- Email: [email protected]
Sensitivity and specificity of the swedish interactive threshold algorithm for glaucomatous visual field defects1
Sensitivity and Specificity of the Swedish Interactive Threshold Algorithm for Glaucomatous Visual Field Defects Donald L. Budenz, MD,1 Paul Rhee, OD,1 William J. Feuer, MS,1 John McSoley, OD,1 Chris A. Johnson, PhD,2 Douglas R. Anderson, MD1 Purpose: To determine the sensitivity and specificity of two new visual field algorithms in detecting glaucomatous visual field defects: (1) Swedish interactive threshold algorithm (SITA) standard and (2) SITA fast. Design: Prospective observational case series. Participants: Ninety normal subjects and 82 glaucoma patients. Testing: Central 30° fields were performed with the Humphrey visual field analyzer 30-2 program (Humphrey Systems, Dublin, CA) using full threshold, SITA standard, and SITA fast algorithms on the same day for two or more sessions within a 1-month period. Main Outcome Measures: Sensitivity and specificity in detecting glaucomatous visual field defects with SITA standard and SITA fast using full threshold testing as the reference standard. Results: The sensitivity of SITA standard and SITA fast in detecting glaucomatous defects overall was 98% and 95%, respectively. In the subset of mild glaucomatous field defects (26 patients), sensitivity of SITA standard was 92% versus 85% with SITA fast. Sensitivity was 100% for both algorithms in moderate to severe glaucomatous defects. Specificity for glaucoma defects using SITA standard and SITA fast was 96% for both algorithms. SITA standard reduced test-taking time from full threshold by 52% in normal subjects and 47% in glaucoma patients (P ⬍ 0.001). SITA fast reduced test-taking time by 72% in normal subjects and 65% in glaucoma patients (P ⬍ 0.001). Mean deviation values were 0.4 dB and 0.8 dB better in SITA standard and SITA fast fields, respectively, in normal subjects (P ⬍ 0.001), and 0.7 dB and 1.2 dB in SITA standard and SITA fast fields, respectively, in glaucoma patients (P ⬍ 0.001) compared with full threshold values. Conclusions: The new algorithms for measuring visual fields, SITA standard and SITA fast, have excellent sensitivity and specificity for glaucomatous visual field loss with considerable savings in time. Ophthalmology 2002;109:1052–1058 © 2002 by the American Academy of Ophthalmology. Visual field analysis and optic nerve visualization are critical features used in the diagnosis and management of glaucoma. Full threshold white-on-white automated static perimetry1,2 is currently the gold standard for the diagnosis, grading, and detection of progression of glaucomatous visual field defects.3– 8 However, the standard full threshold method for measuring the visual field is very time consuming for patients and is subject to fatigue effect, which has Originally received: July 24, 2001. Accepted: October 9, 2001.
Manuscript no. 210664.
1
Bascom Palmer Eye Institute, Department of Ophthalmology, University of Miami School of Medicine, Miami, Florida.
2
Department of Ophthalmology, University of California at Davis, Sacramento, California. Current affiliation: Devers Eye Institute, Portland, Oregon. Presented in part at the Association for Research in Vision and Ophthalmology annual meeting, Fort Lauderdale, Florida, May 2001. Supported by an unrestricted grant from Research to Prevent Blindness, New York, New York. In addition, Dr. Johnson receives research support from Humphrey Systems, Dublin, California. None of the authors has a commercial interest in Humphrey Systems. Reprint requests to Donald L. Budenz, MD, Bascom Palmer Eye Institute, 900 NW 17th Street, Miami, FL 33136.
1052
© 2002 by the American Academy of Ophthalmology Published by Elsevier Science Inc.
been shown to result in poorer results.9 –11 This effect may be even more pronounced in glaucoma patients.10,12 The Swedish interactive threshold algorithm (SITA) is a new computer program that has been developed for the Humphrey visual field analyzer II (Humphrey Systems, Dublin, CA), which reduces test-taking time. The SITA standard program (SS) has been shown to reduce test-taking time by approximately 50%13–16 and the SITA fast program (SF) by 70% compared with full threshold testing.15,17,18 Overall, the number of stimuli actually presented is reduced by 29% in normal fields and 26% in glaucomatous fields.19 This is accomplished using a combination of techniques, including the use of information about surrounding points, information about threshold values in age-matched controls and glaucoma patients at each location, changes in the pacing of the test, elimination of retest trials for the 10 points used to calculate short-term fluctuation in the full threshold algorithm, changing the way in which false positive and false negative reliability parameters are determined, and the use of a maximum likelihood procedure for estimating threshold.18 –20 The difference between the SS and SF programs is that the SF program uses a wider confidence interval for the determination of threshold estiISSN 0161-6420/02/$–see front matter PII S0161-6420(02)01047-3
Budenz et al 䡠 Sensitivity and Specificiy of SITA Table 1. Hodapp, Anderson, Parrish Visual Field Severity Score Criteria for Early Defect 1 Mean deviation no worse than ⫺6 dB 2 On pattern deviation plot, fewer than 25% of points depressed below the 5% level, and fewer than 15% of points depressed below the 1% level 3 No point within central 5° with sensitivity ⬍15 dB Criteria for Moderate Defect 1 Mean deviation worse than ⫺6 dB but no worse than ⫺12 dB 2 On pattern deviation plot, fewer than 50% of points depressed below the 5% level, and fewer than 25% of points depressed below the 1% level 3 No point within central 5° with sensitivity of ⱕ0 dB 4 Only one hemifield containing a point with sensitivity ⬍15 dB within 5° of fixation Criteria for Severe Defect 1 Mean deviation worse than ⫺12 dB 2 On pattern deviation plot, more than 50% of points depressed below the 5% level, and more than 25% of points depressed below the 1% level 3 Any point within central 5° with sensitivity of ⱕ0 dB 4 Both hemifields containing a point or points with sensitivity ⬍15 dB within 5° of fixation
mation, permitting the determination of threshold to stop sooner than in the SS program.18 Because of the time savings that SITA provides, it has been suggested that this testing algorithm should replace the full threshold algorithm,14 which has been the gold standard for detecting and following glaucomatous visual field defects for over 15 years. However, few studies have been performed to determine the specificity and sensitivity of SITA in diagnosing glaucomatous visual field defects.15,16 The purpose of the current study is to determine the sensitivity and specificity of SITA standard and SITA fast in detecting glaucomatous visual field defects using the full threshold algorithm as the reference, or “gold” standard.
