpopulation was -10.27 dB (total range, - 3 1 . 5 to +0.76 dB). The mean superior NFL in patients with glauco ma was 1.4 (95% range, 0.8 to 2.0), whereas the mean superior NFL in matched normal subjects was statistically significantly higher: 2.2 (95% range, 1.1 to 3.3; P = .0004) (Figure 3, top). The mean inferior NFL in patients with glaucoma was identical to the superior NFL (1.4), but with a slightly smaller 95% range of 0.9 to 1.9 (Figure 3, bottom). The mean inferior NFL in normal subjects was again statistically significantly higher: 2.2 (95% range, 1.1 to 3.3; P = .0004). For patients with glaucoma, the mean S/I was 1.05 (95% range, 0.71 to 1.39), whereas for normal subjects, this was 1.01, with, however, a much smaller 95% range of 0.89 to 1.13.
DISCUSSION IN THIS STUDY, WE FOUND A SENSITIVITY OF 96% AND A
second percentile. Six glaucomatous eyes had NFL readings within normal limits. Therefore, when the cutoff point was set at an abnormal S/I or when the superior or inferior NFL was below the second percentile, the sensitivity of scanning laser polarimetry was 96%. Of 150 normal eyes (56 white male and 94 white female subjects; mean age, 49 years [total range, 21 to 88 years]), four (2.7%) had abnormal S/I ratios and 13 had a superior or inferior NFL below the fifth percentile. Seven (4-7%) of these were even below the second percentile. The specificity of scan ning laser polarimetry was 93%. Based on Bayes' theorem of conditional probabilities17 and a preva lence rate of glaucoma of 1% in the mass population more than 40 years old,18'20 the predictive value for glaucoma in an abnormal polarimetric reading (posi
Vot. 123, No. 1
specificity of 93% for the detection of glaucoma using scanning laser polarimetry. These numbers are better than those reported for tonometry, optic disk evalua tion, NFL red-free fundus photographs, the GHT, and even NFL height measurements (Appendix). In previous studies, doubts had been cast on the poten tial of scanning laser polarimetry as a screening instrument, mainly because of the great variability in NFL retardation found between normal subjects.3,5 However, the sensitivity reported here indicates that glaucomatous NFL loss exceeds age-related NFL loss to an extent readily detectable. Although the sensitiv ity and specificity for scanning laser polarimetry were very good, they might have been overstated because these numbers were based on glaucoma service cases rather than on the general population. Moreover, the
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
65
theorem is applied for the G H T with a sensitivity of 94% and specificity of 90%, the positive predictive value is only 9%. Both positive predictive values are poor because of the low prevalence rate of glaucoma in the general population. Therefore, the use of scanning laser polarimetry as a single screening procedure for glaucoma should be reserved for highly selected settings with a higher prevalence rate of glaucoma—for example, patients of older age, black subjects, patients with a family history of glaucoma, and patients with diabetes or other vascular diseases.21 The Public Health Service of the United States has suggested that any medical screening program, to be considered a cost-effective measure, must detect at least a 2% pathology in the general population.22 However, when polarimetry is combined with subse quent static perimetry (GHT), the positive predictive value increases to 53%.
superior NFL
inferior NFL
Figure 3. The mean superior nerve fiber layer (NFL) retardation in 100 patients with glaucoma (A) and 100 matched normal subjects (O). (Top) Superior nerve fiber layer retardation was statistically significantly higher in normal subjects (2.2; 95% range, 1.1 to 3.3) compared with glaucomatous patients (1.4; 95% range, 0.8 to 2.0). (Bottom) Inferior nerve fiber layer retardation. The mean inferior nerve fiber layer retardation was also significantly higher in normal subjects (2.2; 95% range, 1.1 to 3.3) compared with glaucomatous patients (1.4; 95% range, 0.9 to 1.9).
