Combining Structural and FunctionalTesting for Detection of Glaucoma

Neha N. Shah, MD, Christopher Bowd, PhD, Felipe A. Medeiros, MD, Robert N. Weinreb, MD,Pamela A. Sample, PhD, Esther M. Hoffmann, MD, Linda M. Zangwill, PhD

Purpose: To assess whether the combined use of both structural and functional tests improves discrimi-nation between healthy and glaucomatous eyes over either type of testing alone.

Design: Observational cross-sectional study.Participants: One hundred twenty-three eyes of 123 participants enrolled in the Diagnostic Innovations in

Glaucoma Study.Methods: Because both structural and functional tests were evaluated, 2 definitions of glaucoma were used:

glaucomatous visual field (VF) damage based on repeatable abnormal standard automated perimetry results(n � 43) and glaucomatous optic disc (glaucomatous optic neuropathy [GON]) based on masked assessmentof optic disc stereophotographs (n � 65). Participants had scanning laser polarimetry, optical coherencetomography (OCT), and confocal scanning laser ophthalmoscopy imaging in addition to frequency-doublingtechnology (FDT) perimetry and short-wavelength automated perimetry (SWAP) testing completed within a6-month interval.

Main Outcome Measures: For each glaucoma definition, sensitivities and specificities were calculated forthe best structural parameters, FDT perimetry pattern standard deviation (PSD), and SWAP PSD/glaucomahemifield test, and then for each possible pairwise combination of one structural and one functional parameter.

Results: The best structural parameters for discriminating between glaucoma and control eyes were nervefiber indicator for scanning laser polarimetry, inferior average retinal nerve fiber layer thickness for OCT, andMoorfields regression classification for confocal scanning laser ophthalmoscopy. Sensitivities and specificitiesfor detecting glaucomatous VF damage for scanning laser polarimetry, OCT, confocal scanning laser ophthal-moscopy, FDT perimetry, and SWAP were 41.9% and 98.3%, 58.1% and 98.3%, 58.1% and 84.5%, 44.2% and98.3%, and 65.1% and 86.2%, respectively. Adding FDT perimetry to each of the best structural parameters ledto a significant (P�0.05) increase in sensitivity without a significant change in specificity compared with structuralparameters alone. Adding SWAP to each of the best structural parameters led to a significant increase insensitivity and also a significant decrease in specificity compared with each structural parameter alone.

Conclusions: A combination of parameters from structural tests and functional tests can improve thesensitivity of glaucoma detection. The sensitivity and specificity desired for a particular clinical application willdetermine the selection of the particular diagnostic tests. Ophthalmology 2006;113:1593–1602 © 2006 by the

American Academy of Ophthalmology.

Glaucoma is characterized by structural changes to the opticnerve head and retinal nerve fiber layer (RNFL) and func-tional changes, particularly visual field (VF) loss. Recentrandomized clinical trials including the Ocular Hyperten-sion Treatment Study and the European Glaucoma Preven-tion Study report that the first damage in patients withglaucoma can be either structural or functional.1,2 Thissuggests that using structural and functional testing in com-bination may improve glaucoma detection.

Originally received: December 7, 2005.Accepted: June 13, 2006. Manuscript no. 2005-1198.From Hamilton Glaucoma Center, Department of Ophthalmology, Univer-sity of California at San Diego, La Jolla, California.Supported in part by the National Institutes of Health, Bethesda, Mary-land (grant nos.: 5M01 RR000827 [NNS], EY11008 [LMZ], EY08208[PAS]). Participant retention incentive grants in the form of glaucomamedication at no cost: Alcon Laboratories Inc., Fort Worth, Texas;

Allergan, Irvine, California; Pfizer Inc., New York, New York; and

© 2006 by the American Academy of OphthalmologyPublished by Elsevier Inc.

Numerous functional and structural tests for glaucomadiagnosis are available, with newer testing methods over-coming some of the limitations of established techniques.As an example, standard automated perimetry (SAP), whenused alone, has poor sensitivity for detecting glaucoma.In contrast, visual function–specific testing, with eitherfrequency-doubling technology (FDT) perimetry or short-wavelength automated perimetry (SWAP), has been dem-onstrated to be more sensitive and can detect VF loss earlier

SANTEN Inc., Napa, California (PAS). Financial support from CarlZeiss Meditec, Dublin, California (FAM, RNW, PAS, LMZ); Heidel-berg Engineering, Heidelberg, Germany (FAM, RNW, LMZ); Welch-Allyn, Skaneateles Falls, New York (PAS); and Haag-Streit, Mason,Ohio (PAS). Honoraria from Heidelberg Engineering (FAM, RNW,LMZ) and Carl Zeiss Meditec (RNW).

Correspondence to Linda M. Zangwill, PhD, Hamilton Glaucoma Center,Department of Ophthalmology, University of California, San Diego, La

Jolla, CA 92093-0946. E-mail: Zangwill@glaucoma.ucsd.edu.

1593ISSN 0161-6420/06/$–see front matterdoi:10.1016/j.ophtha.2006.06.004

Ophthalmology Volume 113, Number 9, September 2006

than SAP.3–9 As another example, evaluation of optic discstereophotographs in clinical practice has been reported to bepoorly reproducible and subjective.10,11 In contrast, quantita-tive imaging instruments, including the scanning laser polar-imeter, optical coherence tomograph, and confocal scanninglaser ophthalmoscope, can provide highly reproducible andobjective structural measures of the optic disc and retina.Furthermore, each of these 3 quantitative imaging methodshas been shown to be comparable to expert assessment of opticdisc stereophotographs in their ability to discriminate betweenglaucomatous and healthy eyes.12 Despite having a number ofadvantages over the more traditional diagnostic tests, there hasbeen only limited agreement among these newer diagnostictests in discriminating glaucomatous from healthy eyes.12–14

The goal of the current study was to assess whether thecombined use of structural and functional tests improvesdiscrimination between healthy and glaucomatous eyes overeither type of testing alone.

Materials and Methods

Subjects

One hundred twenty-three participants, each with one eye chosenat random for study, were included in this cross-sectional obser-vational study. All participants were enrolled prospectively in thelongitudinal Diagnostic Innovations in Glaucoma Study, at theHamilton Glaucoma Center, University of California, San Diego.

Inclusion Criteria for the Diagnostic Innovations in Glau-coma Study. All subjects underwent a complete ophthalmologicexamination, including slit-lamp biomicroscopy, intraocular pres-sure (IOP) measurement, dilated stereoscopic fundus examination,and stereophotography of the optic nerve heads. Simultaneousstereoscopic photographs were obtained for all subjects and wereof adequate quality for the subject to be included. All subjects hadopen angles, best-corrected visual acuity (VA) of 20/40 or better,spherical refraction within �5.0 diopters (D), and cylinder correc-tion within �3.0 D. A family history of glaucoma was allowed.

