Screening Tests for Detecting Open-Angle Glaucoma: Systematic Review and Meta-analysis

Abstract

purpose. To assess the comparative accuracy of potential screening tests for open angle glaucoma (OAG).

methods. Medline, Embase, Biosis (to November 2005), Science Citation Index (to December 2005), and The Cochrane Library (Issue 4, 2005) were searched. Studies assessing candidate screening tests for detecting OAG in persons older than 40 years that reported true and false positives and negatives were included. Meta-analysis was undertaken using the hierarchical summary receiver operating characteristic model.

results. Forty studies enrolling over 48,000 people reported nine tests. Most tests were reported by only a few studies. Frequency-doubling technology (FDT; C-20-1) was significantly more sensitive than ophthalmoscopy (30, 95% credible interval [CrI] 0–62) and Goldmann applanation tonometry (GAT; 45, 95% CrI 17–68), whereas threshold standard automated perimetry (SAP) and Heidelberg Retinal Tomograph (HRT II) were both more sensitive than GAT (41, 95% CrI 14–64 and 39, 95% CrI 3–64, respectively). GAT was more specific than both FDT C-20-5 (19, 95% CrI 0-53) and threshold SAP (14, 95% CrI 1-37). Judging performance by diagnostic odds ratio, FDT, oculokinetic perimetry, and HRT II are promising tests. Ophthalmoscopy, SAP, retinal photography, and GAT had relatively poor performance as single tests. These findings are based on heterogeneous data of limited quality and as such are associated with considerable uncertainty.

conclusions. No test or group of tests was clearly superior for glaucoma screening. Further research is needed to evaluate the comparative accuracy of the most promising tests.

Glaucoma describes a group of eye diseases in which there is progressive damage to the optic nerve, leading to impaired vision and, in some cases, blindness if untreated. Glaucoma is the leading cause of irreversible blindness worldwide, ^{1 2} with open-angle glaucoma (OAG) the most common form. ¹ Late detection is a major risk factor for blindness. ^{1 3 4 5} It is estimated from population surveys that in developed countries, more than 50% of prevalent OAG is undetected, ⁶ and this estimate is likely to be higher in developing countries. Recent evidence suggests that treatment is effective at delaying progression ^{7 8} ; thus, population-based screening of OAG is under consideration. ^{6 9 10 11} For screening to be considered, several criteria have to be met regarding the condition, the test, and the screening program. ⁹  

Tests for glaucoma involve an assessment of structural changes at the optic nerve head, functional visual loss by visual field testing, and the level of the intraocular pressure (IOP). There are many potential tests or combinations of tests for detecting glaucoma; however, to date, no single test or combination of tests has been identified as optimal in screening for glaucoma. 

The purpose of this study was to assess the comparative accuracy of candidate screening tests. 

Methods

Search Strategy

Highly sensitive electronic searches, using both controlled vocabulary and free text terms, were undertaken. We searched the following electronic databases: Medline (1966 to week 3, November 2005), Medline In Process (February 23 and December 6, 2005), Embase (1980 to 2005 week 49), Science Citation Index (1981 to December 3, 2005), Biosis (1985 to November 30, 2005), and Cochrane Central Register of Controlled Trials (CENTRAL; The Cochrane Library, Issue 4, 2005). In addition, full-text electronic searches of the American Journal of Ophthalmology (1998 to November 2005), Ophthalmology (1998 to November 2005), British Journal of Ophthalmology (1998 to November 2005), Investigative Ophthalmology and Visual Science (1998 to November 2005), and the Journal of Glaucoma (2001 to November 2005) were undertaken. Searches were restricted to English language publications. The reference lists of included studies were scanned to identify additional potentially relevant reports. Full details of the sources searched and search strategies used are available elsewhere ⁶ or can be obtained by contacting the authors. 

Inclusion and Exclusion Criteria

We included studies that assessed the accuracy of tests for detecting OAG in people older than 40 years who were likely to be representative of a screening situation (i.e., no selection and no previous tests had been performed) or of a group of patients with suspected glaucoma (i.e., patients identified from prior testing as possibly having glaucoma or as having, e.g., high IOP, or another risk factor for glaucoma but with an unconfirmed diagnosis). Both randomized (where participants were randomized to one or more tests) and observational (both cohort and case–control) studies were included. The reference standard was either confirmed OAG on follow-up or ophthalmologist-diagnosed OAG, as reported by the study. This latter reference standard required a clinical judgment by an ophthalmologist, including an evaluation of the optic nerve and a measure of visual function. In addition, the study had either to report or to allow the calculation of true and false positives and negatives. 

Non-English-language reports were excluded, as were conference abstracts. Case reports and studies investigating technical aspects of a test were excluded. Case–control studies in which the control group consisted of people with no ocular disease or specifically excluded people with other ocular disease, so that the spectrum of disease and nondisease was unlike that to be encountered in a screening situation, were also excluded. The spectrum of disease expected would be similar to the spectrum of the disease of the general population (e.g., more patients with mild glaucoma, fewer patients with severe glaucoma). 

The candidate tests fell within three broad categories: (1) structure (ophthalmoscopy, optic disc photography, retinal nerve fiber layer [RNFL] photography, Heidelberg retinal tomography [HRT] version II [Heidelberg Engineering, Heidelberg, Germany], GDx VCC retinal nerve fiber layer [RNFL] analyzer [Carl Zeiss Meditec, Oberkochen, Germany], optical coherence tomography [OCT], and retinal thickness analyzer [RTA]); (2) function (oculokinetic perimetry [OKP], white-on-white standard automated perimetry [SAP] including suprathreshold and threshold, short wave-length automated perimetry [SWAP], frequency-doubling technology [FDT], and motion-detection perimetry [MDP]); and (3) IOP (Goldmann applanation tonometry [GAT]; noncontact tonometry [NCT]; TonoPen; Reichert, Vienna, Austria). 

