Questions directed to Chris Johnson in his absence

Question: What are reasonable reliability scores for the SITA-Standard paradigm, that is, what percentage of false-negatives and as such should raise a flag and suggest that a particular field is suspect in terms of patient reliability?

Dr Johnson: Unlike the Full Threshold and Fastpac test strategies, which reported false-positive and false-negative responses as a fraction, the new SITA test strategies report them as percentages. The Full Threshold and Fastpac strategies ‘flag’ excessive false-positive and false-negative errors by placing a double X by the fraction and printing a ‘low patient reliability’ message on the printout. The criterion for ‘flagging’ false-positives and -negatives was greater than 33% errors. Several investigations of visual field reliability confirmed that this was a good criterion, because only 5% or less of individuals with normal vision would produce greater than 33% false-positive and -negative errors. The new SITA strategies use a different method for determining false-positive errors. Instead of performing ‘catch’ trials, the SITA strategies evaluate the percentage of responses made by the patient that fall outside an acceptable response time window. Responses that fall outside this window are considered to be false-positives. Moreover, the SITA strategies do not ‘flag’ false-positives and -negatives that are greater than 33%. Because of differences in test time and the different method of determining falsepositives, it is not clear that the 33% value applies to the SITA strategies in the same manner as for the Full Threshold and Fastpac strategies. However, in the absence of any other information, 33% is probably a good rule of thumb to use clinically. It should be noted that excessive false-positives are useful for identifying ‘trigger happy’ patients who press the response button too often. False-nega- tives can reflect inattention or fatigue in a patient with a normal visual field or mild visual field loss. However, patients with moderate to severe visual field loss may generate excessive false-negatives because damaged visual field locations have much higher variability than locations with normal sensitivity. Thus, excessive false-negatives in a patient with considerable visual field loss may not necessarily indicate an unreliable patient.

Question: When obtaining baseline fields, two sets, which fields do you suggest?: 24-2 versus 30-2, SAP, SITA-Standard or SITA-Fast, SWAP, etc., and why?

Dr Johnson: I don’t believe there is a single answer that would necessarily apply in all cases. For glaucoma patients, a SITA-Standard 24-2 test procedure would probably be the most common visual field procedure to select as the baseline. If the first 24-2 visual field revealed a subtle nasal step or some sensitivity loss for the most peripheral locations, then a 30-2 test pattern would probably be more

Glaucoma in the New Millennium, pp. 207–209

Proceedings of the 50th Annual Symposium of the New Orleans Academy of Ophthalmology, New Orleans, LA, USA, April 6-8, 2001

edited by Jonathan Nussdorf

© 2003 Kugler Publications, The Hague, The Netherlands

appropriate for establishing a baseline. If the patient’s first visual field was normal, but there was a suspicion of possible glaucoma (either because of risk factors, optic disc appearance or other clinical information), then it would probably be best to establish a baseline for SWAP. If the patient cannot tolerate conventional visual field examinations, then SITA-Fast might be the best choice for establishing a baseline. If the physician is interested in a very rapid screening for possible visual field loss, then frequency doubling technology (FDT) perimetry would probably be the method of choice.

Question: How do we follow SITA fields?

Dr Johnson: At the present time, the best thing to do is probably to compare the total deviation and pattern deviation probability plots for successive visual fields. As a rule of thumb, I would suggest that a change in the probability level (or the appearance of new abnormal locations) of four to five locations can be used as a criterion for a suspicion of progression. However, progression at these locations should then be confirmed by one or two additional visual field tests. I am told that a ‘glaucoma change probability’ and related analyses will soon be available for SITA-Standard and SITA-Fast. When these become available, they should improve the ability to follow SITA visual fields.

Question: We have noticed that our reliability scores on crummy visual field takers are ‘good’, but the fields are poor and out of proportion to nerve damage, etc. Then, after six to seven fields (one to two years of training), the test suddenly becomes ‘normal’. Is it the case that ‘good’ reliability scores miss these patients and give a false sense of high reliability? Is it true that poor reliability scores can be trusted but we cannot blindly trust ‘good’ reliability scores, and that all fields should be suspect and viewed in terms of anatomical data and the ‘black box’ that produced the psychophysical data (the patient)?

