- Email: [email protected]
The Heidelberg Retina Tomograph Glaucoma Probability Score
The Heidelberg Retina Tomograph Glaucoma Probability Score Reproducibility and Measurement of Progression Nicholas G. Strouthidis, MD, MRCOphth,1,2 Shaban Demirel, OD, PhD,2 Ryo Asaoka, MD, PhD,1,3 Claudio Cossio-Zuniga, MD,1 David F. Garway-Heath, MD, FRCOphth1 Purpose: To evaluate the reproducibility of the Heidelberg retina tomograph (HRT) Glaucoma Probability Score (GPS) and assess its potential for monitoring progression. Design: Evaluation of diagnostic tests in a randomized, controlled clinical trial. Participants: For reproducibility, we included 43 ocular hypertensive (OHT) and 31 glaucoma subjects. For progression, we included 198 OHT and 21 control subjects. Methods: To study reproducibility, global GPS values were generated for HRT images acquired in a test–retest study. Images were acquired at 2 visits within 6 weeks of each other, by 2 different observers. To study progression, GPS values were generated for HRT images acquired prospectively (1993–2001). Linear regression of GPS against time was performed, with progression defined as a significant negative slope (P⬍0.05). Criterion specificity was estimated from the number of improving subjects (significant positive slope) and the number of progressing controls. Visual field (VF) progression in the same subjects was assessed using 3-omitting pointwise linear regression of sensitivity over time. Main Outcome Measures: Reproducibility of GPS was assessed using Bland–Altman analysis (mean difference, 95% limits of agreement). Progression was assessed by the number of OHT subjects identified as progressing, and by agreement with VF progression. Results: Reproducibility of GPS was better at its extremes (⫺0.01⫾0.20 for GPS 0 – 0.30, and 0.02⫾0.09 for GPS 0.78 –1.00) than in its mid range (0.07⫾0.54 for GPS 0.30 – 0.78). Estimated criterion specificity ranged from 95.2% (95% confidence interval, 76.1%–99.9%) to 96.8% (93.2%–98.5%). Twenty-five OHT subjects (12.6%) progressed by GPS, with 11 of the 25 (5.6%) also progressing by VF; 26 subjects (13.1%) progressed by VF alone. Conclusions: Changes in HRT GPS values between 0.30 and 0.78 should be interpreted with caution because the index has poorer reproducibility in this range. The global GPS progression algorithm performs at least as well as previously described rim area-based HRT progression strategies. Financial Disclosure(s): Proprietary or commercial disclosure may be found after the references. Ophthalmology 2010;117:724 –729 © 2010 by the American Academy of Ophthalmology.
There are 2 principal potential clinical applications of optic nerve head (ONH) imaging devices in the management of patients with glaucoma. The first is to assist the clinician in identifying whether or not a particular ONH is abnormal (glaucomatous) or is within normal limits. The second is to facilitate monitoring of disease progression through the assessment of ONH structural changes over time. The Heidelberg retina tomograph (HRT; Heidelberg Engineering, Heidelberg, Germany) is a confocal scanning laser ophthalmoscope that has been commercially available in various iterations for over a decade. The Glaucoma Probability Score (GPS) is a diagnostic classification system incorporated into the HRT’s current operational software (Explorer Version 3.0 onward). The term “diagnostic” should be used with caution. Classification systems such as the GPS and the Moorfields regression analysis (MRA; also available in the HRT Explorer software) provide the operator with information regarding the likelihood of a particular ONH being
724
© 2010 by the American Academy of Ophthalmology Published by Elsevier Inc.
abnormal. The classification should not be used in isolation to make a diagnosis of glaucoma, but rather all clinical features should be taken into consideration.1,2 The GPS differentiates between normal and glaucomatous eyes by initially generating a 3-dimensional mathematical model of ONH shape.3 The model contains 3 optic disc parameters (cup size, cup depth, and rim steepness) and 2 retinal nerve fiber layer (RNFL) parameters (horizontal RNFL curvature and vertical RNFL curvature). As glaucoma develops, the GPS model assumes that the RNFL and optic disc morphology changes such that the RNFL curvature lessens (flattens) as it thins, that the cup enlarges and deepens, and that the slope of the neuroretinal rim steepens. The parameters from the best 3-dimensional model fit to the data from an individual eye are used as inputs for a relevance vector machine (RVM). The output of the RVM can be compared with normative standards, enabling estimation of the probability that the ONH is outside normal limits. The ISSN 0161-6420/10/$–see front matter doi:10.1016/j.ophtha.2009.09.036
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score GPS does not require contour line placement, unlike the MRA,4 making it an operator-independent algorithm, at least with respect to defining the disc margin. In addition to the numerical GPS “score,” a graphical representation of the GPS is also generated with red crosses indicating an “outside normal limits” disc or disc sector, yellow exclamation marks indicating a “borderline” disc or disc sector, and green check marks indicating a “within normal limits” disc or disc sector. As the GPS generates data that a clinician may use to inform clinical decision making, it is imperative that the information provided be consistent. If the GPS classification were to change between repeated tests in the absence of glaucomatous progression, the information provided would be unreliable and its usefulness in clinical practice diminished. It also follows that, as the disease progresses, the GPS should increase as the likelihood that the ONH has become abnormal increases. If this assumption is correct, then it is reasonable to assume that changes in the GPS over time may be used to monitor glaucomatous progression. The purpose of this study was to assess the reproducibility of the GPS and to explore its potential as a measure of glaucomatous progression.
Methods Assessment of Glaucoma Probability Score Reproducibility Interobserver/intervisit global GPS reproducibility was estimated using HRT mean topographies acquired as part of an HRT test– retest study, which has been described in detail elsewhere.5,6 The reason why interobserver/intervisit GPS reproducibility was assessed is that this most likely reflects clinical practice whereby longitudinal HRT imaging may be performed by different technicians at each visit. In summary, 43 eyes with ocular hypertension (OHT) and 31 with primary open-angle glaucoma were recruited from a research clinic at Moorfields Eye Hospital. The characteristics of subjects included in the test–retest study are shown in Table 1. HRT Classic images (10° scan width) and HRT-II images (15° scan width) were acquired by 2 experienced operators (observers 1 and 2) on the same date (visit 1) and on a second date (visit 2) within 6 weeks of visit 1. Scan focus and scan depth were kept constant across visits for each subject. Subjects had no previous history of intraocular surgery and had all experienced ONH imaging using the HRT Classic. The acquired HRT Classic single topographies were imported as HRTport files into Heidelberg Explorer (Version 3.1.2.0, Heidelberg Engineering) with which mean topography images and global GPS were generated. The HRT-II mean topographies were generated using the same software. To assess interobserver/intervisit reproducibility, global GPS numerical scores were compared for the images acquired by observer 1 at visit 1 and by observer 2 at visit 2 using Bland–Altman plots.7 Reproducibility of the GPS was quantified using the bias (mean difference) and 95% limits of agreement.
