Plummer DJ, Banker A, Taskintuna I, Azen SP, Sample PA, LaBree L, Freeman WR. The Utility of Entoptic Perimetry as a Screening Test for Cytomegalovirus Retinitis. Arch Ophthalmol. 1999;117(2):202-207. doi:10.1001/archopht.117.2.202
To determine the sensitivity and specificity of entoptic perimetry as a noninvasive test for detecting retinal damage due to peripheral cytomegalovirus (CMV) retinitis.
A masked study comparing entoptic perimetry with fundus photography under 4 experimental conditions (determined by increasing pixel sizes) on 2 separate testing sessions.
Acquired immunodeficiency syndrome Ocular Research Unit at the University of California, San Diego.
Twenty-four human immunodeficiency virus–positive and 8 human immunodeficiency virus–negative subjects; 21 eyes with documented CMV retinitis, and 26 eyes that were retinitis free.
For each testing session, screening method, and condition, the presence of CMV retinitis was determined for each meridian (ie, clock hour), each quadrant (consisting of 3 meridians), and each eye (consisting of all meridians); the amount of retinitis was defined as the percentage of meridians or quadrants with CMV retinitis.
Entoptic perimetry was as sensitive and specific as fundus photography in determining the presence of CMV retinitis. Determination of the amount of CMV retinitis tended to be underestimated by perimetry for larger pixel sizes.
Entoptic perimetry may be an effective and inexpensive alternative to fundus photography for CMV retinitis in hospitals and community clinics.
DETECTION OF cytomegalovirus (CMV) retinitis and other potentially treatable infectious retinopathies early in the course of the opportunistic infection is essential for the prevention of severe vision loss in patients with acquired immunodeficiency syndrome.1 Damage due to CMV retinitis may be insidious, particularly because it often affects the peripheral retina first and patients are often asymptomatic until irreversible destruction of the central retina (macula) and loss of visual acuity occur.2- 5 Lesions due to CMV leave large gliotic scars in areas of retinal destruction, producing absolute scotomata.6 A procedure that can measure the extent and locations of retinal scotomata therefore would be expected to give a relatively precise determination of the severity of CMV infection.
Human immunodeficiency virus (HIV)–positive individuals with peripheral CMV retinitis generally do not initially complain of any changes in peripheral vision.3 This is a common problem in all diseases affecting the peripheral visual field (eg, glaucoma). By the time the patient detects changes in vision due to CMV retinitis, damage from infection may be close to or within the central retina (macula) and, as a consequence, central vision may be permanently reduced.
Entoptic perimetry offers a means of following the course of peripheral CMV retinitis. Entoptic or "snow-field" perimetry uses a simple computer monitor filled with random particle motion.7- 10 People with normal retinas (eg, no retinal lesions) and a visual acuity of 20/20 will see random motion of pixels, or "dots," on the screen, which they often report as "snow" or "random motion particles." However, patients with peripheral retinal lesions trace the borders of the areas where they do not see the random motion particles
We developed a clinically useful screening test for CMV retinitis based on entoptic perimetry.7 We found that with this test, patients with CMV retinitis can see the scotoma in the infected eye. In a pilot study, we demonstrated that entoptic perimetry had a high sensitivity and specificity to detect lesions due to CMV retinitis within a 30° radius from the fovea (60° field). Other screening methods, such as the Amsler test, are effective only within the central 10° radius of vision. Recently, Teich and Saltzman11 have increased the effectiveness of the Amsler test with a new stimulus out to a 22.5° radius (45° field), but they reported a sensitivity of only 65%, less than what we reported with entoptic perimetry in our pilot study.7
In this article, we evaluate the utility of entoptic perimetry as a potential screening test for CMV retinitis in a clinical setting. Specifically, we evaluate the sensitivity, specificity, and reproducibility of entoptic perimetry under different experimental conditions characterized by stimulus size.
Because motivation and attention may be factors in determining the reliability and validity of entoptic perimetry, we analyzed data on both a per-eye and per-patient basis. For the per-patient basis, we randomly tested 1 eye of each patient.