Materials and Methods This study was approved by the Human Subjects Subcommittee of the Institutional Review Boards at the University of Miami and the University of California, Davis. Patients older than 17 years of age with known glaucoma, defined as characteristic cupping of the optic nerve and glaucomatous visual field defects in at least one eye, regardless of intraocular pressure level, were invited to participate. Glaucoma patients were required to be experienced visual field takers, having been tested on two or more prior occasions on the Humphrey visual field analyzer (Humphrey Systems). Patients were excluded if the visual acuity in the eye to be tested was less than 20/40. If both eyes had glaucoma, the eye to be tested was selected randomly by the investigator. Normal subjects older than 17 years of age were invited to participate if they met the same visual acuity criteria. An attempt was made to include older patients to approximate the age of the glaucoma patients being studied. Subjects in the eye care field were ineligible for participation to avoid biasing of testing. Normal subjects were excluded if they had an intraocular pressure ⬎ 22 mmHg in either eye; history of amblyopia; refractive error ⬎ than 5 diopters of sphere
or 2.5 diopters of cylinder; suspicious appearing optic nerves; history of any disease, surgery, or trauma to the eye being tested; abnormal pupillary examination or history of miotic use, or other medications that might affect pupil size; history of systemic medication use (e.g., plaquenil) that might affect the visual field; history of cerebrovascular event or diabetic retinopathy; or insulin dependent diabetes mellitus. Subjects recruited as normal who had a visual field defect associated with an identifiable cause (such as chorioretinal scar) were excluded. If both eyes were normal, the eye to be tested was selected randomly by the investigator. After obtaining written informed consent, all patients underwent the following visual field testing protocol, during at least two visits on separate days that were within 1 month of each other. At each visit, a standard full threshold (FT) test using program 30-2 and a size III white stimulus on a white background (31.5 apostilbs or 10 cd/m2) was performed using the Humphrey field analyzer I perimeter (Humphrey Systems). Calculations of the total and pattern deviation plots and global indices (mean deviation [MD] and corrected pattern deviation) were derived using STATPAC, version 9.31. Using the Humphrey field analyzer II (Humphrey Systems), SS, and SF tests using program 30-2 and a size III white stimulus on a white background were performed. Calculations of the total and pattern deviation plots and global indices (MD and pattern standard deviation [PSD]) were derived using STATPAC for SITA, version A10.1. The order of testing was alternated between patients to equalize fatigue effects among testing algorithms. However, the order of tests was kept constant for each patient over all testing sessions. The subject’s best-corrected distance refraction and age-appropriate near add was used in the lens holder and the same prescription was used at each session. Patients were required to take at least a 15-minute rest between visual field tests. Tests using a particular algorithm were performed with the same visual field machine, and each subject had the same visual field technician for all tests. Glaucoma patients on topical miotic therapy were required to be tested on miotics at all visits, and every attempt was made to perform testing an equivalent time after instillation of the miotic. Subjects who required a dilated fundus examination were dilated and examined after visual field testing was performed. An attempt was made to perform testing on glaucomatous eyes with a wide range of visual field severity based on evaluation of prior full threshold testing. Visual fields with any abnormal reliability parameter (fixation losses ⬎33%, false-positive responses ⬎33%, or false-negative responses ⬎33%) were excluded and repeat testing was performed on a subsequent date. If a subject failed to produce two reliable fields using any one algorithm within 1 month of each other, they were excluded from the study. A complete, dilated eye examination was required for all subjects within 6 months of the study as part of the eligibility determination. Normal subjects, if not seen within 6 months of the study initiation, were examined after the second session of visual field testing and excluded if any of the exclusion criteria were met as described above. Minimal criteria for glaucomatous visual field defect were as follows:21,22 glaucoma hemifield test (GHT) outside normal limits, corrected PSD (for FT algorithm) or PSD (for SS and SF algorithms) with p values ⬍5%, or a cluster of three or more points in the pattern deviation plot in a single hemifield (superior or inferior) with P values ⬍5%, one of which must have a P value ⬍1%. Any one of the preceding criteria, if found again on repeat testing, was considered sufficient evidence of a glaucomatous visual field defect. Patients were not required to meet the same criteria on repeat testing, but were required to meet any one of the three criteria on two full threshold tests within 1 month. Full threshold fields were used to classify patients into mild, moderate, or severe glaucomatous defects using two separate systems. First, the Hodapp, Anderson, Parrish (HAP) system23–25 (see
1053
Ophthalmology Volume 109, Number 6, June 2002
Figure 1. Session 1 of a patient with glaucoma as identified with the full threshold (FT) algorithm but not Swedish interactive threshold algorithm (SITA) standard or SITA fast. The pointwise criterion and glaucoma hemifield test (GHT) were met in FT testing, but none of the three criteria for minimal abnormality were met in SITA standard or SITA fast testing. CPSD ⫽ corrected pattern standard deviation; MD ⫽ mean deviation; PSD ⫽ pattern standard deviation; SF ⫽ SITA fast.
Table 1) was used on the FT fields. Second, fields were classified into mild, moderate, and severe using a k-means cluster analysis26 based on the MD for the FT fields. Demographic and clinical characteristics between the two groups were compared using chi-square tests and nonpaired two-tailed student’s t tests for categorical and continuous variables, respectively. Sensitivity and specificity of SS and SF were determined for the entire group and then separately for mild, moderate, and severe defects using the FT fields as the reference standard. Time for test completion, MD, and PSD were averaged between the two tests and compared using a paired two-tailed t test. Data analysis was performed using SPSS for Windows (release 10.0; SPSS Inc., Chicago, IL).
Results Ninety normal eyes of 90 normal subjects and 82 glaucomatous eyes of 82 patients met the inclusion and exclusion criteria. Twen-
1054
ty-seven of 90 normal subjects were male (30%) and 63 of 90 were female (70%). Forty-three of 82 glaucoma patients were male (52%) and 39 were female (48%). There were more females in the normal group than in the glaucoma group (P ⫽ 0.003). The mean age of the normal subjects was 52.9 years (median, 55; range, 19 – 89, standard deviation [SD], 16.2) and the mean age of the glaucoma patients was 68.6 years (median, 70; range, 38 – 89; SD, 10.6). The glaucoma patients were significantly older than the normal subjects (P ⬍ 0.001). Sensitivity and specificity of SS and SF are presented in Table 2. Both algorithms had low false-positive and false-negative rates. With regard to sensitivity, SS and SF agreed on 77 of 82 patients, SS detected a glaucomatous defect that SF missed in 3 patients, and SF detected a glaucomatous defect that SS missed in 1 patient. Both SITA algorithms failed to detect glaucomatous defects in one case (Figs 1 and 2). When the glaucomatous visual field defects were divided into mild, moderate, and severe by the HAP criteria,
Budenz et al 䡠 Sensitivity and Specificiy of SITA
Figure 2. Session 2 of patient shown in Figure 1. All three criteria for minimal abnormality were met in the full threshold (FT) field but none in Swedish interactive threshold algorithm (SITA) standard or SITA fast. Note that the area of the defect, as shown on the pattern deviation plot, is in a different location of the visual field than in the FT field in Figure 1. While these, and prior, FT tests showed an early glaucoma defect, subsequent testing with SITA standard has failed to show any glaucomatous defects over the past 4 years. CPSD ⫽ corrected pattern standard deviation; GHT ⫽ glaucoma hemifield test; MD ⫽ mean deviation; PSD ⫽ pattern standard deviation; SF ⫽ SITA fast.
the sensitivity in detection of mild defects was 92% for SS and 85% for SF. In moderate and severe defects, sensitivity was 100% for both SS and SF. Complete results are provided in Table 3. The average MD for mild, moderate, and severe fields by the HAP criteria were ⫺4.06 dB (SD, 1.78), ⫺7.63 dB (SD, 6.14), and ⫺16.74 dB (SD, 4.91), respectively. Using cluster analysis to divide fields by MD, SS and SF had a sensitivity of 95% and 91%, respectively in mild defects, and both were 100% sensitive in moderate and severe defects (Table 4). The average MD for mild, moderate, and severe fields in the cluster analysis were ⫺4.62 dB (SD, 2.29), ⫺14.17 dB (SD, 2.29), and ⫺23.32 dB (SD, 3.03), respectively.
