Discriminating between Normal and Glaucoma-Damaged Eyes with the Heidelberg Retina Tomograph 3

Transcription

1 Discriminating between Normal and Glaucoma-Damaged Eyes with the Heidelberg Retina Tomograph 3 Antonio Ferreras, MD, PhD, 1 Luís E. Pablo, MD, PhD, 1 José M. Larrosa, MD, PhD, 1 Vicente Polo, MD, PhD, 1 Ana B. Pajarín, MD, PhD, 2 Francisco M. Honrubia, MD, PhD 1 Purpose: To determine and validate the diagnostic ability of a linear discriminant function (LDF) based on global stereometric parameters obtained using the Heidelberg Retina Tomograph version 3 (HRT3) for discriminating between healthy eyes and eyes with glaucomatous visual field loss. Design: Cross-sectional study. Participants: The authors prospectively selected 81 consecutive healthy subjects and 85 consecutive patients with open-angle glaucoma. Another prospective sample of 225 normal eyes and 210 glaucoma eyes was used to evaluate how well the LDF performed in another population. Methods: Participants were divided into 2 groups depending on the results of standard automated perimetry and intraocular pressure. All participants underwent imaging of the optic nerve head with the HRT3. Main Outcome Measures: The LDF was calculated according to the stepwise logistic regression results of global optic nerve head parameters and glaucoma probability score numerical values. The diagnostic accuracy of the LDF and other parameters included in the software of the HRT3 was evaluated in another independent population. Results: Based on the results of the stepwise binary logistic regression analysis, the function proposed was LDF contour line modulation temporal superior 9.41 cup shape measure 4.07 rim area. The areas under the receiver operating characteristic curve were for the LDF, for the Frederick S. Mikelberg (FSM) discriminant function, and for the Reinhard O. W. Burk (RB) discriminant function. There were no significant differences between these values. The LDF and the FSM and RB discriminant functions yielded sensitivities of 74.2%, 70.4%, and 67.6%, respectively, at a fixed specificity of 85%. Conclusions: Compared with the HRT-provided parameters, the proposed LDF exhibited higher diagnostic ability than most available analyses. The LDF had a better sensitivity and specificity balance than the FSM and RB discriminant functions, regardless of optic disc size. Ophthalmology 2008;115: by the American Academy of Ophthalmology. Primary open-angle glaucoma is defined as a progressive multifactorial optic neuropathy in which there is a characteristic acquired atrophy of the optic nerve and loss of retinal ganglion cells and their axons. 1,2 Damage to the retinal nerve fiber layer usually is followed by changes in the optic nerve head shape and specific visual field defects. The detection of early structural changes at the optic disc is Originally received: March 23, Final revision: June 8, Accepted: June 20, Available online: September 17, Manuscript no Department of Ophthalmology, Miguel Servet University Hospital, Zaragoza, Spain. 2 Family Medicine, Centro de Salud San Pablo, Zaragoza, Spain. The authors have no conflicts of interest related to the article. Correspondence to Antonio Ferreras, MD, PhD, Department of Ophthalmology, Miguel Servet University Hospital, Isabel la Católica 1 3, Zaragoza, Spain. aferreras@msn.com. a key point in the diagnosis and follow-up of glaucoma. The evaluation of optic disc stereophotographs and planimetry is highly subjective and requires experienced evaluators. The Heidelberg Retina Tomograph (HRT; Heidelberg Engineering, Heidelberg, Germany) is a confocal scanning laser ophthalmoscope that provides rapid, quantitative, and reproducible 3,4 measurements of the optic nerve head Recently, the HRT version 3 (HRT3) was introduced with a new software version called Advanced Glaucoma Analysis 3.0, which is an enhanced version of the previous HRT glaucoma software. The HRT3 provides larger, ethnicselectable normative databases and includes new data analysis tools such as the glaucoma probability score (GPS) classification The purpose of the present study was to search for an optimal combination of optic nerve head parameters for improving the diagnostic ability of the HRT3. Other authors have proposed discriminant analysis formulas for previous 2008 by the American Academy of Ophthalmology ISSN /08/$ see front matter Published by Elsevier Inc. doi: /j.ophtha