Informed consent was obtained from all participants, and allmethods were approved by the University of California, San Di-ego, Human Subjects Committee. The study adhered to the Dec-laration of Helsinki for research involving human subjects.

Exclusion Criteria for the Diagnostic Innovations in Glau-coma Study. Subjects were excluded if they had a history ofintraocular surgery (except for uncomplicated cataract or glau-coma surgery). We also excluded all subjects with nonglaucoma-tous secondary causes of elevated IOP (e.g., iridocyclitis, trauma),

Table 1. Characte

SAP Definition (n � 10

Glaucoma(n � 43)

Control(n � 58)

Race (% white) 86 88Gender (% male) 53 38Age (yrs) [mean (95% CI)] 68.3 (64.8–71.7) 58.6 (55.6–61.5

CI � confidence interval; SAP � standard automated perimetry.Ninety-two participants overlap between SAP and stereophotograph definparticipants were identified as controls (nonglaucomatous) by both defini

*Using chi-square or t test.

1594

other intraocular eye disease, other diseases affecting the VF (e.g.,pituitary lesions, demyelinating diseases, human immunodefi-ciency virus positive, AIDS, diabetic retinopathy), medicationsknown to affect VF sensitivity, or problems other than glaucomaaffecting color vision.

A gold standard for diagnosis of glaucoma is a test that is widelyaccepted as being accurate in diagnosis and therefore can be used asa standard against which other diagnostic strategies are judged. Be-cause glaucoma cannot be defined outside of the context of structureand function, and we are evaluating both structural and functionaltesting in this study, we risk biasing our results if only a structural oronly a functional gold standard definition for glaucoma is used. Forthis reason, we conducted 2 parallel analyses: one using a functionaldefinition for glaucoma—namely, repeatable glaucomatous fieldloss by SAP—and the other using a structural definition—namely,glaucomatous optic neuropathy (GON) based on assessment ofoptic disc stereophotographs.

Participants included in the current study had scanning laserpolarimetry (GDx VCC, software version 5.5.0.14, Carl ZeissMeditec, Inc., Dublin, CA), optical coherence tomography (StratusOCT, software version 4.0, Carl Zeiss Meditec), confocal scanninglaser ophthalmoscopy (HRT II, acquisition software version1.7.0.2, Heidelberg Engineering, Heidelberg, Germany), FDT pe-rimetry (N-30 thresholding program, software version 3.00.1,Welch-Allyn, Skaneateles Falls, NY), and SWAP testing (Hum-phrey Field Analyzer, software version rev. 3.4.7 SSE [0003], CarlZeiss Meditec) within a 6-month interval (n � 123). Participantswho also had simultaneous stereoscopic optic disc photographstaken within the 6-month interval were included in the stereopho-tograph definition analysis group (n � 114), and those participantswho had reliable SAP using the 24-2 Swedish Interactive Thresh-old Algorithm (Carl Zeiss Meditec) testing within the 6-monthinterval were included in the SAP definition analysis group (n �101); 92 participants overlap between the 2 analysis groups. Areliable SAP test required fixation loss, false-negative, and false-positive errors � 25%, unless false negatives could be explainedby significant field loss.

Participant eyes were classified into either one or both SAP andstereophotograph analysis groups. Of 123 eyes, 101 were includedin the SAP analysis group, and 114 were included in the stereo-photograph analysis group (Table 1, Fig 1). For each analysisgroup, eyes were further classified by glaucoma diagnosis based onthe definition for that group.

For the SAP analysis group, the 43 eyes classified as glaucomahad glaucomatous VF damage defined as 2 consecutively abnor-mal SAP test results, with at least 1 of the 2 abnormal tests beingwithin the 6-month interval. Abnormal SAP results were defined aspattern standard deviation (PSD) outside the 95% normal confi-dence limits or glaucoma hemifield test (GHT) outside normallimits. Abnormal points were not required to be contiguous. The

s of Study Groups

Stereophotograph Definition (n � 114)

P Value*Glaucoma(n � 65)

Control(n � 49) P Value*

0.78 88 86 0.760.12 42 37 0.60

�0.001 65.5 (62.5–68.5) 60.1 (56.7–63.6) 0.02

groups; 32 participants were identified as glaucoma by both definitions: 44

ristic

1)

)

itiontions.

elengt

Shah et al � Combining Structural and Functional Testing for Detection of Glaucoma

58 eyes used as controls had no glaucomatous VF damage and nohistory of IOP � 22 mmHg. Optic disc appearance was not usedas a diagnostic criterion for eyes in the SAP analysis group.

For the stereophotograph analysis group, the 65 eyes classifiedas glaucoma had GON based on masked consensus grading ofsimultaneous stereoscopic optic disc photographs by 2 expertgraders. Adjudication by a third expert grader was completed incases of disagreement. Glaucomatous optic neuropathy was de-fined as focal rim notching, rim thinning, or RNFL abnormality.Because only one eye per participant was included, asymmetry ofthe cup-to-disc ratio between eyes was not a criterion for GON. The49 eyes used as controls had no GON on stereophotograph assessmentand no history of IOP � 22 mmHg. Standard automated perimetrywas not used as a diagnostic criterion for eyes in the stereophotographanalysis group. Ninety-two participants met the criteria for inclusionin both analyses. Ocular hypertensive eyes were not included in eitherthe SAP analysis or the stereophotograph analysis.

Instrumentation

GDx VCC Scanning Laser Polarimeter. The GDx VCC is ascanning laser polarimeter with variable corneal compensation. Itmeasures retardation of light that has double passed through thebirefringent fibers of the RNFL. Details of this instrument havebeen described elsewhere.15 With variable corneal compensation,which replaces the fixed corneal compensation found on earlierversions of this instrument, the macula functions as an intraocularpolarimeter to measure and neutralize eye-specific corneal polar-ization axis and magnitude.16

Each participant had 1 to 3 images of the fundus obtained fromthe GDx VCC in a single sitting. Only those images passingquality control standards as assessed by an expert examiner wereincluded. Based on the manufacturer’s suggestions, good imageshad a focused and evenly illuminated reflectance image with a

SAPGDx VCC Stratus OCT HRT II FDTSWAP

Analysis Group 1: Glaucoma based on functional

(SAP) definition

All tests within

6-month interval

Glaucomatous Visual Field Damage

• Repeatable abnormal standard automated perimetry (SAP), SITA 24-2 program

• GHT outside of normal limits or PSD < 5%

Normal• No repeatable abnormal

SAP• No history of elevated IOP

Figure 1. Overview of standard automated perimetry (SAP) gold standdoubling technology; GHT � glaucoma hemifield test; IOP � intraocularSITA � Swedish Interactive Threshold Algorithm; SWAP � short-wav

centered optic disc, a residual anterior segment retardation of �15

nm, and a typical scan score of �25. Multiple good images, whenavailable, were merged automatically into one mean image byGDx VCC software.