Table 5 is a summary of test-taking time for each of the 3 algorithms. The SITA standard reduced test-taking time by 52% in normal subjects and 47% in glaucoma patients. The SITA fast reduced test-taking time by 72% in normal subjects and 65% in glaucoma patients. Both SS and SF reduced test-taking time at a statistically significant level in normal and glaucoma subjects (P ⬍ 0.001). Results of global indices produced by the STATPAC program of each algorithm are presented in Tables 6 and 7. Mean deviation values were 0.4 dB and 0.8 dB better in SS and SF fields, respectively, compared with FT in normal subjects (P ⬍ 0.001), and 0.7 dB and 1.2 dB in SS and SF fields, respectively, compared
1055
Ophthalmology Volume 109, Number 6, June 2002 Table 2. Sensitivity and Specificity of SITA Standard and SITA Fast Algorithms
Sensitivity Specificity
SITA Standard
SITA Fast
80/82 98% 86/90 96%
78/82 95% 86/90 96%
SITA ⫽ Swedish interactive threshold algorithm.
Discussion Despite the widespread adoption of the SITAs for the detection and follow-up of glaucomatous visual field defects, relatively little has been published on the sensitivity of these two algorithms compared with full threshold perimetry, long considered the “gold standard” in visual field testing. The sensitivity and specificity of any new test, compared with a gold standard, will always fall short of 100%, as in our study. This does not mean that full threshold testing is better at detecting glaucoma than the SITA algorithms. In fact, it is possible that the faster SITA algorithms could be more sensitive to the detection of early glaucoma than the full threshold algorithm. Because the diagnosis of glaucoma relies on a constellation of clinical signs (progressive characteristic optic nerve cupping usually accompanied by characteristic visual field defects), currently there is no gold standard for the diagnosis of glaucoma. However, in deciding whether individual patients have characteristic visual field loss, full threshold automated perimetry is currently considered the gold standard in clinical trials and clinical practice. Any new perimetric algorithm must be compared with this standard. Sekhar et al14 compared the sensitivity of these two algorithms compared with full threshold. In their study of 48 glaucoma patients using full threshold as the gold standard, the SITA standard algorithm yielded a sensitivity of 95%, and the SITA fast yielded a sensitivity of 93%. Their criteria for abnormality did not include the corrected pattern standard deviation/PSD or point-wise criteria used in the present study, but only the GHT; and a GHT that was borderline was considered positive in that study. Despite these differences, the sensitivity of the two new algorithms was similar to the 98% and 95% sensitivity obtained with the SITA standard and SITA fast in the current study. No attempt was Table 3. Sensitivity by the Hodapp, Anderson, Parrish Severity Score
1056
Mild Moderate Severe
No.
SITA Standard
SITA Fast
43 27 12
41/43 95% 27/27 100% 12/12 100%
39/43 91% 27/27 100% 12/12 100%
SITA ⫽ Swedish interactive threshold algorithm.
with FT in glaucoma patients (P ⬍ 0.001). Pattern SD values were measured as slightly better in the normal group using SS and SF compared with FT (P ⬍ 0.001), but were not statistically different in glaucoma patients.
Mild Moderate Severe
Table 4. Sensitivity by Cluster Analysis Severity Score
No.
SITA Standard
SITA Fast
26 20 36
24/26 92% 20/20 100% 36/36 100%
22/26 85% 20/20 100% 36/36 100%
made to describe or separate their visual fields by severity, making it difficult to discern the severity of defects in their patients. In a cohort composed of earlier glaucoma defects, for example, a lower sensitivity would be expected. In the current study, two methods were used to characterize the severity of visual field defects. The standard clinical criteria used by us22,23 and Prevent Blindness America24 showed better sensitivity with the SITA standard algorithm than the SITA fast algorithm in mild glaucoma defects (92% vs. 85%). However, the cluster analysis, which is a statistical method of separating fields into three groups by MD, yielded comparable sensitivity between the SITA standard and SITA fast algorithms (95% vs. 91%) in patients with mild defects. The slightly lower sensitivity to mild defects in the HAP severity groupings may be due to the fact that the HAP criteria includes a grading of defects based partly on proximity of the defect to fixation (Table 1). For example, a patient with a small but deep (⬍15 dB in 1 location) defect within 5° of fixation would be classified as moderate with the HAP grading scale, even if the defect had an MD less than ⫺6 dB. Sharma et al15 reported the sensitivity and specificity of SITA standard in 102 patients. Their study included patients with ocular hypertension and glaucoma suspects. Only one test was performed on each subject using full threshold and SITA standard programs and seven different criteria for abnormality were assessed. Sensitivity for detecting a glaucoma defect ranged from 83% to 93%, depending on the criteria used for identifying glaucomatous defects, and specificity ranged from 79% to 96%. Time saved with SITA standard has been shown to be approximately 50%13–16 and approximately 70%14,17,18 with SITA fast compared with full threshold. Our results were similar for both normal subjects (52% with SS and 72% with SF) and glaucoma patients (47% with SS and 65% with SF). As in previous studies,16,18,27 MD was shown to be Table 5. Test-taking Time in Normal and Glaucoma Patients
Normal Glaucoma
Full Threshold
SITA Standard
SITA Fast
13.7 ⫾ 1.4 16.5 ⫾ 2.8
6.6 ⫾ 0.7(P⬍.001)* 8.8 ⫾ 1.4(P⬍.001)*
3.9 ⫾ 1.0(P⬍.001)* 5.8 ⫾ 1.3(P⬍.001)*
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean minutes ⫾ standard deviation. *P values are compared with full threshold.
Budenz et al 䡠 Sensitivity and Specificiy of SITA Table 6. Mean Deviation (dB) in Normal and Glaucoma Patients Full Threshold
SITA Standard
Table 7. Pattern Standard Deviation (dB) in Normal and Glaucoma Patients
SITA Fast
Full Threshold 2.0 ⫾ 0.4 8.4 ⫾ 3.3
SITA Standard
SITA Fast
Normal ⫺0.4 ⫾ 1.4 0.03 ⫾ 1.3 (P⬍.001)* 0.4 ⫾ 1.4 (P⬍.001)* Glaucoma ⫺10.6 ⫾ 7.1 ⫺9.9 ⫾ 7.4 (P⬍.001)* ⫺9.4 ⫾ 6.9 (P⬍.001)*
Normal Glaucoma
1.8 ⫾ 0.5 (P⬍.001)* 1.6 ⫾ 0.4 (P⬍.001)* 8.6 ⫾ 4.1 (P⫽.35)* 8.2 ⫾ 3.9 (P⫽.18)*
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean ⫾ standard deviation. *P values are compared with full threshold.
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean ⫾ standard deviation. *P values are compared with full threshold.
