| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
|
|
|
|||||||
|
|||||||||
Electronic Letters to:
|
|
Electronic letters published:
![[Read eLetter]](/icons/misc/sectionDOWN.gif){kind=icon}
The A. J. Anderson TOP Algorithm Simulation
Author Response: The A. J. Anderson TOP Algorithm Simulation
|
|
|||
|
Manuel A. González de la Rosa University of La Laguna
Send letter to journal:
mgdelarosa{at}jet.es Manuel A. González de la Rosa
|
I have read A.J. Anderson’s publication entitled "Spatial Resolution of the Tendency-Oriented Perimetry Algorithm"1 with great interest. I have questions about the originality of the study design, and the validity of its methods, results, and conclusions. A. The method used in this study is not correct. The author considers that each location in the visual field has a sensitivity that is independent from neighboring points. Therefore, the visual field defects that are judged to be "true" by the author are arbitrarily created and are not similar to those encountered in clinical practice.2 In addition, a similar simulation applied to the standard bracketing strategy may also miss absolute scotomas of less than 6 degrees. Program 32 detects the blind spot almost in every instance because, fortuitously, one of its points coincides with the blind spot. However, when program 31 was used, with the same 6 degrees separation, the blind spot was unnoticed in one-third of the examinations. What really happens is that scotomas like the blind spot do not exist in clinical practice, except in macular pathology. We developed macular-TOP (which has been enclosed with Octopus perimeters for years) for that purpose. The author acknowledges another source of the problem. He affirms that he made a free interpretation of the algorithm because he felt that its description is not clear enough in the articles that we have published. We have always given additional information to any scientist or clinician who has requested it, and we would be pleased to provide Dr. Anderson with such information if he wishes to dispel his doubts. As an example, the TOP program has two internal algorithms for identifying false negatives which, as Dr. Anderson says, can influence the results, mostly when they occur during the examination of the first two matrices. B. The paper is not original: the smoothness of abrupt scotoma edges and the consequent LV reduction when using TOP are not newly reported in this paper. We described them since the very first publication of the method,3 and they have been confirmed later by others.4 Just as in this paper, other researchers have previously supposed that these characteristics could limit the clinical usefulness of the algorithm,5,6 but this theoretical hypothesis has never actually been demonstrated. It has also been described how the SITA strategy underestimates glaucoma defects.7 If the author had focused on studying this aspect, he would have had more difficulties to find a useful description in the literature than using TOP. The use of simulation to analyze this problem is not new either. We carried it out in 1996 in order to evaluate the method before applying it to patients.8 Later on, Prof. Bebie designed the PeriSim simulation program (Interzeag AG. Köniz/Bern, Switzerland) that is much more precise than ours. Prof. Bebie’s simulation program has important characteristics that are absent in the paper under comment: it correctly simulates TOP; the normal visual field is not simulated with uniform values but with real values that are corrected for the subject’s age; the neurological "fatigue effect" which reduces retinal sensitivity as a function of test duration and the blind spot are simulated; it is applied to real patient visual fields; the patient’s degree of collaboration can be simulated; the "see curve" can be modulated; averaged statistics of the simulated local error can be obtained simulating several results in order to estimate the influence of short fluctuation, etc. C. The limitations of the simulation used by Anderson can be highlighted by the opposite contradictory results of several clinical studies evaluating the value of LV-TOP. In a recent multicenter study carried out on 406 cases,9 we observed that the LV in TOP has a behavior that is much more coherent with clinical reality than in the conventional strategy during the earliest stages of glaucoma. To be precise, its dependence with the MD value is much closer. Higher LV values are frequently observed in conventional perimetry, even in normal subjects, which decreases its diagnostic usefulness. On the other hand, Fabre’s excellent results10 obtained with TOP in patients with early glaucoma, seem to be in relation to having used the LV as discriminator for the diagnosis. During the past year, we have been carrying out a