2 Ophthalmology Volume 115, Number 5, May 2008 versions of the HRT. The Frederick S. Mikelberg (FSM) discriminant function 5 and Reinhard O. W. Burk (RB) discriminant function 18 were designed some years ago and are still included in the software of this device. These formulas and others 8,9 have a worse diagnostic performance when they are applied to different populations from those in which they were calculated. Recently, Shunmugam and Azuara-Blanco 19 evaluated the quality of the reporting of diagnostic accuracy using the HRT. They concluded that the quality was suboptimal and suggested the Standards for Reporting of Diagnostic Accuracy initiative 20 for appraising the strengths and weaknesses of diagnostic ability studies. Thus, the design of the current study followed the 25 items of the Standards for Reporting of Diagnostic Accuracy guidelines, which are aimed at increasing the quality of reporting diagnostic accuracy research. To the authors knowledge, this is the first study to assess the diagnostic ability of a linear discriminant function (LDF) designed for the HRT3. The strength of this study lies in the validation of the LDF in an independent sample. Patients and Methods Patients and Measurement Protocol The prospective study protocol was approved by the ethics committee of Miguel Servet University Hospital, and informed written consent was obtained from all participants. The design of the study followed the tenets of the Declaration of Helsinki for biomedical research. From April, 2006, through December, 2006, 2 samples (one population for obtaining the LDF and a second independent population for testing the LDF) of consecutive healthy control subjects and glaucoma patients were preenrolled prospectively from the Department of Ophthalmology of Miguel Servet University Hospital and two outpatient clinics under the area of influence of the hospital. Normal eyes were recruited consecutively from among patients referred for refraction who underwent routine examination without abnormal ocular findings, from among hospital staff, and from among relatives of patients in the hospital. Patients with glaucoma were recruited consecutively from an ongoing longitudinal follow-up study at the Miguel Servet University Hospital. Three of the subjects did not provide informed consent, 11 subjects did not complete all of the required tests, and 7 subjects were unable to perform at least 1 of the tests included in the study protocol (7 of them did not provide a reliable standard automated perimetry [SAP] evaluation, and in 5 subjects, it was impossible to acquire HRT scans with good quality scores). Finally, 601 eyes from white persons were included in the statistical analysis. One eye from each subject was chosen randomly for the study, unless only 1 eye met the inclusion criteria. All participants had to meet the following inclusion criteria: best-corrected visual acuity of 20/40 or better, refractive error within 5.00 diopters (D) equivalent sphere and 2 D astigmatism, transparent ocular media (nuclear color or opalescence, cortical or posterior subcapsular lens opacity 1) according to the Lens Opacities Classification System III system, 21 and an open anterior chamber angle. Subjects with previous intraocular surgery, diabetes or other systemic diseases, history of ocular or neurologic disease, or current use of a medication that could affect visual field sensitivity were excluded. Participants underwent a full ophthalmologic examination: clinical history, visual acuity, biomicroscopy of the anterior segment using a slit lamp, gonioscopy, Goldmann applanation tonometry, central corneal ultrasonic pachymetry (model DGH 500, DGH Technology, Exton, PA), and ophthalmoscopy of the posterior segment. Simultaneous stereophotographs of the optic disc were obtained after mydriasis. At least 2 reliable SAP tests were performed using a Humphrey Field analyzer, model 745 (Zeiss Humphrey Systems, Dublin, CA), with the Swedish interactive threshold algorithm standard 24-2 test. Near addition was added to the subject s refractive correction. If fixation losses and false-positive or false-negative rates were more than 20%, the test was repeated. The second reliable perimetry test obtained was used in this study to minimize the learning effect. 22,23 Abnormal SAP results were defined as a reproducible glaucomatous visual field loss in the absence of any other abnormalities to explain the defect, with a pattern standard deviation significantly elevated beyond the 5% level and/or a glaucoma hemifield test outside normal limits. The subjects completed the perimetry tests before any clinical examination or structural test. Each perimetry was measured on different days to avoid a fatigue effect. Topographic analysis of the optic nerve head was performed using a confocal scanning laser ophthalmoscope, the HRT3 with a diode laser (670-nm wavelength). The HRT provides topographic measurements of the optic nerve head derived from 16 to 64 optical sections to a depth of 4 mm, depending on the longitudinal field of view. 24 The spherical equivalent refractive error of each eye was adjusted in the dioptric ring of the HRT. After keratometric readings were entered into the software (to correct for magnification errors), topographic