Article Text
PDF
Abstract
Purpose To compare detection of retinal nerve fibre layer changes using GDx guided progression analysis (GPA) fast mode (which assumes fixed variability of a reference population) and extended mode (which measures individual variability), and to determine how they compare with photography and conventional visual field-based methods for identifying glaucoma progression.
Methods 172 eyes from 117 participants in the Diagnostic Innovations in Glaucoma Study (12 healthy, 108 glaucoma suspects and 52 glaucoma eyes) with ≥4 GDx VCC visits and ≥3 good quality GDx VCC scans at each visit were included.
Results Agreement between the GDx GPA fast mode and GDx GPA extended mode was limited. The GDx fast mode and extended mode detected 15 and 18 eyes, respectively, as ‘likely progression’, but only seven of them agreed. The conventional reference standard (stereophotograph-based optic disc and/or visual field progression) identified nine eyes as progressing, of which two eyes were also identified by the GDx fast mode and three eyes by the extended mode. In the GDx fast mode, we found that the progression detection varied depending on which two scans were included in the baseline and follow-up images.
Conclusion There was limited agreement between the GDx fast mode and the GDx extended mode for progression detection, and between different scans included in the GDx fast mode progression analysis. Longer follow-up is needed to determine the proportion of eyes classified as ‘likely progression’ by the GDx analysis that are early change and the proportion that are false positive results.
- glaucoma
- imaging
- diagnostic tests/investigation
- glaucoma
- field of vision
- imaging
- Intraocular pressure
- optic nerve
- diagnostic tests/investigation
- epidemiology
Statistics from Altmetric.com
- glaucoma
- imaging
- diagnostic tests/investigation
- glaucoma
- field of vision
- imaging
- Intraocular pressure
- optic nerve
- diagnostic tests/investigation
- epidemiology
Introduction
Over the last decade, imaging instruments have been studied and employed for glaucoma diagnosis and optic disc and RNFL progression detection.1–9 One of these instruments, the scanning laser polarimeter (SLP; GDx VCC, Carl Zeiss Meditec Inc. Dublin, CA, USA) has been shown to be effective in differentiating between glaucomatous and healthy eyes.8 10–12 Recently, the rate of retinal nerve fibre layer (RNFL) change measured with the GDx VCC has been shown to be faster in eyes progressing by visual field and stereophotographs than in eyes that remain stable.13 Although this study showed that the GDx RNFL measurements can be used to measure the rate of RNFL change in groups of eyes, evaluation of tools to assess progression in individual patients in clinical practice is still needed.
The GDx now includes a new software algorithm, the guided progression analysis (GPA), for detecting glaucoma progression. GPA uses two different modes, the ‘fast mode’ and the ‘extended mode’, to compare follow-up images with baseline images to determine the likelihood of progression. An advantage of the extended mode is that it uses measured variability from the three scans of an individual's eye. In contrast, the fast mode uses variability from an external cohort to compare the two baseline scans with the final two follow-up scans. It has been shown that the GDx fast mode detected glaucoma progression in a significant number of eyes showing progression by conventional methods.14 Based on these results, the authors suggested that the GDx fast mode could be used to complement clinical evaluation in detection of glaucoma progression.14 The aim of the current study was to compare the agreement in detection of glaucoma progression between the GDx GPA fast mode and GDx GPA extended mode, and to determine how they compare with conventional methods for detecting glaucoma progression.
Methods
Participants included in this prospective study were enrolled from the Diagnostic Innovations in Glaucoma Study.
All participants met Diagnostic Innovations in Glaucoma Study inclusion and exclusion criteria: best corrected acuity worse than 20/40, spherical refraction within +/−5 D, cylinder correction within +/−3 D, good quality optic disc stereophotographs and reliable SITA standard automated perimetry (SAP) 24-2 tests (≤33% fixation losses, false negatives and false positives).13 15 Ophthalmologic examinations and testing were completed annually and participants were excluded if they no longer met inclusion criteria.13 15
Subjects were required to have at least four GDx VCC visits with three good quality scans at each visit so that the GDx extended mode could be calculated.
Diagnostic classification
All patients with at least two repeatable abnormal SAP visual fields at baseline were classified as glaucomatous. Abnormal visual fields were defined as a pattern SD (PSD) with p<0.05 or a glaucoma hemifield test ‘outside normal limits’. The glaucoma suspects did not have abnormal baseline SAP visual fields but had at least one of the following indications in at least one eye: glaucomatous optic disc appearance based on masked review of stereophotographs by two experienced graders; elevated intraocular pressure (IOP) (>21 mm Hg); or were under treatment with IOP-lowering medication.
