Normal Intersubject Threshold Variability and Normal Limits of the SITA SWAP and Full Threshold SWAP Perimetric Programs

Abstract

purpose. To compare intersubject variability and normal limits of threshold values between the new Swedish interactive test algorithm short wavelength automated perimetry (SITA SWAP) and the older Full Threshold SWAP programs (Carl Zeiss Meditec, Dublin, CA).

methods. Normal reduction of differential light sensitivity with age, age-corrected thresholds, intersubject variability, and normal limits of sensitivity were calculated from SITA SWAP and Full Threshold SWAP fields obtained in 53 normal subjects between 20 and 72 years of age.

results. Age influence on threshold sensitivity was the same with the two SWAP programs. On average, sensitivity decreased by 0.13 dB per year of age. Age-corrected normal threshold sensitivity was significantly higher (P < 0.0001) for SITA SWAP than for Full Threshold SWAP. The means for a subject 45.4 years of age were 28.8 dB with SITA SWAP and 24.4 dB with Full Threshold SWAP. Intersubject variance was 22% smaller with SITA SWAP than with Full Threshold SWAP. Normal limits at the P < 5% significance level were, on average, 14% narrower with SITA SWAP than with Full Threshold SWAP using Total Deviations from age-corrected normal thresholds and 11% narrower when applying Pattern Deviation, which is intended to adjust for general depression or elevation of the field.

conclusions. SITA SWAP test results from normal eyes showed higher sensitivities than results from the older Full Threshold SWAP. This represents an increase of the dynamic range, which implies that more patients can be tested with SWAP. The smaller intersubject variability with SITA SWAP means narrower normal limits and may be associated with more sensitive probability maps.

Short-wavelength automated perimetry (SWAP) can detect glaucomatous visual field loss earlier than conventional white-on-white perimetry (WWP). ^{1 2 3 4 5} However, limitations in the technology make SWAP difficult and impractical in everyday practice. One problem with SWAP has been the length of time it takes to administer the test. ⁶ By modifying the SITA program originally developed for WWP ^{7 8} for SWAP, we obtained radically reduced test times and maintained the same precision of threshold estimates, defined as threshold test–retest variability, compared with Full Threshold SWAP and Fastpac SWAP (all terms relate to the Humphrey Field Analyzer; Carl Zeiss Meditec, Dublin, CA). ⁹  

Other problems associated with SWAP have been the reduced dynamic range and larger threshold intersubject variability than for WWP. ^{6 10} Small intersubject variability is desirable, because it results in narrow normal limits, which means that shallow defects also will be recognized and flagged as significantly depressed in probability maps. Probability maps, which are essential for interpretation of visual field test results ¹¹ are entirely based on such intersubject variability in normal populations. ¹² In Figure 1 , two fields of the same eye are displayed, one Full Threshold SWAP and one Swedish interactive test algorithm (SITA) Fast WWP. The difference in gray-tone maps of raw threshold data is mainly a scaling problem. The gray-tones originally designed for the Full Threshold WWP test have been transferred to SWAP without adjusting for differences in stimulus contrast between WWP and SWAP. The probability maps, which display significant deviations from age-corrected thresholds, appear more similar. Ocular media absorption of the SWAP stimulus, particularly in older subjects, partly explains the larger SWAP threshold intersubject variability in normal eyes compared with WWP. ^{13 14} The Pattern Deviation concept ¹¹ included in the Statpac interpretation program ¹² was designed to reduce effects of cataract, which gives a general and similar reduction of sensitivity across the field. Thus, interindividual variability caused by cataract is reduced in Pattern Deviation data. Despite transformation of raw thresholds to Pattern Deviations, the normal limits applied in the probability maps available for the Full Threshold SWAP test on the HFA are wider than those for WWP. This indicates that SWAP variability is larger than for WWP and that this is not caused by lens status only. This is in agreement with data from Wild et al., ⁶ who found that intersubject variability was 1.9 times larger with SWAP than with WWP after correction for ocular media absorption. 