images were obtained through dilated pupils and were analyzed using the Advanced Glaucoma Analysis 3.0 software. All scans had to have an interscan standard deviation of less than 30 m. The margin of the optic discs was traced manually by the same glaucoma specialist (while viewing the stereophotographs under a stereoscopic viewer) who was masked to the patients identity and clinical history (AF), defining the inner edge of the Elschnig s ring with at least a 4-point contour line. The HRT3 software displays several windows in which the topographic results are detailed: stereometric parameters, the Moorfields regression analysis (MRA) classification, the GPS classification, and interactive measurements. All the ophthalmic examinations, perimetry tests, and the topographic analysis were performed within 1 month from the subject s date of enrollment into the study. Classification into Groups Healthy eyes had an intraocular pressure (IOP) of less than 21 mmhg (on at least 3 readings on different days), no history of increased IOP, and a normal SAP, regardless of the appearance of the optic disc. The glaucoma group comprised subjects with primary open-angle glaucoma, pseudoexfoliative glaucoma, and pigmentary glaucoma. Glaucomatous eyes had an IOP of more than 21 mmhg and typical SAP defects, regardless of the appearance of the optic disc. The eyes were classified by 2 glaucoma specialists (AF, LEP) masked to patient identity and clinical history. Any disagreement was resolved by consensus. The total sample was divided randomly into 2 populations, one for calculating the discriminant analysis and the other for evaluating the performance of the LDF in an independent group. Statistical Analysis All statistical analyses were calculated using SPSS software version 15.0 (SPSS, Inc., Chicago, IL) and MedCalc software version 776

3 Ferreras et al HRT3 and Logistic Regression Analysis for Glaucoma Diagnosis (MedCalc Software, Mariakerke, Belgium). Binomial (or binary) logistic regression is useful for predicting the presence or absence of a characteristic or outcome based on values of a set of predictor variables and is applicable to a broader range of research situations than discriminant analysis. It is similar to a linear regression model but is suited to models where the dependent variable is dichotomous (healthy or diseased). Logistic regression coefficients can be used to estimate odds ratios for each of the independent variables in the model. The relative importance of each independent variable was assessed by stepwise binary logistic regression analysis using the forward Wald method. The Wald chi-square statistic tested the unique contribution of each independent variable (predictor variable), in the context of the other predictors (holding constant the other predictors), eliminating any overlap between them. The stepwise probability test determined the criteria by which variables were entered into and removed from the model. In this study, the dependent variable was whether a patient had glaucoma and the independent variables were: disc area, cup area, rim area, cup-todisc area ratio, rim-to-disc area ratio, cup volume, rim volume, mean cup depth, maximum cup depth, height variation contour, cup shape measure (CSM), mean retinal nerve fiber layer (RNFL) thickness, RNFL cross-sectional area, horizontal cup-to-disc ratio, vertical cup-to-disc ratio, maximum contour elevation, maximum contour depression, contour line modulation (CLM) temporal superior, CLM temporal inferior, average variability, reference height, linear cup-to-disc ratio, and numerical values of global GPS, GPS temporal, GPS temporal superior, GPS temporal inferior, GPS nasal, GPS nasal superior, and GPS nasal inferior. The stepwise probability value was fixed at 0.01 for entry and 0.05 for removal. The stepwise procedure identified the optic nerve head parameter that accounted for the greatest amount of error, then included the next best variable, and so on. At the first iteration, the rim area was selected. At the second iteration, the CLM temporal superior was added to the model, and at the third iteration, the CSM was added to the model. The significant parameters of the HRT3 then were combined to generate a new variable in such a way that the measurable differences between the groups were maximized by means of the LDF. In summary, the LDF is a score formed by taking a weighted sum of the predictor variables. The LDF was defined as LDF CLM temporal superior CSM rim area. The diagnostic ability of the LDF and the discriminant functions included in the software of the HRT3 (FSM 5 and RB 18 ) were checked in a second population that was stratified by disc area size. The optic discs were classified as small (disc area less than 1.6 mm 2 ), average (disc area mm 2 ), or large (disc area more than 2.6 mm 2 ). 