study in a wider sample of patients, in which we have verified the high diagnostic ability of LV-TOP for glaucoma diagnosis (González de la Rosa M et al. 2002 IPS Meeting). We have recently finished an extensive multicenter study that has reasserted LV-TOP’s potential (González de la Rosa M, González-Hernández M, Garcia Feijoo J, Morales J, Azuara-Blanco A, manuscript submitted, 2003), which widely surpasses the one obtained for the same index with the conventional strategy (González-Hernández M, Morales J, Azuara-Blanco A, Garcia Sanchez J, González de la Rosa M, manuscript submitted, 2003). We cannot be more precise about the results of these papers and their interpretation, because they are being reviewed for publication in another journal. D. For all of the above reasons, I believe that the conclusions of Anderson’s paper are completely unjustified. At the end of the paper, authors question the clinical usefulness of the TOP algorithm and they recommend its limited use for screening. In our opinion, those conclusions cannot be derived from Anderson’s study. The only conclusion that can be derived from this paper is that when a simulation contradicts actual real experience, what is to be questioned is the simulation model and not reality. Manuel A. González de la Rosa University of La Laguna
References 1. Anderson AJ. Spatial resolution of the tendency-oriented perimetry algorithm. Invest Ophthalmol Vis Sci. 2003;44:1962-1968.
|
|||
|
|
|||
|
Andrew J. Anderson Discoveries in Sight
Send letter to journal:
ajanderson{at}hotmail.com Andrew J. Anderson
|
Although it is true that there are probabilistic relationships between the sensitivity of neighboring points of the visual field in a particular type of ocular disease, these relationships differ between disease types (e.g. chorio-retinal versus nerve-fiber layer versus central visual pathway disease). Indeed, it is often the distinctive patterns of these various field defects that suggest a particular type of diagnosis. If a perimetric test is to be applicable to the gamut of eye disease for which perimetry is performed, then the sensitivity of one point does not bear an easy relationship with its neighbors, except in areas where retinal sensitivity is normal. I do not agree with Professor González de la Rosa’s claim that defects like the blind spot (i.e. deep, localized defects) occur only in macular disease: such defects frequently arise in glaucoma, chorio-retinal disease, and central visual pathway disease. The TOP procedure presented in my paper1 purposely incorporated only the fundamental features of the algorithm, thereby allowing readers to understand the somewhat complex workings of TOP. Incorporation of the additional simulation parameters mentioned by Professor González de la Rosa would not change the underlying response characteristics of the TOP algorithm, whose response anomalies are both systematic and of large magnitude. I concur with Professor González de la Rosa that both the blurring of scotomata edges in TOP and the significantly reduced index LV have been noted previously, and that others have used computer simulations to investigate TOP. I discuss these studies in my paper. I am unaware of any work that has systematically evaluated the causes of the anomalous thresholds returned from the TOP procedure, however, hence the motivation to undertake my study. Contrary to Professor González de la Rosa’s assertion, my results are not in conflict with current literature describing the sensitivity and specificity of TOP for disease detection. A test’s sensitivity and specificity (i.e. its ability to screen for disease) may be independent of its ability to provide accurate threshold data. For example, supra-threshold procedures can reliably screen for disease, but provide little information about perimetric thresholds. However tempting it may be, it is a mistake to interpret the potentially good screening performance of the index LV as meaning that the LV index is accurate. This is true even if sensitivity and specificity were 100%. I concede that it may yet be shown that an accurate spatial map of sensitivity is not the best way to detect and/or monitor ocular disease using psychophysical data, although I know of no evidence to suggest that this is so. If one accepts, however, that a basic requirement for threshold perimetry is the faithful rendering of sensitivity as a function of location, then TOP, in which location and sensitivity are irreversibly convolved, cannot be recommended for determining perimetric thresholds. Andrew J. Anderson Discoveries in Sight
References 1. Anderson, AJ. Spatial resolution of the tendency-oriented perimetry algorithm. Invest Ophthalmol Vis Sci. 2003;44:1962-1968. |
|||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
eLetters: