| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 From the Departments of Ophthalmology and Neurology, University of Iowa, College of Medicine, Iowa City; and the 2 Department of Psychology, Eastern Illinois University, Charleston.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionReferences |
|---|
METHODS. Eighteen volunteers in two age groups participated in a series of three test sessions. Each session consisted of a Humphrey Field Analyzer 30-2 test, a questionnaire, and a customized test program using a Humphrey perimeter to construct frequency of seeing (FOS) curves from which thresholds were calculated, and a descriptive measure of response criterion was derived. The FOS curves were obtained at a central and a peripheral test location within the same test session. The three test sessions differed only by the instructions given. The instructions were adapted from those listed in the manufacturer’s instruction manual and were designed to influence participants to respond to the stimuli in a conservative, liberal, or neutral manner.
RESULTS. For the 30-2 threshold test, a significant difference in mean deviation was found among the three instruction types (P = 0.001) and between the two age groups (P = 0.001). Although differences were small in the younger subjects (2.04 dB), the means for the responses from liberal to conservative differed by 6.57 dB in the older subjects. Thresholds obtained in a peripheral location by the customized threshold test were found to differ significantly between the age groups (younger group mean, 31.0 dB; older group mean, 27.2 dB) and among the instruction types (liberal mean, 30.9 dB; conservative mean, 28.1 dB; and neutral mean, 30.3 dB; P < 0.001). The descriptive measurement of response criterion suggests that a difference in criteria occurred as a result of the instructions in both peripheral and central locations for both age groups (P = 0.0001). In addition, according to self-reports, liberal instructions caused participants to be more likely to respond, whereas the conservative instructions caused them to be more reluctant to respond.
CONCLUSIONS. Perimetrists’ instructions can significantly affect obtained automated perimetry thresholds.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Although cognitive factors such as response biases (i.e., a participant’s willingness to report detection of a stimulus) can affect threshold measurements, there is no reliable method to quantify or describe the effects of these response biases on measurements of visual sensitivity in automated perimetry. In the field of psychophysics an attempt to quantify the effects of response bias lead to the development of signal detection theory. The main assumption of signal detection theory is that all stimuli are embedded in noise (both internal and external to the system). Thresholds are at best momentary and are not merely a function of the sensitivity of the system detecting the stimuli. Consequently, any attempt to measure the sensitivity of a detection system with a threshold, according to the theory, may be confounded by response biases. The criterion measure c was developed by psychophysicists as a method to measure and quantify these nonstimulus factors (including response biases, attention, motivation) that affect responding to the stimuli.1 2
It is proposed, by signal detection theory, that response biases can result in responses from participants or decision makers that range from very liberal (responding when uncertain that a stimulus has been presented) to very conservative (responding only when certain a stimulus has been presented). This response criterion, which varies among individuals, can easily be changed in response to the demands of a specific situation and is defined automatically each time a response to a perimetric stimulus is required.2 This suggests that changes in a response criterion during or between testing sessions can influence sensitivity measurements and therefore the predictive value of perimetry testing.
In clinical vision testing, the impact on visual sensitivity created by response bias has been studied for high-pass resolution perimetry (ring test) and contrast sensitivity testing.3 4 5 6 7 8 However, although response bias may account for a portion of perimetric variability, the topic has received limited attention in perimetry literature. In a typical testing situation, there is uncertainty about how much response biases are affecting the observer. According to the principals of signal detection theory, it can be predicted that, with differential light sensitivity, automated perimetry adopting a conservative criterion would result in a constricted visual field or abnormally low sensitivity in the absence of visual system damage. A liberal response criterion, therefore, would result in abnormally high visual sensitivity that would not be truly indicative of visual system function. In both situations, the reliability of the visual field test would be reduced.
To understand better the effects of response bias on perimetric thresholds, we manipulated the participant’s natural or neutral criterion by using specifically worded instructions to help set either a liberal or a conservative response criterion. In addition, to study the impact of response biases on age differences in automated perimetry outcomes, we included two groups of participants ranging from 24 to 35 and 62 to 71 years of age.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
Each participant met the following criteria for inclusion in the study: best corrected Snellen visual acuity of at least 20/20, refractive error no greater than 5 D sphere and 3 D cylinder, undilated pupil size of at least 3 mm, and no history of eye disease other than refractive error in the selected eye. Potential participants who were experienced with automated perimetry were excluded. The age of the participants in the younger group ranged from 24 to 35 years (mean, 29.7 years) and in the older group from 62 to 71 years (mean, 67.5 years). There were 12 participants in the younger group and 6 in the older group and an equal number of men and women in each age group.