25 The receiver operating characteristic (ROC) curves were plotted for the FSM and RB discriminant functions and were compared with the proposed LDF. The cutoff points were calculated by the MedCalc software as the points with the best sensitivity and specificity balance. Differences between the ROC curves were tested to compare the area under the ROC curves (AUCs) using the Hanley-McNeil method. 26 Sensitivities at 85% and 95% (5% false-positive rate) fixed specificities, and positive and negative likelihood ratios (LRs) also were calculated for the LDF and the FSM and RB discriminant functions depending on the disc area size. Results Table 1 shows the clinical characteristics of the groups of both populations enrolled in the study. The sample used for obtaining the LDF consisted of 166 subjects divided into 81 normal eyes and 85 glaucomatous eyes (71 with primary open-angle glaucoma, 10 with pseudoexfoliative glaucoma, and 4 with pigmentary glaucoma). The mean age was years for the normal group and years for the glaucoma group. The second sample used for validating the LDF included 225 normal eyes and 210 glaucomatous eyes (163 with primary openangle glaucoma, 34 with pseudoexfoliative glaucoma, and 13 with pigmentary glaucoma). The mean age of the normal group was years and the mean age of the glaucomatous group was years. Age, central corneal thickness, and disc area (measured with the HRT3) did not differ significantly (P 0.05) between the groups in both samples. There were significant differences between the normal and glaucoma groups in both samples in mean baseline IOP, vertical cup-to-disc ratio evaluated in stereophotographs, mean deviation, and pattern standard deviation of SAP using Student s t test (Table 1). In the population used for calculating the LDF, the LDF had higher sensitivity and specificity balance (81.1% and 93.8%, respectively) than the MRA overall and GPS global classifications (Table 2 [available at but exhibited similar Table 1. Clinical Characteristics of Both Populations Included in the Study Population for Calculating the Linear Discriminant Function Mean Normal Group Standard Deviation Glaucomatous Group Mean Standard Deviation P Value* Population for Checking the Linear Discriminant Function Mean Normal Group Standard Deviation Glaucomatous Group Mean Standard Deviation P Value* Age Mean IOP C/D Pachymetry MD of SAP PSD of SAP Disc area N C/D vertical cup-to-disc ratio in stereophotographs; IOP basal intraocular pressure (without treatment); MD mean deviation; PSD pattern standard deviation; SAP standard automated perimetry. *Significant differences (P 0.05) in Student s t test results between normal and glaucomatous groups for each population. 777

4 Table 4. Areas under the Receiver Operating Characteristic Curve for Each Global Topographic Parameter and Glaucoma Probability Score Global Numerical Value Obtained with the Heidelberg Retina Tomograph 3, and Sensitivities at 85% and 95% Fixed Specificities, in the Sample for Testing the Diagnostic Ability of the Linear Discriminant Function Heidelberg Retina Tomograph Global Topographic Parameters Ophthalmology Volume 115, Number 5, May 2008 Area under the Receiver Operating Characteristic Curve 95% Confidence Interval P Value Sensitivity and Specificity Specificity 85% Specificity 95% LDF Cup-to-disc area ratio Rim-to-disc area ratio Vertical cup-to-disc ratio Rim volume Mean RNFL thickness FSM discriminant function value Rim area RNFL cross-sectional area RB discriminant function value CLM temporal inferior Horizontal cup-to-disc ratio Cup area Cup shape measure GPS global (numerical value) Cup volume CLM temporal superior Maximum contour elevation Average variability (SD) Mean cup depth Height variation contour Maximum contour depression Disc area Reference height Maximum cup depth CLM contour line modulation; FSM Frederick S. Mikelberg; GPS global glaucoma probability score; LDF linear discriminant function; RB Reinhard O. W. Burk; RNFL retinal nerve fiber layer; SD standard deviation. diagnostic ability than the discriminant functions developed from previous versions of the HRT. The LDF had the highest positive LR (13.08), and the LDF and the FSM discriminant function had the lowest negative LR (0.2 for both). The LDF had an area under the curve (AUC) of (Fig 1 [available at the FSM discriminant function had an AUC of 0.899, and the RB discriminant function had an AUC of The stereometric parameters with the largest AUCs were the cup-to-disc area ratio (0.901), the rim-to-disc area ratio (0.901), and the vertical cup-to-disc ratio (0.909). There was no significant difference between the AUCs of the 3 discriminant functions and the best stereometric parameters. The LDF and the FSM and RB discriminant functions presented similar diagnostic abilities (Table 3 [available at %, 82.3%, and 82.3% sensitivities