Stubeda et al. retrospectively compared 5 methods for analyzing individual Standard Automated Perimetry findings in 1230 eyes of 1230 patients with suspect or known glaucoma. The authors applied novel methods of estimating the sensitivities and specificities of the Glaucoma Hemifield Test (GHT),1 the Hodapp-Anderson-Parrish 2 (HAP2) method,2 the method of Foster et al. (Foster),3 the method used in the United Kingdom Glaucoma Treatment Study (UKGTS),4 and the method used in the Low-pressure Glaucoma Treatment Study (LoGTS).5 Fields were staged on the basis of Mean Deviation normal limits, and then separately analyzed after staging on the basis of OCT findings. Confirmatory testing of positive findings was not performed.
The authors did not test healthy subjects in order to assess specificity, but instead created a specificity proxy by defining as normal all fields having an MD not significant at p<10% or worse, or, alternatively, an OCT score indicating no detected structural damage. The authors also avoided use of a gold standard for diagnosis, citing lack of a uniformly accepted standard, and used levels of functional and structural damage as their reference standards. Thus, we will adopt the authors' terminology of proxy sensitivity and proxy specificity.
A finding of Outside Normal Limits was required for positive GHT result. The Foster analysis required satisfaction of two criteria, one of which was a positive GHT, suggesting that Foster should be more specific and less sensitive than GHT. A positive HAP2 required meeting any of three criteria, one of which was a positive GHT, suggesting that HAP2 should have lower specificity and higher sensitivity than the GHT. The UKGTS required meeting any of three criteria, suggesting that it too may have lower specificity and higher sensitivity than GHT. LoGTS required meeting either of two criteria, both of which are based upon the number of points at specific numbers of dB below age-normal.6 Because LoGTS ignores the fact that the range of SAP normal visual sensitivity changes by a factor of three between the macula and 27 degrees of eccentricity, one might expect it to underperform the other methods.
As one might expect, the authors found that the proxy sensitivities of the five methods differed very little when applied to fields defined by the authors as having the highest functional loss, but differed quite markedly in suspect/subtle disease, regardless of whether staged on the basis of fields or OCT findings. As theoretically expected, the authors found HAP2 and UKGTS to have the highest proxy sensitivities and the lowest proxy specificities. GHT and Foster had similar midrange proxy sensitivities and specificities, and LoGTS consistently showed the lowest proxy sensitivity but, unexpectedly, the highest proxy specificity.
It is perhaps worth noting that the authors' proxy false positive rates for Foster, GHT, and HAP2 were roughly twice the rates reported by Wu and colleagues using normal controls.7 Katz and colleagues found GHT false positive rates based upon normal controls that differed similarly from the authors' findings.8 Budenz and colleagues found HAP2 to have a false positive rate of 4%, compared to the authors' findings of ≥43%, however, Budenz required confirmation testing of all positive findings.9 Part of these differences might have been due to the fact that the authors compared methods using a single visual field examination, while the LoGTS, HAP2 and UKGTS require that their criteria be found in 2 consecutive fields and others have also required the same for GHT.7,9,12
It seems likely that false positive findings in the current paper were artifactually high for the simple reason that early glaucomatous field loss frequently is associated with MD values that are not statistically outside normal limits, even at p<10%
It seems likely that false positive findings in the current paper were artifactually high for the simple reason that early glaucomatous field loss frequently is associated with MD values that are not statistically outside normal limits, even at p<10%.10 One might also argue that defining eyes as abnormal only if they are detected by a metric as insensitive to early glaucomatous field loss as MD may not provide a realistic assessment of a method's sensitivity to subtle loss. Similar observations might be made regarding glaucomatous field loss in eyes having normal OCT findings.
We also should note that the authors' highest functional loss category was defined to include all fields having MDs that were statistically significant at p<0.5%, which is reached in the SITA test strategies when an MD is worse than approximately -6 dB. This is a level of visual field damage that usually is staged as early to moderate.11 In this study, the median MD was -7.3db, with an interquartile range of -11.3 to -5.3 dB.
More accurate specificity data might be obtained from analysis of long-term data in prospective studies. For example, patient eligibility in the EMGT required two consecutive field tests, both having GHT findings of Outside Normal Limits in the same GHT sector(s), or GHT findings of Borderline with supporting disc findings. Long-term follow-up showed a specificity of 97%.12 Long-term analyses of UKGTS and/or LoGTS results might provide similar information.5,13
Still, for the most part, the authors' findings qualitatively confirm a number of theoretical expectations, as described earlier.