Thirty-two subjects participated in the study; 24 (75%) were HIV positive and 8 (25%) were HIV negative (Table 1). Of the 24 HIV-positive subjects, 17 (71%) had documented CMV retinitis in 1 or both eyes, while 7 (29%) were CMV retinitis free. All subjects were recruited from ongoing studies at the University of California San Diego AIDS Ocular Research Unit, La Jolla. Subjects were tested as part of their normal examinations for treatment of CMV retinitis. Participation was entirely voluntary and informed consent was obtained. Subjects with retinitis directly affecting the fovea or optic nerve were not included in this study.
Forty-seven eyes in HIV-positive subjects were tested in 2 separate sessions; 21 eyes (45%) with documented CMV retinitis, and 26 eyes (55%) that were retinitis free (Table 1). One eye that had no light perception owing to CMV retinitis was not tested. Sixteen eyes of HIV-negative subjects were tested in 2 separate sessions. In HIV-positive subjects without CMV retinitis and HIV-negative subjects without CMV retinitis, we found no evidence of cotton-wool patches or other noninfectious retinopathological findings at the time of testing. Also shown in Table 1 are the numbers of eyes used in the per-patient analyses.
Subjects were first tested for CMV retinitis with entoptic perimetry. All subjects fully understood the testing procedure. Testing involved selecting an eye at random, covering the other eye with a patch, and having the subject look at a computer monitor (RasterOps 21 Correct-Color Monitor; RasterOps Corp, Santa Clara, Calif). Proper head position was ensured by bolting the headrest from a slitlamp onto a table in front of the computer monitor. The supports of the headrest were 30.5 cm from the screen. This setup allowed subjects to rest the chin and forehead on the supports.
Eyes were not dilated during the testing procedure, and subjects wore their normal (near) corrective lenses when viewing the stimuli. Subjects were allowed to view the stimuli for as long as necessary, but never required more than 1 minute per test.
A 27.94 × 43.18-cm transparent sheet was placed over the monitor, and a black dot was marked as a fixation point in the center of the transparency. The experimenter started a computer program that filled the screen with monochromatic random-motion particles. The participant fixated on the black target placed on the screen and was asked to trace the borders of any areas in the periphery where the dots were no longer flickering. The tracing was done with a permanent ink marker. If the participant was not satisfied with the tracing, the procedure was repeated.
The participant repeated this procedure 4 times, beginning with the 1-pixel condition and ending with the 16-pixel condition. For each condition, patients viewed stimuli that varied in spatial frequency (see "Stimuli" section that follows). The subjects were reminded that they needed to treat each condition based purely on their perception, not on the previous tests. The patch was then placed over the tested eye and the entire procedure was repeated with the fellow eye.
After the testing procedure, subjects received an ophthalmologic examination. Cytomegalovirus retinitis was diagnosed by indirect ophthalmoscopy, and a diagram of locations of lesions was made by a qualified ophthalmologist (A.B.). Presence and locations of these lesions was confirmed by fundus photography, thereby providing documentation of the location of any lesions on the retina within 1 hour of testing. Fundus photographs were taken to include all areas of retinal abnormalities as previously described.4,12 In all cases, the observations made by indirect ophthalmoscopy were identical to findings in the fundus photographs.
All subjects had 1 follow-up visit in which the procedures of the initial visit were repeated, including ophthalmologic examination and fundus photography. For all patients, fundus photography findings were identical to indirect ophthalmoscopic observations. The operator of the entoptic perimetry program was unaware of the ophthalmologic findings at the initial visit. The 2 testing sessions were separated by 2 to 4 weeks.