at a fixed specificity of 85%, respectively. In the sample used to test the performance of the LDF, the AUCs were (the LDF), (FSM), and (RB; Table 4, Fig 2). There were no significant differences between them. Other stereometric parameters such as cup-to-disc area ratio, rim-to-disc area ratio, vertical cup-to-disc ratio, and rim volume had similar AUCs compared with the LDF and the FSM and RB discriminant functions. The disc area, the maximum cup depth, and the reference height did not have significant AUCs. The discriminant functions yielded sensitivities of 74.2% (the LDF), 70.4% (FSM), and 67.6% (RB) at a fixed specificity of 85%. The ROC curves were plotted only for the discriminant function values because the MRA and color-coded GPS classifications are not continuous variables. The small number of categories in these cases might have led to an underestimation of the ROC curve area. 27 In this population, the best sensitivity and specificity balance was 76.6% to 83.5% for the color-coded MRA overall (cutoff point, borderline) and 66.6% to 84.4% for the color-coded GPS global (cutoff point, outside normal limits), respectively. Figure 2. Receiver operating characteristic (ROC) curves of the linear discriminant function (LDF), Frederick S. Mikelberg (FSM) discriminant function, and Reinhard O. W. Burk (RB) discriminant function between healthy eyes and glaucomatous patients in the population used for validating the LDF. Area under the ROC curve (AUC) for LDF, (95% confidence interval [CI], ); AUC for FSM, (95% CI, ); AUC for RB, (95% CI, ). 778

5 Ferreras et al HRT3 and Logistic Regression Analysis for Glaucoma Diagnosis Small discs (less than 1.6 mm 2 ) exhibited the highest specificities for the 3 discriminant functions, and larger discs (more than 2.6 mm 2 ) presented the greatest AUCs (Table 5 [available at For small discs, the LDF had a larger AUC than did the FSM discriminant function (P 0.031). For average size discs, the LDF had a larger AUC than did the RB discriminant function (P 0.026). There were no significant differences between the AUCs of the 3 discriminant functions for large discs. The FSM discriminant function in small discs and the FSM and the RB discriminant functions in large discs had the highest positive LR (37, 13, and 10, respectively). The LDF in large discs had the lowest negative LR (0.1). The LDF showed the best sensitivity and specificity balance in all subgroups depending on disc area size for discriminating between healthy and glaucomatous eyes. Discussion The identification of optic nerve head damage is crucial for glaucoma diagnosis. Because of the wide variations in optic disc appearance in the normal population, a cross-sectional assessment of optic disc stereophotographs may not be sufficient. Nevertheless, the use of scanning laser devices, such as the HRT, can improve the accuracy of optic nerve head evaluation. The HRT is easy to perform, provides quantitative data, and does not usually require mydriasis (depending on pupil size). The ability of the HRT to detect glaucomatous changes of the optic nerve head has been validated widely. Several studies 8,10,12,14,28 32 have been performed to determine the best parameters and have led to the development of discriminant functions 5,9,18 for improving the diagnostic ability of previous versions of the HRT. The HRT3 includes new software for calculating the topographic parameters of the optic disc and new data analysis tools such as the GPS classification In general, up to a 4% difference can be expected in the data when comparing the HRT2 with the HRT3. A horizontal scaling error, in which horizontal HRT2 stereometric measurements are enlarged by 4%, was reported by the manufacturer; this was corrected in the HRT3. 33 The new version 3.0 software for the HRT glaucoma module has enhanced alignment algorithms and software analysis that can be applied to past examinations. Therefore, the optimal combination of structural parameters using this new device must be updated and validated. In this study, disc areas measured with the HRT were similar between the normal and glaucoma groups. This is important when comparing groups because several HRT parameters, such as rim area, cup area, rim volume, and so forth, are related directly to disc size. 8,9 Moreover, the optic disc morphologic features correlate with the evaluated test (HRT stereometric parameters); therefore, the healthy and glaucomatous eyes were classified regardless of optic disc appearance to avoid an overestimation of the sensitivity and specificity of the HRT3. 