Assessing Progression Using the Glaucoma Probability Score Progression was assessed in 198 OHT subjects and 21 control subjects followed prospectively with the HRT Classic and visual
Table 1. Characteristics of the Ocular Hypertensive (OHT) and Primary Open-Angle Glaucoma (POAG) Subjects Included in the Test–Retest Study Characteristics of Subjects
OHT
POAG
Number of eyes 43 31 Age (yrs) 67.8 (20.4–81.7) 71.5 (45.1–84.8) Baseline mean defect ⫺0.73 (⫺3.31 to 1.32) ⫺3.88 (⫺11.31 to ⫺1.37) (dB) Baseline global rim 1.28 (0.61–1.99) 1.03 (0.43–1.84) area (HRT classic) Baseline global rim 1.15 (0.53–2.23) 1.06 (0.37–1.71) area (HRT II) Image quality 22 (10–101) 26 (10–143) throughout study (MPHSD–HRT classic) Image quality 19 (10–78) 22 (10–140) throughout study (MPHSD–HRT II) MPHSD ⫽ mean pixel height standard deviation. Values are median (range).
field (VF) testing (1993–2001). No HRT-II images were included in these longitudinal series. These 2 groups of subjects have been described in detail elsewhere.8 Briefly, these subjects were originally recruited to a prospective randomized trial comparing betaxolol versus placebo that took place at Moorfields Eye Hospital between 1993 and 1998,9 and continue to be followed until the present day. We defined OHT as an untreated intraocular pressure ⬎22 mmHg and ⬍35 mmHg on ⱖ2 occasions within a 2-week period and a baseline mean Advanced Glaucoma Intervention Study VF score of 0 (Humphrey Field Analyzer, full-threshold algorithm, 24-2 test pattern).10 Control subjects were recruited from senior citizens groups or were the spouses or friends of subjects in the OHT cohort; they were not attending the eye clinic as patients and were not seeking care or undergoing checkups.11 Controls had a baseline intraocular pressure of ⬍21 mmHg, normal baseline VF test results (same criteria as in the OHT group), and were excluded if there was a self-reported family history of glaucoma or any coexistent ocular or neurologic pathology. In the current study, the same eye was selected for analysis as had been randomized in the original study. The OHT eye randomization was stratified according to risk of glaucomatous conversion, classified according to pattern electroretinogram results, intraocular pressure, and cup-to-disc ratio at the time of recruitment.9 Each subject had a minimum of 5 HRT mean topographies, with images of all quality (as measured by mean pixel height standard deviation) included. The reason for selecting images of all quality was to closely reflect the scenario encountered in clinical practice and to limit the possibility of excluding potentially useful clinical data. However, images were excluded on the grounds of image quality if the contour lines exported from the baseline mean topography could not be satisfactorily aligned in the follow-up mean topography. Images that were so grainy as to preclude accurate placement of the contour line were excluded. In total, 8 mean topographies were excluded for these reasons. The baseline HRT image was the first mean topography available for each subject and the baseline VF was taken as the VF test coinciding with, or nearest in acquisition date to, the baseline HRT mean topography. We conducted VF testing (24-2, full threshold program) with the Humphrey Field Analyzer (Carl Zeiss Meditec, Dublin, CA) every 4 months. We performed HRT Classic imaging annually for the first 2 years of
725
Ophthalmology Volume 117, Number 4, April 2010 Table 2. Characteristics of the Ocular Hypertensive (OHT) and Control Subjects Included in the Progression Assessment Characteristics of Subjects
OHT
Control
Number of subjects Age (yrs) Length of follow-up (yrs) Number of HRT examinations Number of visual field examinations Baseline mean defect (dB) Baseline global rim area (mm2) Image quality throughout study (MPHSD)
198 60 (32–79) 6.0 (2.3–7.2) 10 (5–16) 17 (5–33)
21 65 (41–77) 5.3 (3.1–6.8) 9 (8–11) 9 (7–14)
⫹0.1 (⫹3.0–2.7) 1.24 (0.63–2.31) 20 (7–186)
⫹0.1 (⫹2.6–2.4) 1.35 (0.86–2.51) 23 (9–80)
HRT ⫽ Heidelberg retina tomograph; MPHSD ⫽ mean pixel height standard deviation. Values are median (range).
the study and every 4 months thereafter. Three single topographies were acquired at each visit. The characteristics of all subjects used to assess progression are summarized in Table 2. For the current study, HRT Classic single topographies were exported as HRTport files into Heidelberg Explorer (Version 3.1.2.0, Heidelberg Engineering) and mean topographies were generated. Global GPS values were collected for each visit in a subject’s longitudinal series. A linear regression of GPS over time was performed for each subject using the measurements from baseline until the end of the study period. Progression was defined as a significant positive slope of GPS over time (P⬍0.05), with improvement defined as a significant negative slope (P⬍0.05). In the absence of a gold standard for measuring disease progression, 2 proxy measures were used to estimate specificity. The first estimate was derived from the total number of subjects (out of 219) demonstrating improvement. The second estimate was derived from the number of controls (out of 21) demonstrating progression. Progression in the OHT cohort using the GPS strategy was compared with 2 previously described progression algorithms applied to the same subjects—a VF trend analysis (3-omitting pointwise linear regression of sensitivity over time)12 and an HRT trend analysis (linear regression of rim area, using the 320-m reference plane, over time).8
Figure 2. Bland–Altman plot of interobserver, intervisit global Glaucoma Probability Score for Heidelberg Retina Tomograph (HRT)-II images. SD ⫽ standard deviation.
This study adhered to the tenets of the Declaration of Helsinki and had local ethical committee approval. In addition, subjects’ gave their consent to be studied after the risks and benefits of their participation had been explained to them. Statistical analyses were performed using Medcalc Version 7.4.2.0 (Medcalc Software, Mariakerke, Belgium) and using R (R Foundation for Statistical Computing, Vienna, Austria).
Results Reproducibility There was no significant difference between the interobserver differences and zero for both the HRT Classic (mean difference, 0.02; P ⫽ 0.22, paired samples t-test) and the HRT-II (mean difference, 0.001; P ⫽ 0.95, paired samples t-test). A Bland– Altman plot of interobserver/intervisit global GPS for HRT Classic images is depicted in Figure 1 and for HRT-II images in Figure 2. Repeated GPS measurements showed evidence of heteroskedasticity, in that the variance about repeat tests is not equal across the range of the variable. There seemed to be high reproducibility (tight distribution around the mean difference) at low and high GPS scores and poorer reproducibility (wide distribution around the mean difference) in between. The distribution in the middle range was narrower for HRT-II images compared with the HRT Classic images. All subject eyes were ranked according to their mean global GPS for the 2 observations and then binned approximately equally into low, medium, and high GPS tertiles (n ⫽ 25, 24, and 25, respectively). The low mean GPS range was 0 to 0.30, the medium ⬎0.30 to 0.78, and the high mean GPS was ⬎0.78 to 1.00. The biases (mean differences) and 95% limits of agreement generated for these 3 categories are shown in Table 3.