slightly better in the SITA algorithms compared with full threshold. Wild et al18 compared MD in normal and glaucoma16 subjects using full threshold, SITA standard, SITA fast, and FASTPAC algorithms and found marginally better values with the SITA algorithms than with the full threshold or FASTPAC algorithms. In a study of 330 normal subjects, Bengtsson and Heijl27 found that MD was 1.2 dB higher using SITA standard than full threshold, and MD was 1.6 dB higher using SITA fast than full threshold. In the present study, PSD values were slightly better using the two SITA programs compared with full threshold in normal subjects, but were not significantly different in glaucoma patients, suggesting that this parameter may be compared if follow-up fields are obtained with a SITA test in a patient who has previously undergone testing with full threshold. This is one of several parameters used in assessing progression of glaucoma; an increasing PSD value is a sign of increasing localized field loss by far the most common type in glaucomatous progression. Heijl et al28 performed a retrospective study of 31 glaucoma patients who had full threshold and SITA standard algorithms, and found that comparisons of fields based on percentile deviations (total or pattern deviation probabilities) provided the best basis for crossalgorithm comparisons. Statistical help based on a direct empirical method for making cross-strategy comparisons has been developed, but is still in clinical evaluation.18 In the absence of such help, we prefer to obtain full threshold tests as follow-up tests in patients with glaucomatous field loss previously followed with full threshold, unless the patient is willing to obtain a SITA test to establish a new baseline. The HAP severity score,23 based on clinical observation of the size, depth, and proximity of defect to fixation, is very easy to use and has been adopted by Prevent Blindness America.24 It has also been used to determine the sensitivity and specificity of frequency doubling technology.25 The distribution of severity was fairly equal, suggesting a reasonable distribution of glaucomatous defects in this study population. The cluster analysis performed on MD, dividing defects into mild, moderate, and severe, revealed a much higher proportion of mild defects (43) compared with moderate (27) and severe (12) defects. Given this distribution of MDs, both algorithms performed extremely well in the detection of glaucomatous defects. In this study, we were careful in making comparisons between normal and glaucoma subjects for several reasons. First, the glaucoma patients were, on average, 15.7 years older than normals, despite efforts to recruit normal subjects
who were approximately in the same age group of glaucoma patients. Second, the normal subjects in this study were inexperienced visual field takers, which may affect MD and reliability parameters. Previous attempts to shorten test-taking time in automated perimetry include the use of suprathreshold testing and FASTPAC. Suprathreshold testing may be appropriate for screening purposes in neuroophthalmic disease,29 but the sensitivity and specificity has been shown to be relatively poor in glaucoma.30,31 The FASTPAC is similar to the full threshold testing method except that changes in stimulus intensity are made in 3 dB steps rather than 2 dB steps, and threshold is only crossed once. Some investigators have found that FASTPAC has similar sensitivity and specificity compared with full threshold testing.30 Others have found that FASTPAC underestimates the severity of glaucomatous defects and is subject to greater intratest variability,32,33 which may cause problems with the detection of early relative field defects and with the detection of progression. Whichever strategy is used to test a glaucoma patient, we recommend that follow-up visual fields be performed with the same strategy, because comparing tests performed with different strategies may result in mistakes in judging progression. In summary, the current study demonstrates excellent sensitivity and specificity for the new SITA algorithms in detecting glaucomatous visual field defects using full threshold as the gold standard. Because of the considerable time savings with these newer test algorithms, it may be that SITA is producing a more accurate representation of the visual field in glaucoma patients because reduction in testtaking time results in less fatigue and reduced retinal fatigue.10 –12 Future prospective longitudinal studies are needed to determine whether these new testing algorithms detect glaucomatous change before longer, full threshold testing. However, there is substantial evidence that suggests the SITA algorithms provide high enough sensitivity and specificity to replace full threshold and FASTPAC testing algorithms in the detection of glaucomatous visual field defects.
References 1. Heijl A. The Humphrey Field Analyzer, construction and concepts. In: Heijl A, Greve, EL, eds. Sixth International Visual Field Symposium. Boston: W. Junk, 1985, 77– 84. Doc Ophthalmol Proc Ser; 42.
1057
Ophthalmology Volume 109, Number 6, June 2002 2. Haley MJ, ed. The Field Analyzer Primer, 2nd ed. San Leandro, CA: Allergan Humphrey, 1986. 3. Schulzer M. Errors in the diagnosis of visual field progression in normal-tension glaucoma. The Normal-Tension Glaucoma Study Group. Ophthalmology 1994;101:1589 –94. 4. The Advanced Glaucoma Intervention Study (AGIS): 1. Study design and methods and baseline characteristics of study patients. Control Clin Trials 1994;15:299 –325. 5. Advanced Glaucoma Intervention Study. 2. Visual field test scoring and reliability. Ophthalmology 1994;101:1445–55. 6. Musch DC, Lichter PR, Guire KE, Standardi CL. The Collaborative Initial Glaucoma Treatment Study: study design, methods, and baseline characteristics of enrolled patients. Ophthalmology 1999;106:653– 62. 7. Gordon MO, Kass MA. The Ocular Hypertension Treatment Study: design and baseline description of the participants. Arch Ophthalmol 1999;117:573– 83. 8. Keltner JL, Johnson CA, Quigg JM, et al. Confirmation of visual field abnormalities in the Ocular Hypertension Treatment Study. Arch Ophthalmol 2000;118:1187–94. 9. Johnson CA, Adams CW, Lewis RA. Fatigue effects in automated perimetry. Appl Opt 1988;27:1030 –7. 10. Searle AET, Wild JM, Shaw DE, O’Neill EC. Time-related variation in normal automated static perimetry. Ophthalmology 1991;98:701–7. 11. Hudson C, Wild JM, O’Neill EC. Fatigue effects during a single session of automated static threshold perimetry. Invest Ophthalmol Vis Sci 1994;35:268 – 80. 12. Heijl A, Drance SM. Changes in differential threshold in patients with glaucoma during prolonged perimetry. Br J Ophthalmol 1983;67:512– 6. 13. Bengtsson B, Heijl A. Evaluation of a new perimetric threshold strategy, SITA, in patients with manifest and suspect glaucoma. Acta Ophthalmol Scand 1998;76:268 –72. 14. Sekhar GC, Naduvilath TJ, Lakkai M, et al. Sensitivity of Swedish interactive threshold algorithm compared with standard full threshold algorithm in Humphrey visual field testing. Ophthalmology 2000;107:1303– 8. 15. Sharma AK, Goldberg I, Graham SL, Mohsin M. Comparison of the Humphrey Swedish interactive thresholding algorithm (SITA) and full threshold strategies. J Glaucoma 2000;9: 20 –7. 16. Wild JM, Pacey IE, O’Neill EC, Cunliffe. The SITA perimetric threshold algorithms in glaucoma. Invest Ophthalmol Vis Sci 1999;40:1998 –2009. 17. Bengston B, Heijl A. SITA Fast, a new rapid perimetric threshold test. Description of methods and evaluation in patients with manifest and suspect glaucoma. Acta Ophthalmol Scand 1998;76:431–7. 18. Wild JM, Pacey IE, Hancock SA, Cunliffe IA. Between-
1058
19. 20.
21. 22. 23. 24. 25. 26. 27.
28. 29. 30. 31. 32.
33.
algorithm, between-individual differences in normal perimetric sensitivity: Full Threshold, FASTPAC, and SITA. Invest Ophthalmol Vis Sci 1999;40:1152– 61. Bengston B, Olsson J, Heijl A, Rootze´ n H. A new generation of algorithms for computerized threshold perimetry, SITA. Acta Ophthalmol Scand 1997;75:368 –75. Olsson J, Bengtsson B, Heijl A, Rootze´ n H. An improved method to estimate frequency of false positive answers in computerized perimetry. Acta Ophthalmol Scand 1997;75: 181–3. Katz J, Sommer A, Gaasterland DE, Anderson DR. Comparison of analytic algortihms for detecting glaucomatous visual field loss. Arch Ophthalmol 1991;109:1684 –9. Anderson DR. Automated Static Perimetry. St. Louis. Mosby Year Book, 1992;123. Hodapp E, Parrish RK II, Anderson DR. Clinical Decisions in Glaucoma. St. Louis: Mosby,1993;52– 61. Sponsel WE, Ritch R, Stamper R, et al. Prevent Blindness America visual field screening study. Am J Ophthalmol 1995; 120:699 –708. Sponsel WE, Arango S, Trigo Y, Mensah J. Clinical classification of glaucomatous visual field loss by frequency doubling perimetry. Am J Ophthalmol 1998;125:830 – 6. Johnson RA, Wichern DW. Applied Multivariate Statistical Analysis, 3rd ed. Englewood Cliffs, NJ: Prentice Hall, 1992; 596 – 602. Bengtsson B, Heijl A. Inter-subject variability and normal limits of the SITA Standard, SITA Fast, and the Humphrey Full Threshold computerized perimetry strategies, SITA STATPAC. Acta Ophthalmol Scand 1999;77:125–9. Heijl A, Bengtsson B, Patella VM. Glaucoma follow-up when converting from long to short perimetric threshold tests. Arch Ophthalmol 2000;118:489 –93. Siatkowski RM, Lam BL, Anderson DR, et al. Automated suprathreshold static perimetry screening for detecting neuroophthalmologic disease. Ophthalmology 1996;103:907–17. Katz J, Tielsch JM, Quigley HA, et al. Automated suprathreshold screening for glaucoma: the Baltimore Eye Survey. Invest Ophthalmol Vis Sci 1993;34:3271–7. Mills RP, Barnebey HS, Migliazzo CV, Li Y. Does saving time using FASTPAC or suprathreshold testing reduce quality of visual fields? Ophthalmology 1994;101:1596 – 603. Flanagan JG, Wild JM, Trope GE. Evaluation of FASTPAC, a new strategy for threshold estimation with the Humphrey Field Analyzer, in a glaucomatous population. Ophthalmology 1993;100:949 –54. Schaumberger M, Scha¨ fer B, Lachenmayr BJ. Glaucomatous visual fields. FASTPAC versus full threshold strategy of the Humphrey Field Analyzer. Invest Ophthalmol Vis Sci 1995; 36:1390 –7.