Stimuli were generated by a personal computer (Macintosh Quadra 840 AV; Apple Corp, Cupertino, Calif) using software written by us with a commercial software design program (Think C 5.0; Symantec Corp, Cupertino). The screen refresh rate was 60 Hz. The average luminance of the test screen was 48 candela (cd)/m2; luminance of the black areas was 0.3 cd/m2, and luminance of the white areas was 95 cd/m2. Each pixel was 0.3 mm, subtending 0.07° of visual angle. We used 4 stimulus conditions that varied the size of the pixels on the screen. The sizes of the square-shaped pixels were 1 × 1, 2 × 2, 4 × 4, and 16 × 16.
The locations of the retinitis determined from fundus drawings and fundus photographs were compared with the tracings of perceived scotomata made by subjects during testing. Readings of fundus photographs were performed by a qualified ophthalmologist with 5 years' experience caring for patients with CMV retinitis (A.B.); readings of the tracings were scored by an experienced psychophysicist (D. J. P.). Fundus photographs and retinal drawings were compared against the tracings in a double-masked manner, ie, the ophthalmologist was masked to the results of entoptic perimetry testing, and the psychophysical experimenter was masked to the fundus photography results. This procedure was carried out for each eye and session. In addition, for each eye and each session, entoptic perimetry tracings were scored for each of the 4 pixel sizes.
Perimetry tracings were scored as follows: for a given eye and pixel size, a meridian (ie, clock hour) was given a positive score for CMV retinitis if the subjects perceived a disturbance in the visual field crossing that meridian; otherwise, it was given a negative score. For a given eye and pixel size, a quadrant (consisting of 3 meridians) was given a positive score for CMV retinitis if a disturbance in the visual field was perceived for at least 1 meridian within the quadrant; otherwise it was given a negative score. A similar scoring scheme was applied to the fundus photographs. However, for a given eye and testing session, the single retinitis score for the fundus photographs was compared with all 4 retinitis scores (corresponding with the 4 pixel sizes) obtained from the entoptic perimetry tracings.
For each eye, each condition, and each session, sensitivity was calculated as the ratio of the number of eyes scored as positive by perimetry to the number of eyes scored as positive by fundus photography. Specificity was calculated as the ratio of the number of eyes scored as negative by perimetry to the number of eyes scored as negative by fundus photography. The percentage of eyes correctly classified as positive or negative was also calculated. Differences in the percentages of eyes correctly classified across pixel sizes were tested statistically using 1-way analysis of variance with repeated measures.
Analyses by meridian (or quadrant), conducted only in eyes with CMV retinitis, involved calculating the "percent involvement" for each method. Percent involvement was defined as the percentage of positive meridians (quadrants) in each eye as determined by fundus photography or perimetry. Differences in estimates of percent involvement between methods used the paired Student t test. For a given method, differences in percent involvement across conditions used 1-way analysis of variance with repeated measures. All analyses are conducted on a per-eye and per-patient basis.
For the 8 subjects who were HIV negative, the specificity of entoptic perimetry was 100% regardless of condition (ie, pixel size).
Table 2 presents the sensitivity, specificity, and number of correct diagnoses of entoptic perimetry in 47 eyes (21 with CMV retinitis, 26 without CMV retinitis) in 24 HIV-positive subjects for sessions 1 and 2, respectively. On a per-eye basis, compared with fundus photography, the 2 methods were equally sensitive (range, 0.86-1.00) and specific (range, 1.00 for all conditions) with regard to the determination of CMV retinitis. Analyses by condition indicated that for the first testing session, the percentage of correct diagnoses decreased from 100% to 94% with increasing pixel size (P=.29) (Table 2, session 1). No difference in diagnostic accuracy was found for the second session (Table 2, session 2). Similar results were obtained when data were analyzed on a per-patient basis.
Table 3 presents the results of the analyses of percent involvement by quadrant and by meridian for 21 eyes with CMV retinitis in 17 HIV-positive patients with CMV retinitis in 1 or both eyes. For session 1, the percent involvement as determined by perimetry for the larger pixel sizes (eg, 4 pixels) significantly underestimated the percent involvement as determined by fundus photography (quadrant, P<.05; meridian, P<.01) (Table 3, session 1). Further, as the pixel size increased from 1 to 16, there was decreasing accuracy in determining percent involvement by perimetry within meridians (P=.004). For session 2, the percent involvement as determined by perimetry for meridians underestimated the percent involvement as determined by the fundus photographs only for 16 pixels (P<.001) (Table 3, session 2) for all eyes. Similar results were obtained when data were analyzed on a per-patient basis.
Table 4 compares the quadrant and meridian results (averaged for all conditions) between the 2 sessions for eyes with CMV retinitis. There were significant differences in the determination of percent involvement between the 2 screening methods in session 1 (quadrant, P=.009; meridian, P=.02). No significant differences in the determination of percent involvement between the screening methods were found in session 2. In addition, there were significant differences in the determination of percent involvement by entoptic perimetry between the 2 sessions (quadrant, P=.05; meridian, P=.04; for all eyes), which indicated a learning curve. No differences in the determination of percent involvement by fundus photography between the 2 sessions were found. Similar results were obtained when data were analyzed on a per-patient basis.
Also shown in Table 3 and Table 4 are summary statistics describing the average difference in percent involvement between the 2 methods. Shown in Table 3 are the mean (SD), median, and range of the difference in score. Although on average the difference between the 3 methods was small (eg, the median tended to be close to 0), the range did demonstrate some large differences in percent involvement. Similar results were obtained when data were analyzed on a per-patient basis.
We reviewed the fundus photographs and entoptic perimetry tracings of all patients. Most patients had CMV retinitis beyond the tested central field or had papillomacular or peripapillary involvement that would produce large distal field defects. Among all eyes tested, there were 2 eyes that had retinal lesions that did not extend beyond the central 30° (radius) of vision and were not located within the central 10° where the lesions would be visually symptomatic. In both cases, the location of the lesion as measured by fundus photography corresponded well with location of the entoptic perimetry disturbance.
Table 5 presents an analysis of the extent of agreement between fundus photography and entoptic perimetry. For each patient, the locations of the lesions (as designated by the meridian) were compared between the 2 methods. In general, there was good agreement ±1 meridian (57%-79% for session 1, and 65%-81% for session 2). There was excellent agreement ±2 meridians (67%-81% for session 1 and 85%-91% for session 2). Furthermore, we found no systematic pattern of disagreement between the 2 methods by location on the retina (eg, superotemporal).
Our preliminary work with entoptic perimetry suggested that entoptic perimetry may be useful in screening HIV-positive patients with a stimulus that can be adopted by nonophthalmologic medical professionals and adapted to a television screen displaying patterns from a videotape. That report showed promising sensitivity and specificity; however, the sample size was small, the stimulus was not rigidly optimized, there was no measure of repeatability, and there was a high variability in the localization of lesions, possibly because we did not rigidly control patient head position.
The goal of the current study was to evaluate a larger number of patients, to determine the optimal stimulus, to control patient positioning, to determine the reproducibility of results on consecutive visits, and to determine the ability of entoptic perimetry not only to detect CMV retinitis, but also to localize the disease. The computer-generated stimuli were shown to detect disease in the central 30° radius of the retina, which corresponds to studies of the ocular complications of AIDS zone 1 and the posterior part of zone 2. The ability to accurately detect CMV retinitis within this area would be a major advance in screening HIV-positive individuals at risk for CMV retinitis because it would allow detection of retinitis prior to involvement of retinal areas of immediate threat to vision and also because it could improve detection of CMV.
This study produced several important results. First, we found that entoptic perimetry has 100% sensitivity and specificity in detecting CMV retinitis. Furthermore, sensitivity and specificity remain at a reasonable level while the coarseness of the stimuli is increased (ie, the 2- and 4-pixel conditions). This establishes the potential utility of entoptic perimetry as a tool for screening HIV-positive patients for CMV retinitis in the primary care and community clinic setting. The larger pixel sizes approximate the resolution of commercially available large-screen television sets. The effectiveness of this stimulus suggests that patients afflicted with CMV retinitis will be able to perceive peripheral retinal scotomata with the use of a large-screen television or even a projection while maintaining a close viewing distance to increase visual field size. In a community clinic setting, a television and VCR may supplement indirect ophthalmoscopy for rapid screening of large numbers of patients while retaining a high sensitivity and specificity.