Parameters examined were calculated automatically by GDxVCC software and have been described previously.17 These in-cluded superior ratio, inferior ratio, superior maximum, inferiormaximum, superior average, inferior average, superior normalizedarea (area under the temporal–superior–nasal–inferior–temporal[TSNIT] curve in the superior quadrant), inferior normalized area,TSNIT average, TSNIT standard deviation (SD), and nerve fiberindicator (NFI).

Diagnostic categorization was based on comparison with themanufacturer’s internal normative database or suggested cutoffs.The parameters TSNIT average, TSNIT SD, superior average, andinferior average have corresponding probability measures of ab-normality (provided on GDx VCC printouts) based on comparisonof measurements with an internal normative database of 540normal eyes. The probabilities, each indicated by a different coloron the printout, include levels of P�0.5%, P�1%, P�2%, P�5%,and P�5%. In this study, we considered P�0.5% to be outsidenormal limits (ONL); the 3 probabilities P�1%, P�2%, andP�5% to be borderline; and P�5% to be within normal limits(WNL). The parameter NFI does not have the probability ofabnormality values provided but does have cutoffs suggested bythe manufacturer: 0 to 30, WNL; 31 to 50, borderline; and 51 to100, ONL.

Stratus Optical Coherence Tomograph. Stratus OCT mea-sures peripapillary RNFL thickness by computing the delay timeof light backscatter from the RNFL compared with light reflectedoff of a reference mirror. Information regarding thickness of retinalstructures is obtained from the pattern of interference of these 2light beams.18

Each participant had 3 images obtained with Stratus OCT usingthe fast RNFL algorithm. Each set of 3 images was merged

StereophotographyGDx VCC Stratus OCT HRT II FDTSWAP

Analysis Group 2: Glaucoma based on structural (stereophotograph) definition

All tests within

6-month interval

laucomatous Optic Neuropathy (GON)

Masked assessment of simultaneous stereophotographs by two expert graders Rim thinning, excavation, or RNFL defect

Normal• No GON on assessment of

stereophotograph• No history of elevated IOP

nd stereophotography gold standard analysis groups. FDT � frequency-ure; PSD � pattern standard deviation; RNFL � retinal nerve fiber layer;h automated perimetry.

G

•

•

ard apress

automatically into a mean image by OCT software. Each image

1595

Ophthalmology Volume 113, Number 9, September 2006

consists of 256 A-scans along a 3.4-mm-diameter ring surroundingthe optic disc. Only images passing quality control standards asassessed by an expert examiner were included. Good images hadan imaging ring centered on the optic disc, focused fundus image,and signal strength � 7.

Parapapillary RNFL thickness parameters calculated by StratusOCT software were examined. These included average RNFLthickness (360° measure), temporal quadrant thickness (316°–45°), superior quadrant thickness (46°–135°), nasal quadrant thick-ness (136°–225°), and inferior quadrant thickness (226°–315°).

Diagnostic categorization was provided by Stratus OCTsoftware-based comparison of thickness parameter values with themanufacturer’s internal normative database of 328 eyes. Parame-ters for a given eye are noted on the Stratus OCT printout as towhether they fall outside of the 99% confidence interval (CI)(ONL), between the 95% and 99% CIs (borderline), or within the95% CI (WNL).

HRT II Confocal Scanning Laser Ophthalmoscope. TheHRT II sequentially scans the retinal surface in horizontal andvertical directions at multiple planes using a diode laser. Scans areused to construct a 3-dimensional topographic image using con-focal scanning principles.19

Each participant had 3 topographic images of the optic discobtained during one sitting. These images were aligned automat-ically and combined to generate a single mean topographic image.Magnification errors were corrected by using participants’ cornealcurvature measurements. Only those images passing standard-ized quality control standards were included. Good images hada focused and centered reflectance image and SD � 50 �m. Theoptic disc margin was defined by a contour line drawn byagreement of 2 experienced examiners while referring to opticdisc stereophotographs.

We examined parameters that are calculated by HRT II soft-ware using Moorfields regression analysis,20 which compares aparticular eye’s rim area with a predicted rim area for a given discarea and age. Moorfields regression classification (MRC) param-eters corresponding to 6 optic disc sectors were examined: tem-poral (316°–44°), temporal superior (45°–90°), nasal superior(91°–135°), nasal (136°–225°), nasal inferior (226°–270°), andtemporal inferior (271°–315°). Also examined were MRC global,which evaluates the overall rim area, and the parameter MRCresult, which will be ONL if any 1 of the 6 sector or MRC globalparameters is ONL and borderline if any of the 6 sector or MRCglobal parameters is borderline and none are ONL.

Diagnostic categorization of MRC parameters is based onconfidence limits of regression analysis from 112 normal eyes. AnMRC parameter is WNL if it falls within the 95% CI, borderlineif it falls between the 95% and 99.9% CI limits, and ONL if it fallsoutside the 99.9% CI.

Short-Wavelength Automated Perimetry. Short-wavelengthautomated perimetry involves a 440-nm narrow-band stimuluspresented on a 100 candela per square meter yellow background21

to stress short-wavelength cones and small bistratified blue–yellowganglion cells preferentially.22,23

Participants had SWAP testing done using the 24-2 full-thresholdprogram, which presents target stimuli to the subject in 1 of 54locations separated by 6° surrounding the horizontal and verticalmeridians.23 Only SWAP tests meeting reliability standards wereincluded in the study. A reliable test required fixation loss, false-negative, and false-positive errors � 25%, unless false negativescould be explained by significant field loss.

To determine abnormality on visual function, we combined theglobal parameters of PSD and the GHT result. This combinationhas been shown to have good specificity for glaucoma detection23

and is often used in combination for diagnosis for improved

sensitivity.1 Diagnostic categorization of eyes using PSD and GHT

1596

is based on comparison with age-corrected normative data from348 normal subjects from the Hamilton Glaucoma Center data-base, none of whom were part of this study’s control groups.Categorization by PSD was based on PSD probability levels:P�5%, P�5%, P�2%, P�1%, and P�0.5%. For this study, PSDprobability levels of P�1% and P�0.5% were considered ONL;levels of P�2% and P�5%, borderline; and P�5%, WNL. AGHT result of “outside normal limits” was considered ONL; aresult of “borderline” was considered borderline; and a result of“generalized reduction of sensitivity” or “within normal limits”was considered WNL. Participants were considered to be ONL bythe SWAP PSD/GHT meta-parameter if either PSD or GHT orboth met ONL criteria; borderline by SWAP PSD/GHT if either orboth met borderline criteria, with neither meeting ONL criteria;and WNL by SWAP PSD/GHT if both PSD and GHT met WNLcriteria.