Progression
Figure 1. Bland–Altman plot of interobserver, intervisit global Glaucoma Probability Score for Heidelberg Retina Tomograph (HRT) Classic images. SD ⫽ standard deviation.
726
The mean (range) of GPS change was 0.006/year (⫺0.11 to 0.12). The estimated specificity for the GPS progression criterion ranged between 95% and 97% (Table 4). Twenty-five OHT subjects (12.6%) progressed by global GPS. The estimated specificity of the criteria for 3-omitting pointwise linear regression of VF results, and of HRT rim area linearly regressed over time, in this OHT
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score Table 3. Biases (Mean Differences) and 95% Limits of Agreement for Interobserver, Intervisit Global Glaucoma Probability Score (GPS) Using the Heidelberg Retina Tomograph (HRT) Classic and HRT-II GPS Range
HRT Classic
HRT-II
0–0.3 ⬎0.3–0.78 ⬎0.78–1.00
⫺0.01⫾0.20 0.07⫾0.54 0.02⫾0.09
⫺0.01⫾0.12 0.01⫾0.37 ⫺0.01⫾0.11
cohort, have been published previously and are similar to that of the GPS strategy (95%–98% for both strategies).8 A comparison of the frequency of identified progression between the 3 different trend analyses was likely therefore to be valid because the estimated specificities are so closely matched. A Venn diagram comparing the percentage of OHT subjects progressing by each strategy is shown in Figure 3. Eleven subjects (5.5%) progressed by both GPS and by VF. One OHT eye was “outside” of the morphologic limits of the GPS model so could not be used to generate a slope of GPS; this subject was, however, included in the progression analysis and assumed to not be progressing. All ONHs in the test–retest component of this study were within the morphologic limits of the GPS model. There was no relationship between global GPS and mean pixel height standard deviation in the 2001 mean topographies examined longitudinally in the OHT and control subjects (Pearson correlation coefficient r ⫽ 0.03; P ⫽ 0.166).
Discussion The ability of the GPS to discriminate between normal and glaucomatous ONHs has been assessed extensively in recent literature and there is a general consensus that it is at least as effective as MRA and has the advantage that it is operator independent. However, GPS is more likely than MRA to misclassify large normal ONHs as abnormal.13–19 There is evidence to suggest that a within normal limits GPS classification is more useful than the MRA to confirm that a disc is normal, whereas an outside normal limits MRA classification is more useful than the GPS in confirming that a disc is abnormal.19 Disease severity has also been shown to influence GPS classification, with GPS having a higher sensitivity and lower specificity than MRA in patients with mild glaucomatous VF damage, whereas MRA better discriminates subjects with severe glaucomatous VF damage.17 Alencar et al20 recently showed that baseline GPS can be used to predict which glaucoma suspects will go on to display VF deterioration and optic disc change, the latter having a similar predictive value to expert evaluation of stereophotographs. In all of these studies, the GPS classifications are based on the manufacturer’s own suggested Table 4. Specificity Estimates for the Glaucoma Probability Score Progression Strategy Status
Specificity Estimate
95% Confidence Interval
No significant improvement No progression in controls
212/219 (96.8%) 20/21 (95.2%)
93.2%–98.5% 76.1%–99.9%
Figure 3. Venn diagram comparing the number (percentage) of ocular hypertensive subjects identified as progressing by Glaucoma Probability Score (GPS) trend analysis, rim area (RA) trend analysis and visual field (VF) 3-omitting pointwise linear regression. Specificity is matched at approximately 97% for the 3 strategies.
cutoff values, with 0 to 0.27 classified as within normal limits, 0.28 to 0.64 as borderline, and 0.65 to 1.00 as outside normal limits.20 Borderline classifications should be interpreted with a degree of caution, because this study demonstrates that the GPS has poorer reproducibility in this range. This suggests that eyes with scores close to the upper and lower cutoffs may well be differently classified on repeated testing. This is unfortunate; eyes with very low or very high GPS values present the least diagnostic confusion and it is in the mid range where the clinician would benefit most from reliable information. When interpreting the GPS in clinical practice, it would probably be prudent to assess the classification alongside the actual numerical score. This would help to establish whether a “within normal limits” or “outside normal limits” classification is at the fringes of the borderline cutoff and should therefore be given less credence than if they were well away from the boundary values. The performance of both HRT Classic and HRT-II is poorer in the mid range compared with the high GPS range, suggesting that one should be cautious interpreting mid range global GPS scores, regardless of whether HRT Classic or HRT-II acquisitions are used. It is difficult to fully explain the pattern of GPS reproducibility observed in this study without precise knowledge of the proprietary RVM and learning data sets used in the HRT. It is possible that the RVM may place relevance on a particular parameter with a high degree of intertest variability. The RVM effectively operates in 5 dimensions if it uses all 5 parameters from the 3-dimensional model so the mapping between any ONH structural change and a given GPS change is likely to be highly nonlinear. This is somewhat analogous to the relationship between VF sensitivity and disease progression.21 As glaucoma progresses and VF sensitivity decreases, variability increases. The relationship between disease severity and variability holds true until very
727
Ophthalmology Volume 117, Number 4, April 2010 advanced field loss (the “floor effect”), at which point sensitivity values become less variable because they are anchored at the lower dynamic range of the perimeters, or indeed, the disease has reached end stage. Our assessment of GPS reproducibility is made with the assumption that no disease progression should have occurred in the 6-week interim between baseline and followup imaging during the test–retest study. Indeed, there was no difference between the interobserver differences and zero, suggesting that no disease-related change occurred on average in the 6-week time period. However, when interpreting these results it is important to consider the possibility, although remote, that the global GPS scores may have changed between visits because of ONH structural change, rather than owing to measurement error alone. Despite the apparent limitations regarding the reproducibility of the GPS in its mid range, the results of our study suggest that it can be used to measure progression in OHT eyes. In particular, it seems to perform at least as well as an HRT rim area trend analysis at the same level of estimated specificity, identifying a similar number of subjects as progressing (25 by GPS, 24 by rim area). The agreement between these 2 HRT progression estimates was, however, poor (only 7 subjects progressed by both GPS and rim area). Rim area is not a parameter that contributes to the calculated value of GPS. Some parameters assessed in the ONH model (cup size and rim steepness) reflect changes in rim area, whereas other morphologic changes (RNFL curvature, cup depth) do not necessarily reflect rim area change. This could explain why there is some agreement between GPS and rim area change but in the majority the 2 measures are identifying different eyes as progressing. Another potential explanation is that, because the rim area is reference plane dependent, reference height variability may be masking change in some eyes. Agreement between HRT and VF progression was higher for the GPS trend analysis than for the rim area trend analysis (11 subjects compared with 7 subjects). The poor agreement with VF is disappointing, but is consistent with other published studies using both an alternative HRT progression strategy and an alternative imaging device.22,23 The level of agreement between structural and functional progression measured in this study is disappointing. This perhaps suggests that there is temporal dissociation between structural and functional change in glaucomatous eyes or that the currently available structural and functional tests provide insight into fundamentally different facets of glaucomatous pathophysiology. Another possibility is that measurement error (noise) in structural and functional data collected from patients may be masking progression in some eyes while also giving rise to spurious/ unreliable rates of loss in others. The results of this study, in terms of GPS progression, may not be fully applicable to eyes with manifest glaucoma. In the current study, which was performed in OHT eyes, a fairly wide range of GPS slopes (⫺0.11 to 0.12 per year) was recorded. In glaucomatous eyes, the baseline GPS values are likely to be high, meaning that the degree of change identifiable as “progression” is likely to be small. Although this is countered by the fact that reproducibility is
728
higher at this stage of disease (lower “noise”), the ability to detect change (“signal”) may be reduced. In this study, we have shown that the GPS has good reproducibility at the extremes of the score but poorer reproducibility in the “borderline” range. In clinical practice, GPS classification should be assessed alongside the actual numerical score. The GPS trend analysis described in this study is at least as effective at identifying OHT subjects as progressing as an HRT rim area trend analysis at the same level of specificity, although it largely identified different subjects as progressing. The poor agreement with VF progression, as well as rim area progression, suggests that GPS progression should not be used in isolation to detect glaucomatous progression.