predictive value of a positive predictive value test was quite disappointing: 12%. However, this was still better than the equivalent for the GHT: when Bayes'
66
What does this process of serial testing mean in practice? Suppose we were to screen a population of 100,000 adults with an estimated 1% glaucoma prevalence. For serial testing, any screening procedure is most efficient when the test with the highest sensitivity and specificity is used first.17 Hence, by starting with polarimetry, 960 of the 100,000 patients (sensitivity, 96%) would be correctly diagnosed. Forty patients would be missed. With a specificity of 93%, 6,930 normal subjects would have abnormal polarimetric readings, leading to wrong referral. Followed by static perimetry (GHT), however, only an addi tional 1,595 of the 100,000 adults would be referred and another 58 glaucomatous patients would be missed. Therefore, the low rate of referral occurs at the expense of a total of 98 (9.8%) missed glaucoma tous cases. If scanning laser polarimetry were not available, the initial screening for glaucoma with visual field testing would result in wrong referrals of 9.9% (3% more than initial screening with polarime try). Moreover, the acquisition of polarimetric read ings is much faster3 (including patient instructions, about 10 minutes) and objective, whereas visual fields require longer testing and are subjective. Moreover, in a study assessing various analytic strategies, almost 30% of the subjects failed to meet the reliability criteria for the HFA central 30-2 test.14 Hence, scanning laser polarimetry might be useful as the first test when the general population is screened for
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
glaucoma; automated perimetry may function as a second, additive test. One limitation of our method of data analysis is that it requires an operator to select the squares to represent each segment. This might introduce great interoperator variability. However, this method was necessary for us to exclude the areas with blood vessels and to correct for the lack of absolute calibra tion in the Nerve Fiber Analyzer I. In its newest version, the Nerve Fiber Analyzer II, the calibration problem has been dealt with; and, with the newest software, it will be possible to assess areas without blood vessels in an automated manner. Scanning laser polarimetry has a higher sensitivity than the G H T does, that is, 96% compared with 94% (Appendix). This may concur with our previous finding that 58% of subjects with ocular hypertension had an abnormal polarimetric reading.5 Our data therefore suggested that high intraocular pressure had caused early NFL loss before visual field defects had become apparent. Nerve fiber layer or optic disk abnormalities preceding visual field defects were in deed reported by Quigley and associates,23 who found substantial loss of axons in patients with ocular hypertension. Whether these findings suggest that current definitions of glaucoma and ocular hyperten sion should be modified and whether ocular hyper tensive patients with abnormal polarimetric readings should be treated as glaucomatous patients are ques tions yet to be answered. As did Weinreb and associates,6 we found a poor correlation between NFL retardation and visual field indices. The amount of variation explained by the linear regression models, however, was somewhat better in their study. Our variation may have been adversely affected by the great variability in NFL retardation at a mean defect of about 0 dB, or a G H T score of zero. Quigley and associates24 assessed the relation between decibel sensitivity loss on perimetry and the percent of normal ganglion cells at precisely measured locations. They concluded that a 5-dB loss required 20% to 50% ganglion cell death. However, in locations with a 0-dB sensitivity, 3 % to 36% of normal density of ganglion cells was still found.24 Although these cells might be physically present, their function in terms of retinal sensitivity on perimetry is absent.24 Therefore, the relation between