Frequency-Doubling Technology Perimetry. This perimetryuses a modified binary search algorithm to measure stimuluscontrast detection. It makes use of the frequency-doubling effectthat occurs when a coarse sinusoidal grating is counterphase flick-ered rapidly. In FDT perimetry, individual 10° squares with 0.25cycle/degree vertical sinusoidal gratings, counterphase flickeringat 25 hertz, make up the target stimuli.24 The stimulus parametersof FDT perimetry are designed primarily to target the magnocel-lular ganglion cells.4,21,24

Participants had FDT perimetry testing done using the N-30thresholding program, which presents stimuli to a subject in 1 of19 test locations in the central 30° of the VF.24 Only FDT perim-etry tests meeting reliability standards were included in the study.A reliable test required fixation loss, false-negative, and false-positive errors � 25%, unless false negatives could be explainedby significant field loss.

The FDT perimetry global parameter investigated in this studywas PSD, as GHT is not available for the N-30 test pattern. Patternstandard deviation has been shown to have good sensitivity andspecificity for separating eyes with GON from healthy eyes.4

Diagnostic categorization of eyes is based on comparison of at-tained PSD values, after age correction, with those in the manu-facturer’s normative database. Possible probability levels includeP�5%, P�5%, P�2%, P�1%, and P�0.5%. For this study, PSDprobability levels of P�1% and P�0.5% were considered ONL;P�2% and P�5%, borderline; and P�5%, WNL.

Statistical AnalysisTwo parallel analyses were conducted, each using a differentdefinition for glaucoma diagnosis: (1) functional diagnosisbased on SAP and (2) structural diagnosis based on optic discstereophotography.

For both analyses, receiver operator characteristic (ROC)curves were generated for each structural parameter investigated.Receiver operator characteristic curves were generated for contin-uous numerical data for GDx VCC and Stratus OCT parametersand for diagnostic categorizations for HRT II MRC parameters.We assigned WNL a value of 1; borderline, 2; and ONL, 3. Areasunder the ROC curve (AUROCs) were compared across parame-ters for each structural instrument. The parameter from each struc-tural instrument with the highest AUROC for glaucoma diagnosisbased on SAP (Table 2) was included in further analysis.

Functional parameters used in both analyses were those gener-ally accepted, as described above, to have good sensitivity andspecificity for glaucoma detection1,23: PSD and GHT in combina-tion (PSD/GHT meta-parameter) for SWAP, and PSD for FDTperimetry.

Sensitivity, specificity, and likelihood ratios were calculated for

diagnostic classification of eyes by the best structural parameter

is ind

Shah et al � Combining Structural and Functional Testing for Detection of Glaucoma

(i.e., parameter with the highest AUROC) for each structuralinstrument, PSD for FDT perimetry, and the PSD/GHT meta-parameter for SWAP. This was done for both SAP and stereopho-tograph definition analyses. In calculation of sensitivity and spec-ificity, WNL and borderline classifications were both considerednonglaucoma, and the ONL classification was considered glau-coma. Likelihood ratios were calculated for all 3 classifications(WNL, borderline, and ONL).

After sensitivities and specificities for glaucoma diagnosis werecalculated for structural and functional parameters alone, theywere calculated for all possible pairwise combinations of onestructural with one functional parameter. For each combination, aneye was considered glaucomatous if it was diagnostically classifiedas glaucoma by either the structural or the functional parameter inthe combination. Next, likelihood ratios for glaucoma diagnosisfor both SAP and stereophotograph definition analysis groups werecalculated for each combination pairing of structural and func-tional parameters. Likelihood ratios were calculated for all 3classifications (WNL, borderline, ONL) for each combination pair-ing. For likelihood ratio calculations, the combination was ONL ifeither the structural or the functional parameter or both were ONL,borderline if either the structural or the functional parameter orboth were borderline and neither were ONL, and WNL if bothstructural and functional parameters were WNL.

Ninety-five percent CIs around sensitivities and specificitieswere calculated using the Wilson score method as described byNewcombe.25 Sensitivities and specificities were compared us-ing the McNemar test. Statistical analysis was completed usingJMP statistical software (version 5.1.1, SAS Institute Inc., Cary,

Table 2. Areas under the Receiver OperatinDiagnosis Based on Standard Automated

Based on Stereophotography fo

Structural Instrument Parameter

GDx VCC Nerve fiber indicatTSNIT standard dInferior normalizedTSNIT averageSuperior normalizeInferior averageSuperior maximumSuperior averageSuperior ratioInferior maximumInferior ratio

Stratus OCT Inferior thicknessAverage thicknessSuperior thicknessNasal thicknessTemporal thicknes

HRT II MRC resultMRC temporal infMRC nasal inferioMRC globalMRC temporal supMRC nasal superioMRC nasalMRC temporal

MRC � Moorfields regression classification; TSNIT �The highest AUROC from each structural instrument

NC).

Results

Participant characteristics by analysis group are described in Table 1.There were no significant differences (P�0.05) in race or betweenglaucoma and control participants based on SAP or stereophoto-graphs. There was, however, a significant difference in age be-tween glaucoma and control participants.

Areas under the ROC curves for various structural parametersare shown in Table 2. For analysis 1, glaucoma based on SAP, thebest parameter for GDx VCC was NFI, the best parameter forStratus OCT was inferior thickness, and the best Moorfields re-gression parameter for HRT II was MRC result. For analysis 2,glaucoma based on stereophotography, the best GDx VCC param-eter was NFI, the best Stratus OCT parameter was average thick-ness, and the best HRT II Moorfields regression parameter wasMRC result. Best parameters from the SAP analysis group werechosen for further analysis.

Analysis 1: Standard AutomatedPerimetry DefinitionUpon combination with FDT perimetry PSD, sensitivities fordetection of glaucoma increased from 41.9% to 62.8% for GDxVCC NFI, 58.1% to 72.1% for Stratus OCT inferior thickness, and58.1% to 74.4% for HRT II MRC result (Table 3). Sensitivitiesafter combination were significantly higher (P�0.05) than sensi-tivities for structural parameters alone and for FDT perimetry PSDalone.