References 1. Garway-Heath DF. Early diagnosis in glaucoma. Prog Brain Res 2008;173:47–57. 2. Garway-Heath DF, Friedman DS. How should results from clinical tests be integrated into the diagnostic process? Ophthalmology 2006;113:1479 – 80. 3. Swindale NV, Stjepanovic G, Chin A, Mikelberg FS. Automated analysis of normal and glaucomatous optic nerve head topography images. Invest Ophthalmol Vis Sci 2000;41: 1730 – 42. 4. Wollstein G, Garway-Heath DF, Hitchings RA. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology 1998;105:1557– 63. 5. Strouthidis NG, White ET, Owen VM, et al. Factors affecting the test-retest variability of Heidelberg retina tomograph and Heidelberg retina tomograph II measurements. Br J Ophthalmol 2005;89:1427–32. 6. Strouthidis NG, White ET, Owen VM, et al. Improving the repeatability of Heidelberg retina tomograph and Heidelberg retina tomograph II rim area measurements. Br J Ophthalmol 2005;89:1433–7. 7. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. 8. Strouthidis NG, Scott A, Peter NM, Garway-Heath DF. Optic disc and visual field progression in ocular hypertensive subjects: detection rates, specificity, and agreement. Invest Ophthalmol Vis Sci 2006;47:2904 –10. 9. Kamal D, Garway-Heath D, Ruben S, et al. Results of the betaxolol versus placebo treatment trial in ocular hypertension. Graefes Arch Clin Exp Ophthalmol 2003;241:196 –203. 10. Advanced Glaucoma Intervention Study. 2. Visual field test scoring and reliability. Ophthalmology 1994;101:1445–55. 11. Kamal DS, Viswanathan AC, Garway-Heath DF, et al. Detection of optic disc change with the Heidelberg retina tomograph before confirmed visual field change in ocular hypertensives converting to early glaucoma. Br J Ophthalmol 1999;83: 290 – 4. 12. Gardiner SK, Crabb DP. Examination of different pointwise linear regression methods for determining visual field progression. Invest Ophthalmol Vis Sci 2002;43:1400 –7. 13. De Leon-Ortega JE, Sakata LM, Monheit BE, et al. Comparison of diagnostic accuracy of Heidelberg Retina Tomograph II and Heidelberg Retina Tomograph 3 to discriminate glaucomatous and nonglaucomatous eyes. Am J Ophthalmol 2007; 144:525–32.
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score 14. Oddone F, Centofanti M, Rossetti L, et al. Exploring the Heidelberg Retinal Tomograph 3 diagnostic accuracy across disc sizes and glaucoma stages: a multicenter study. Ophthalmology 2008;115:1358 – 65. 15. Harizman N, Zelefsky JR, Ilitchev E, et al. Detection of glaucoma using operator-dependent versus operator-independent classification in the Heidelberg retinal tomograph-III. Br J Ophthalmol 2006;90:1390 –2. 16. Burgansky-Eliash Z, Wollstein G, Bilonick RA, et al. Glaucoma detection with the Heidelberg retina tomograph 3. Ophthalmology 2007;114:466 –71. 17. Ferreras A, Pajarin AB, Polo V, et al. Diagnostic ability of Heidelberg Retina Tomograph 3 classifications: glaucoma probability score versus Moorfields regression analysis. Ophthalmology 2007;114:1981–7. 18. Coops A, Henson DB, Kwartz AJ, Artes PH. Automated analysis of Heidelberg retina tomograph optic disc images by glaucoma probability score. Invest Ophthalmol Vis Sci 2006; 47:5348 –55.
19. Zangwill LM, Jain S, Racette L, et al. The effect of disc size and severity of disease on the diagnostic accuracy of the Heidelberg Retina Tomograph glaucoma probability score. Invest Ophthalmol Vis Sci 2007;48:2653– 60. 20. Alencar LM, Bowd C, Weinreb RN, et al. Comparison of HRT-3 glaucoma probability score and subjective stereophotograph assessment for prediction of progression in glaucoma. Invest Ophthalmol Vis Sci 2008;49:1898 – 906. 21. Heijl A, Lindgren A, Lindgren G. Test-retest variability in glaucomatous visual fields. Am J Ophthalmol 1989;108: 130 –5. 22. Artes PH, Chauhan BC. Longitudinal changes in the visual field and optic disc in glaucoma. Prog Retin Eye Res 2005; 24:333–54. 23. Wollstein G, Schuman JS, Price LL, et al. Optical coherence tomography longitudinal evaluation of retinal nerve fiber layer thickness in glaucoma. Arch Ophthalmol 2005; 123:464 –70.
Footnotes and Financial Disclosures Originally received: April 17, 2009. Final revision: September 19, 2009. Accepted: September 22, 2009. Available online: January 4, 2010.