Manuscript no. 210664.
1
Bascom Palmer Eye Institute, Department of Ophthalmology, University of Miami School of Medicine, Miami, Florida.
2
Department of Ophthalmology, University of California at Davis, Sacramento, California. Current affiliation: Devers Eye Institute, Portland, Oregon. Presented in part at the Association for Research in Vision and Ophthalmology annual meeting, Fort Lauderdale, Florida, May 2001. Supported by an unrestricted grant from Research to Prevent Blindness, New York, New York. In addition, Dr. Johnson receives research support from Humphrey Systems, Dublin, California. None of the authors has a commercial interest in Humphrey Systems. Reprint requests to Donald L. Budenz, MD, Bascom Palmer Eye Institute, 900 NW 17th Street, Miami, FL 33136.
1052
© 2002 by the American Academy of Ophthalmology Published by Elsevier Science Inc.
been shown to result in poorer results.9 –11 This effect may be even more pronounced in glaucoma patients.10,12 The Swedish interactive threshold algorithm (SITA) is a new computer program that has been developed for the Humphrey visual field analyzer II (Humphrey Systems, Dublin, CA), which reduces test-taking time. The SITA standard program (SS) has been shown to reduce test-taking time by approximately 50%13–16 and the SITA fast program (SF) by 70% compared with full threshold testing.15,17,18 Overall, the number of stimuli actually presented is reduced by 29% in normal fields and 26% in glaucomatous fields.19 This is accomplished using a combination of techniques, including the use of information about surrounding points, information about threshold values in age-matched controls and glaucoma patients at each location, changes in the pacing of the test, elimination of retest trials for the 10 points used to calculate short-term fluctuation in the full threshold algorithm, changing the way in which false positive and false negative reliability parameters are determined, and the use of a maximum likelihood procedure for estimating threshold.18 –20 The difference between the SS and SF programs is that the SF program uses a wider confidence interval for the determination of threshold estiISSN 0161-6420/02/$–see front matter PII S0161-6420(02)01047-3
Budenz et al 䡠 Sensitivity and Specificiy of SITA Table 1. Hodapp, Anderson, Parrish Visual Field Severity Score Criteria for Early Defect 1 Mean deviation no worse than ⫺6 dB 2 On pattern deviation plot, fewer than 25% of points depressed below the 5% level, and fewer than 15% of points depressed below the 1% level 3 No point within central 5° with sensitivity ⬍15 dB Criteria for Moderate Defect 1 Mean deviation worse than ⫺6 dB but no worse than ⫺12 dB 2 On pattern deviation plot, fewer than 50% of points depressed below the 5% level, and fewer than 25% of points depressed below the 1% level 3 No point within central 5° with sensitivity of ⱕ0 dB 4 Only one hemifield containing a point with sensitivity ⬍15 dB within 5° of fixation Criteria for Severe Defect 1 Mean deviation worse than ⫺12 dB 2 On pattern deviation plot, more than 50% of points depressed below the 5% level, and more than 25% of points depressed below the 1% level 3 Any point within central 5° with sensitivity of ⱕ0 dB 4 Both hemifields containing a point or points with sensitivity ⬍15 dB within 5° of fixation
mation, permitting the determination of threshold to stop sooner than in the SS program.18 Because of the time savings that SITA provides, it has been suggested that this testing algorithm should replace the full threshold algorithm,14 which has been the gold standard for detecting and following glaucomatous visual field defects for over 15 years. However, few studies have been performed to determine the specificity and sensitivity of SITA in diagnosing glaucomatous visual field defects.15,16 The purpose of the current study is to determine the sensitivity and specificity of SITA standard and SITA fast in detecting glaucomatous visual field defects using the full threshold algorithm as the reference, or “gold” standard.
Materials and Methods This study was approved by the Human Subjects Subcommittee of the Institutional Review Boards at the University of Miami and the University of California, Davis. Patients older than 17 years of age with known glaucoma, defined as characteristic cupping of the optic nerve and glaucomatous visual field defects in at least one eye, regardless of intraocular pressure level, were invited to participate. Glaucoma patients were required to be experienced visual field takers, having been tested on two or more prior occasions on the Humphrey visual field analyzer (Humphrey Systems). Patients were excluded if the visual acuity in the eye to be tested was less than 20/40. If both eyes had glaucoma, the eye to be tested was selected randomly by the investigator. Normal subjects older than 17 years of age were invited to participate if they met the same visual acuity criteria. An attempt was made to include older patients to approximate the age of the glaucoma patients being studied. Subjects in the eye care field were ineligible for participation to avoid biasing of testing. Normal subjects were excluded if they had an intraocular pressure ⬎ 22 mmHg in either eye; history of amblyopia; refractive error ⬎ than 5 diopters of sphere
or 2.5 diopters of cylinder; suspicious appearing optic nerves; history of any disease, surgery, or trauma to the eye being tested; abnormal pupillary examination or history of miotic use, or other medications that might affect pupil size; history of systemic medication use (e.g., plaquenil) that might affect the visual field; history of cerebrovascular event or diabetic retinopathy; or insulin dependent diabetes mellitus. Subjects recruited as normal who had a visual field defect associated with an identifiable cause (such as chorioretinal scar) were excluded. If both eyes were normal, the eye to be tested was selected randomly by the investigator. After obtaining written informed consent, all patients underwent the following visual field testing protocol, during at least two visits on separate days that were within 1 month of each other. At each visit, a standard full threshold (FT) test using program 30-2 and a size III white stimulus on a white background (31.5 apostilbs or 10 cd/m2) was performed using the Humphrey field analyzer I perimeter (Humphrey Systems). Calculations of the total and pattern deviation plots and global indices (mean deviation [MD] and corrected pattern deviation) were derived using STATPAC, version 9.31. Using the Humphrey field analyzer II (Humphrey Systems), SS, and SF tests using program 30-2 and a size III white stimulus on a white background were performed. Calculations of the total and pattern deviation plots and global indices (MD and pattern standard deviation [PSD]) were derived using STATPAC for SITA, version A10.1. The order of testing was alternated between patients to equalize fatigue effects among testing algorithms. However, the order of tests was kept constant for each patient over all testing sessions. The subject’s best-corrected distance refraction and age-appropriate near add was used in the lens holder and the same prescription was used at each session. Patients were required to take at least a 15-minute rest between visual field tests. Tests using a particular algorithm were performed with the same visual field machine, and each subject had the same visual field technician for all tests. Glaucoma patients on topical miotic therapy were required to be tested on miotics at all visits, and every attempt was made to perform testing an equivalent time after instillation of the miotic. Subjects who required a dilated fundus examination were dilated and examined after visual field testing was performed. An attempt was made to perform testing on glaucomatous eyes with a wide range of visual field severity based on evaluation of prior full threshold testing. Visual fields with any abnormal reliability parameter (fixation losses ⬎33%, false-positive responses ⬎33%, or false-negative responses ⬎33%) were excluded and repeat testing was performed on a subsequent date. If a subject failed to produce two reliable fields using any one