34,35 Other authors, 5,36,37 using multiple linear regression and discriminant analyses, reported that the rim area, the maximum cup depth, the height variation contour, and the CSM are the most important predictors of glaucomatous visual field defects. In the current model, 2 of these parameters (rim area and CSM) were identified by the stepwise procedure as the best optic nerve head parameters for discriminating between healthy and glaucomatous eyes. Mikelberg et al 5 used rim volume, CSM, and height variation contour in their formula. Bathija et al 9 proposed using rim area, height variation contour, CSM, and RNFL thickness as the best variables to differentiate healthy from glaucomatous eyes. Different sensitivity and specificity values for discriminant functions of the HRT have been described depending on the cutoff point applied. In the original populations, the FSM discriminant function 5 exhibited 87% sensitivity and 84% specificity, whereas the formula of Bathija et al 9 yielded 78% sensitivity and 88% specificity. The new HRT software adds a new automated classification called GPS. The GPS is an approach to optic disc analysis that eliminates operator-dependent factors, which are considered to be important sources of variability. In this study, the GPS numerical values were included in the stepwise binary logistic regression analysis, but all of them were excluded in the first step because other stereometric parameters were better suited to detect glaucomatous changes in the optic nerve head. The LDF represents the best combination of HRT3 parameters to differentiate between normal and glaucomatous eyes. No other combination of HRT3 parameters had better diagnostic ability. The LR is the probability that a given test result would occur in a patient with the disease compared with the likelihood that the same result would occur in a patient without the disease. 38 An LR value close to 1 indicates insignificant effects, whereas LR values higher than 10 or lower than 0.1 indicate large changes in disease probability. In the total sample, the LDF, the FSM discriminant function, and the GPS global had positive values higher than 10, but they ranged from 4 to more than 30, depending on disc area size. The FSM discriminant function presented the highest positive LR (more than 10) in small and large discs, and this indicates that abnormal results are associated with important posttest effects. However, the FSM discriminant function yielded a worse negative LR for those disc sizes, thus normal results are associated with a small change in the posttest probability of disease and therefore are limited in their ability to exclude the presence of glaucoma. For the 3 discriminant functions evaluated, negative LRs ranged from 0.11 to The LDF showed the lowest negative LR in small (0.19) and large (0.11) disc sizes. Clinicians must take disc size into account when interpreting the HRT outcomes. 8,39 Iester et al 8 reported that the sensitivity and specificity of the FSM function increased when disc area was more than 2 mm 2. In the current study, the 3 discriminant functions classified small and large optic discs more conservatively (with lower sensitivity and greater specificity) than medium discs. Larger optic discs were associated with increased sensitivity, whereas small optic discs were associated with increased specificity. Cupto-disc area ratio, rim-to-disc area ratio, vertical cup-to-disc ratio, and rim volume presented similar diagnostic abilities as the LDF, but these parameters are influenced directly by disc size, 8,9 and therefore their performance may be worse for small and large optic discs. In this study, the control group had no previous visual field testing experience, although most of the glaucoma 779

6 Ophthalmology Volume 115, Number 5, May 2008 patients had undergone SAP tests. All participants underwent at least 2 reliable SAP tests to minimize the influence of the learning effect. Normal subjects did not need to undergo a second round of SAP when the results from the first round were normal, because training is expected to influence the results in the direction of improvement. Obviously, the severity of visual field loss has an important influence on imaging instrument sensitivity. 39 More severe disease is associated with increased sensitivity. In this study, most glaucomatous eyes presented mild damage based on the Hodapp-Parrish-Anderson score. 