We have reported the results of comparisons of test times and threshold reproducibility between SITA SWAP and Full Threshold SWAP. ⁹ The current article addresses the other SWAP aspects that we have mentioned: large intersubject variability and reduced differential light sensitivity. The purpose of this study was to calculate preliminary normal limits for the new SITA SWAP program and to compare the results with those obtained with the older Full Threshold SWAP program using the same normal database. 

Methods

Personnel at the Department of Ophthalmology, Malmö University Hospital and relatives and friends of personnel were invited to participate in the study. Invited subjects had to have no or very limited experience with SWAP perimetry and/or conventional WWP. The study was conducted in accordance with the tenets of the Declaration of Helsinki. All study subjects gave informed consent, and the Ethics Committee of the medical Faculty of the University of Lund approved the study. 

Men and women between 20 and 80 years of age with no serious eye disease, eye trauma, intraocular surgery, amblyopia, or ocular finding or medication that could affect the visual field, were eligible. Subjects with a history of stroke, diabetes, or other systemic disease that could influence the visual field were not included. 

One eye of each subject was randomly selected. All subjects underwent an ophthalmic examination including refraction and determination of best visual acuity (VA). Intraocular pressure (IOP) was measured by applanation tonometry, lens status graded by the Lens Opacities Classification System II (LOCS II), ¹⁵ and color fundus photographs were obtained with a fundus camera (TRC-NW3; Topcon, Tokyo, Japan) through a dilated pupil. Color vision was tested using Standard Pseudoisochromatic Plates part 3 (SPP3; Igaku-Shoin, Tokyo, Japan). A congenital color vision deficit was noted but was not considered to be an exclusion criterion. It has been reported that color contrast thresholds in tritan color axes are not influenced by the presence of congenital red-green defects. ¹⁶  

Visual field testing was performed at two separate visits for each subject. The minimum intervisit interval was 1 day and the maximum 2 weeks. Two tests, one 24-2 SITA SWAP test and one 24-2 Full Threshold SWAP test, were obtained at each of the two visits. Test order was randomized between individuals so that approximately 50% started with SITA SWAP. All subjects had a 15-minute break between the two SWAP tests. In addition, one conventional WWP test with SITA Fast was obtained 15 minutes or more after the two SWAP tests at the second visit. 

Subjects were excluded after examination if they had corrected VA less than 0.7; IOP of 22 mm Hg or more; unexpected eye disease or other systemic disease that could affect the visual field; suspect or pathologic optic discs; lens opacities classified by LOCS II as the sum of nuclear, cortical, and posterior subcapsular more than 2; posterior subcapsular only more than 1; or fields showing clearly artifactual test results—for example, clover leaf pattern or rim defects with sensitivities less than 0 dB. Frequencies of False-Positive or False-Negative answers had to be less than 15%, and Fixation Losses less than 20%, as determined by the blind spot method in the three tests (i.e., SITA Fast WWP, SITA SWAP, and Full Threshold SWAP). Cutoff limits for false answers were set on empiric bases—that is, the 95th percentile for normal subjects (Bengtsson B, et al. IOVS 2000;41:ARVOAbstract 2539). Subjects were excluded if Fixation Losses exceeded 20% in both the SWAP and in the WWP test. SWAP alone could not be used for determining the patients’ fixation ability because the large SWAP stimulus, Goldmann size V, often tends to overestimate Fixation Losses with the blind spot method. Such a large stimulus is often seen when exposed in the blind spot, also when fixation seems to be perfect. Each patient’s first visit was considered to be a training session, ^{17 18} and only test results from the second visit were used for calculation of results. 