algorithm within 1 month of each other, they were excluded from the study. A complete, dilated eye examination was required for all subjects within 6 months of the study as part of the eligibility determination. Normal subjects, if not seen within 6 months of the study initiation, were examined after the second session of visual field testing and excluded if any of the exclusion criteria were met as described above. Minimal criteria for glaucomatous visual field defect were as follows:21,22 glaucoma hemifield test (GHT) outside normal limits, corrected PSD (for FT algorithm) or PSD (for SS and SF algorithms) with p values ⬍5%, or a cluster of three or more points in the pattern deviation plot in a single hemifield (superior or inferior) with P values ⬍5%, one of which must have a P value ⬍1%. Any one of the preceding criteria, if found again on repeat testing, was considered sufficient evidence of a glaucomatous visual field defect. Patients were not required to meet the same criteria on repeat testing, but were required to meet any one of the three criteria on two full threshold tests within 1 month. Full threshold fields were used to classify patients into mild, moderate, or severe glaucomatous defects using two separate systems. First, the Hodapp, Anderson, Parrish (HAP) system23–25 (see
1053
Ophthalmology Volume 109, Number 6, June 2002
Figure 1. Session 1 of a patient with glaucoma as identified with the full threshold (FT) algorithm but not Swedish interactive threshold algorithm (SITA) standard or SITA fast. The pointwise criterion and glaucoma hemifield test (GHT) were met in FT testing, but none of the three criteria for minimal abnormality were met in SITA standard or SITA fast testing. CPSD ⫽ corrected pattern standard deviation; MD ⫽ mean deviation; PSD ⫽ pattern standard deviation; SF ⫽ SITA fast.
Table 1) was used on the FT fields. Second, fields were classified into mild, moderate, and severe using a k-means cluster analysis26 based on the MD for the FT fields. Demographic and clinical characteristics between the two groups were compared using chi-square tests and nonpaired two-tailed student’s t tests for categorical and continuous variables, respectively. Sensitivity and specificity of SS and SF were determined for the entire group and then separately for mild, moderate, and severe defects using the FT fields as the reference standard. Time for test completion, MD, and PSD were averaged between the two tests and compared using a paired two-tailed t test. Data analysis was performed using SPSS for Windows (release 10.0; SPSS Inc., Chicago, IL).
Results Ninety normal eyes of 90 normal subjects and 82 glaucomatous eyes of 82 patients met the inclusion and exclusion criteria. Twen-
1054
ty-seven of 90 normal subjects were male (30%) and 63 of 90 were female (70%). Forty-three of 82 glaucoma patients were male (52%) and 39 were female (48%). There were more females in the normal group than in the glaucoma group (P ⫽ 0.003). The mean age of the normal subjects was 52.9 years (median, 55; range, 19 – 89, standard deviation [SD], 16.2) and the mean age of the glaucoma patients was 68.6 years (median, 70; range, 38 – 89; SD, 10.6). The glaucoma patients were significantly older than the normal subjects (P ⬍ 0.001). Sensitivity and specificity of SS and SF are presented in Table 2. Both algorithms had low false-positive and false-negative rates. With regard to sensitivity, SS and SF agreed on 77 of 82 patients, SS detected a glaucomatous defect that SF missed in 3 patients, and SF detected a glaucomatous defect that SS missed in 1 patient. Both SITA algorithms failed to detect glaucomatous defects in one case (Figs 1 and 2). When the glaucomatous visual field defects were divided into mild, moderate, and severe by the HAP criteria,
Budenz et al 䡠 Sensitivity and Specificiy of SITA
Figure 2. Session 2 of patient shown in Figure 1. All three criteria for minimal abnormality were met in the full threshold (FT) field but none in Swedish interactive threshold algorithm (SITA) standard or SITA fast. Note that the area of the defect, as shown on the pattern deviation plot, is in a different location of the visual field than in the FT field in Figure 1. While these, and prior, FT tests showed an early glaucoma defect, subsequent testing with SITA standard has failed to show any glaucomatous defects over the past 4 years. CPSD ⫽ corrected pattern standard deviation; GHT ⫽ glaucoma hemifield test; MD ⫽ mean deviation; PSD ⫽ pattern standard deviation; SF ⫽ SITA fast.
the sensitivity in detection of mild defects was 92% for SS and 85% for SF. In moderate and severe defects, sensitivity was 100% for both SS and SF. Complete results are provided in Table 3. The average MD for mild, moderate, and severe fields by the HAP criteria were ⫺4.06 dB (SD, 1.78), ⫺7.63 dB (SD, 6.14), and ⫺16.74 dB (SD, 4.91), respectively. Using cluster analysis to divide fields by MD, SS and SF had a sensitivity of 95% and 91%, respectively in mild defects, and both were 100% sensitive in moderate and severe defects (Table 4). The average MD for mild, moderate, and severe fields in the cluster analysis were ⫺4.62 dB (SD, 2.29), ⫺14.17 dB (SD, 2.29), and ⫺23.32 dB (SD, 3.03), respectively.
Table 5 is a summary of test-taking time for each of the 3 algorithms. The SITA standard reduced test-taking time by 52% in normal subjects and 47% in glaucoma patients. The SITA fast reduced test-taking time by 72% in normal subjects and 65% in glaucoma patients. Both SS and SF reduced test-taking time at a statistically significant level in normal and glaucoma subjects (P ⬍ 0.001). Results of global indices produced by the STATPAC program of each algorithm are presented in Tables 6 and 7. Mean deviation values were 0.4 dB and 0.8 dB better in SS and SF fields, respectively, compared with FT in normal subjects (P ⬍ 0.001), and 0.7 dB and 1.2 dB in SS and SF fields, respectively, compared
1055
Ophthalmology Volume 109, Number 6, June 2002 Table 2. Sensitivity and Specificity of SITA Standard and SITA Fast Algorithms
Sensitivity Specificity
SITA Standard
SITA Fast
80/82 98% 86/90 96%
78/82 95% 86/90 96%
SITA ⫽ Swedish interactive threshold algorithm.
Discussion Despite the widespread adoption of the SITAs for the detection and follow-up of glaucomatous visual field defects, relatively little has been published on the sensitivity of these two algorithms compared with full threshold perimetry, long considered the “gold standard” in visual field testing. The sensitivity and specificity of any new test, compared with a gold standard, will always fall short of 100%, as in our study. This does not mean that full threshold testing is better at detecting glaucoma than the SITA algorithms. In fact, it is possible that the faster SITA algorithms could be more sensitive to the detection of early glaucoma than the full threshold algorithm. Because the diagnosis of glaucoma relies on a constellation of clinical signs (progressive characteristic optic nerve cupping usually accompanied by characteristic visual field defects), currently there is no gold standard for the diagnosis of glaucoma. However, in deciding whether individual patients have characteristic visual field loss, full threshold automated perimetry is currently considered the gold standard in clinical trials and clinical practice. Any new perimetric algorithm must be compared with this standard. Sekhar et al14 compared the sensitivity of these two algorithms compared with full threshold. In their study of 48 glaucoma patients using full threshold as the gold standard, the SITA standard algorithm yielded a sensitivity of 95%, and the SITA fast yielded a sensitivity of 93%. Their criteria for abnormality did not include the corrected pattern standard deviation/PSD or point-wise criteria used in the present study, but only the GHT; and a GHT that was borderline was considered positive in that study. Despite these differences, the sensitivity of the two new algorithms was similar to the 98% and 95% sensitivity obtained with the SITA standard and SITA fast in the current study. No attempt was Table 3. Sensitivity by the Hodapp, Anderson, Parrish Severity Score
1056
Mild Moderate Severe
No.
