iain

ebo

imuesentomParticipants: A total of 34 participants with moderate to severe glaucoma; mean deviation at their last clinicvisit averaged 10.90 dB (range, 20.94 to 3.38 dB). A total of 75 of the 136 locations tested had a perimetricsensitivity of 19 dB.

Methods: Frequency-of-seeing curves were constructed at 4 nonadjacent visual eld locations by the Methodof Constant Stimuli (MOCS), using 35 stimulus presentations at each of 7 contrasts. Locations were chosen a prioriand included at least 2 with glaucomatous damage but a sensitivity of 6 dB. Cumulative Gaussian curves were tto the data, rst assuming a 5% false-negative rate and subsequently allowing the asymptotic maximum responseprobability to be a free parameter.

Main Outcome Measures: The strength of the relation (R2) between perimetric sensitivity (mean of last 2clinic visits) and MOCS sensitivity (from the experiment) for all locations with perimetric sensitivity within 4 dB ofeach selected value, at 0.5 dB intervals.

Results: Bins centered at sensitivities 19 dB always had R2> 0.1. All bins centered at sensitivities 15 dBhad R2< 0.1, an indication that sensitivities are unreliable. No consistent conclusions could be drawn between 15and 19 dB. At 57 of the 81 locations with perimetric sensitivity

respond to on 50% of presentations. To maintain an acceptable for the atter FOS curves and thus the higher variability that

Swanson et al reported that in nonhuman primates, semi-

Ophthalmology Volume -, Number -, Month 2014test duration and avoid overly fatiguing the patient, thissensitivity is typically estimated on the basis of

visual eld test results (conducted on an HFA perimeter). The mostrecent clinic visit occurred an average of 169 days before the study

Gaussian distribution, such that F(N) 0 and F(N) 1. CS

Gardiner et al Accuracy of Perimetry at Low Sensitivities in Glaucomavisit. At least 2 of the chosen locations had signicantly reducedsensitivity that was no less than 6 dB (to ensure that some functionremained at that location), with the remaining locations in differentregions of the visual eld to promote xation stability. Testingseveral locations within the visual eld also ensures spatial un-certainty, which increases the slope of the FOS curve by preventingattention being focused on a single location28 and makes the testconditions more similar to clinical perimetry.

The FOS curves were assessed using the Method of ConstantStimuli (MOCS) on an Octopus perimeter.29 At each test location,perimetric sensitivity was dened as the mean of the sensitivitiesmeasured at that location during the subjects 2 most recent clinicalvisual eld examinations. For the 2 less damaged locations of the4, 7 contrasts were chosen for testing, set at 3-dB intervals centeredat the perimetric sensitivity (i.e., so that the range 9 dB from thisvalue is covered). If perimetric sensitivity was

Ophthalmology Volume -, Number -, Month 2014increased (lower on the decibel scale). The MOCS sensitivities forthese locations were 29.7 and 31.1 dB, respectively. However, forthe 2 more damaged locations (3, 15) and (15, 3) on thebottom row, the response probability never approaches 100%,despite the fact that it is clearly greater than zero and so somefunction remains. The tted sensitivities assuming a 5% false-negative rate (dashed lines) are 4.1 and 6.3 dB, despite thefact that during the subjects last clinic visit the pointwise sensi-tivities measured by perimetry were 15 and 14 dB, respectively. Inthe secondary analysis allowing for response saturation (dottedlines), the tted asymptotic maximum response probability is 66%for location (3, 15) (bottom left). This indicates that unless thefalse-negative error rate is extremely high (which seems unlikelygiven that the subject produces response probabilities reaching100% at the healthier locations), response saturation is likely takingplace. For location (15, 3) (bottom right), the tted asymptoticmaximum response probability is just 10%. In this case, theresponse threshold (i.e., 50% probability of response) is neverattained, and thus contrast sensitivity in its most common clinicalformulation is undened.

Figure 2 plots the MOCS sensitivity (tted to the experimentaldata assuming a false-negative rate of 5%, i.e., according to the null

Figure 1. Response probabilities for a sample study subject at 4 tested locationsof-seeing (FOS) curve as tted using the primary analysis, in which the maximuhigh (assuming a 5% false-negative rate). The dotted line indicates the FOS curvdescribed in the Methods section.