Dr Johnson: There are several reasons why we might get a normal visual field after several abnormal ones. One possibility is a learning effect. A few patients require practice before becoming good visual field testers. Another possibility is fatigue during the first few tests. An additional possibility is that some glaucoma patients have high long-term variability (whether due to the glaucomatous disease process or to psychophysical or cognitive reasons). In longitudinal clinical studies of glaucoma and ocular hypertension, we have seen patients reach an endpoint of visual field progression only to return to a normal visual field some time later. With regard to the reliability indices for the Humphrey Field Analyzer, I believe that they tend to underestimate false-positive and false-negative responses. Because of this, I tend to play down the importance of the reliability indices. I find the ‘gaze tracking’ option on the new Humphrey Field Analyzer to be more informative. By looking at the pattern of the gaze tracking printout, you can determine whether a patient was getting tired during the latter portion of the test, whether the patient was having difficulty staying awake, whether the patient was fixating properly, and whether or not the patient was moving their head during the test. I find this to be a better indicator of the patient’s level of cooperation and attention during a visual field exam.

Question: Please make a comment about the data and conclusions by J. Katz and

their comparison of criteria for visual field progression (Arch Ophthalmol 117:11371142, 1999).

Dr Johnson: I am in agreement with Joanne Katz’s general findings. I believe that detecting progression of glaucomatous visual field loss is one of the major challenges in the clinical management of glaucoma patients, as well as in defining endpoints for multicenter clinical trials. Each one of the multicenter clinical trials in glaucoma and ocular hypertension (Advanced Glaucoma Intervention Study (AGIS), Collaborative Initial Glaucoma Treatment Study (CIGTS), Ocular Hypertension Treatment Study (OHTS), Normal Tension Glaucoma Study (NTGS), Early Manifest Glaucoma Treatment Study (EMGT), Primary Treatment Trial (PTTMoorfields)) has different criteria for determining visual field progression. The only thing they have in common is that, to declare an endpoint (visual field progression), a number of confirming visual fields are required.

Dr Paul Spry and I have recently written a comprehensive review of visual field progression in glaucoma for the Survey of Ophthalmology 47:158-173, 2002 that discusses the various problems, the different approaches, and their advantages and disadvantages, and what is needed in the future to improve our ability to determine glaucomatous visual field progression. We have also used computer simulation to evaluate the performance of various approaches for determining visual field progression. With this simulation, we are able to generate a series of visual fields that have different rates of progression (or no progression at all), along with various amounts of shortand long-term variability. This allows us to study various visual field progression criteria for a variety of conditions and to compare their performance (sensitivity and specificity, as well as time to first detecting progression). Our results are similar to those of Joanne Katz. CIGTS and EMGT detect progression about equally, but there is agreement in only about half to twothirds of the cases. AGIS detects visual field progression about half as often as CIGTS and EMGT criteria. AGIS is quite insensitive for detection of visual field progression. Progressor detects progression at about the same rate as CIGTS and EMGT criteria, but again there is agreement in only about 50-75% of the cases. Because Joanne Katz analyzed real patient data, she has no ‘gold standard’ (i.e., there is no definite answer as to whether a patient truly did or did not progress, or what their ‘true’ visual field is). Therefore, she was able to make relative comparisons, but could not determine absolute sensitivity and specificity values for the various criteria. In our computer simulation, we can specify the beginning and ending visual fields, so we know whether or not visual field progression has occurred, and can therefore directly measure sensitivity and specificity. Our work has yielded several general conclusions. Firstly, any of the current methods of detecting visual field progression requires at least seven or eight visual fields over a multi-year time period in order to have reasonable sensitivity and specificity values for distinguishing variability from true progression. Secondly, although some criteria perform moderately better than others, none of them do a completely satisfactory job. Thirdly, all the current criteria perform poorly when there is a small number of visual fields. Finally, refinement of current procedures is unlikely to improve performance dramatically. We need improvements in three areas: 1. more reliable visual field tests; 2. visual field tests that are more sensitive to glaucomatous damage; and 3. development of new, innovative analysis methods (possible candidates might include neural networks, fuzzy logic, chaos analysis, and related techniques).