Threshold data from left eyes were inverted in relation to right eye data. The two test points located in the blind spot area were excluded from all calculations. Linear regression analyses were performed to estimate age slopes of sensitivity for each test point location. The slopes were then used to calculate age-corrected thresholds. It would be natural to calculate normal random intersubject variability from point-wise distributions of the age-corrected thresholds for each test strategy, but such an approach would also be very sensitive to outliers. To reduce the influence of possible outliers intersubject variability was calculated as sums of squares of residuals (i.e., the sum of the squared difference between the expected age-corrected normal threshold and the measured threshold for each individual and test program, including all test points: S_ijk ², where i is individual, j is test program, and k is test point number. 

To compare intersubject variability between the two test programs individual ratios were calculated  

\[\mathrm{S}_{\mathrm{i}1.}^{2}\mathrm{/S}_{\mathrm{i}2.}^{2}\]

where S_i1. ² is the sum of squares of all points (.) for individual i with program 1 (SITA SWAP), and S_i2. ² is the sum of squares of all points (.) for individual i with program 2 (Full Threshold SWAP). 

The individual ratios did not follow normal (Gaussian) distributions. Therefore, a transformation using the natural logarithm was applied (Fig. 2) , before performing a one-sample t-test of ratios. 

Total Deviations, defined as the deviation in dB from the calculated age corrected normal threshold value, were calculated both for SITA SWAP and Full Threshold SWAP. Pattern Deviations were defined in the same way as in the original Statpac program ¹² (i.e., by using the 85th percentile of the Total Deviation in each single test to adjust for general depression or elevation of the field). Distributions of Total Deviations and Pattern Deviations were plotted for each test point and SWAP program. The 5th percentile of all deviation distributions was calculated using the method of linear interpolation. In this way, the empiric normal limits at the P < 5% level for both SITA SWAP and Full Threshold SWAP were determined. 

Results

Sixty-six subjects were recruited. Thirteen of these were rejected, because they met at least one of the exclusion criteria (Table 1) . Thus 53 subjects, 38 women and 15 men, were included. The mean age was 45 years, ranging from 20 to 72. VA ranged from 0.7 to 1.0; 49 subjects had a VA of 1.0. 

All but three subjects answered correctly to all plates of the SPP3 color vision test. The test confirmed reported congenital red-green color deficits in two subjects. Both performed better perimetrically than the group average, but mean sensitivities were well within the 95% confidence interval. Another subject, a 51-year-old healthy woman, was not able to give a correct answer to one of the plates testing blue-yellow sensitivity. Her VA was 1.0, the lens was clear, and the disc and the central fundus were normal, and she was not taking medications known to affect color vision. The threshold data of both SITA SWAP and Full Threshold SWAP were slightly worse than the group average, but within the 95% confidence interval. She was retested 3 months later and passed the same color vision test without any errors. These three subjects were included in the database. 

The mean sensitivity of a normal visual field for a subject 45.4 years of age was 24.4 dB with Full Threshold SWAP and 28.8 dB with SITA SWAP. This difference was significant (P < 0.0001; Student’s t-test) and of similar magnitude at all locations across the field. Age slopes of sensitivities were linear and the same for the two SWAP programs (Fig. 3) . The mean square error of residuals around the regression slope was 8.44 for Full Threshold and 5.18 for SITA SWAP, indicating smaller intersubject variability with SITA. 

Intersubject variability calculated as individual ratios of point-wise sums of squares of residuals was smaller with SITA SWAP than with Full Threshold SWAP (Fig. 2) . The variance was 22% smaller, on average, with SITA SWAP. 