There was a small decrease in specificities upon combination of

aracteristic Curve (AUROCs) for Glaucomaetry (SAP) and for Glaucoma Diagnosis

uctural Instrument Parameters

AUROC

SAPDefinition

StereophotographDefinition

0.90 0.80on 0.86 0.77

0.85 0.750.84 0.750.84 0.760.83 0.720.83 0.790.83 0.760.79 0.770.78 0.680.76 0.730.88 0.770.86 0.790.79 0.750.70 0.690.67 0.590.75 0.800.74 0.720.74 0.750.69 0.670.67 0.660.66 0.650.60 0.650.59 0.61

oral–superior–nasal–inferior–temporal.icated in bold.

g ChPerimr Str

oreviati

area

d area

s

eriorr

eriorr

temp

structural parameters with FDT perimetry PSD. Specificities de-

1597

ation

Ophthalmology Volume 113, Number 9, September 2006

creased from 98.3% to 96.6% for GDx VCC NFI, 98.3% to 96.6%for Stratus OCT inferior thickness, and 84.5% to 82.8% for HRTII MRC result. Specificities after combination were not signifi-cantly lower than specificities for each structural parameter alone.There was, however, a significant decrease in specificity comparedwith FDT perimetry PSD alone upon combination with HRT IIMRC result, but not upon combination with Stratus OCT inferiorthickness or GDx VCC NFI.

There was a greater increase in sensitivities and greater de-crease in specificities upon combination of structural parameterswith SWAP PSD/GHT relative to combination with FDT perim-etry PSD. Upon combination with SWAP PSD/GHT, sensitivitiesincreased from 41.9% to 76.7% for GDx VCC NFI, 58.1% to76.7% for Stratus OCT inferior thickness, and from 58.1% to76.7% for HRT II MRC result. Upon combination of structuralparameters with SWAP PSD/GHT, there was a significant increasein sensitivity compared with each structural parameter alone, butno significant increase in sensitivity compared with SWAP PSD/GHT alone.

Specificities decreased from 98.3% to 84.5% for GDx VCCNFI, 98.3% to 86.2% for Stratus OCT inferior thickness, and84.5% to 72.4% for HRT II MRC result. Upon combination withSWAP, there was a significant decrease in specificity comparedwith each structural parameter alone. There was a significantdecrease in specificity relative to SWAP alone upon combinationwith HRT II MRC result, but not upon combination with eitherOCT inferior thickness or GDx VCC NFI.

Likelihood ratios (Table 4) for each category WNL, borderline,and ONL can be interpreted in the context of a system thatcategorizes likelihood ratios by the extent to which they affectposttest probability of disease.26 According to this system, likeli-hood ratios between 0.5 and 2 have an insignificant effect onposttest probability of disease. Likelihood ratios between 0.2 and0.5 or between 2 and 5 have a small effect on posttest probability.Ratios between 0.1 and 0.2 or between 5 and 10 have a moderateeffect, and ratios � 0.1 or � 10 have a large effect on posttestprobability of disease.

Applying this likelihood ratio classification system when add-ing FDT perimetry to GDx NFI and OCT inferior thickness anal-ysis 1 WNL outcomes resulted in the likelihood ratios changing

Table 3. Sensitivities and Specificities (%) with 95on Standard Automat

Sensitivity Specificity

44.Structural parameters

GDx VCC NFI 41.9 (28.4–56.7) 98.3 (90.9–99.7) 62.

Stratus OCTinferior thickness

58.1 (43.3–71.6) 98.3 (90.9–99.7) 72.

HRT IIMRC result

58.1 (43.3–71.6) 84.5 (73.1–91.6) 74.

FDT � frequency-doubling technology; GHT � glaucoma hemifield tePSD � pattern standard deviation; SWAP � short-wavelength automate*P value based on comparison with functional test alone in the combina†P value based on comparison with structural parameter alone in combin

from 0.389 and 0.311, respectively (a small effect), to 0.172 and

1598

0.197, respectively (a moderate effect) (Table 4). A similar butmore limited change in likelihood ratios occurred with the additionof FDT perimetry to HRT II WNL MRC outcomes (from 0.376 to0.202). The addition of SWAP to GDx, OCT, and HRT II WNLoutcomes also resulted in a decrease in likelihood ratios and,therefore, a greater effect on posttest probabilities. In contrast, theaddition of FDT perimetry or SWAP to GDx, OCT, and HRT IIONL outcomes resulted in a lowering of the likelihood ratios andsome reduced effect on the posttest probability. However, after theaddition of FDT perimetry, the likelihood ratios for ONL outcomesthat had large effects on posttest probability remained large,and the likelihood ratio that had a small effect remained small.With the addition of SWAP, the likelihood ratio of GDx ONLoutcome decreased from 24.3 (large effect) to 4.49 (small effect).Similarly, with the addition of SWAP the OCT ONL outcomelikelihood ratio changed from 33.8 (large effect) to 5.56 (moderateeffect).

Analysis 2: Stereophotograph Definition

A similar pattern of improvement in sensitivity was seen uponaddition of FDT perimetry and SWAP parameters to structuralparameters using stereophotography as the definition of glaucoma(Table 5). In short, sensitivities after combination with FDT pe-rimetry increased significantly compared with each parameteralone, whereas specificities remained relatively constant.

As with the SAP definition analysis, there was a greater in-crease in sensitivities and greater decrease in specificities uponcombination of structural parameters with SWAP PSD/GHT com-pared with combination with FDT perimetry PSD. Upon combi-nation of structural parameters with SWAP PSD/GHT, there wasa significant increase in sensitivity relative to each structuralparameter alone. There was a significant increase in sensitivityover SWAP PSD/GHT alone upon combination with HRT II MRCresult, but not upon combination with either OCT inferior thick-ness or GDx VCC NFI. Upon combination of each structuralparameter with SWAP, there was a significant decrease in speci-ficity compared with each structural parameter alone, but no sig-

onfidence Intervals for Glaucoma Diagnosis Basederimetry (Analysis 1)

Functional Parameters

FDT PSD SWAP PSD/GHT

tivity Specificity Sensitivity Specificity

.4–58.9) 98.3 (90.9–99.7) 65.1 (50.2–77.6) 86.2 (75.1–92.8)

.9–75.6) 96.6 (88.3–99.1) 76.7 (62.3–86.9) 84.5 (73.1–91.6)33* �0.99* 0.0736* �0.99*77† �0.99† 0.0003† 0.0133†

.3–83.3) 96.6 (88.3–99.1) 76.7 (62.3–86.9) 86.2 (75.1–92.8)15* �0.99* 0.0736* �0.99*12† �0.99† 0.0133† 0.0233†

.8–85.1) 82.8 (71.1–90.4) 76.7 (62.3–86.9) 72.4 (59.8–82.3)09* 0.0077* 0.0736* 0.0133*33† �0.99† 0.0133† 0.0233†

RC � Moorfields regression classification; NFI � nerve fiber indicator;imetry.