Nicholas G. Strouthidis - research support - Heidelberg Engineering (S)
Manuscript no. 2009-538.
1
NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom. 2 3
Devers Eye Institute, Legacy Health System, Portland, Oregon.
Department of Ophthalmology, Hamamatsu University School of Medicine, Hamamatsu City, Japan. Financial Disclosure(s): The authors have made the following disclosures: David F. Garway-Heath - research support - Heidelberg Engineering, OptoVue and Carl Zeiss Meditec; consultant - Carl Zeiss Meditec (C and S)
David F. Garway-Heath has received a proportion of his funding from the Department of Health’s National Institute for Health Research Biomedical Research Centre at Moorfields Eye Hospital and the UCL Institute of Ophthalmology. The views expressed in this publication are those of the authors and not necessarily those of the Department of Health. Nicholas G. Strouthidis is funded by an unrestricted educational grant from Heidelberg Engineering and by a Royal College of Ophthalmologists/Pfizer Travel Fellowship. Correspondence: Nicholas G. Strouthidis, MD, MRCOphth, Glaucoma Research Unit, Moorfields Eye Hospital, 162 City Road, London, EC1V 2PD. E-mail: [email protected].
729
There are 2 principal potential clinical applications of optic nerve head (ONH) imaging devices in the management of patients with glaucoma. The first is to assist the clinician in identifying whether or not a particular ONH is abnormal (glaucomatous) or is within normal limits. The second is to facilitate monitoring of disease progression through the assessment of ONH structural changes over time. The Heidelberg retina tomograph (HRT; Heidelberg Engineering, Heidelberg, Germany) is a confocal scanning laser ophthalmoscope that has been commercially available in various iterations for over a decade. The Glaucoma Probability Score (GPS) is a diagnostic classification system incorporated into the HRT’s current operational software (Explorer Version 3.0 onward). The term “diagnostic” should be used with caution. Classification systems such as the GPS and the Moorfields regression analysis (MRA; also available in the HRT Explorer software) provide the operator with information regarding the likelihood of a particular ONH being
724
© 2010 by the American Academy of Ophthalmology Published by Elsevier Inc.
abnormal. The classification should not be used in isolation to make a diagnosis of glaucoma, but rather all clinical features should be taken into consideration.1,2 The GPS differentiates between normal and glaucomatous eyes by initially generating a 3-dimensional mathematical model of ONH shape.3 The model contains 3 optic disc parameters (cup size, cup depth, and rim steepness) and 2 retinal nerve fiber layer (RNFL) parameters (horizontal RNFL curvature and vertical RNFL curvature). As glaucoma develops, the GPS model assumes that the RNFL and optic disc morphology changes such that the RNFL curvature lessens (flattens) as it thins, that the cup enlarges and deepens, and that the slope of the neuroretinal rim steepens. The parameters from the best 3-dimensional model fit to the data from an individual eye are used as inputs for a relevance vector machine (RVM). The output of the RVM can be compared with normative standards, enabling estimation of the probability that the ONH is outside normal limits. The ISSN 0161-6420/10/$–see front matter doi:10.1016/j.ophtha.2009.09.036
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score GPS does not require contour line placement, unlike the MRA,4 making it an operator-independent algorithm, at least with respect to defining the disc margin. In addition to the numerical GPS “score,” a graphical representation of the GPS is also generated with red crosses indicating an “outside normal limits” disc or disc sector, yellow exclamation marks indicating a “borderline” disc or disc sector, and green check marks indicating a “within normal limits” disc or disc sector. As the GPS generates data that a clinician may use to inform clinical decision making, it is imperative that the information provided be consistent. If the GPS classification were to change between repeated tests in the absence of glaucomatous progression, the information provided would be unreliable and its usefulness in clinical practice diminished. It also follows that, as the disease progresses, the GPS should increase as the likelihood that the ONH has become abnormal increases. If this assumption is correct, then it is reasonable to assume that changes in the GPS over time may be used to monitor glaucomatous progression. The purpose of this study was to assess the reproducibility of the GPS and to explore its potential as a measure of glaucomatous progression.
Methods Assessment of Glaucoma Probability Score Reproducibility Interobserver/intervisit global GPS reproducibility was estimated using HRT mean topographies acquired as part of an HRT test– retest study, which has been described in detail elsewhere.5,6 The reason why interobserver/intervisit GPS reproducibility was assessed is that this most likely reflects clinical practice whereby longitudinal HRT imaging may be performed by different technicians at each visit. In summary, 43 eyes with ocular hypertension (OHT) and 31 with primary open-angle glaucoma were recruited from a research clinic at Moorfields Eye Hospital. The characteristics of subjects included in the test–retest study are shown in Table 1. HRT Classic images (10° scan width) and HRT-II images (15° scan width) were acquired by 2 experienced operators (observers 1 and 2) on the same date (visit 1) and on a second date (visit 2) within 6 weeks of visit 1. Scan focus and scan depth were kept constant across visits for each subject. Subjects had no previous history of intraocular surgery and had all experienced ONH imaging using the HRT Classic. The acquired HRT Classic single topographies were imported as HRTport files into Heidelberg Explorer (Version 3.1.2.0, Heidelberg Engineering) with which mean topography images and global GPS were generated. The HRT-II mean topographies were generated using the same software. To assess interobserver/intervisit reproducibility, global GPS numerical scores were compared for the images acquired by observer 1 at visit 1 and by observer 2 at visit 2 using Bland–Altman plots.7 Reproducibility of the GPS was quantified using the bias (mean difference) and 95% limits of agreement.
Assessing Progression Using the Glaucoma Probability Score Progression was assessed in 198 OHT subjects and 21 control subjects followed prospectively with the HRT Classic and visual
Table 1. Characteristics of the Ocular Hypertensive (OHT) and Primary Open-Angle Glaucoma (POAG) Subjects Included in the Test–Retest Study Characteristics of Subjects
OHT
POAG
Number of eyes 43 31 Age (yrs) 67.8 (20.4–81.7) 71.5 (45.1–84.8) Baseline mean defect ⫺0.73 (⫺3.31 to 1.32) ⫺3.88 (⫺11.31 to ⫺1.37) (dB) Baseline global rim 1.28 (0.61–1.99) 1.03 (0.43–1.84) area (HRT classic) Baseline global rim 1.15 (0.53–2.23) 1.06 (0.37–1.71) area (HRT II) Image quality 22 (10–101) 26 (10–143) throughout study (MPHSD–HRT classic) Image quality 19 (10–78) 22 (10–140) throughout study (MPHSD–HRT II) MPHSD ⫽ mean pixel height standard deviation. Values are median (range).