Our results also suggest that entoptic perimetry is effective at localizing scotomata on both a by-quadrant and a by-meridian basis when compared with fundus photographs at either the finest resolutions (1- and 2-pixel sizes) or after the initial visit when patients become familiar with the procedure at any pixel size. This is essential if entoptic perimetry is to have any possible role in following scotomata in those patients with diagnosed and CMV retinitis. Our results suggest that patients should be able to monitor their own retinitis using a television and VCR. Entoptic perimetry identifies retina that has been destroyed by CMV retinitis regardless of whether the area is active or healed. We hope that this technique may eventually be useful for identifying the progression of the disease. For this purpose, we would predict that the lesion would increase in size during activity regardless of the overall size of the retinal area already affected.
As mentioned earlier, in every case, the scotoma was in the same area and the entoptic tracing could be superimposed onto the fundus photograph. The statistical technique we used for analyzing the data was to divide the photograph and tracing into meridians, and there was a high correspondence between retinal lesions and perceived entoptic field disturbances. Ongoing work is warranted to determine the limits of accuracy in localizing lesions; however, qualitatively, the location of a lesion corresponds well with entoptic tracing. We have yet to determine the threshold but are currently doing research in this area. In all cases, entoptic perimetry detected all CMV lesions.
We also compared results from the 2 sessions, which are presented in Table 4. We found that subjects more accurately estimate their scotomata as measured by fundus photography on subsequent visits than on the first visit, which we called a "learning effect." This may be applied to screening in community-based studies, as it is possible that 2 entoptic perimetry sessions may provide improved detection of CMV retinitis.
For both the per-patient and the per-eye analyses, our results indicate that there are 3 main results: (1) patients seem to experience a learning effect between the 2 sessions; (2) use of the smaller pixel sizes allows patients to more accurately perceive the extent of retinal scotomata; and (3) the overall sensitivities and specificities are the same by location.
We believe that entoptic perimetry is a rapid, inexpensive, initial screening tool, and this study establishes the potential utility of entoptic perimetry for screening HIV-positive patients for CMV retinitis. Medical personnel who are not opthalmologists may be able to administer it for initial detection of CMV retinitis within the central 30° radius of fixation or for subsequent monitoring. Once a scotoma is detected, diagnosis and precise localization can be made with indirect ophthalmoscopy. We recommend that in a community-based clinic setting where patients may not be seen frequently, primary care physicians should test patients with entoptic perimetry and analyze their results by quadrants so that they are able to refer patients to an ophthalmologist with a specific diagnosis. For screening, we recommend the smaller pixel sizes because they have a much higher sensitivity and they seem to be less affected by learning effects. Standard television monitors should provide adequate resolution. Although the fact that the 4 pixel sizes were not randomized is a potential weakness of our study, our results suggest that the best sensitivity was obtained with the smaller pixel dots that were tested first. This makes it unlikely that patients experienced a learning effect with the 16-pixel stimulus. Our results already suggest that it is less sensitive than the smaller stimuli. Future studies will evaluate the use of entoptic perimetry on a large population in a primary care setting with a VCR and television.
Accepted for publication September 16, 1998.
This study was supported by grant R95-SD-082 (Dr Plummer) from the Universitywide AIDS Research Program; grant EY11961 (Dr Plummer), grant 07366 (Dr Freeman), and core grant for Vision Research EY03040 (Dr Azen and Ms LaBree) from the National Eye Institute, National Institutes of Health, Bethesda, Md; and a departmental grant from Research to Prevent Blindness Inc, New York, NY.
Corresponding author: Daniel J. Plummer, PhD, Shiley Eye Center, Department of Ophthalmology, School of Medicine, University of California San Diego, La Jolla, CA 92093-0946 (e-mail: email@example.com).