.

% Ced P

Sensi

2 (30

8 (470.010.00

1 (570.000.04

4 (590.000.02

st; Md pertion.

nificant decrease in specificity compared with SWAP alone.

Shah et al � Combining Structural and Functional Testing for Detection of Glaucoma

Discussion

Regardless of the definition of glaucoma employed in thecurrent study, the combination of FDT perimetry PSD withvarious structural parameters increased sensitivity, with a lim-ited decrease in specificity, relative to structural parametersalone and FDT perimetry PSD alone. Combination of param-eters from structural tests with SWAP PSD/GHT led to agreater increase in sensitivity than combination with FDT

Table 4. Likelihood Ratios for Glaucoma Diagnosis BaseStereophotography (Analysis 2) for Structural and Function

with Frequency-Doubling Technology (FDT) or

SAP Definition (Analy

ParameterAlone

Parameter withFDT PSD

GDx VCC NFIWNL 0.389 0.172Borderline 2.698 1.499ONL 24.337 18.2

Stratus OCT inferior thicknessWNL 0.311 0.197Borderline 1.618 0.843ONL 33.802 20.896

HRT MRC resultWNL 0.376 0.202Borderline 1.349 0.843ONL 3.746 4.317

FDT PSDWNL 0.305 —Borderline 4.045 —ONL 25.692 —

SWAP PSD/GHTWNL 0.337 —Borderline 1.125 —ONL 4.553 —

GHT � glaucoma hemifield test; MRC � Moorfields regression classificatistandard deviation; WNL � within normal limits.

Table 5. Sensitivities and Specificities (%) with 95on Optic Disc Stereo

Sensitivity Specificity

27.Structural parameters

GDx VCC NFI 27.7 (18.3–39.6) 100 (92.7–100) 41.

Stratus OCTinferior thickness

33.9 (23.5–46.0) 98.0 (89.3–99.6) 43.

HRT IIMRC result

49.2 (37.5–61.1) 89.8 (78.2–95.6) 60.

FDT � frequency doubling technology; GHT � glaucoma hemifield tePSD � pattern standard deviation; SWAP � short-wavelength automate*P value based on comparison with functional test alone in combination†

P value based on comparison with structural test alone in combination.

perimetry PSD, but also a greater decrease in specificity thancombination with FDT perimetry PSD relative to structuralparameters alone. For combination with SWAP PSD/GHT,there was, in general, no significant increase in sensitivity ordecrease in specificity compared with SWAP PSD/GHT alone.

To our knowledge, this is the first study that examined thediagnostic accuracy of combining visual function–specific testresults with optical imaging structural test results for detec-tion of glaucoma. Other investigators have combined the

Standard Automated Perimetry (SAP) (Analysis 1) andrameters Alone and Structural Parameters in Combinationrt-Wavelength Automated Perimetry (SWAP)

Stereophotograph Definition (Analysis 2)

ameter withP PSD/GHT

ParameterAlone

Parameter withFDT PSD

Parameter withSWAP PSD/GHT

0.213 0.508 0.367 0.5420.490 4.023 2.414 0.7544.945 Infinity 20.363 3.445

0.173 0.565 0.467 0.6080.613 1.885 1.659 0.5485.565 16.593 10.559 3.445

0.204 0.305 0.245 0.3020.749 6.034 3.268 1.132.781 4.826 4.902 2.81

— 0.574 — —— 4.525 — —— 13.574 — —

— 0.754 — —— 0.283 — —— 3.015 — —

FI � nerve fiber indicator; ONL � outside normal limits; PSD � pattern

onfidence Intervals for Glaucoma Diagnosis Basedography (Analysis 2)

Functional Parameters

FDT PSD SWAP PSD/GHT

tivity Specificity Sensitivity Specificity

.3–39.6) 98.0 (89.3–99.6) 43.1 (31.8–55.2) 85.7 (73.3–92.9)

.4–53.7) 98.0 (89.3–99.6) 49.2 (37.5–61.1) 85.7 (73.3–92.9)077* �0.99* 0.1336* �0.99*077† �0.99† 0.0005† 0.0233†

.8–55.2) 95.9 (86.3–98.9) 49.2 (37.5–61.1) 85.7 (73.3–92.9)044* �0.99* 0.1336* �0.99*412† �0.99† 0.0044† 0.0412†

.9–71.0) 87.8 (75.8–94.3) 63.1 (50.9–73.8) 77.6 (64.1–87.0)001* 0.0736* 0.0009* 0.1336*233† �0.99† 0.0077† 0.0412†

RC � Moorfields regression classification; NFI � nerve fiber indicator;imetry.

d onal PaSho

sis 1)

ParSWA

on; N

% Cphot

Sensi

7 (18

5 (300.00.0

1 (310.00.0

0 (47�0.0

0.0

st; Md per.

1599

Ophthalmology Volume 113, Number 9, September 2006

results of early-generation computed optic image analysis andSAP27 or combined results obtained with earlier-generationoptical imaging instruments and optic disc photographs.12

Caprioli27 found that a discriminant function that combinedstructural and functional parameters showed better discrim-ination between glaucomatous and nonglaucomatous eyesthan discriminant functions for structural or functional pa-rameters alone. Greaney et al12 found that a combination ofvarious optical imaging instruments and stereophotographsprovided discrimination between healthy and glaucomatouseyes superior to that of imaging instruments and stereopho-tographs alone. Additionally, there have been several stud-ies showing limited agreement among techniques in detec-tion of glaucoma patients (Invest Ophthalmol Vis Sci46:e-abstract 3636, 2005).14,28,29 The latter 3 of these stud-ies found poor agreement particularly between structuraland functional tests in identifying patients with glaucoma-tous progression. Poor agreement suggests that combiningstructural and functional tests may improve detection ofglaucomatous eyes.

To simulate glaucoma diagnosis in clinical practice, wealso examined diagnostic classifications based on manufacturer-recommended cutoffs and/or using a normative database. Todetermine the best parameters for GDx VCC and StratusOCT, we used the raw numerical data for each parameter togenerate ROC curves. The best parameters were those withthe largest AUROC for each instrument. We also calculatedsensitivity, specificity, and likelihood ratios with these bestparameters for the instrument-based diagnostic classifica-tions. For HRT II parameters, unlike with GDx VCC andStratus OCT parameters, we generated ROC curves for theMRC classification of WNL, borderline, and ONL, ratherthan for a continuous variable. This can result in underes-timation of the AUROC.30 Because we only directly com-pared the AUROC for each HRT II parameter to the areaunder the curve for other HRT II parameters calculated in asimilar manner, this method for estimation and comparisonwas appropriate. However, these AUROC data for HRT IIshould not be used to compare HRT II diagnostic accuracywith that of other methods. Other studies have been de-signed for this direct comparison.17

The visual function testing parameters included in thisstudy, GHT and PSD, both have been demonstrated to havehigh specificity in glaucoma detection for SAP andSWAP.23 Under Diagnostic Innovations in Glaucoma Studyinclusion criteria, we have included VA cutoffs met byparticipants. Lens status does influence visual function test-ing results, particularly SWAP. As a result, we do notinclude mean deviation in our analysis, but rely on indicesreflecting localized loss, PSD, and GHT.31 As cataractsaffect the VF globally, this approach has gained wide ac-ceptance, especially when evaluating visual function tests,which are all affected to some extent by lens changes.