field (VF) testing (1993–2001). No HRT-II images were included in these longitudinal series. These 2 groups of subjects have been described in detail elsewhere.8 Briefly, these subjects were originally recruited to a prospective randomized trial comparing betaxolol versus placebo that took place at Moorfields Eye Hospital between 1993 and 1998,9 and continue to be followed until the present day. We defined OHT as an untreated intraocular pressure ⬎22 mmHg and ⬍35 mmHg on ⱖ2 occasions within a 2-week period and a baseline mean Advanced Glaucoma Intervention Study VF score of 0 (Humphrey Field Analyzer, full-threshold algorithm, 24-2 test pattern).10 Control subjects were recruited from senior citizens groups or were the spouses or friends of subjects in the OHT cohort; they were not attending the eye clinic as patients and were not seeking care or undergoing checkups.11 Controls had a baseline intraocular pressure of ⬍21 mmHg, normal baseline VF test results (same criteria as in the OHT group), and were excluded if there was a self-reported family history of glaucoma or any coexistent ocular or neurologic pathology. In the current study, the same eye was selected for analysis as had been randomized in the original study. The OHT eye randomization was stratified according to risk of glaucomatous conversion, classified according to pattern electroretinogram results, intraocular pressure, and cup-to-disc ratio at the time of recruitment.9 Each subject had a minimum of 5 HRT mean topographies, with images of all quality (as measured by mean pixel height standard deviation) included. The reason for selecting images of all quality was to closely reflect the scenario encountered in clinical practice and to limit the possibility of excluding potentially useful clinical data. However, images were excluded on the grounds of image quality if the contour lines exported from the baseline mean topography could not be satisfactorily aligned in the follow-up mean topography. Images that were so grainy as to preclude accurate placement of the contour line were excluded. In total, 8 mean topographies were excluded for these reasons. The baseline HRT image was the first mean topography available for each subject and the baseline VF was taken as the VF test coinciding with, or nearest in acquisition date to, the baseline HRT mean topography. We conducted VF testing (24-2, full threshold program) with the Humphrey Field Analyzer (Carl Zeiss Meditec, Dublin, CA) every 4 months. We performed HRT Classic imaging annually for the first 2 years of
725
Ophthalmology Volume 117, Number 4, April 2010 Table 2. Characteristics of the Ocular Hypertensive (OHT) and Control Subjects Included in the Progression Assessment Characteristics of Subjects
OHT
Control
Number of subjects Age (yrs) Length of follow-up (yrs) Number of HRT examinations Number of visual field examinations Baseline mean defect (dB) Baseline global rim area (mm2) Image quality throughout study (MPHSD)
198 60 (32–79) 6.0 (2.3–7.2) 10 (5–16) 17 (5–33)
21 65 (41–77) 5.3 (3.1–6.8) 9 (8–11) 9 (7–14)
⫹0.1 (⫹3.0–2.7) 1.24 (0.63–2.31) 20 (7–186)
⫹0.1 (⫹2.6–2.4) 1.35 (0.86–2.51) 23 (9–80)
HRT ⫽ Heidelberg retina tomograph; MPHSD ⫽ mean pixel height standard deviation. Values are median (range).
the study and every 4 months thereafter. Three single topographies were acquired at each visit. The characteristics of all subjects used to assess progression are summarized in Table 2. For the current study, HRT Classic single topographies were exported as HRTport files into Heidelberg Explorer (Version 3.1.2.0, Heidelberg Engineering) and mean topographies were generated. Global GPS values were collected for each visit in a subject’s longitudinal series. A linear regression of GPS over time was performed for each subject using the measurements from baseline until the end of the study period. Progression was defined as a significant positive slope of GPS over time (P⬍0.05), with improvement defined as a significant negative slope (P⬍0.05). In the absence of a gold standard for measuring disease progression, 2 proxy measures were used to estimate specificity. The first estimate was derived from the total number of subjects (out of 219) demonstrating improvement. The second estimate was derived from the number of controls (out of 21) demonstrating progression. Progression in the OHT cohort using the GPS strategy was compared with 2 previously described progression algorithms applied to the same subjects—a VF trend analysis (3-omitting pointwise linear regression of sensitivity over time)12 and an HRT trend analysis (linear regression of rim area, using the 320-m reference plane, over time).8
Figure 2. Bland–Altman plot of interobserver, intervisit global Glaucoma Probability Score for Heidelberg Retina Tomograph (HRT)-II images. SD ⫽ standard deviation.
This study adhered to the tenets of the Declaration of Helsinki and had local ethical committee approval. In addition, subjects’ gave their consent to be studied after the risks and benefits of their participation had been explained to them. Statistical analyses were performed using Medcalc Version 7.4.2.0 (Medcalc Software, Mariakerke, Belgium) and using R (R Foundation for Statistical Computing, Vienna, Austria).
Results Reproducibility There was no significant difference between the interobserver differences and zero for both the HRT Classic (mean difference, 0.02; P ⫽ 0.22, paired samples t-test) and the HRT-II (mean difference, 0.001; P ⫽ 0.95, paired samples t-test). A Bland– Altman plot of interobserver/intervisit global GPS for HRT Classic images is depicted in Figure 1 and for HRT-II images in Figure 2. Repeated GPS measurements showed evidence of heteroskedasticity, in that the variance about repeat tests is not equal across the range of the variable. There seemed to be high reproducibility (tight distribution around the mean difference) at low and high GPS scores and poorer reproducibility (wide distribution around the mean difference) in between. The distribution in the middle range was narrower for HRT-II images compared with the HRT Classic images. All subject eyes were ranked according to their mean global GPS for the 2 observations and then binned approximately equally into low, medium, and high GPS tertiles (n ⫽ 25, 24, and 25, respectively). The low mean GPS range was 0 to 0.30, the medium ⬎0.30 to 0.78, and the high mean GPS was ⬎0.78 to 1.00. The biases (mean differences) and 95% limits of agreement generated for these 3 categories are shown in Table 3.
Progression
Figure 1. Bland–Altman plot of interobserver, intervisit global Glaucoma Probability Score for Heidelberg Retina Tomograph (HRT) Classic images. SD ⫽ standard deviation.