VOL.123, No. 1
structure and function of the retinal nerve fibers— that is, between optic disk or NFL properties and visual field indices—is not yet clear-cut. Scanning laser polarimetry would be of great clinical potential for the follow-up of patients if it were capable of detecting change over time more readily than any other technique. We imagine that functional loss precedes anatomic loss of retinal ganglion cells. However, it was estimated in histologic monkey studies that a 0-dB sensitivity on static perimetry may occur only when the overlapping receptive field centers of at least six ganglion cells showed atrophy.24 Therefore, a number of axons will be lost before we will be able to detect the loss psychophysically, so psychophysical tests as perimetry are probably not sensitive enough to detect early loss of ganglion cells. If, in the longitudinal follow-up of patients with ocular hypertension, scanning laser polarimetry were proved to be highly sensitive to early changes in the NFL, then it might become the primary diagnostic tool in the clinical management of glaucoma. Finally, we found a significant difference in mean NFL retardation measurements between patients with glaucoma and matched control subjects. The mean superior NFL or inferior NFL in patients with glauco ma was about 30% lower compared with those in normal subjects. Weinreb and associates3 reported a mean retardation difference of about 13% between glaucomatous and normal eyes. Their glaucomatous population, however, exhibited an average mean defect in their visual fields of 3.1 dB, whereas in our population, the mean defect was greater: 10.3 dB. In addition, Weinreb and associates3 included the highly variable areas with blood vessels in their measure ments, whereas we excluded them. Their inclusion of highly variable areas might have reduced the contrast between glaucomatous and normal subjects. Even though there appeared to be considerable overlap between the retardation values of the glaucomatous and normal populations, the sensitivity and specifici ty of scanning laser polarimetry were quite satisfacto ry. The large overlap could be largely attributed to the better ratio of the nerve fiber bundles, that is, either superior or inferior, in the patients with glaucoma, whereas the worse of the two still allowed the patients to be correctly classified.
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
67
Appendix Reported Sensitivities and Specificities of Various Diagnostic Methods for Screening of Populations for Glaucoma Diagnostic Method
Sensitivity (%) Specificity (%)
Scanning laser polarimetry* Glaucoma Hemifield Test14 Optic disk evaluation25 Red-free nerve fiber layer photographs26 Nerve fiber layer height measurements27 Tonometry28
96
93
94
90
85
90
64
84
83
88
70
30
* Present study.
With a sensitivity of 96% and a specificity of 93% in a clinical population, scanning laser polarimetry may prove to be a valuable first test in screening programs for glaucoma. Whether scanning laser po larimetry is also suitable for the longitudinal followup of patients with glaucoma and subjects with ocular hypertension requires further study.
REFERENCES 1. Dreher AW, Reiter K. Retinal laser ellipsometry: a new method for measuring the retinal nerve fiber layer thickness distribution? Clin Vision Sci 1992;7:481-488. 2. Weinreb RN, Dreher AW, Coleman A, Quigley H, Shaw B, Reiter K. Histopathologic validation of Fourier-ellipsometry measurements of retinal nerve fiber layer thickness. Arch Ophthalmol 1990;108:557-560. 3. Weinreb RN, Shakiba S, Zangwill L. Scanning laser polar imetry to measure the nerve fiber layer of normal and glaucomatous eyes. Am J Ophthalmol 1995;119:627-636. 4. Chi Q, Tomita G, Inazumi K, Hayakawa T, Ido T, Kitazawa Y. Evaluation of the effect of aging on the retinal nerve fiber
68