4hypothesis) against perimetric sensitivity (the average of the sen-sitivities measured by perimetry at the 2 most recent clinic visits)for all 4 locations of all 34 subjects. At higher perimetric sensi-tivities, the association is strong and perimetry seems to reect theMOCS sensitivity, with a comparatively small amount of vari-ability. However, at lower perimetric sensitivities, the associationbreaks down and perimetry tends to overestimate the MOCSsensitivity. Perhaps more important, it does so by an unpredictableamount. Figure 3 shows the correlation between perimetric andMOCS sensitivities within sliding bins of width 8 dB. For eachbin centered at sensitivities 19 dB (e.g., the bins covering theranges 15e23 dB, 15.5e23.5 dB), the relation has R2> 0.1 andis signicant at the 5% level. At sensitivities below this level thecorrelation between the 2 is lower and indeed is generally noteven statistically signicant despite sample sizes of at least 39tested locations in each of the bins. For each bin centered atsensitivities 15 dB, the relation has R2< 0.1 and is neversignicant at the 5% level.

Figure 4 shows the percentage of presentations during MOCStesting for which the subject responded to the highest luminancestimulus presented at each of the 2 most damaged locations(according to their perimetric sensitivity). This maximal stimulus

(at positions as labeled in degrees). The dashed line indicates the frequency-m response probability would be 95% if contrast could be made sufcientlye t allowing this asymptotic maximum to vary, as in the secondary analysis

of these results is that an apparent change in sensitivitywithin the range from 0 to 15 to 19 dB may not be indicative

Gardiner et al Accuracy of Perimetry at Low Sensitivities in Glaucomawas 0 dB on the Octopus perimeter used for testing, equivalent to3.7 dB on an HFA perimeter (the units reported here). At 44% oflocations, this probability was

Ophthalmology Volume -, Number -, Month 2014presentation during testing, or a location with true threshold

Gardiner et al Accuracy of Perimetry at Low Sensitivities in Glaucomamany regions of the upper hemield, sensitivities were

sensitivity at the second test date could have been as high as (Fig 3). This criterion is useful but essentially arbitrary, and

Ophthalmology Volume -, Number -, Month 201419 dB. Indeed, on the third test the reported sensitivity atthat location had increased to 17 dB.

Within glaucoma research, many studies have relied onperimetric sensitivities from severely damaged locations,which may be unreliable. Our interpretation of the resultssuggests restricting analyses to locations with sensitivities19 dB. For example, there has been debate over the beststatistical models to deal with the oor effect occurring at0 dB.32e34 However, our ndings suggest that a oormight actually exist between 15 and 19 dB. Studies of thestructureefunction relation35,36 would be affected by theinclusion of regions with severe glaucomatous damagewhose sensitivities are unreliable, and it remains to be seenwhether this has any material impact on the conclusions. Theimplications of our ndings on other studies will varydepending on the study population used. The Ocular Hy-pertension Treatment Study, for instance, consisted of sub-jects with visual elds that were within normal limits atbaseline,37 and so the proportion of locations that progressedbeyond 19 dB was likely small; the implications of ndingsfrom that study are unlikely to change.

Our results also suggest the possible need for alternateclinical methods for assessing progressive functionalchange. For example, severely damaged locations (in thiscase,

recommend this as the oor of future testing algorithms that could have decreased the slopes of the FOS curves.28 Testing

Gardiner et al Accuracy of Perimetry at Low Sensitivities in Glaucomarely on these stimuli.Although variability has been suggested as a precursor of

visual eld damage in glaucoma,40 the high variability ofperimetric sensitivities has more typically been thought ofas a problem to be battled rather than a potential source ofinformation. However, gaining an improved understandingof the reasons for this variability could aid efforts to reduceit, whether that is by postprocessing of the data,41 differenttest algorithms,42,43 or different test stimuli.44 It may alsoshed light on aspects of the pathophysiology, such as thepossible presence of living yet dysfunctional RGCs.45,46 IfRGCs were dysfunctional in a manner that caused a pro-portionately reduced response to any given stimulus contrast,this might effectively shift the FOS curve toward the left(toward higher contrasts/lower sensitivity). Response prob-abilities would still eventually reach 100%, but at a greatercontrast than would be the case for healthy cells. There wasno evidence for this hypothesis in our results because theresponse probability frequently does not reach 100% withinthe contrast range produced bymodern increment perimeters.Our data do not rule out other forms of dysfunction. Forexample, the maximal response of a dysfunctional RGCcould be reduced.

