Aims To validate the EyeSuite version of German Adaptive Threshold Estimation (GATE), a new thresholding algorithm for automated static perimetry.
Methods Specification of agreement and its clinical evaluation as validation criteria. Comparison of local differential luminance sensitivity (DLS) and test time values between the prototype version of GATE (GATEp) and a clinical trial version, implemented in EyeSuite (GATEe), by means of modified Bland–Altman plots. All examinations were performed on the Octopus 900 perimeter (Haag-Streit Inc., Köniz, Switzerland). Visually impaired patients (anterior ischaemic optic neuropathy [n=3], glaucomatous optic neuropathy [n=15], (post-)chiasmal visual pathway lesion [n=6], retinitis pigmentosa [n=6]) were either tested with grid 30A (30° excentricity, 83 test locations) or grid 84NO (90° excentricity, 109 test locations, patients with RP only).
Results The comparison of local DLS values showed good-to-acceptable agreement between GATEp and GATEe (bias <2 dB, limits of agreement [LOA] <5 dB) and very good repeatability for GATEp (bias <0.5 dB, LOA<3 dB). Median test times for GATEp and GATEe were 7.8 and 8.8 min for the 30° grid and 6.7 and 7.8 min for the 90° grid.
Conclusions GATEp and GATE, implemented in the commercially available EyeSuite software package (GATEe), show good agreement regarding local differential luminance sensitivity. GATEe can thus be also recommended for clinical practice.
Clinical trial number NCT01265628.
- Diagnostic tests/Investigation
- Field of vision
- Visual pathway
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Static automated perimetry is currently the mainstay of assessment of visual function, especially in glaucoma diagnostics and surveillance. Several algorithms have been developed in order to shorten examination duration to reduce fatigue effect,1 such as Swedish interactive thresholding algorithm (SITA),2 tendency-oriented perimetry,3 the dynamic strategy by Weber,4 continuous light increment perimetry5 and SPARK.6 In 2009, German Adaptive Threshold Estimation (GATE), another new fast-thresholding algorithm based on a modified 4-2-dB staircase strategy for static visual field examinations, was launched.7 GATE achieved comparable results to a conventional full threshold (FT) strategy and SITA, with a shorter examination time than FT and similar test times as SITA. In contrast to SITA, however, thresholds may be obtained for any test point arrangement and any ophthalmological pathology. In addition, GATE offers the possibility of condensed grid testing since examinations are relatively short. The results proved to be comparable to currently used strategies.7
Until now, GATE has only been available in a prototype version for laboratory use (GATEp). In order to introduce this fast-thresholding algorithm into clinical practice, it has therefore been incorporated into the commercially available EyeSuite Perimetry (GATEe) software by Haag-Streit Inc., Köniz, Switzerland. The primary objective of this study was to assess the agreement between this incorporated software version GATEe and the original version GATEp regarding local differential luminance sensitivity (DLS). The secondary outcomes were to compare the repeatability of GATEp and the test duration of the systems.
Subjects and methods
A total of 30 patients (17 male, 13 female, average age 58.7 years, age range 21–78) suffering from anterior ischaemic optic neuropathy (AION; n=3, median mean sensitivity (MS): 17.9 dB), glaucomatous optic neuropathy (n=15, MS 17.6 dB), chiasmal or postchiasmal lesions of the visual pathway (n=6, MS 20.0 dB) or retinitis pigmentosa (RP; n=6, MS 5.2 dB) each underwent six perimetric examinations on one eye. All patients were compensated for their participation. Two patients with glaucoma had to be excluded from the analysis because no GATEe data could be obtained because of problems with the driving laptop. Selection of patients was achieved via chart review or during a scheduled clinic visit.
The inclusion criteria were as follows: physical, intellectual and linguistic abilities in order to understand the test requirements, willingness to comply with the protocol, >18 years old and informed consent. For the study eye, spherical ametropia max ±8D (dioptres), cylindrical ametropia max ±3D, distant visual acuity better than 10/20, isocoria, pupil diameter > 3 mm.
The exclusion criteria were suspected lack of compliance, pregnancy, nursing, diabetic retinopathy, asthma, HIV-positive or AIDS, history of epilepsy or significant psychiatric disease (eg, dementia), history of stroke (except for patients with [post-]chiasmal visual pathway lesions), medications known to affect the visual field sensitivity, acute ocular infections (eg, keratitis, conjunctivitis, uveitis), severely dry eyes, miotic drugs, amblyopia, squint, nystagmus, albinism, keratoconus, intraocular surgery (except for uncomplicated cataract or glaucoma surgery) performed <3 months prior to screening, history or presence of macular disease and/or macular oedema, relevant opacities of central refractive media (cornea, lens, vitreous body), ocular trauma and any ocular pathology in either eye that may interfere with the ability to obtain visual fields, disc imaging or accurate intraocular pressure readings.
Instruments and strategies
All tests were carried out between November 2010 and June 2011 with Octopus 900 perimeters ( Haag-Streit Inc., Köniz, Switzerland). GATEp was executed by serial device 104, GATEe by serial device 894.
White on white static perimetry was performed with a background luminance of 10 cd/m2 with a pericentral diamond fixation target, Goldmann stimuli size III and a stimulus duration of 200 ms. The interstimulus intervals were unequal (GATEp: 1200 ms; GATEe: 1500 ms) due to a misleadingly labelled scale of the EyeSuite software and the percentage of catch trials was different (GATEp: 2%; GATEe: 10%). Furthermore, five test locations were tested twice for short-term-fluctuation evaluation with the GATEp version only. Fixation was checked via an infrared video camera and corrected if necessary. Corrective lenses were used according to the manufacturer's recommendations.
The GATE algorithm is described in details elsewhere.7 Essentially, it consists of two parts: the initial examination (GATE-i) begins with the testing of five predefined seed points. If deviating from the age-corrected normal hill of vision, all other starting stimuli intensities of the visual field will be adjusted. For all following examinations (GATE), the starting values are based on previously accessed local thresholds resulting in even shorter test times than for GATE-i. DLS values are estimated via a modified 4-2-dB staircase. In order to terminate the testing at a given test point, two yes-no-answer reversals are needed. The DLS is defined as the intensity between the brightest stimulus not seen and the dimmest stimulus seen. In areas of deep or absolute defects, premature termination is possible if a stimulus of maximum brightness is not confirmed or if the initial stimulus has not been seen. In these instances, test time can be reduced without loss of information.
Two visits within 14 days were scheduled for each patient. On both visits, GATEp was applied twice (GATEp1, GATEp2, serial device 104), GATEe only once (serial device 894). The sequence of the examinations was randomised for each visit. GATE-i was executed on the first visit, whereas the regular GATE strategy was used on the second visit. Two different grids with test point arrangements according to a polar coordinate system—respecting the vertical and horizontal median—were used because they better respect the retinal receptor and ganglion cell arrangement than rectangular grids.8 ,9 Furthermore, they allow for condensation of test points towards the visual field centre or within regions of interest9 that can be processed by the GATE software.7 Grid 30A (central 30° visual field, 83 test locations TL) for the patients with AION, glaucoma and visual pathway lesions; grid 84NO (whole 90° visual field, 109 TL) for patients with RA only—see online supplementary figures S1 and S2. For AION and patients with glaucoma, either the affected or the worse eye was chosen as study eye, whereas for patients with RP or chiasmal lesions the study eye was chosen by randomisation. The study complied with the tenets of the Declaration of Helsinki and was approved by the independent Ethics Committee of the Faculty of Medicine, Tuebingen University. Informed consent was obtained from all patients. The study was registered (http://www.clinicaltrials.gov; unique identifier NCT01265628).
Differential luminance sensitivity differences
Two factors are crucial in the decision of whether a new method may be used interchangeably with an already established method: the amount of agreement between the methods and its clinical evaluation. The approach to comparing methods as recommended by Bland and Altman10 was applied in order to assess statistical agreement of first GATEp1–GATEp2 (ie, retest reliability) and second GATEp1–GATEe. Modified Bland–Altman plots for these two comparisons and the two grids were drawn as follows: average DLS values of the examinations of all included patients were plotted against their differences. The median difference of DLS values between these examinations was depicted by a horizontal line and rated as the bias. Furthermore, the 2.5 centile and the 97.5 centile of the DLS differences (also depicted by horizontal lines) were established in order to specify the so-called limits of agreement (LOA). Like this, 95% of the differences between the measurements were assumed to lie within these limits.
Statistical agreement was therefore specified by the bias (representing a possible systematic error) and the LOA (representing the spread of differences between the measurements) and categorised by the following clinical evaluation criteria: for LOA ≤3 dB and a bias ≤0.5 dB, a very good agreement was assumed. Good agreement was stated for LOA ≤±4 dB and a bias of ≤±1 dB, acceptable agreement for LOA ≤±5 dB and a bias of ≤±2 dB, respectively. LOA >5 dB or a bias > 3 dB indicate bad agreement and were rated as not acceptable. These criteria have been defined considering the final step size of 2 dB of the GATE algorithm,7 ,11 the measurement accuracy of 0.5 dB of the Octopus 900 perimeters and an assumed short-term fluctuation of 1.5 dB (normative value for Octopus 10112). Since the criteria for the LOA are based on the assumption of a bias of 0 dB, LOA were corrected by the bias before evaluation (cLOA). Furthermore, according to the recent literature,13–17 maximally two adjacent test locations per examination were allowed to exceed deviations of ±5 dB (test points at the edges of the examined visual field region were not considered). If this would be the case and the clinical evaluation criteria would allow stating of acceptable to very good agreement, the two software versions could be recommended to be used interchangeably.
Examination durations of GATE-i and GATE were analysed and compared for GATEp and GATEe. Median test times were assessed and their range was specified by the 2.5 and the 97.5 centile.
Differential luminance sensitivity
The specification of statistical agreement of GATEp was assessed by the comparison of GATEp1–GATEp2. For results for grid 30A, see figure 1 (grid 84NO: online supplementary figure S3).
Table 1 shows a summary of the maximum, minimum and median differences and the LOA of the examinations performed with GATEp as taken from the Bland–Altman plots. The LOA corrected by the bias are called cLOA.
The modified Bland–Altman plots comparing the DLS results of GATEp1 versus GATEp2, grid 30A and grid 84NO all show very good agreement. The median values of the differences for grid 30A indicate a small bias of approximately 0.4 dB. For grid 84NO, no bias is found. If regarding the LOA separately for GATE-i and GATE (see online supplementary table S1), they are larger for GATE-i, representing a greater variability in the initial examinations.
Comparison of GATEp1 versus GATEe
Online supplementary figures S4 and S5 show the Bland–Altman plots for the comparison of GATEp1–GATEe.
Table 2 summarises the information that can be drawn from these Bland–Altman plots. A bias of 1.5 dB for grid 30A indicates systematically higher threshold values for GATEp1 compared with GATEe. The same is observed for grid 84NO, but only to a small extent (0.5 dB). The differences between GATEp1 and GATEe—depicted by the range of the LOA—are greater than those between GATEp1 and GATEp2.
For grid 30A, GATEp1 and GATEe show acceptable to very good agreement (84NO: very good). All points exceeding the 5dB deviation were checked for adjacency and location in the visual field and found to meet the criteria based on recent literature as mentioned above. Again, the range of LOA is greater for GATE-i (see online supplementary table S2) than for GATE.
Test duration results are shown in table 3. GATE proved to be faster than GATE-i by 1.5 min in median. Test times for grid 84NO and GATE were about 1 min shorter than for grid 30A. Overall median test duration for GATEp was 8.6 min (2.5 centile: 5.5 min, 97.5 centile: 11.6 min), for GATEe 9.3 min (6.3 min, 12.4 min).
Test–retest variability of threshold perimetry is known to vary greatly dependent on the sensitivity loss of a given test location.18–20 Taking into consideration that all included patients suffered from moderate to severe visual field loss, the agreement between the two methods is sufficient from a clinical point of view.
The fact that the agreement between GATEp and GATEe was worse than the repeatability of GATEp itself is not surprising since agreement of two methods is limited by the repeatability of these methods10 ,21 and variability is known to increase with decreasing sensitivity of patients.22 We observed a bias of approximately 1.5 dB for grid 30A, indicating a systematic tendency of GATEp to assess higher DLS values than GATEe, which may be due to the methodological differences between the two procedures. This bias is smaller for grid 84NO (approximately 0.5 dB) probably due to a ceiling effect resulting from the extended visual field losses due to advanced RP. Biases of approximately 1 dB have been found for comparisons between SITA algorithms and the FT strategy.19 ,23
The risk to bias an examination towards previous findings is especially high in cases of immediate, pronounced change. However, in the follow-up of most (mostly chronic) ophthalmological diseases, this is rarely the case and assuming normal conditions in these patients and neglecting previous findings may result in a considerable prolongation of threshold examination and thus provoke fatigue.
A worsening of at least 5–10 dB at 2–3 adjacent test locations of the non-central visual field has been defined as a progression or new manifestation of glaucoma in several studies.13 ,14 For example, Anderson and colleagues defined a progression of glaucoma as follows: a worsening of ≥3 points by ≥10 dB in an existing defect or a worsening of ≥2 new adjacent points by 10 dB.15 In the Advanced Glaucoma Intervention Study (AGIS), a minimum depression of 9 dB in peripheral test locations and a depression of 5 dB in paracentral points are needed to elevate the score that then may indicate a progression of visual field defects.16 This is why we also checked for adjacency of test points exceeding a DLS difference of 5 dB. Maximum differences between GATEp and GATEe were far below 10 dB, and the upper (ie, 97.5%) LOA values were below 5.3 dB for all test conditions.
The observed bias of test times was probably due to differences of the settings. A simulation of test times that we performed assuming identical settings for both software versions indicated that similar test times would have been achieved. For SITA standard, average or median test times of about 6–8 min were achieved by patients with visual field loss for a 24–2 pattern (52 TL, 0.12–0.15 min/TL) in several studies. Test times for healthy subjects were shorter: about 5 min duration.7 ,18 ,20 In this study with visually impaired patients, the median test times for GATEe were 8.8 min for grid 30A (83 TL, 0.11 min/TL) and 7.6 min for grid 84NO (109 TL, 0.07 min/TL). GATE might therefore even be faster than SITA standard, when testing the same patients with the same grid. For SITA, test times usually increase with increasing visual field defects,7 ,18 which is not observed for GATE. On the contrary, patients with RP with great visual field loss showed even faster test times than the other patients probably due to the comparatively high proportion of locations with absolute field loss.
In conclusion, the prototype version of the fast thresholding algorithm (GATEp) showed a very good repeatability. GATEe, implemented in the commercially available EyeSuite software package, showed very good-to-acceptable agreement with the GATEp algorithm in terms of measured DLS thresholds. Examination durations of GATEp were slightly shorter than for GATEe. The overall results of this study suggest that GATEp and GATEe can be used interchangeably. This thresholding algorithm with short examination duration is not restricted to glaucomatous field loss and allows for individually tailored locally condensed grids, and thus offers promising future options for detection and follow-up of visual field loss of various origin—also in clinical use.
Thanks to Jonathan Sunkersing for the linguistic revision of the manuscript. Thanks to Regine Grund and Andrea Mast for their practical support.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
Contributors All authors contributed substantially to the conception or design of the work; AFL, US and EK to the acquisition; US, AFL, CM and MM to the analysis and interpretation of data for the work. AFL was responsible for drafting the work, and all others for revising it critically for important intellectual content. They gave final approval of the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. RG and AM gave practical support in the acquisition, and JS contributed by revising the manuscript linguistically.
Funding This work was supported by Haag-Streit Inc., Köniz, Switzerland, who has sponsored the Octopus 900 perimeters, the EyeSuite Perimetry software and patients’ compensations. They were involved in the conception of the study design, but they had no part in the collection, analysis and interpretation of the data, in writing this report or the decision to submit the paper for publication.
Competing interests AFL reports grants and non-financial support from Haag-Streit Inc., Köniz, Switzerland, during the conduct of the study. CM has nothing to disclose; EK reports grants and non-financial support from Haag-Streit Inc., Köniz, Switzerland, during the conduct of the study. MM reports grants and non-financial support from Haag-Streit Inc., Köniz, Switzerland, during the conduct of the study; personal fees from Haag-Streit Inc., Köniz, Switzerland, outside the submitted work. US reports grants and non-financial support from Haag-Streit Inc., Köniz, Switzerland, during the conduct of the study; personal fees and non-financial support from Haag-Streit Inc., Köniz, Switzerland, other from Pharm-Allergan Inc., other from Pfizer Inc., other from MSD Inc., outside the submitted work.
Ethics approval Independent Ethics Committee of the Faculty of Medicine, Tuebingen University.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Additional until now unpublished data from the study are available to Annette Luithardt and the Visual Pathway team, University Eye Hospital Tübingen.