40 The mean deviations of the SAPs were lower than 6 db for both samples, and therefore in populations of more patients with moderate and severe glaucoma, higher sensitivity and specificity balance for the discriminant functions is expected. The ethnic characteristics of the validation sample were similar to those of the sample used for obtaining the LDF, and this fact might have biased toward the current LDF when compared with other LDFs. The quality of the data obtained by the imaging devices is influenced by the media opacity, retinal pigment epithelium status, instrument variability, and positioning and centering of the images. These limitations must be taken into account in clinical practice. The diagnostic performance of the 3 discriminant functions evaluated was very similar in the second population, but in general, the LDF showed a better sensitivity and specificity balance. The results in the second sample confirmed those obtained in the original population that was used to develop the LDF. References 1. American Academy of Ophthalmology Glaucoma Panel. Preferred practice pattern. Primary open-angle glaucoma. San Francisco: American Academy of Ophthalmology; 2005:3. Available at: Accessed June 14, Quigley HA. Neuronal death in glaucoma. Prog Retin Eye Res 1999;18: Rohrschneider K, Burk RO, Kruse FE, Völcker HE. Reproducibility of the optic nerve head topography with a new laser tomographic scanning device. Ophthalmology 1994; 101: Janknecht P, Funk J. Optic nerve head analyser and Heidelberg retina tomograph: accuracy and reproducibility of topographic measurements in a model eye and in volunteers. Br J Ophthalmol 1994;78: Mikelberg FS, Parfitt CM, Swindale NV, et al. Ability of the Heidelberg retina tomograph to detect early glaucomatous visual field loss. J Glaucoma 1995;4: Brigatti L, Caprioli J. Correlation of visual field with scanning confocal laser optic disc measurements in glaucoma. Arch Ophthalmol 1995;113: Uchida H, Brigatti L, Caprioli J. Detection of structural damage from glaucoma with confocal laser image analysis. Invest Ophthalmol Vis Sci 1996;37: Iester M, Mikelberg FS, Drance SM. The effect of optic disc size on diagnostic precision with the Heidelberg retina tomograph. Ophthalmology 1997;104: Bathija R, Zangwill L, Berry CC, et al. Detection of early glaucomatous structural damage with confocal scanning laser tomography. J Glaucoma 1998;7: Wollstein G, Garway-Heath DF, Hitchings RA. Identification of early glaucoma cases with the scanning laser ophthalmoscope. Ophthalmology 1998;105: 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: Zangwill LM, Bowd C, Berry CC, et al. Discriminating between normal and glaucomatous eyes using the Heidelberg retina tomograph, GDx nerve fiber analyzer, and optical coherence tomograph. Arch Ophthalmol 2001;119: Greaney MJ, Hoffman DC, Garway-Heath DF, et al. Comparison of optic nerve imaging methods to distinguish normal eyes from those with glaucoma. Invest Ophthalmol Vis Sci 2002;43: Medeiros FA, Zangwill LM, Bowd C, Weinreb RN. Comparison of the GDx VCC scanning laser polarimeter, HRT II confocal scanning laser ophthalmoscope, and Stratus OCT optical coherence tomograph for the detection of glaucoma. Arch Ophthalmol 2004;122: 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: 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: 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: Burk RO, Noack H, Rohrschneider K, Volcker HE. Prediction of glaucomatous visual field defects by reference plane independent three-dimensional optic nerve head parameters. In: Wall M, Wild JM, eds. Perimetry Update 1998/1999: Proceedings of the XIIIth International Perimetric Society Meeting, Gardone Riviera (BS), Italy, September 6 9, The Hague: Kugler; 1999: Shunmugam M, Azuara-Blanco A. The quality of reporting of diagnostic accuracy studies in glaucoma using the Heidelberg retina tomograph. Invest Ophthalmol Vis Sci 2006;47: Bossuyt PM, Reitsma JB, Bruns DE, et al. The STARD statement for reporting studies of diagnostic accuracy: explanation and elaboration. Clin Chem 2003;49: Chylack LT Jr, Wolfe JK, Singer DM, et al. The Lens Opacities Classification System III. Arch Ophthalmol 1993;111: Heijl A, Lindgren A, Lindgren G. Test-retest variability in glaucomatous visual fields. Am J Ophthalmol 1989;108: Chauhan BC, Johnson CA. Test-retest variability of frequencydoubling perimetry and conventional perimetry in glaucoma patients and normal subjects. Invest Ophthalmol Vis Sci 1999; 40: Girkin CA. Principles of confocal scanning laser ophthalmoscopy for the clinician. In: Fingeret M, Flanagan JG, Liebmann JM, eds. The Essential HRT Primer. Heidelberg, Germany: Heidelberg Engineering; 2005: Fingeret M. Using the Heidelberg retina tomograph II (HRT II): image acquisition and accessing the data. In: Fingeret M, Flanagan JG, Liebmann JM, eds. The Essential HRT Primer. Heidelberg, Germany: Heidelberg Engineering; 2005: Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148: Wagner RR, Beiden SV, Metz CE. Continuous versus cate- 780