Distributions of Total Deviations and Pattern Deviations were negatively skewed. Distributions at peripheral points were more skewed than at paracentral points (Fig. 4) , just as previously found for WW fields. ¹² Average skewness of all points was −0.51 dB for Total Deviation and −0.72 dB for Pattern Deviation. Distributions of both Total and Pattern Deviations were narrower with SITA SWAP than with Full Threshold SWAP. The Total Deviation 5th percentile indicating the P < 5% normal limit was, on average, 14% closer to 0 dB, the age-corrected normal deviation value, with SITA SWAP than with Full Threshold SWAP. The average numerical difference was 1.05 dB, indicating that defects have to be more than 1 dB deeper with Full Threshold to be flagged as significantly depressed in the Total Deviation probability map at the investigated P < 5% level. The numerical difference was significant (P < 0.0001 paired t-test; distribution of differences of percentiles at all test points between SITA SWAP and Full Threshold SWAP was Gaussian (Kolmogorov-Smirnov P = 0.87). Similarly, the 5th percentile of Pattern Deviation was 11% closer to the normal value with SITA SWAP, mean difference 0.84 dB (P < 0.0001 paired t-test; distribution of differences was Gaussian; Kolmogorov-Smirnov P = 0.61). 

Discussion

The dynamic range increased with 4 dB with SITA SWAP compared with Full Threshold SWAP. This effect of SITA is desirable, because it makes meaningful SWAP testing possible for more patients. One reason for the increased available range is differences in estimation of threshold values. In SITA the threshold values are defined as that stimulus intensity that has a 50% chance to be detected, ⁷ whereas Full Threshold defines the threshold as the intensity of the last perceived stimulus at the end of a staircase procedure in which intensities are altered in 4-dB steps until a first reversal and then in 2-dB steps. ¹⁹ Expected threshold estimates would be more similar if Full Threshold had defined the threshold as the mean of the last seen and not seen stimulus intensity. Reduced visual fatigue ²⁰ of the shorter SITA test is also a likely contributor to the higher sensitivity. Visual fatigue reduces threshold sensitivity along with increasing test time. 

Normal limits for SITA SWAP were narrower than for Full Threshold SWAP. This result was not surprising, because the same relationship has been seen when comparing the SITA WWP limits with those of Full Threshold WWP. Those limits were based on normal data collected in a multicenter fashion for comparison of SITA and Full Threshold WWP. ²¹ The results of the present study were based on a considerably smaller number of normal subjects from just one center, but our purpose was not to establish definite normal limits for SITA SWAP, merely to compare SITA SWAP limits with Full Threshold SWAP limits. 

Although distributions of Total Deviation and Pattern Deviation were bell shaped, we used point-wise empiric distributions to estimate the 5th percentiles. In fact, the distributions were negatively skewed. The tails of the distribution are of particular interest when establishing normal limits, assuming that normal Gaussian distribution would result in false normal limits at points with negatively skewed distributions (Fig. 4) . 

Normal databases ideally should be randomly selected from the population that will be studied. There is always a risk associated with the use of healthy volunteers (e.g., hospital staff) who might be well acquainted with the test procedure. Their performance might be supernormal and produce very narrow limits, thus giving the test too high a sensitivity, but poor specificity. Well aware of that risk, we invited only staff with no or very limited experience with perimetry. Presumably healthy friends and relatives to hospital staff were also invited. Among these subjects, the study examination revealed previously undetected early glaucoma in both eyes of one subject, with typical rim notches and corresponding visual field defects. Others were excluded because of high frequencies of False-Positive or False-Negative answers. For exclusion based on deficient fixation, we required Fixation Losses of more than 20%, not only in the SWAP tests, but also in the WWP test, in that SWAP often tends to overestimate Fixation Losses because it employs the blind spot method using stimulus size Goldmann V. Such large stimuli are often seen by the viewer when exposed in the blind spot, even when the operator judges fixation to be perfect. 

Three subjects were excluded because of significant cataract. We defined significant cataract as a sum of LOCS II grading more than 2 or more than 1 if the cataract was of the posterior subcapsular type, because it has been shown that subcapsular cataracts have larger effects on SWAP fields than other types of cataract. ²² Nevertheless, all types of cataract affect SWAP sensitivity. 

In analogy with earlier Statpac programs including normal limits for perimetric threshold values, we used normal subjects with some but not extensive experience with automated perimetry. Limits for glaucoma progression should be based on glaucoma subjects in change probability calculation, ²³ but limits based on normal subjects are applied when differentiating between normal and abnormal visual fields. 