SITA Standard
SITA Fast
43 27 12
41/43 95% 27/27 100% 12/12 100%
39/43 91% 27/27 100% 12/12 100%
SITA ⫽ Swedish interactive threshold algorithm.
with FT in glaucoma patients (P ⬍ 0.001). Pattern SD values were measured as slightly better in the normal group using SS and SF compared with FT (P ⬍ 0.001), but were not statistically different in glaucoma patients.
Mild Moderate Severe
Table 4. Sensitivity by Cluster Analysis Severity Score
No.
SITA Standard
SITA Fast
26 20 36
24/26 92% 20/20 100% 36/36 100%
22/26 85% 20/20 100% 36/36 100%
made to describe or separate their visual fields by severity, making it difficult to discern the severity of defects in their patients. In a cohort composed of earlier glaucoma defects, for example, a lower sensitivity would be expected. In the current study, two methods were used to characterize the severity of visual field defects. The standard clinical criteria used by us22,23 and Prevent Blindness America24 showed better sensitivity with the SITA standard algorithm than the SITA fast algorithm in mild glaucoma defects (92% vs. 85%). However, the cluster analysis, which is a statistical method of separating fields into three groups by MD, yielded comparable sensitivity between the SITA standard and SITA fast algorithms (95% vs. 91%) in patients with mild defects. The slightly lower sensitivity to mild defects in the HAP severity groupings may be due to the fact that the HAP criteria includes a grading of defects based partly on proximity of the defect to fixation (Table 1). For example, a patient with a small but deep (⬍15 dB in 1 location) defect within 5° of fixation would be classified as moderate with the HAP grading scale, even if the defect had an MD less than ⫺6 dB. Sharma et al15 reported the sensitivity and specificity of SITA standard in 102 patients. Their study included patients with ocular hypertension and glaucoma suspects. Only one test was performed on each subject using full threshold and SITA standard programs and seven different criteria for abnormality were assessed. Sensitivity for detecting a glaucoma defect ranged from 83% to 93%, depending on the criteria used for identifying glaucomatous defects, and specificity ranged from 79% to 96%. Time saved with SITA standard has been shown to be approximately 50%13–16 and approximately 70%14,17,18 with SITA fast compared with full threshold. Our results were similar for both normal subjects (52% with SS and 72% with SF) and glaucoma patients (47% with SS and 65% with SF). As in previous studies,16,18,27 MD was shown to be Table 5. Test-taking Time in Normal and Glaucoma Patients
Normal Glaucoma
Full Threshold
SITA Standard
SITA Fast
13.7 ⫾ 1.4 16.5 ⫾ 2.8
6.6 ⫾ 0.7(P⬍.001)* 8.8 ⫾ 1.4(P⬍.001)*
3.9 ⫾ 1.0(P⬍.001)* 5.8 ⫾ 1.3(P⬍.001)*
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean minutes ⫾ standard deviation. *P values are compared with full threshold.
Budenz et al 䡠 Sensitivity and Specificiy of SITA Table 6. Mean Deviation (dB) in Normal and Glaucoma Patients Full Threshold
SITA Standard
Table 7. Pattern Standard Deviation (dB) in Normal and Glaucoma Patients
SITA Fast
Full Threshold 2.0 ⫾ 0.4 8.4 ⫾ 3.3
SITA Standard
SITA Fast
Normal ⫺0.4 ⫾ 1.4 0.03 ⫾ 1.3 (P⬍.001)* 0.4 ⫾ 1.4 (P⬍.001)* Glaucoma ⫺10.6 ⫾ 7.1 ⫺9.9 ⫾ 7.4 (P⬍.001)* ⫺9.4 ⫾ 6.9 (P⬍.001)*
Normal Glaucoma
1.8 ⫾ 0.5 (P⬍.001)* 1.6 ⫾ 0.4 (P⬍.001)* 8.6 ⫾ 4.1 (P⫽.35)* 8.2 ⫾ 3.9 (P⫽.18)*
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean ⫾ standard deviation. *P values are compared with full threshold.
SITA ⫽ Swedish interactive threshold algorithm. All values expressed as mean ⫾ standard deviation. *P values are compared with full threshold.
slightly better in the SITA algorithms compared with full threshold. Wild et al18 compared MD in normal and glaucoma16 subjects using full threshold, SITA standard, SITA fast, and FASTPAC algorithms and found marginally better values with the SITA algorithms than with the full threshold or FASTPAC algorithms. In a study of 330 normal subjects, Bengtsson and Heijl27 found that MD was 1.2 dB higher using SITA standard than full threshold, and MD was 1.6 dB higher using SITA fast than full threshold. In the present study, PSD values were slightly better using the two SITA programs compared with full threshold in normal subjects, but were not significantly different in glaucoma patients, suggesting that this parameter may be compared if follow-up fields are obtained with a SITA test in a patient who has previously undergone testing with full threshold. This is one of several parameters used in assessing progression of glaucoma; an increasing PSD value is a sign of increasing localized field loss by far the most common type in glaucomatous progression. Heijl et al28 performed a retrospective study of 31 glaucoma patients who had full threshold and SITA standard algorithms, and found that comparisons of fields based on percentile deviations (total or pattern deviation probabilities) provided the best basis for crossalgorithm comparisons. Statistical help based on a direct empirical method for making cross-strategy comparisons has been developed, but is still in clinical evaluation.18 In the absence of such help, we prefer to obtain full threshold tests as follow-up tests in patients with glaucomatous field loss previously followed with full threshold, unless the patient is willing to obtain a SITA test to establish a new baseline. The HAP severity score,23 based on clinical observation of the size, depth, and proximity of defect to fixation, is very easy to use and has been adopted by Prevent Blindness America.24 It has also been used to determine the sensitivity and specificity of frequency doubling technology.25 The distribution of severity was fairly equal, suggesting a reasonable distribution of glaucomatous defects in this study population. The cluster analysis performed on MD, dividing defects into mild, moderate, and severe, revealed a much higher proportion of mild defects (43) compared with moderate (27) and severe (12) defects. Given this distribution of MDs, both algorithms performed extremely well in the detection of glaucomatous defects. In this study, we were careful in making comparisons between normal and glaucoma subjects for several reasons. First, the glaucoma patients were, on average, 15.7 years older than normals, despite efforts to recruit normal subjects
who were approximately in the same age group of glaucoma patients. Second, the normal subjects in this study were inexperienced visual field takers, which may affect MD and reliability parameters. Previous attempts to shorten test-taking time in automated perimetry include the use of suprathreshold testing and FASTPAC. Suprathreshold testing may be appropriate for screening purposes in neuroophthalmic disease,29 but the sensitivity and specificity has been shown to be relatively poor in glaucoma.30,31 The FASTPAC is similar to the full threshold testing method except that changes in stimulus intensity are made in 3 dB steps rather than 2 dB steps, and threshold is only crossed once. Some investigators have found that FASTPAC has similar sensitivity and specificity compared with full threshold testing.30 Others have found that FASTPAC underestimates the severity of glaucomatous defects and is subject to greater intratest variability,32,33 which may cause problems with the detection of early relative field defects and with the detection of progression. Whichever strategy is used to test a glaucoma patient, we recommend that follow-up visual fields be performed with the same strategy, because comparing tests performed with different strategies may result in mistakes in judging progression. In summary, the current study demonstrates excellent sensitivity and specificity for the new SITA algorithms in detecting glaucomatous visual field defects using full threshold as the gold standard. Because of the considerable time savings with these newer test algorithms, it may be that SITA is producing a more accurate representation of the visual field in glaucoma patients because reduction in testtaking time results in less fatigue and reduced retinal fatigue.10 –12 Future prospective longitudinal studies are needed to determine whether these new testing algorithms detect glaucomatous change before longer, full threshold testing. However, there is substantial evidence that suggests the SITA algorithms provide high enough sensitivity and specificity to replace full threshold and FASTPAC testing algorithms in the detection of glaucomatous visual field defects.