726
The mean (range) of GPS change was 0.006/year (⫺0.11 to 0.12). The estimated specificity for the GPS progression criterion ranged between 95% and 97% (Table 4). Twenty-five OHT subjects (12.6%) progressed by global GPS. The estimated specificity of the criteria for 3-omitting pointwise linear regression of VF results, and of HRT rim area linearly regressed over time, in this OHT
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score Table 3. Biases (Mean Differences) and 95% Limits of Agreement for Interobserver, Intervisit Global Glaucoma Probability Score (GPS) Using the Heidelberg Retina Tomograph (HRT) Classic and HRT-II GPS Range
HRT Classic
HRT-II
0–0.3 ⬎0.3–0.78 ⬎0.78–1.00
⫺0.01⫾0.20 0.07⫾0.54 0.02⫾0.09
⫺0.01⫾0.12 0.01⫾0.37 ⫺0.01⫾0.11
cohort, have been published previously and are similar to that of the GPS strategy (95%–98% for both strategies).8 A comparison of the frequency of identified progression between the 3 different trend analyses was likely therefore to be valid because the estimated specificities are so closely matched. A Venn diagram comparing the percentage of OHT subjects progressing by each strategy is shown in Figure 3. Eleven subjects (5.5%) progressed by both GPS and by VF. One OHT eye was “outside” of the morphologic limits of the GPS model so could not be used to generate a slope of GPS; this subject was, however, included in the progression analysis and assumed to not be progressing. All ONHs in the test–retest component of this study were within the morphologic limits of the GPS model. There was no relationship between global GPS and mean pixel height standard deviation in the 2001 mean topographies examined longitudinally in the OHT and control subjects (Pearson correlation coefficient r ⫽ 0.03; P ⫽ 0.166).
Discussion The ability of the GPS to discriminate between normal and glaucomatous ONHs has been assessed extensively in recent literature and there is a general consensus that it is at least as effective as MRA and has the advantage that it is operator independent. However, GPS is more likely than MRA to misclassify large normal ONHs as abnormal.13–19 There is evidence to suggest that a within normal limits GPS classification is more useful than the MRA to confirm that a disc is normal, whereas an outside normal limits MRA classification is more useful than the GPS in confirming that a disc is abnormal.19 Disease severity has also been shown to influence GPS classification, with GPS having a higher sensitivity and lower specificity than MRA in patients with mild glaucomatous VF damage, whereas MRA better discriminates subjects with severe glaucomatous VF damage.17 Alencar et al20 recently showed that baseline GPS can be used to predict which glaucoma suspects will go on to display VF deterioration and optic disc change, the latter having a similar predictive value to expert evaluation of stereophotographs. In all of these studies, the GPS classifications are based on the manufacturer’s own suggested Table 4. Specificity Estimates for the Glaucoma Probability Score Progression Strategy Status
Specificity Estimate
95% Confidence Interval
No significant improvement No progression in controls
212/219 (96.8%) 20/21 (95.2%)
93.2%–98.5% 76.1%–99.9%
Figure 3. Venn diagram comparing the number (percentage) of ocular hypertensive subjects identified as progressing by Glaucoma Probability Score (GPS) trend analysis, rim area (RA) trend analysis and visual field (VF) 3-omitting pointwise linear regression. Specificity is matched at approximately 97% for the 3 strategies.
cutoff values, with 0 to 0.27 classified as within normal limits, 0.28 to 0.64 as borderline, and 0.65 to 1.00 as outside normal limits.20 Borderline classifications should be interpreted with a degree of caution, because this study demonstrates that the GPS has poorer reproducibility in this range. This suggests that eyes with scores close to the upper and lower cutoffs may well be differently classified on repeated testing. This is unfortunate; eyes with very low or very high GPS values present the least diagnostic confusion and it is in the mid range where the clinician would benefit most from reliable information. When interpreting the GPS in clinical practice, it would probably be prudent to assess the classification alongside the actual numerical score. This would help to establish whether a “within normal limits” or “outside normal limits” classification is at the fringes of the borderline cutoff and should therefore be given less credence than if they were well away from the boundary values. The performance of both HRT Classic and HRT-II is poorer in the mid range compared with the high GPS range, suggesting that one should be cautious interpreting mid range global GPS scores, regardless of whether HRT Classic or HRT-II acquisitions are used. It is difficult to fully explain the pattern of GPS reproducibility observed in this study without precise knowledge of the proprietary RVM and learning data sets used in the HRT. It is possible that the RVM may place relevance on a particular parameter with a high degree of intertest variability. The RVM effectively operates in 5 dimensions if it uses all 5 parameters from the 3-dimensional model so the mapping between any ONH structural change and a given GPS change is likely to be highly nonlinear. This is somewhat analogous to the relationship between VF sensitivity and disease progression.21 As glaucoma progresses and VF sensitivity decreases, variability increases. The relationship between disease severity and variability holds true until very
727
Ophthalmology Volume 117, Number 4, April 2010 advanced field loss (the “floor effect”), at which point sensitivity values become less variable because they are anchored at the lower dynamic range of the perimeters, or indeed, the disease has reached end stage. Our assessment of GPS reproducibility is made with the assumption that no disease progression should have occurred in the 6-week interim between baseline and followup imaging during the test–retest study. Indeed, there was no difference between the interobserver differences and zero, suggesting that no disease-related change occurred on average in the 6-week time period. However, when interpreting these results it is important to consider the possibility, although remote, that the global GPS scores may have changed between visits because of ONH structural change, rather than owing to measurement error alone. Despite the apparent limitations regarding the reproducibility of the GPS in its mid range, the results of our study suggest that it can be used to measure progression in OHT eyes. In particular, it seems to perform at least as well as an HRT rim area trend analysis at the same level of estimated specificity, identifying a similar number of subjects as progressing (25 by GPS, 24 by rim area). The agreement between these 2 HRT progression estimates was, however, poor (only 7 subjects progressed by both GPS and rim area). Rim area is not a parameter that contributes to the calculated value of GPS. Some parameters assessed in the ONH model (cup size and rim steepness) reflect changes in rim area, whereas other morphologic changes (RNFL curvature, cup depth) do not necessarily reflect rim area change. This could explain why there is some agreement between GPS and rim area change but in the majority the 2 measures are identifying different eyes as progressing. Another potential explanation is that, because the rim area is reference plane dependent, reference height variability may be masking change in some eyes. Agreement between HRT and VF progression was higher for the GPS trend analysis than for the rim area trend analysis (11 subjects compared with 7 subjects). The poor agreement with VF is disappointing, but is consistent with other published studies using both an alternative HRT progression strategy and an alternative imaging device.22,23 The level of agreement between structural and functional progression measured in this study is disappointing. This perhaps suggests that there is temporal dissociation between structural and functional change in glaucomatous eyes or that the currently available structural and functional tests provide insight into fundamentally different facets of glaucomatous pathophysiology. Another possibility is that measurement error (noise) in structural and functional data collected from patients may be masking progression in some eyes while also giving rise to spurious/ unreliable rates of loss in others. The results of this study, in terms of GPS progression, may not be fully applicable to eyes with manifest glaucoma. In the current study, which was performed in OHT eyes, a fairly wide range of GPS slopes (⫺0.11 to 0.12 per year) was recorded. In glaucomatous eyes, the baseline GPS values are likely to be high, meaning that the degree of change identifiable as “progression” is likely to be small. Although this is countered by the fact that reproducibility is
728
higher at this stage of disease (lower “noise”), the ability to detect change (“signal”) may be reduced. In this study, we have shown that the GPS has good reproducibility at the extremes of the score but poorer reproducibility in the “borderline” range. In clinical practice, GPS classification should be assessed alongside the actual numerical score. The GPS trend analysis described in this study is at least as effective at identifying OHT subjects as progressing as an HRT rim area trend analysis at the same level of specificity, although it largely identified different subjects as progressing. The poor agreement with VF progression, as well as rim area progression, suggests that GPS progression should not be used in isolation to detect glaucomatous progression.