If a population-based random selection has not been used to create a normal sample, whether the normal database is representative of the target population can always be debated. Our normal subjects showed considerably (22%) narrower P < 5% normal limits with Full Threshold SWAP than did those applied in the commercially available Full Threshold SWAP Statpac of the Humphrey Field Analyzer. This comparison may suggest that our subjects performed better than an average population. A comparison with the standard limits for SITA WWP tests indicated, on the contrary, that the subjects used in the present study may be very representative of a normal population Thus, the established P < 5% limits of the SITA Fast WWP program were almost identical with the SITA Fast WWP P < 5% limits obtained in the present study. The established SITA Fast WWP limits were based on 333 normal subjects collected at 10 different centers all over the world. ²¹  

In conclusion, our results showed that SITA SWAP had larger dynamic range and narrower normal limits than the older Full Threshold SWAP program. This means that more patients can be tested with SWAP and that shallower depressions are needed with SITA SWAP for statistical and clinical significance. Short test times and sensitive probability maps can facilitate the use of SWAP for detection of early glaucomatous visual field defects in clinical practice. 

 Supported by funds from Carl Zeiss Meditec, Dublin, California, to Lund University.

 Submitted for publication November 27, 2002; revised April 15 and June 13, 2003; accepted June 25, 2003.

 Disclosure: B. Bengtsson, Carl Zeiss Meditec (F, C, P); A. Heijl, Carl Zeiss Meditec (F, C, P)

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

Two fields of the same eye obtained on the same day: one Full Threshold SWAP (A) and one SITA Fast WWP (B). The gray-tone maps of raw threshold values, as they are displayed on the printout of the perimeter, appear quite different. The probability maps are more similar.

Figure 2.

 

Distribution of individual ratios of variance between SITA SWAP and Full Threshold SWAP (top). After applying the natural logarithm (ln) for transformation of the variance ratios the distribution turned out to be Gaussian (bottom; Kolmogorov-Smirnov P = 0.79). The mean ln ratio differed significantly from 0 (one group t-test P = 0.005), indicating that the SITA SWAP variance was significantly smaller than the Full Threshold SWAP variance.

Table 1.

 

Excluded Subjects

	Subjects (n)
Sum of LOCS II gradings >2 or LOCS II posterior subcapsular >1	3
Fixation loss	6*
False-positive answers	4, †
False-negative answers	1, ‡
Unexpected eye disease	1
Total number	13

Figure 3.

 

Age-dependent threshold sensitivity decay, plotted as individual Mean Sensitivity over age, show parallel slopes for SITA SWAP (top) and Full Threshold SWAP (bottom). The slopes were −0.13 dB per year of age for both programs, whereas the intercept was 34.6 dB for SITA SWAP and 30.2 dB for Full Threshold SWAP, indicating a height difference of slightly more than 4 dB. The larger dispersion around the regression line in the Full Threshold SWAP plot (bottom) is in agreement with the statistical analyses showing larger intersubject variability.

Figure 4.

 

(A) Distributions of deviations from age-corrected thresholds (Total Deviations) of all included subjects at two test points located at 3° and 24° from fixation. Full Threshold (FT) SWAP had wider distributions than SITA SWAP. This was true at almost all locations, resulting in narrower normal limits with SITA SWAP than with Full Threshold SWAP. If assuming normal (Gaussian) distribution at the SITA SWAP peripheral point (the top right histogram), the 5% normal limit would be at −5.7 dB, whereas the empirically derived 5% limit is at −6.5 dB, despite the fact that skewness at this point was only −0.15, less than at most test point locations. (B) Distributions of Pattern Deviation values at the same two test points located at 3° and 24° from fixation. Also, Full Threshold (FT) SWAP gave wider distributions and thereby wider normal limits than SITA SWAP.

The authors thank Johnny Olsson, PhD, Lund, Sweden, for providing expertise in the statistical analyses.