The inaccuracy of perimetry when assessing sensitivity atdamaged locations is not caused by the choice of testing al-gorithm (in this case, the SITA20). For many of the locationstested, the asymptotic maximum response probability was50%, if there is only a small increase in responseprobability with contrast (which would be expectedbecause RGC saturation is an asymptotic rather thanabsolute phenomenon), that increase is so small that atesting algorithm would require an unrealistically long testduration to reliably determine the response threshold. Inthis study, testing took up to half an hour to quantifyperformance at just 4 locations. Our conclusions also applyequally to different perimeters. Because of the instrument-specic nature of the decibel scale, the cutoff we have cho-sen would be different; 19 dB on the HFA (the units used inthis report) is equivalent to 15.3 dB on the Octopus 900perimeter, or 400% contrast. It should be noted that theperimetric sensitivity used in this study was the mean of the 2most recent clinical visual eld tests, and if a single test hadbeen used the results most likely would have been morevariable than shown here.

Study Limitations

In this study, we made every attempt to make the FOS curvetesting mimic clinical perimetry as closely as possible. Themost signicant difference is that only 4 locations were testedinstead of 54 (in a 24-2 visual eld) or 68 (in a 10-2 visualeld). This will have reduced spatial uncertainty, and thistook place as part of a longer session lasting up to 1 hour intotal, and so some fatigue effects may be present, which couldlower the MOCS sensitivity. Allowing breaks between runsshould have minimized these effects, and the majority ofsubjects took at least 1 longer break of 10 minutes or morepart way through the testing sequence.

In conclusion, this study found that in eyes with glau-coma, clear media, and no other ocular comorbidities,pointwise sensitivities less than approximately 15 to 19 dBfrom clinical perimetry showed little correlation with thetrue functional status at that location. In an eye with rela-tively clear media, the only reliable information that suchlocations provide may be that the sensitivity is likely to bebetween 0 and 15 to 19 dB. These ndings provide apossible explanation for the high variability observed at lowsensitivities when using perimetry. It may be useful toanalyze data from research studies on the basis of 19 dBbeing the lower limit of the reliable stimulus range ofstandard automated perimetry, rather than using the 0 dBlower limit of the technical stimulus range of the instrument.Clinically, threshold values less than 15 to 19 dB should beinterpreted with caution because they may not be reliable forassessing the true level of damage or glaucomatousprogression.

References

1. Artes PH, Hutchison DM, Nicolela MT, et al. Threshold andvariability properties of matrix frequency-doubling technologyand standard automated perimetry in glaucoma. Invest Oph-thalmol Vis Sci 2005;46:24517.

2. Blumenthal EZ, Sample PA, Berry CC, et al. Evaluatingseveral sources of variability for standard and SWAP visualelds in glaucoma patients, suspects, and normals. Ophthal-mology 2000;110:1895902.

3. Chauhan BC, House PH. Intratest variability in conventionaland high-pass resolution perimetry. Ophthalmology 1991;98:7983.

4. Gardiner SK, Johnson CA, Demirel S. The effect of testvariability on the structureefunction relationship in earlyglaucoma. Graefes Arch Clin Exp Ophthalmol 2012;250:185161.

5. Gilpin LB, Stewart WC, Hunt HH, Broom CD. Thresholdvariability using different Goldmann stimulus sizes. ActaOphthalmol 1990;68:6746.

6. Heijl A, Lindgren G, Olsson J. Normal variability of staticperimetric threshold values across the central visual eld. ArchOphthalmol 1987;105:15449.

7. Heijl A, Lindgren A, Lindgren G. Testretest variability inglaucomatous visual elds. Am J Ophthalmol 1989;108:1305.

8. Henson DB, Chaudry S, Artes PH, et al. Response variabilityin the visual eld: comparison of optic neuritis, glaucoma,ocular hypertension, and normal eyes. Invest Ophthalmol VisSci 2000;41:41721.

9. Jampel HD, Vitale S, Ding Y, et al. Test-retest variability instructural and functional parameters of glaucoma damage inthe Glaucoma Imaging Longitudinal Study. J Glaucoma2006;15:1527.

10. Piltz JR, Starita RJ. Test-retest variability in glaucomatousvisual elds [letter]. Am J Ophthalmol 1990;109:10910.

9

11. Spry PG, Johnson CA, McKendrick AM, Turpin A. Vari-ability components of standard automated perimetry and

29. Laming D, Laming JF. Hegelmaier: on memory for the lengthof a line. Psychol Res 1992;54:2339.

30. Turpin A, Artes PH, McKendrick AM. The Open Perimetry

Ophthalmology Volume -, Number -, Month 2014visual eld progression. Br J Ophthalmol 2002;86:5604.14. Gardiner SK, Demirel S, De Moraes CG, et al; Ocular Hy-

pertension Treatment Study Group. Series length used duringtrend analysis affects sensitivity to changes in progression ratein the Ocular Hypertension Treatment Study. Invest Oph-thalmol Vis Sci 2013;54:12529.

15. Katz J, Gilbert D, Quigley HA, Sommer A. Estimating pro-gression of visual eld loss in glaucoma. Ophthalmology1997;104:101725.

16. Nouri-Mahdavi K, Brigatti L, Weitzman M, Caprioli J.Comparison of methods to detect visual eld progression inglaucoma. Ophthalmology 1997;104:122836.

17. Chauhan BC, Garway-Heath DF, Goni FJ, et al. Practicalrecommendations for measuring rates of visual eld change inglaucoma. Br J Ophthalmol 2008;92:56973.

18. Keltner JL, Johnson CA, Quigg JM, et al; Ocular HypertensionTreatment Study Group. Conrmation of visual eld abnor-malities in the Ocular Hypertension Treatment Study. ArchOphthalmol 2000;118:118794.

19. Gardiner SK, Swanson WH, Demirel S, et al. A two-stageneural spiking model of visual contrast detection in perimetry.Vision Res 2008;48:185969.

20. Bengtsson B, Olsson J, Heijl A, Rootzen H. A new generationof algorithms for computerized threshold perimetry, SITA.Acta Ophthalmol Scand 1997;75:36875.

21. Wall M, Woodward KR, Doyle CK, Zamba G. The effective dy-namic ranges of standard automated perimetry sizes III and V andmotion and matrix perimetry. Arch Ophthalmol 2010;128:5706.

22. Chauhan BC, Johnson CA. Test-retest variability of frequency-doubling perimetry and conventional perimetry in glaucomapatients and normal subjects. Invest Ophthalmol Vis Sci1999;40:64856.

23. Kaplan E, Shapley RM. The primate retina contains two typesof ganglion cells, with high and low contrast sensitivity. ProcNatl Acad Sci U S A 1986;83:27557.

24. Swanson WH, Sun H, Lee BB, Cao D. Responses of primateretinal ganglion cells to perimetric stimuli. Invest OphthalmolVis Sci 2011;52:76471.

25. Johnson KA, Goody RS. The original Michaelis constant:translation of the 1913 MichaeliseMenten paper. Biochem-istry 2011;50:82649.

26. Baylor DA, Hodgkin AL, Lamb TD. Reconstruction of theelectrical responses of turtle cones to ashes and steps of light.J Physiol 1974;242:75991.

27. Hodgkin AL, Huxley AF. A quantitative description ofmembrane current and its application to conduction andexcitation in nerve. J Physiol 1952;117:50044.

28. Pelli D. Uncertainty explains many aspects of visual contrastdetection and discrimination. J Opt Soc AmA 1985;2:150831.

Footnotes and Financial Disclosures

Originally received: August 30, 2013.Final revision: January 14, 2014.Accepted: January 16, 2014.Available online: ---. Manuscript no. 2013-1493.

1032. Caprioli J, Mock D, Bitrian E, et al. A method to measure andpredict rates of regional visual eld decay in glaucoma. InvestOphthalmol Vis Sci 2011;52:476573.

33. Russell RA, Crabb DP. On alternative methods for measuringvisual eld decay: Tobit linear regression [letter]. InvestOphthalmol Vis Sci 2011;52:953940.

34. Caprioli J, Mock D, Bitrian E, et al. Author response: onalternative methods for measuring visual eld decay: Tobitlinear regression. Invest Ophthalmol Vis Sci 2012;53:118.

35. Hood D, Kardon R. A framework for comparing structural andfunctional measures of glaucomatous damage. Prog Retin EyeRes 2007;26:688710.

36. Harwerth RS, Wheat JL, Fredette MJ, Anderson DR. Linkingstructure and function in glaucoma. Prog Retin Eye Res2010;29:24971.

37. Gordon MO, Kass MA, Ocular Hypertension Treatment StudyGroup. The Ocular Hypertension Treatment Study: design andbaseline description of the participants. Arch Ophthalmol1999;117:57383.

38. Tabernero J, Ohlendorf A, Fischer MD, et al. Peripheralrefraction proles in subjects with low foveal refractive errors[report online]. Optom Vis Sci 2011;88:E38894.

39. de Waard PW, IJspeert JK, van den Berg TJ, de Jong PT.Intraocular light scattering in age-related cataracts. InvestOphthalmol Vis Sci 1992;33:61825.

40. Werner EB, Drance SM. Increased scatter of responses as aprecursor of visual eld changes in glaucoma. Can J Oph-thalmol 1977;12:1402.

41. Gardiner SK, Crabb DP, Fitzke FW, Hitchings RA. Reducingnoise in suspected glaucomatous visual elds by using a newspatial lter. Vision Res 2004;44:83948.

42. Turpin A, McKendrick AM, Johnson CA, Vingrys AJ. Prop-erties of perimetric threshold estimates from full threshold,ZEST, and SITA-like strategies, as determined by computersimulation. Invest Ophthalmol Vis Sci 2003;44:478795.

43. Malik R, Swanson WH, Garway-Heath DF. Development andevaluation of a linear staircase strategy for the measurement ofperimetric sensitivity. Vision Res 2006;46:295667.

44. Hot A, Dul MW, Swanson WH. Development and evaluationof a contrast sensitivity perimetry test for patients with glau-coma. Invest Ophthalmol Vis Sci 2008;49:304957.

45. Weber AJ, Harman CD. Structure-function relations of parasolcells in the normal and glaucomatous primate retina. InvestOphthalmol Vis Sci 2005;46:3197207.

46. Fortune B, Burgoyne CF, Cull GA, et al. Structural andfunctional abnormalities of retinal ganglion cells measuredin vivo at the onset of optic nerve head surface change inexperimental glaucoma. Invest Ophthalmol Vis Sci 2012;53:393950.

1 Devers Eye Institute, Legacy Research Institute, Portland, Oregon.2 Optometry and Vision Science, Indiana University, Bloomington, Indiana.automated perimetry with stimulus size III and V, matrix andmotion perimetry. Invest Ophthalmol Vis Sci 2008;50:9749.

13. Gardiner SK, Crabb DP. Frequency of testing for detecting

January 4, 2014.31. Liang KY, Zeger SK. Longitudinal data analysis using

generalized linear models. Biometrika 1986;73:1322.frequency-doubling technology perimetry. Invest OphthalmolVis Sci 2001;42:140410.

12. Wall M, Woodward KR, Doyle CK, Artes PH. Repeatabilityof automated perimetry: a comparison between standardInterface: an enabling tool for clinical visual psychophysics.J Vis [serial online] 2012;12:22, 1e5. Available at: http://www.journalofvision.org/content/12/11/22.long. Accessed