All participants had an ophthalmic examination consisting of a confrontation visual field examination and a direct ophthalmoscopic examination. Potential participants were asked to complete a vision and health history questionnaire to exclude volunteers with visual dysfunction. We did not inform study participants of the nature of the study before participation. Participants were told only that they were to follow the specific instructions given each day to the best of their ability. After the final test session, participants were debriefed about the intent of the study. Each participant was given a full-field threshold test (Humphrey 30-2; Humphrey Instruments, San Leandro, CA) using the test instructions we defined as neutral, found in the owner’s manual.9 This practice test was intended to screen for visual field defects and to decrease the effects of perimetric learning.
Apparatus and Stimuli
Humphrey Full-Field Threshold Test.
We used the full-field threshold test of the field analyzer (model 640,
program 30-2; Humphrey) with a white Goldmann size III stimulus varying
over a 4-log-unit range. This white background was calibrated to 31.5
apostlib (asb). Each participant’s appropriate near correction was
used with an additional refraction performed at the perimeter.
Threshold values at test locations throughout the central 30° field
were determined with a 4 dB–2 dB staircase procedure. For each
participant, the number of false-positive responses (responses when no
stimuli were presented), false-negative responses (no response to a
9-dB brighter stimulus than had previously been seen), fixation losses,
and short-term fluctuation at selected test locations were calculated
by the test instrument. After completion of the full-field threshold
test, each participant was given a minimum of 10 minutes’ rest before
the second threshold test.
Customized Threshold Test.
After the rest period, two additional thresholds were obtained to
confirm the effect of instructions with a customized threshold test
program using the method of limits. An IBM-compatible personal computer
(486/66 mHz; Hewlett–Packard, Palo Alto, CA) was used to control the
field analyzer externally. FOS curves were obtained for two
predetermined locations chosen to represent central and peripheral
locations.10
White Goldmann size III stimuli, varying in
intensity from 40 to 20 dB (in 2-dB increments) were presented 20 times
each in random order on a background of 31.5 asb at two locations
(-27°, 3° and 3°, -3°). Eighty-eight catch trials (17% of
the total number of trials), in which no stimulus was presented, were
included as a false-positive response measure. The technician monitored
the participant’s fixation, and the appropriate near correction was
used. If the participant did not respond within 2 seconds, the stimulus
was considered not seen.
FOS curves were then constructed to obtain a threshold for each subject by calculating cumulative gaussian functions and a least-squares fit by computer (Excel solver function; Microsoft, Redmond, WA; the FOS curves were fitted with a cumulative gaussian function). This function changed the mean ± SD of the curve fit until the R2 value was maximized. The resultant mean of the fitted distribution corresponds to the 50% correct threshold for the test location (defined as the stimulus intensity corresponding to the 50% FOS point of the fitted curve). The SD of the cumulative gaussian function (an index of the maximum slope of the FOS function) and the coefficient of determination or goodness of fit (R2) were calculated for each FOS curve.
Questionnaire.
We designed a survey to ascertain how and to what degree several
factors influenced participants’ responses while taking the full-field
threshold test. Participants were instructed to rate the extent to
which anxiety, the test instructions, and their desire to follow the
instructions correctly affected test performance by choosing a number
on the scale between 0 (strongly disagree) and 10 (strongly agree). The
specific wording of the questionnaire items can be found in Table 1
.
|
During the first testing session, the standard instructions recommended by the manufacturer’s instruction manual were used to allow participants to respond according to their natural or neutral response criterion.9 The verbatim instructions used are listed in Table 2 . All participants received the neutral instructions for the first session, whereas the remaining two sessions were counterbalanced between liberal and conservative instructions. The instructions were read from cue cards by a single experienced technician (KEK) before each threshold test to ensure that the same instructions were given to each participant. It is important to note that the only instructions given before and during test sessions were those listed in Table 2 . A shortened version of the initial instructions was repeated during the short intratest breaks to remind participants of the instructions or in the event of poor test performance.
|
To obtain thresholds, FOS curves were derived by plotting the proportion of correct responses as a function of each stimulus intensity for each test location separately from the calculations used to measure response criterion. Threshold was defined as the stimulus intensity corresponding to the 50%-seen point of the fitted curve.
Results from the customized threshold test (threshold and separate criterion calculations), questionnaires and Humphrey 30-2 (mean deviation [MD], false-negative responses, false-positive responses, fixation losses, and short-term fluctuations) were analyzed using analysis of variance (ANOVA) with a 2 (age) x 3 (test instructions) split plot design. Student–Newman–Keuls was used for post hoc test analyses. Statistical significance was set at P < 0.05. All statistical analyses were performed by computer (SAS ver. 6.12 software; SAS, Cary, NC).
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
|
|
|
Differences were found between the two age groups regardless of instruction type in the number of false-negative response errors calculated by the Humphrey 30-2 test. The Humphrey Field Analyzer includes trials 9 dB brighter than a previously seen stimulus as its normal testing procedure for false-positive catch trials. The average number of these false-negative responses for the older group was 11 but only 5 for the younger group (P = 0.026). There was also a difference between the older and younger groups for the short-term fluctuation measure (P = 0.002), with the older group (1.74 dB) having a slightly higher value than the younger group (1.27 dB). The younger group had a greater number of false-positive response errors for both the neutral (young, three errors; old, two errors) and liberal (young, seven errors; old, five errors) instruction sets, but these differences in the number of errors failed to reach statistical significance when the two age groups were compared (P = 0.86).
The Glaucoma Hemifield Test (GHT) classified Humphrey Field Analyzer 30-2 test results as abnormal for three of the six older participants for the conservative instruction set and three as abnormal with the neutral set. In addition, four of the neutral instruction program 30-2 results were classified as borderline according to the GHT. None of the tests for the younger group, for any instruction type, was classified by the GHT as either borderline or abnormal. As shown in Table 3 the older participants for the conservative and neutral instruction types had MD, pattern standard deviation (PSD), and corrected pattern standard deviation (CPSD) indices flagged with P values of less than 10% more frequently than the younger group.
Customized Threshold Test
In the peripheral test location (-27°, 3°) significant
differences in threshold (50% seen intensity) were obtained among
instruction types (P = 0.0001) and between the two age
groups (P = 0.0001). The fits
(R2) for the cumulative gaussian
functions used to calculate the thresholds ranged from 0.94 to 1.0
(Table 4)
. Liberal instructions led to the highest visual sensitivity
(mean, 30.9 ± 3.05 dB), whereas conservative instructions (mean,
28.1 ± 2.49 dB) led to the lowest. The older adults overall
averaged lower visual sensitivity (mean, 27.2 ± 1.92 dB) than the
young participants (mean, 31.0 ± 2.29 dB). Figure 3 shows the mean threshold values for each participant in both the older
and younger age group at the both the central and peripheral test
locations for all three instruction types.
|
|
The descriptive measurement of response criterion suggested differences in the degree of response biases affecting participants in both the peripheral and central locations for all instruction types and between age groups (Fig. 4) . In the peripheral location a significant difference was found between the conservative (mean, 1.3 c) and both neutral (mean, 1.0 c) and liberal (mean, 0.87 c) instructions (P = 0.0001). Centrally, however, only differences in criterion between the conservative and liberal instructions were significant (P = 0.039). At both test locations, the older group responded more conservatively than the younger participants regardless, of instruction type (P = 0.0001, Fig. 4 ).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
The impact of instructions on threshold values is of obvious interest for clinical perimetry, because the ability to detect generalized improvement or worsening of the visual field is limited by long-term or interest variability. This long-term variability has been partitioned into homogeneous and heterogeneous components12 resulting from the fluctuation of repeated measurements at the same test location at different times. The heterogeneous component refers to local fluctuations, whereas the homogeneous component refers to effects involving the entire visual field manifested by a change in total mean deviation. Our findings suggest that instructions given by a perimetrist are responsible for some of the homogeneous component of long-term variability.
Overall, all participants had higher visual sensitivity values on both the Humphrey 30-2 threshold test and on the customized method of limits test when given liberal instructions and lower sensitivity with the conservative instructions. Although the differences in thresholds obtained from the younger group were statistically significant, from a clinical perspective, in most cases the results were not of the magnitude that commonly signals clinical change (liberal mean, 0.55 ± 0.98 dB; conservative mean, -1.55 ± 1.4 dB; and neutral mean, -0.34 ± 0.86 dB). However, a clinically significant change (Fig. 2) could easily be mistakenly concluded in some of our older healthy participants (liberal mean, -0.22 ± 1.2 dB; conservative mean, -4.49 ± 5.6 dB; and neutral mean, -2.08 ± 2.7 dB) based on the instruction set used.
The number of false-negative response errors obtained from the 30-2 threshold test were not significantly associated with instruction type. This may be due in part to our small sample size (n = 18). However, in both age groups, no false-positive response errors were recorded after the conservative instructions, a greater number for the neutral instructions, and the largest number for the liberal instructions. Consequently, none of the 18 participants was found to have abnormally high false-positive response rates on either the program 30-2 or the customized threshold test (5 false-positive responses for the neutral instructions and 12 for liberal instructions in both age groups combined). There are several possible implications for clinical perimetry from these findings: Based on the standard cutoff criteria for reliability, patients do not become unreliable, even when heavily encouraged to respond using liberal instructions; the number of false-positive responses is not a reliable indicator of response bias; liberal instructions given by perimetrists do not cause an excessive number of false-positive responses, but these responses may be the result of other psychological factors; and changes in threshold from response bias, especially in older subjects, can mimic visual field changes due to sensory visual system damage.
The customized test allowed us to obtain a second more precise threshold at both peripheral and central locations along with a false-positive response rate. For the younger group at the central test location, the threshold differences between the neutral and liberal instructions were slight. This suggests that the response bias used by the younger participants tested was more liberal when the test instructions did not give specific cues as to when to respond to a stimulus. In contrast, the older participants were found to demonstrate a more conservative response bias than the younger group for the perimetric threshold task, regardless of instruction type or location within the visual field. It is important to note that even when participants were following liberal instructions, their false-alarm rate was not higher than the miss rate, as the true definition of liberal would indicate. Because our sample size was small, we cannot conclude that all older participants will respond in a conservative manner to a light stimulus. Inconsistent use of conservative criteria by older adults has been reported in the aging and psychology literature.13 14 15
The methodology used for this study differs from that used in previous work,3 in that response bias was not only measured but specifically manipulated. Although sensitivity is lower in the periphery, the descriptive measurement of c additionally showed that participants in both age groups responded in a more conservative manner in the periphery regardless of the instructions. Although at the central test locations, thresholds did not differ significantly between the neutral and liberal instructions in the younger group and between the neutral and conservative instructions in the older group, the quantified measure of response criterion used by the participants corresponded with the instructions that were given. The liberal instructions and conservative instructions were correlated with liberal and conservative response criterion, respectively.
Although perimetry is automated, interaction with the patient remains important. The criterion differences demonstrated between the instruction sets and their differential effect on the central and peripheral regions of the visual field for both age groups illustrates the need for perimetrists to give instructions that emphasize a liberal response criteria for the peripheral area. Failure to encourage a liberal type of response in the periphery may result in a constricted visual field, because of a conservative response criterion. Our formulation for criterion (c) resulted in a range of values with zero at the center; a criterion of c = 0 indicated that the proportion of false-positive responses would be equal to that of false-negative responses and thus no response bias. A criterion of c < 0 indicated a liberal response tendency—false-positive responses would exceed false negative ones; a criterion of c > 0 indicated a conservative response tendency—false-positive responses would be fewer than false negative ones.2 Although we obtained various criterion differences, examination of the c values indicates that most of our observers responded in a conservative manner, even when liberal test instructions were administered. Our results included only visually normal subjects of two age groups. The impact of response criterion on the visual fields of patients with visual field loss remains uncertain.
In conclusion, our results show response bias can have a statistically and clinically significant effect on perimetric thresholds. A problem inherent in automated perimetry is that although there are recommended instructions, there is large interperimetrist variability in instructions used. There is no standardization of instructions. Our study suggests perimetrists should use a standard set of instructions to reduce the response bias effect. This will lessen the chance that changes in threshold from response bias will imitate visual field changes due to sensory visual system damage.
|
Footnotes |
|---|
Supported by VA Merit Review, and an unrestricted grant to the Department of Ophthalmology, University of Iowa, from Research to Prevent Blindness (MW).
Submitted for publication October 21, 1999; revised December 17, 1999; accepted January 11, 2000.
Commercial relationships policy: N.
Corresponding author: Michael Wall, Department of Neurology, University of Iowa, College of Medicine, 200 Hawkins Drive 2007 RCP, Iowa City, IA 52242-1053. michael-wall{at}uiowa.edu
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. Wall, K. R. Woodward, C. K. Doyle, and P. H. Artes Repeatability of Automated Perimetry: A Comparison between Standard Automated Perimetry with Stimulus Size III and V, Matrix, and Motion Perimetry Invest. Ophthalmol. Vis. Sci., February 1, 2009; 50(2): 974 - 979. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. Wall, K. R. Woodward, and C. F. Brito The Effect of Attention on Conventional Automated Perimetry and Luminance Size Threshold Perimetry Invest. Ophthalmol. Vis. Sci., January 1, 2004; 45(1): 342 - 350. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. M. McKendrick, A. J. Anderson, C. A. Johnson, and B. Fortune Appearance of the Frequency Doubling Stimulus in Normal Subjects and Patients with Glaucoma Invest. Ophthalmol. Vis. Sci., March 1, 2003; 44(3): 1111 - 1116. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H Sherafat, P G D Spry, A Waldock, J M Sparrow, and J P Diamond Effect of a patient training video on visual field test reliability Br. J. Ophthalmol., February 1, 2003; 87(2): 153 - 156. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. J. Anderson and C. A. Johnson Mechanisms Isolated by Frequency-Doubling Technology Perimetry Invest. Ophthalmol. Vis. Sci., February 1, 2002; 43(2): 398 - 401. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. H. Artes, D. McLeod, and D. B. Henson Response Time as a Discriminator between True- and False-Positive Responses in Suprathreshold Perimetry Invest. Ophthalmol. Vis. Sci., January 1, 2002; 43(1): 129 - 132. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