It is important to note that the results presented in thisstudy are applicable to the chosen parameters and cutoffs. Ifinstead we had used cutoffs for structural parameters dif-ferent from those automatically provided or suggested bythe manufacturer, it is possible that different sensitivitiesand specificities for glaucoma diagnosis would have been

obtained. For functional parameters, we arbitrarily chose to

1600

use P�1% as an FDT perimetry cutoff for ONL. Had wechosen a different cutoff for ONL FDT perimetry (e.g.,P�2%), it is possible that there would have been highersensitivities for glaucoma diagnosis. Short-wavelength au-tomated perimetry PSD/GHT produced higher sensitivitiesand lower specificities before and after combination withstructural parameters than FDT perimetry, for which onlyPSD was used. It is possible that using only PSD forSWAP—rather than PSD and GHT together, as other stud-ies have done—would have resulted in lower sensitivitiesand higher specificities.

The best parameters for discrimination between glau-coma and control eyes for GDx VCC and Stratus OCT foranalysis 1 (glaucoma defined by SAP) are the same as thebest parameters reported by Medeiros et al,17 who also usedSAP to define glaucoma. Areas under the ROC curve forthese parameters (0.90 for GDx VCC NFI and 0.88 forStratus OCT inferior thickness) are similar to the AUROCs(0.91 and 0.92, respectively) found by Medeiros et al. Theslightly lower AUROC values found in the current studymay be explained in part by controls in analysis 1 not beingrequired to have a healthy-appearing optic disc and RNFL,as was required of normals in the Medeiros et al study.Some controls in the current study with an abnormal opticdisc and/or RNFL may have led to poorer discriminationbetween glaucoma and nonglaucoma subjects by the imag-ing instruments and contributed to a lower AUROC.

In general, for Stratus OCT and GDx VCC we foundlower AUROCs when glaucoma was defined by stereopho-tography versus when glaucoma was defined by SAP. Thisis similar to what Bowd et al reported when examiningearlier versions of OCT and GDx.14 This may be attributedto SAP criteria identifying those patients with more ad-vanced disease as abnormal, compared with those identifiedas abnormal by stereophotograph criteria.

When combining 2 tests such that the combination oftests is considered ONL if one or both of the componenttests are ONL, we would expect sensitivity to increase andspecificity to decrease because of the greater probability ofan ONL result for the combination, compared with eachcomponent test alone. However, in the present study thedegree to which sensitivity increased and specificity de-creased varied based on the particular tests added together.Accordingly, a clinician seeking to increase diagnostic ac-curacy by adding a functional test to a structural test needsto consider the sensitivity or specificity tradeoffs. If increas-ing sensitivity is desired, as in the case of a patient who isa glaucoma suspect, our study suggests that the clinicianmay benefit from adding SWAP to the structural test. Ifmaintaining a high specificity is desired, as in the case ofpopulation screening for glaucoma, the clinician may con-sider adding FDT perimetry to a structural test. Similarrecommendations can be made for adding a structural test toa functional test. If clinicians have either SWAP or FDTperimetry and are seeking to maximize sensitivity, they maywant to add Stratus OCT inferior or HRT II MRC result, asthese parameters had the highest sensitivities in our studybefore and after combination with functional tests. If thegoal, however, is to maximize specificity at the same time,

our results suggest that it would be more useful to add

Shah et al � Combining Structural and Functional Testing for Detection of Glaucoma

Stratus OCT inferior thickness than HRT II MRC result, asStratus OCT inferior thickness had a higher specificitybefore and after combination than HRT II MRC result.Likelihood ratios for analysis 1 suggest that addition of afunctional test to a structural test would be most helpful toa clinician who is seeking to confirm a WNL result from astructural test. Likelihood ratios for a WNL categorizationfor structural parameters increased from having a smalleffect to having or approaching a moderate effect on post-test probability after addition of FDT perimetry and alsoafter addition of SWAP. Our data also suggest that additionof FDT perimetry or SWAP would not be particularlyhelpful in confirming an ONL result from a GDx or OCTbecause the likelihood ratios were large using the structuraltest alone, and the addition of the functional test reduced thelikelihood ratios.

For analysis 2, in contrast with analysis 1, likelihoodratios do not support much benefit of adding a functionaltest to a structural test to confirm a WNL result from astructural test. This difference between the 2 analyses maybe explained in part by structural test results likely beingmore consistent with the structural definition of glaucomaused in analysis 2 than with the functional definition used inanalysis 1. As with analysis 1, our likelihood ratio data foranalysis 2 suggest that addition of FDT perimetry or SWAPto structural tests would not be particularly helpful in con-firming an ONL result.

An inherent limitation of the current study and similarstudies evaluating structural and functional techniques is thelack of an independent gold standard for glaucoma that isneither structural nor functional in nature. For instance, astructural definition for evaluating a structural diagnostictechnique may lead to an overestimation of the accuracy ofthe structural technique. To control for such biases, weconducted our analysis once using a structural gold standardand then separately using a functional gold standard. Wefound a similar pattern of change in sensitivity and speci-ficity upon combining functional parameters with structuralparameters, regardless of the definition of glaucoma. Thismay be explained in part by the overlap of 92 participants inthe SAP and stereophotograph analyses (Table 1). To in-vestigate the effect of the definition, it theoretically wouldbe useful to have independent populations in each of the 2analysis groups. However, glaucoma is characterized byboth structural and functional changes, so finding indepen-dent analysis groups is difficult. As the present study iscross-sectional, we are unable to evaluate the ability of thecombination of structural and functional tests to predict whichparticipants will convert to glaucoma at a later time. Longitu-dinal studies are needed to evaluate whether a combination ofoptical imaging techniques and visual function-specific meth-ods for detection of glaucoma can improve prediction of indi-viduals who will convert to glaucoma over time.