7 Ferreras et al HRT3 and Logistic Regression Analysis for Glaucoma Diagnosis gorical data for ROC analysis: some quantitative considerations. Acad Radiol 2001;8: Saruhan A, Orgül S, Koçak I, et al. Descriptive information of topographic parameters computed at the optic nerve head with the Heidelberg retina tomograph. J Glaucoma 1998;7: Iester M, Mickelberg FS, Swindale NV, Drance SM. ROC analysis of Heidelberg retina tomograph optic disc shape measures in glaucoma. Can J Ophthalmol 1997;32: Mistlberger A, Liebmann JM, Greenfield DS, et al. Heidelberg retina tomograph and optical coherence tomography in normal, ocular-hypertensive, and glaucomatous eyes. Ophthalmology 1999;106: Vihanninjoki K, Teesalu P, Burk RO, et al. Search for an optimal combination of structural and functional parameters for the diagnosis of glaucoma: multivariate analysis of confocal scanning laser tomograph, blue-on-yellow visual field and retinal nerve fiber layer data. Graefes Arch Clin Exp Ophthalmol 2000;238: Miglior S, Casula M, Guareschi M, et al. Clinical ability of Heidelberg Retina Tomograph examination to detect glaucomatous visual field changes. Ophthalmology 2001;108: Heidelberg Retina Tomograph Glaucoma Module Operating Instructions Version 3.0. Heidelberg, Germany: Heidelberg Engineering; Garway-Heath DF, Hitchings RA. Sources of bias in studies of optic disc and retinal nerve fibre layer morphology. Br J Ophthalmol 1998;82: Medeiros FA. How should diagnostic tests be evaluated in glaucoma? Br J Ophthalmol 2007;91: Iester M, Mickelberg FS, Courtright P, Drance SM. Correlation between the visual field indices and Heidelberg Retina Tomography parameters. J Glaucoma 1997;6: Iester M, Broadway D, Mickelberg FS, Drance SM. A comparison of healthy, ocular hypertensive, and glaucomatous optic disc topographic parameters. J Glaucoma 1997;6: Centre for Evidence-Based Medicine. Likelihood ratios. Available at: Accessed February 24, Medeiros FA, Zangwill LM, Bowd C, et al. Influence of disease severity and optic disc size on the diagnostic performance of imaging instruments in glaucoma. Invest Ophthalmol Vis Sci 2006;47: Hodapp E, Parrish RKII, Anderson DR. Clinical Decisions in Glaucoma. St. Louis: Mosby; 1993:

8 Ophthalmology Volume 115, Number 5, May 2008 Figure 1. Receiver operating characteristic curves of the linear discriminant function (LDF), Frederick S. Mikelberg (FSM) discriminant function, and Reinhard O. W. Burk (RB) discriminant function between healthy eyes and glaucomatous patients in the population used for obtaining the LDF. Table 2. Best Sensitivity and Specificity Balance and Likelihood Ratios of the Linear Discriminant Function, Frederick S. Mikelberg Discriminant Function, Reinhard O. W. Burk Discriminant Function, Moorfields Regression Analysis, and Glaucoma Probability Score to Discriminate between Normal and Glaucomatous Eyes in the Population Used for Obtaining the Linear Discriminant Function Heidelberg Retina Tomograph Parameter Cutoff Point Sensitivity (%) Specificity (%) Positive Likelihood Ratio Negative Likelihood Ratio LDF FSM RB MRA overall Borderline GPS global ONL FSM Frederick S. Mikelberg discriminant function; GPS global glaucoma probability score; LDF linear discriminant function; MRA global Moorfields regression analysis classification; ONL outside normal limits; RB Reinhard O. W. Burk discriminant function. The cutoff points were calculated by the MedCalc software as the points with the best sensitivity and specificity balance. Likelihood ratios 10 and 0.1 are related with large changes in the probability of disease. 781.e1

9 Ferreras et al HRT3 and Logistic Regression Analysis for Glaucoma Diagnosis Table 3. Receiver Operating Characteristic Curve Analyses for the Linear Discriminant Function and the Frederick S. Mikelberg and Reinhard O. W. Burk Discriminant Functions in the Population Used for Obtaining the Linear Discriminant Function Area under the Receiver Operating Characteristic Curve 95% Confidence Interval P Value (Area 0.5) Sensitivity and Specificity Specificity 85% Specificity 95% LDF % (85.1%) 78.8% (95.0%) FSM % (85.1%) 76.4% (95.0%) RB % (85.1%) 68.2% (95.0%) FSM Frederick S. Mikelberg discriminant function; LDF linear discriminant function; RB Reinhard O. W. Burk discriminant function. Table 5. Sensitivity and Specificity of Moorfields Regression Analysis and Glaucoma Probability Score Classifications Depending on Disc Area Size Disc Area Size Small discs, 1.6 mm 2 56 normal eyes 34 glaucomatous eyes Average size discs, mm normal eyes 123 glaucomatous eyes Large discs, 2.6 mm 2 42 normal eyes 53 glaucomatous eyes Heidelberg Retina Tomograph Parameter Area under the Receiver Operating Characteristic Curve (95% Confidence Interval) Cutoff Point Sensitivity (%) Specificity (%) Positive Likelihood Ratio Negative Likelihood Ratio LDF ( ) FSM ( ) RB ( ) LDF ( ) FSM ( ) RB ( ) LDF ( ) FSM ( ) RB ( ) FSM Frederick S. Mikelberg discriminant function; LDF linear discriminant function; RB Reinhard O. W. Burk discriminant function. P values were less than for all areas under the receiver operating characteristic curve. The cutoff points were calculated by the MedCalc software as the points with the highest sensitivity and specificity balance. Likelihood ratios 10 and 0.1 are related to large changes in the probability of disease. 781.e2