Individuals classified as normal were recruited from the general population through advertisement, as well as from University of California, San Diego staff. All normal participants had no abnormal clinical, optic disc photography or SAP findings in both eyes, and no history of increased IOP.
The definition of ‘likely progression’ as defined by the SAP GPA (a significant decrease from baseline pattern deviation of ≥3 identical test points on ≥3 consecutive tests) was used.16 All progression cases were evaluated by a masked glaucoma specialist and did not show evidence of artefacts or non-glaucomatous progression.
Baseline simultaneous stereoscopic optic disc photographs were graded as glaucomatous or normal based on the presence or absence of neuroretinal rim thinning, RNFL thinning or cupping characteristic of glaucoma. To detect progressive glaucomatous optic neuropathy, the photographs closest (within 1 year) to the baseline and last GDx date were reviewed by two experienced masked graders for evidence of progressive glaucomatous change (neuroretinal rim or RNFL thinning, or increased cupping). In cases of disagreement, a third observer adjudicated the decision.
Instrumentation
SLP images were acquired with the GDx VCC (software version 5.0) and analysed with the GDx GPA algorithm (software version 6.0).
The GDx GPA report uses colour-coded checkmarks in the summary box to display results of three different progression detection methods. When significant change is detected, ‘Possible progression’ (yellow checkmark) is flagged; when change is confirmed in two consecutive examinations ‘Likely progression’ (red checkmark) is flagged. An increase in RNFL is reported as ‘Possible increase’ (purple checkmark).
Eyes flagged as ‘likely progression’ in at least one of the three maps were considered as progressing.
The Image Progression Map, designed to be sensitive to focal RNFL loss, displays a fundus image with colour-coded areas showing RNFL change compared with the two baselines. At least 150 adjacent points clusters must show statistically significant change for the map to flag progression,
The TSNIT (temporal, superior, nasal, inferior, temporal) Progression Graph, designed to be sensitive to broader focal change, plots RNFL thickness around the optic disc with the most recent examination in red. Statistically significant changes in ≥4 adjacent segments from the calculation circle (from a total of 64 segments) are required for the map to flag progression.
The Summary Parameter Charts, designed to be most sensitive to diffuse change, plots three parameters over time: TSNIT Average, Superior Average and Inferior Average. If there is a significant linear trend (p<0.05) a regression line is drawn and the rate of change is displayed in μm/year.
The GDx GPA has two different methods for detecting ‘statistically significant’ change: the fast and extended mode. Fast mode requires only single images from each visit, while the extended mode requires at least three images from each visit. In the fast mode, the observed change is compared with variability derived from an independent outside sample population. In the extended mode, the observed change is compared with the test–retest variability measured on that same eye.
The ‘Change From Baseline’ algorithm, which is used in all fast mode maps and in the ‘Image Progression Map’ for the extended mode, is an event-based algorithm that compares change from the mean of the two baselines to measurement variability. The ‘Change from Baseline’ algorithm uses only the first two visits and last two visits to determine change and is most sensitive when the variability between the two baselines is small. The fast mode can detect the earliest changes only after the third test; the first two are used for defining the average baseline value and the third test is compared with this measurement. In contrast, the trend-based algorithm, ‘Statistical Image Mapping’ used in the ‘TSNIT Progression Graph’ and ‘Summary Parameter Chart’ of the extended mode has the advantage that all visits contribute to determining whether change is significant.
Eligible subjects were included based on the ability to run the extended mode (ie, subjects had three good quality images from four different visits). We only used images that were evenly illuminated, well focused, centred and with quality score 7 or higher. Scans with poor registration, SD >11.5 and typical scan score (TSS) <80 were excluded. Scans with TSS ≥80 have been shown to have minimal atypical retardation profiles.17 18 TSS provides a measure of the how typical the RNFL image is and ranges from 0 (very atypical) to 100 (very typical). To ensure comparability of the analysis we included the same baseline and follow-up dates for each method.
Since three GDx scans were acquired at each visit, we were able to analyse whether the choice of scans could influence the GDx GPA fast mode results. Specifically, the fast mode uses two scans for the baseline and two for the last follow-up examination. To compare GDx GPA fast mode results using different scan combinations acquired on the same day, we interchanged the scans included in the baseline and the last follow-up and ran the analysis three ways, using first and second scans acquired (scans 1 and 2), scans 2 and 3, and scans 1 and 3, respectively.
Statistical analysis
Statistical analysis was performed using JMP (V. 8.0.1; SAS Institute Inc.). Random effects models were used to estimate differences in the change in GDx measurements in progressing and non-progressing eyes to account for the lack of independence between eyes.