In conclusion, a combination of parameters from struc-tural tests with FDT perimetry PSD led to an increase insensitivity with a limited decrease in specificity comparedwith structural parameters alone. A combination of param-eters from structural tests with SWAP PSD/GHT led to agreater increase in sensitivity than combination with FDT

perimetry PSD, but also a greater decrease in specificity

than combination with FDT perimetry PSD. The particularstructural and functional parameters chosen for combinationwill depend on the clinical setting for testing and the needfor high sensitivity or high specificity. For glaucoma de-fined by VFs, the likelihood ratio analysis suggests that theaddition of FDT perimetry can be particularly useful toconfirm a normal structural test.

References

1. Kass MA, Heuer DK, Higginbotham EJ, et al. The OcularHypertension Treatment Study: a randomized trial determinesthat topical ocular hypotensive medication delays or preventsthe onset of primary open-angle glaucoma. Arch Ophthalmol2002;120:701–13, discussion 829–30.

2. European Glaucoma Prevention Study (EGPS) Group. Resultsof the European Glaucoma Prevention Study. Ophthalmology2005;112:366–75.

3. Soliman MA, de Jong LA, Ismaeil AA, et al. Standard ach-romatic perimetry, short wavelength automated perimetry, andfrequency doubling technology for detection of glaucomadamage. Ophthalmology 2002;109:444–54.

4. Sample PA, Bosworth CF, Blumenthal EZ, et al. Visual function-specific perimetry for indirect comparison of different gan-glion cell populations in glaucoma. Invest Ophthalmol Vis Sci2000;41:1783–90.

5. Sample PA. Short-wavelength automated perimetry: its role inthe clinic and for understanding ganglion cell function. ProgRetin Eye Res 2000;19:369–83.

6. Medeiros FA, Sample PA, Weinreb RN. Frequency doublingtechnology perimetry abnormalities as predictors of glauco-matous visual field loss. Am J Ophthalmol 2004;137:863–71.

7. Johnson CA, Adams AJ, Casson EJ, Brandt JD. Blue-on-yellow perimetry can predict the development of glaucoma-tous visual field loss. Arch Ophthalmol 1993;111:645–50.

8. Bayer AU, Maag KP, Erb C. Detection of optic neuropathy inglaucomatous eyes with normal standard visual fields using atest battery of short-wavelength automated perimetry and pat-tern electroretinography. Ophthalmology 2002;109:1350–61.

9. Bayer AU, Erb C. Short wavelength automated perimetry,frequency doubling technology perimetry, and pattern electro-retinography for prediction of progressive glaucomatous stan-dard visual field defects. Ophthalmology 2002;109:1009–17.

10. Varma R, Steinmann WC, Scott IU. Expert agreement inevaluating the optic disc for glaucoma. Ophthalmology 1992;99:215–21.

11. Tielsch JM, Katz J, Quigley HA, et al. Intraobserver andinterobserver agreement in measurement of optic disc charac-teristics. Ophthalmology 1988;95:350–6.

12. Greaney MJ, Hoffman DC, Garway-Heath DF, et al. Compar-ison of optic nerve imaging methods to distinguish normaleyes from those with glaucoma. Invest Ophthalmol Vis Sci2002;43:140–5.

13. Zangwill LM, Bowd C, Berry CC, et al. Discriminating be-tween normal and glaucomatous eyes using the HeidelbergRetina Tomograph, GDx Nerve Fiber Analyzer, and OpticalCoherence Tomograph. Arch Ophthalmol 2001;119:985–93.

14. Bowd C, Zangwill LM, Berry CC, et al. Detecting earlyglaucoma by assessment of retinal nerve fiber layer thicknessand visual function. Invest Ophthalmol Vis Sci 2001;42:1993–2003.

15. Weinreb RN, Shakiba S, Zangwill L. Scanning laser polarim-etry to measure the nerve fiber layer of normal and glauco-

matous eyes. Am J Ophthalmol 1995;119:627–36.

1601

Ophthalmology Volume 113, Number 9, September 2006

16. Zhou Q, Weinreb RN. Individualized compensation of ante-rior segment birefringence during scanning laser polarimetry.Invest Ophthalmol Vis Sci 2002;43:2221–8.

17. Medeiros FA, Zangwill LM, Bowd C, Weinreb RN. Compar-ison of the GDx VCC scanning laser polarimeter, HRT IIconfocal scanning laser ophthalmoscope, and Stratus OCToptical coherence tomograph for the detection of glaucoma.Arch Ophthalmol 2004;122:827–37.

18. Huang D, Swanson EA, Lin CP, et al. Optical coherencetomography. Science 1991;254:1178–81.

19. Weinreb RN. Laser scanning tomography to diagnose andmonitor glaucoma. Curr Opin Ophthalmol 1993;4:3–6.

20. Wollstein G, Garway-Heath DF, Hitchings RA. Identificationof early glaucoma cases with the scanning laser ophthalmo-scope. Ophthalmology 1998;105:1557–63.

21. Sample PA. What does functional testing tell us about opticnerve damage? Surv Ophthalmol 2001;45(suppl):S319–24.

22. Sample PA, Johnson CA, Haegerstrom-Portnoy G, Adams AJ.Optimum parameters for short-wavelength automated perim-etry. J Glaucoma 1996;5:375–83.

23. Johnson CA, Sample PA, Cioffi GA, et al. Structure andfunction evaluation (SAFE): I. Criteria for glaucomatousvisual field loss using standard automated perimetry (SAP)and short wavelength automated perimetry (SWAP). Am J

Ophthalmol 2002;134:177–85.

1602

24. Anderson AJ, Johnson CA. Frequency-doubling technologyperimetry. Ophthalmol Clin North Am 2003;16:213–25.

25. Newcombe RG. Two-sided confidence intervals for the singleproportion: comparison of seven methods. Stat Med 1998;17:857–72.

26. Jaeschke R, Guyatt GH, Sackett DL, Evidence-Based Medi-cine Working Group. Users’ guides to the medical literature.III. How to use an article about a diagnostic test. B. What arethe results and will they help me in caring for my patients?JAMA 1994;271:703–7.

27. Caprioli J. Discrimination between normal and glaucomatouseyes. Invest Ophthalmol Vis Sci 1992;33:153–9.

28. Artes PH, Chauhan BC. Longitudinal changes in the visualfield and optic disc in glaucoma. Prog Retin Eye Res 2005;24:333–54.

29. Iester M, Traverso CE, De Feo F, et al. Correlation betweenfrequency doubling technology and Heidelberg retina tomo-graph. J Glaucoma 2005;14:368–74.

30. Zhou XH, Obuchowski NA, McClish DK. Statistical Methodsin Diagnostic Medicine. New York: Wiley-Interscience; 2002:110–37.

31. Sample PA, Martinez GA, Weinreb RN. Short-wavelengthautomated perimetry without lens density testing. Am J Oph-

thalmol 1994;118:632–41.