References 1. Garway-Heath DF. Early diagnosis in glaucoma. Prog Brain Res 2008;173:47–57. 2. Garway-Heath DF, Friedman DS. How should results from clinical tests be integrated into the diagnostic process? Ophthalmology 2006;113:1479 – 80. 3. Swindale NV, Stjepanovic G, Chin A, Mikelberg FS. Automated analysis of normal and glaucomatous optic nerve head topography images. Invest Ophthalmol Vis Sci 2000;41: 1730 – 42. 4. Wollstein G, Garway-Heath DF, Hitchings RA. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology 1998;105:1557– 63. 5. Strouthidis NG, White ET, Owen VM, et al. Factors affecting the test-retest variability of Heidelberg retina tomograph and Heidelberg retina tomograph II measurements. Br J Ophthalmol 2005;89:1427–32. 6. Strouthidis NG, White ET, Owen VM, et al. Improving the repeatability of Heidelberg retina tomograph and Heidelberg retina tomograph II rim area measurements. Br J Ophthalmol 2005;89:1433–7. 7. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10. 8. Strouthidis NG, Scott A, Peter NM, Garway-Heath DF. Optic disc and visual field progression in ocular hypertensive subjects: detection rates, specificity, and agreement. Invest Ophthalmol Vis Sci 2006;47:2904 –10. 9. Kamal D, Garway-Heath D, Ruben S, et al. Results of the betaxolol versus placebo treatment trial in ocular hypertension. Graefes Arch Clin Exp Ophthalmol 2003;241:196 –203. 10. Advanced Glaucoma Intervention Study. 2. Visual field test scoring and reliability. Ophthalmology 1994;101:1445–55. 11. Kamal DS, Viswanathan AC, Garway-Heath DF, et al. Detection of optic disc change with the Heidelberg retina tomograph before confirmed visual field change in ocular hypertensives converting to early glaucoma. Br J Ophthalmol 1999;83: 290 – 4. 12. Gardiner SK, Crabb DP. Examination of different pointwise linear regression methods for determining visual field progression. Invest Ophthalmol Vis Sci 2002;43:1400 –7. 13. De Leon-Ortega JE, Sakata LM, Monheit BE, et al. Comparison of diagnostic accuracy of Heidelberg Retina Tomograph II and Heidelberg Retina Tomograph 3 to discriminate glaucomatous and nonglaucomatous eyes. Am J Ophthalmol 2007; 144:525–32.
Strouthidis et al 䡠 Reproducibility and Progression of the Glaucoma Probability Score 14. Oddone F, Centofanti M, Rossetti L, et al. Exploring the Heidelberg Retinal Tomograph 3 diagnostic accuracy across disc sizes and glaucoma stages: a multicenter study. Ophthalmology 2008;115:1358 – 65. 15. Harizman N, Zelefsky JR, Ilitchev E, et al. Detection of glaucoma using operator-dependent versus operator-independent classification in the Heidelberg retinal tomograph-III. Br J Ophthalmol 2006;90:1390 –2. 16. Burgansky-Eliash Z, Wollstein G, Bilonick RA, et al. Glaucoma detection with the Heidelberg retina tomograph 3. Ophthalmology 2007;114:466 –71. 17. Ferreras A, Pajarin AB, Polo V, et al. Diagnostic ability of Heidelberg Retina Tomograph 3 classifications: glaucoma probability score versus Moorfields regression analysis. Ophthalmology 2007;114:1981–7. 18. Coops A, Henson DB, Kwartz AJ, Artes PH. Automated analysis of Heidelberg retina tomograph optic disc images by glaucoma probability score. Invest Ophthalmol Vis Sci 2006; 47:5348 –55.
19. Zangwill LM, Jain S, Racette L, et al. The effect of disc size and severity of disease on the diagnostic accuracy of the Heidelberg Retina Tomograph glaucoma probability score. Invest Ophthalmol Vis Sci 2007;48:2653– 60. 20. Alencar LM, Bowd C, Weinreb RN, et al. Comparison of HRT-3 glaucoma probability score and subjective stereophotograph assessment for prediction of progression in glaucoma. Invest Ophthalmol Vis Sci 2008;49:1898 – 906. 21. Heijl A, Lindgren A, Lindgren G. Test-retest variability in glaucomatous visual fields. Am J Ophthalmol 1989;108: 130 –5. 22. Artes PH, Chauhan BC. Longitudinal changes in the visual field and optic disc in glaucoma. Prog Retin Eye Res 2005; 24:333–54. 23. Wollstein G, Schuman JS, Price LL, et al. Optical coherence tomography longitudinal evaluation of retinal nerve fiber layer thickness in glaucoma. Arch Ophthalmol 2005; 123:464 –70.
Footnotes and Financial Disclosures Originally received: April 17, 2009. Final revision: September 19, 2009. Accepted: September 22, 2009. Available online: January 4, 2010.
Nicholas G. Strouthidis - research support - Heidelberg Engineering (S)
Manuscript no. 2009-538.
1
NIHR Biomedical Research Centre for Ophthalmology, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom. 2 3
Devers Eye Institute, Legacy Health System, Portland, Oregon.
Department of Ophthalmology, Hamamatsu University School of Medicine, Hamamatsu City, Japan. Financial Disclosure(s): The authors have made the following disclosures: David F. Garway-Heath - research support - Heidelberg Engineering, OptoVue and Carl Zeiss Meditec; consultant - Carl Zeiss Meditec (C and S)
David F. Garway-Heath has received a proportion of his funding from the Department of Health’s National Institute for Health Research Biomedical Research Centre at Moorfields Eye Hospital and the UCL Institute of Ophthalmology. The views expressed in this publication are those of the authors and not necessarily those of the Department of Health. Nicholas G. Strouthidis is funded by an unrestricted educational grant from Heidelberg Engineering and by a Royal College of Ophthalmologists/Pfizer Travel Fellowship. Correspondence: Nicholas G. Strouthidis, MD, MRCOphth, Glaucoma Research Unit, Moorfields Eye Hospital, 162 City Road, London, EC1V 2PD. E-mail: [email protected].
729