Results
We included 172 eyes (108 glaucoma suspect eyes, 52 glaucoma eyes and 12 healthy eyes) from 117 individuals. Seventy (60%) individuals were female. Most participants were Caucasian (78 (67%)) or African-American (36 (31%)). Table 1 shows the baseline demographic information, visual field indices and GDx RNFL parameters for participants at study entry.
Demographics and clinical characteristics, based on the diagnostic classification at baseline, for 172 eyes of 117 study participants*
The GDx fast mode detected 15 eyes as ‘likely progression’ and the GDx extended mode detected 18, but there was agreement in only seven. Two of the fast mode progressing eyes and three of the extended mode progressing eyes agreed with the conventional reference standard (figure 1). Overall agreement for detection of progressing and non-progressing eyes between the conventional methods and the GDx fast mode and the extended mode was observed in 152/172 (88.4%) and 151/172 (87.8%) eyes, respectively. Overall agreement between the fast mode and the extended mode was 153/172 (89.0%). The number of eyes identified as progressing by the Image Progression Map, TSNIT Progression Map and Summary Parameter Chart for the GDx fast mode was 3, 9 and 10, respectively, and for the GDx extended mode was 8, 10 and 10, respectively.
Venn diagram showing agreement between the GDx VCC guided progression analysis (GPA) extended mode, GDx VCC GPA fast mode and conventional methods of change detection by standardised photographic and visual field assessment.
We also evaluated the number of false positives in the group of the 12 healthy eyes (12 subjects) that were followed for an average of 38 (95% CI 29 to 47) months. None of the healthy eyes showed progression by the fast mode, while one eye showed progression by the extended mode.
In the GDx fast mode, we found that progression detection varies depending on which two scans are included in the baseline and follow-up analysis (figure 2). We evaluated three possibilities, using the first and second scan (scans 1 and 2), using the second and third scan (scans 2 and 3), and using the first and third scan (scans 1 and 3). Of the total 25 eyes flagged with ‘likely progression’ using either scan combination, only eight (32%) eyes were identified by all three combinations.
Venn diagram showing agreement between using different individual scans in the baseline and follow-up of the GDx VCC guided progression analysis (GPA) fast mode analysis. Using scans 1 and 2, scans 1 and 3, and scans 2 and 3 identified different eyes as ‘likely progression’.
The change in RNFL parameters over time is presented in table 2. We found a consistent significantly larger RNFL change in progressing eyes compared with non-progressing eyes as defined by GDx fast mode and GDx extended mode (most p values ≤0.002), but not when progression was defined by conventional methods (most p values ≥0.05).
Change in RNFL parameters during the follow-up time, stratified by the method that detected progression*
Figure 3 presents three different case examples illustrating agreement between the GDx fast and extended mode in eyes without SAP and stereophotographs progression. In figure 3A, the fast mode reports ‘likely progression’, whereas the extended mode does not. Figure 3B shows the extended mode detecting ‘likely progression’, but the fast mode does not, and figure 3C shows both modes showing ‘likely progression’.
(A) Print out of the fast mode (left) and extended mode (right) results for the left eye of a 61-year-old male glaucoma suspect with normal visual fields and photographs at baseline. The fast mode reports ‘likely progression’ in the ‘Summary Parameter Charts’, whereas the extended mode does not. No progression was found based on guided progression analysis (GPA) analysis of the visual fields or by standardised assessment of stereophotographs. (B) Print out of the fast mode (left) and extended mode (right) results for the right eye of a 54-year-old male glaucoma patient with glaucomatous appearing optic disc and visual fields at baseline. The ‘Extended Mode’ print out indicates ‘likely progression’ in the ‘TSNIT Progression Graph’, but the fast mode does not. No progression was found based on GPA analysis of the visual fields or by standardised assessment of stereophotographs. (C) Print out of the fast mode (left) and extended mode (right) results for the right eye of a 51-year-old female glaucoma suspect with glaucomatous appearing optic disc and normal visual fields at baseline. Both the ‘Fast Mode’ print out and the ‘Extended Mode’ print out show progression. The ‘Fast Mode’ (left) shows progression in the ‘Image Progression Map, TSNIT Progression Graph’ and the ‘Summary Parameter Charts’. The ‘Extended Mode’ shows ‘likely progression’ on the ‘TSNIT Progression Graph’ only.
Discussion
We found limited agreement between the GDx fast mode and the GDx extended mode for progression detection (figure 1). In addition, our results suggest that the GDx fast mode results can vary depending on the scans selected for analysis in the baseline and the last follow-up examinations. To our knowledge, this is the first report comparing progression detection using the fast and extended mode, and also the first assessing the effect of scan selection on GDx fast mode progression detection.