References 1. Heijl A. The Humphrey Field Analyzer, construction and concepts. In: Heijl A, Greve, EL, eds. Sixth International Visual Field Symposium. Boston: W. Junk, 1985, 77– 84. Doc Ophthalmol Proc Ser; 42.
1057
Ophthalmology Volume 109, Number 6, June 2002 2. Haley MJ, ed. The Field Analyzer Primer, 2nd ed. San Leandro, CA: Allergan Humphrey, 1986. 3. Schulzer M. Errors in the diagnosis of visual field progression in normal-tension glaucoma. The Normal-Tension Glaucoma Study Group. Ophthalmology 1994;101:1589 –94. 4. The Advanced Glaucoma Intervention Study (AGIS): 1. Study design and methods and baseline characteristics of study patients. Control Clin Trials 1994;15:299 –325. 5. Advanced Glaucoma Intervention Study. 2. Visual field test scoring and reliability. Ophthalmology 1994;101:1445–55. 6. Musch DC, Lichter PR, Guire KE, Standardi CL. The Collaborative Initial Glaucoma Treatment Study: study design, methods, and baseline characteristics of enrolled patients. Ophthalmology 1999;106:653– 62. 7. Gordon MO, Kass MA. The Ocular Hypertension Treatment Study: design and baseline description of the participants. Arch Ophthalmol 1999;117:573– 83. 8. Keltner JL, Johnson CA, Quigg JM, et al. Confirmation of visual field abnormalities in the Ocular Hypertension Treatment Study. Arch Ophthalmol 2000;118:1187–94. 9. Johnson CA, Adams CW, Lewis RA. Fatigue effects in automated perimetry. Appl Opt 1988;27:1030 –7. 10. Searle AET, Wild JM, Shaw DE, O’Neill EC. Time-related variation in normal automated static perimetry. Ophthalmology 1991;98:701–7. 11. Hudson C, Wild JM, O’Neill EC. Fatigue effects during a single session of automated static threshold perimetry. Invest Ophthalmol Vis Sci 1994;35:268 – 80. 12. Heijl A, Drance SM. Changes in differential threshold in patients with glaucoma during prolonged perimetry. Br J Ophthalmol 1983;67:512– 6. 13. Bengtsson B, Heijl A. Evaluation of a new perimetric threshold strategy, SITA, in patients with manifest and suspect glaucoma. Acta Ophthalmol Scand 1998;76:268 –72. 14. Sekhar GC, Naduvilath TJ, Lakkai M, et al. Sensitivity of Swedish interactive threshold algorithm compared with standard full threshold algorithm in Humphrey visual field testing. Ophthalmology 2000;107:1303– 8. 15. Sharma AK, Goldberg I, Graham SL, Mohsin M. Comparison of the Humphrey Swedish interactive thresholding algorithm (SITA) and full threshold strategies. J Glaucoma 2000;9: 20 –7. 16. Wild JM, Pacey IE, O’Neill EC, Cunliffe. The SITA perimetric threshold algorithms in glaucoma. Invest Ophthalmol Vis Sci 1999;40:1998 –2009. 17. Bengston B, Heijl A. SITA Fast, a new rapid perimetric threshold test. Description of methods and evaluation in patients with manifest and suspect glaucoma. Acta Ophthalmol Scand 1998;76:431–7. 18. Wild JM, Pacey IE, Hancock SA, Cunliffe IA. Between-
1058
19. 20.
21. 22. 23. 24. 25. 26. 27.
28. 29. 30. 31. 32.
33.
algorithm, between-individual differences in normal perimetric sensitivity: Full Threshold, FASTPAC, and SITA. Invest Ophthalmol Vis Sci 1999;40:1152– 61. Bengston B, Olsson J, Heijl A, Rootze´ n H. A new generation of algorithms for computerized threshold perimetry, SITA. Acta Ophthalmol Scand 1997;75:368 –75. Olsson J, Bengtsson B, Heijl A, Rootze´ n H. An improved method to estimate frequency of false positive answers in computerized perimetry. Acta Ophthalmol Scand 1997;75: 181–3. Katz J, Sommer A, Gaasterland DE, Anderson DR. Comparison of analytic algortihms for detecting glaucomatous visual field loss. Arch Ophthalmol 1991;109:1684 –9. Anderson DR. Automated Static Perimetry. St. Louis. Mosby Year Book, 1992;123. Hodapp E, Parrish RK II, Anderson DR. Clinical Decisions in Glaucoma. St. Louis: Mosby,1993;52– 61. Sponsel WE, Ritch R, Stamper R, et al. Prevent Blindness America visual field screening study. Am J Ophthalmol 1995; 120:699 –708. Sponsel WE, Arango S, Trigo Y, Mensah J. Clinical classification of glaucomatous visual field loss by frequency doubling perimetry. Am J Ophthalmol 1998;125:830 – 6. Johnson RA, Wichern DW. Applied Multivariate Statistical Analysis, 3rd ed. Englewood Cliffs, NJ: Prentice Hall, 1992; 596 – 602. Bengtsson B, Heijl A. Inter-subject variability and normal limits of the SITA Standard, SITA Fast, and the Humphrey Full Threshold computerized perimetry strategies, SITA STATPAC. Acta Ophthalmol Scand 1999;77:125–9. Heijl A, Bengtsson B, Patella VM. Glaucoma follow-up when converting from long to short perimetric threshold tests. Arch Ophthalmol 2000;118:489 –93. Siatkowski RM, Lam BL, Anderson DR, et al. Automated suprathreshold static perimetry screening for detecting neuroophthalmologic disease. Ophthalmology 1996;103:907–17. Katz J, Tielsch JM, Quigley HA, et al. Automated suprathreshold screening for glaucoma: the Baltimore Eye Survey. Invest Ophthalmol Vis Sci 1993;34:3271–7. Mills RP, Barnebey HS, Migliazzo CV, Li Y. Does saving time using FASTPAC or suprathreshold testing reduce quality of visual fields? Ophthalmology 1994;101:1596 – 603. Flanagan JG, Wild JM, Trope GE. Evaluation of FASTPAC, a new strategy for threshold estimation with the Humphrey Field Analyzer, in a glaucomatous population. Ophthalmology 1993;100:949 –54. Schaumberger M, Scha¨ fer B, Lachenmayr BJ. Glaucomatous visual fields. FASTPAC versus full threshold strategy of the Humphrey Field Analyzer. Invest Ophthalmol Vis Sci 1995; 36:1390 –7.