layer thickness using scanning laser polarimetry. ] Glaucoma 1995;4:1-8. Tjon-Fo-Sang MJ, de Vries J, Lemij HG. Measurement by nerve fiber analyzer of retinal nerve fiber layer thickness in normal subjects and patients with ocular hypertension. Am J Ophthalmol 1996;122:220-227. 6. Weinreb RN, Shakiba S, Sample P, et al. Association between quantitative nerve fiber layer measurement and visual field loss in glaucoma. Am ] Ophthalmol 1996; 120:732-738. 7. Heijl A, Lindgren G, Olsson J. A package for the statistical analysis of visual fields. In: Greve EL, Heijl A, editors. Documenta Ophthalmologica Proceedings, Series 49. Sev enth International Visual Field Symposium: 1986. Dordrecht, The Netherlands: Martinus, Nijhoff, Junk. 1987:153-168. Katz J, Sommer A, Witt K. Reliability of visual field results over repeated testing. Ophthalmology 1991;98:70-75. Asman P, Heijl A. Glaucoma hemifield test: automated visual field examination. Arch Ophthalmol 1992;110:812-819. 10 Heijl A, Lindgren G, Olsson J, Asman P. Visual field interpretation with empiric probability maps. Arch Ophthal mol 1989;107:204-208. 11 Heijl A, Asman A. A clinical study of perimetric probability maps. Arch Ophthalmol 1989;107:199-203. 12. Asman P. Glaucoma hemifield test: physiological variability. Acta Ophthalmol 1992;70(suppl):21-26. 13 Asman P. Glaucoma hemifield test: condensation of data. Acta Ophthalmol 1992;70(suppl):27-38. 14 Katz J, Sommer A, Gaasterland DE, Anderson DR. Compar ison of analytic algorithms for detecting glaucomatous visual field loss. Arch Ophthalmol 1991;109:1684-1689. 15. Katz J, Tielsch JM, Quigley HA, Javitt J, Witt K, Sommer A. Automated suprathreshold screening for glaucoma: the Balti more Eye Survey. Invest Ophthalmol Vis Sci 1993;34: 3271-3277. 16. Kosoko O, Sommer A, Auer C. Screening with automated perimetry using a threshold related three-level algorithm. Ophthalmology 1986;93:882-886. 17. Fletcher RH, Fletcher SW, Wagner EH. Clinical epidemiolo gy: the essentials. Baltimore: Williams and Wilkins, 1996: 43-74. 18. Hollows FC, Graham PA. Intraocular pressure, glaucoma, and glaucoma suspects in a defined population. Br J Ophthal mol 1966;50:570-586. 19. Bengtsson B. The prevalence of glaucoma. Br J Ophthalmol 1981;65:46-49. 20. Leske MC. The epidemiology of open-angle glaucoma: a review. Am J Epidemiol 1983;118:116-191. 21 Levi L, Schwartz B. Glaucoma screening in the health care setting. Surv Ophthalmol 1983;28:164-174. 22. Keltner JL, Johnson CA. Screening for visualfieldabnormali ties with automated perimetry. Surv Ophthalmol 1983;28:175-183. 23 Quigley HA, Addicks EM, Green WR. Optic nerve damage in human glaucoma, III: quantitative correlation of nerve fiber loss and visual field defect in glaucoma, ischemic neuropathy, papilledema, and toxic neuropathy. Arch Oph thalmol 1982;100:135-146. 24 Quigley HA, Dunkelberger GR, Green WR. Retinal gan glion cell atrophy correlated with automated perimetry in human eyes with glaucoma. Am J Ophthalmol 1989;107: 453-464. 25 Quigley HA, Schwartz B. Open-angle Glaucoma. In:
AMERICAN JOURNAL OF OPHTHALMOLOGY
JANUARY 1997
Shingleton BJ, Berson FG, Cantor L, Hodapp EA, Lee DA, editors. Basic and Clinical Science course. Section 10. San Francisco: American Academy of Ophthalmology, 1994: 66-69. 26. Wang F, Quigley HA, Tielsch JM. Screening for glaucoma in a medical clinic with photographs of the nerve fiber layer. Arch Ophthalmol 1994;112:796-800.
27. Caprioli J, Miller J. Measurement of relative nerve fiber layer surface height in glaucoma. Ophthalmology 1989;96:633-641. 28. Packer H, Deutsch AR, Deweese MW, Kashgarian M, Lewis PM. Efficiency of screening tests for glaucoma. JAMA 1965;192:693-696.
Authors Interactive™ Comments and questions of 100 words or less regarding this article may be addressed to authors via THE JOURNAL'S Web Site on the Internet. Selected questions and comments along with the authors' statements and responses will be posted on the Authors Interactive™ Bulletin Board, a feature of THE JOURNAL'S home page. Readers may access THE JOURNAL'S home page and the Authors Interactive™ Bulletin Board at http://www.ajo.com/
VOL.123, No. I
SENSITIVITY AND SPECIFICITY OF POLARIMETRY
69