There are several possible explanations for the limited agreement between the GDx fast and extended modes for detecting glaucoma progression. First, the method for calculating the variability differs between the two modes. The GPA extended mode is based on the measured variability of an individual's scans, while the GPA fast mode is based on a fixed variability using an outside reference population. Therefore, if an eye has higher measurement variability than the average variability of reference population used in the fast mode, the fast mode might detect progression whereas the extended mode may not. Conversely, if an eye has lower variability than the reference population, the fast mode might not detect change while the GDx extended mode may identify progression.
The fast and the extended modes of the GDx each have their advantages and limitations. An advantage of the fast mode is that fewer scans are required at each visit, so image acquisition is less time-consuming than for the extended mode. If three scans are not available at each visit, then only the GDx fast mode can be calculated. Limitations of the fast mode include its dependence at least in part on which scans were included in the analysis (figure 2). In addition, GDx change algorithms (both modes) do not account for RNFL loss due to ageing. Ageing changes should be considered when interpreting any automated change detection method. The lack of objective interpretation of photographs and the variability of visual field results are among the limitations of conventional methods for progression detection.
We also found limited agreement between both GDx GPA modes and conventional methods for detecting glaucoma progression. Other investigators have also reported limited agreement between the GDx fast mode and stereophotograph or visual field-based progression detection methods, and between other imaging techniques and conventional methods for progression detection.14 19–23 One explanation is that RNFL loss might be preceded by changes in birefringence from alterations of the cytoarchitecture of RNFL and that the SLP might detect birefringence changes before change in RNFL thickness can be detected.24 This may at least in part explain why the GDx GPA software detects some progressing eyes that are not detected with stereophotograph or visual field assessment and why RNFL parameter changes do not differ between progressing and non-progressing eyes as defined by conventional methods. One study14 reported better agreement between GPA fast mode and conventional methods (50%) than the current study (22%). Possible explanations for this inconsistency include longer length of follow-up and larger sample size than the current study.
Even though the SLP has been available for almost a decade, assessment of progression is difficult because hardware and software improvements have made current scanning protocols incompatible with earlier scans for automated change detection. The newest SLP version, the GDx ECC with enhanced corneal compensator13 25 reduces atypical retardation patterns (ARPs) measured by TSS.18 26 27 The ability of SLP to discriminate between healthy and glaucomatous eyes has been shown to decrease when ARPs are present.6 17 28 Moreover, ARPs are not necessarily stable over time,18 and can therefore either mask true progressing eyes (if ARP increases over time) or cause the GPA to detect false progression (if ARP decreases over time). However, this same study suggests that GPA GDx VCC can be used to detect progression in eyes with minimal ARP. Since we limited our analysis to scans with TSS ≥80, and the baseline TSS in this study was high, with mean TSS score was 96.1 (95% CI 95.2 to 99.0), we believe the influence of change in TSS on progression detection is unlikely.18
In summary, we found limited agreement between the GDx fast and extended mode, and between both GDx modes and conventional methods for progression detection in glaucoma suspects and early to moderate glaucoma patients. Moreover, we found that fast mode results can vary depending on which scans are used in the baseline and the last follow-up examination. Each of the two GDx GPA modes have their strengths and limitations and should be used with visual field testing and a thorough clinical examination for patient management decisions.3 29 30
References
Footnotes
Disclosure RN Weinreb is a consultant to Carl Zeiss Meditec, Inc. and receives research equipment from Carl Zeiss Meditec. FA Medeiros receives research support/materials and honoraria from Carl Zeiss Meditec, Inc. LM Zangwill receives research equipment from Carl Zeiss Meditec, Inc.
Funding This study was supported in part by NEI R01-EY08208 (FAM), NEI R01-11008 (LMZ), and the Danish Eye Research Foundation. Participant retention incentive grants were provided in the form of glaucoma medication at no cost (Alcon Laboratories Inc., Allergan, Pfizer Inc., SANTEN Inc.).
Competing interests RN Weinreb is a consultant to Optovue Inc, Alcon Laboratories, Allergan Inc, Glaxo, and Pfizer Inc, and receives research materials/support from Heidelberg Engineering, GmbH,Optovue, Inc., Topcon Medical Systems, Inc., and Novartis. FA Medeiros is a consultant for Alcon Laboratories, Inc., Allergan, Inc. and Pfizer, receives research support/materials from Alcon, and Pfizer, Inc, and received honoraria from Alcon Laboratories Inc, Allergan Inc, Pfizer Inc, and Reichert, Inc. LM Zangwill receives research equipment from Heidelberg Engineering, GmbH, Optovue, Inc., and Topcon Medical Systems, Inc.
Patient consent Written informed consent was obtained from each participant.
Ethics approval Ethics approval was obtained from the Institutional Review Board of the UCSD Human Research Protection Program.
Provenance and peer review Not commissioned; externally peer reviewed.