Data Abstraction and Quality Assessment

Two reviewers undertook single-data extraction of the included studies. In the event of uncertainty, the other reviewer provided advice and validated the data extraction. 

Two reviewers independently assessed the quality of the included studies using a version of QUADAS adapted for assessing reports of the accuracy of screening tests for OAG. QUADAS is a quality-assessment tool for use in systematic reviews of diagnostic studies. ¹² Disagreements were resolved by consensus or arbitration by a third reviewer. A higher quality study was considered one for which “yes” was checked in response to questions 1 (patient spectrum representative), 3 and 4 (partial and differential verification bias avoided), and 6 and 7 (test review bias and diagnostic review bias avoided) on the adapted QUADAS checklist. 

Statistical Methods

After data extraction a common (most frequently reported) cutoff for each test was selected after discussion by two ophthalmologists (JMB, MARS). Summary receiver operating characteristic (SROC) curves were produced for each test where two or more studies reported estimates of sensitivity and specificity at the common cutoff. Meta-analysis models were fitted using the hierarchical summary receiver operating characteristic (HSROC) model ¹³ in WinBUGS 1.4. ¹⁴ Normally distributed random effects were assumed with noninformative uniform priors. No adjustment was made for the correlation between results from paired studies, as the level of information required is rarely reported. Summary sensitivity, specificity, and diagnostic odds ratios (DORs) at the operating point were reported for each model as the median and 95% credible interval (CrI). A DOR is a single indicator of test performance and is the ratio of the odds of testing positive in those with the disease relative to the odds of testing positive in those without the disease. ¹⁵ It can be calculated from the sensitivity and specificity: DOR = [sensitivity/(1 − sensitivity)]/[(1 − specificity)/specificity]. 

Credible intervals are the Bayesian equivalent of confidence intervals. A simplified model, which assumed a symmetrical ROC shape, was used where limited data caused convergence problems under the full model. Sensitivity analysis was undertaken by examining separately the results of the higher quality studies, using HSROC analysis where more than one higher quality study reported the same test. 

Comparisons between tests were made in two ways: First, studies in which participants were directly compared who either received all tests or were randomized to different tests were identified, and the direct comparisons inspected. Second, an indirect comparison between tests was made, for all tests reported by two or more studies were modeled together in a single HSROC model to formally compare test performance. Pair-wise differences in sensitivity and specificity between tests were assessed from the median difference and corresponding 95% CrI. 

Results

Trial Flow

Figure 1shows the flow of studies through the review. Out of a total of 5918 titles/abstracts screened, 877 potentially relevant full text articles were obtained, with 40 studies, published in 46 reports, meeting the inclusion criteria. 

Study Characteristics and Methodological Quality

Twenty studies were population-based and representative of a screening setting ^{16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39} whereas 20 studies were considered representative of patients with suspected glaucoma referred from primary care, of which 8 were cohort studies ^{40 41 42 43 44 45 46 47} and 12 were case–control studies. ^{48 49 50 51 52 53 54 55 56 57 58 59 60 61} Seven studies ^{18 34 40 43 44 48 58} used the first and best reference standard of OAG confirmed on longitudinal follow-up, whereas the remainder used ophthalmologist-diagnosed OAG. The characteristics of the included studies are shown in Table 1. 

The 40 studies enrolled more than 48,000 people, with more than 39,000 included in the analysis. The studies took place from 1963 to 2004. In 26 studies reporting the sex of the participants, 51% were women. The median (range) age of participants across studies was 60.5 years (13–97 years). The reports included several major population-based prevalence surveys, such as the Baltimore Eye Survey, ^{25 31} the Blue Mountains Eye Study, ²³ the Crete, Greece, Glaucoma Study, ²⁷ the Dalby Population Survey, ¹⁷ the Egna-Neumarkt Study, ¹⁸ the Framingham Eye Study, ⁴³ the Glaucoma Screening Study (GLASS), ^{24 26} the Groningen Longitudinal Glaucoma Study, ^{53 54 59} the Rhondda Valley Study, ²² the Rotterdam Study, ³⁸ the Segovia Study, ¹⁶ and the Visual Impairment Project. ³⁷  

The included studies reported the following tests: ophthalmoscopy (seven studies); optic disc photography (six studies); RNFL photography (four studies); HRT II (three studies); OKP (four studies); SAP (14 studies); FDT (eight studies); GAT (nine studies); and NCT (one study). No reports of GDx VCC, OCT, RTA, SWAP, MDP, or TonoPen were identified that met our inclusion criteria. 

Figure 2summarizes the results of the quality assessment for the 40 included studies. Study quality was variable; only eight studies ^{20 21 30 34 38 39 45 46} met the specified criteria for higher quality studies. 

Quantitative Data Synthesis

Individual Tests.

The sensitivity and specificity of the individual tests included in the HSROC meta-analysis models are shown in Figure 3and Appendix 1, which also includes DORs. 

DORs ranged from 10 for FDT C-20-5 to 181 for FDT C-20-1, with higher DORs indicating a better ability to differentiate between diseased and nondiseased. There was statistical heterogeneity (variability in outcome beyond what would be expected by chance) across studies for most tests. Ophthalmoscopy, retinal photography (optic disc photography and RNFL photography), GAT, SAP (threshold and suprathreshold), and FDT C-20-5 were all relatively poorly performing tests based on lower DORs (range, 10–30). 

Eight studies met the criteria for higher quality studies, including six population-based studies and two cohort studies, and test accuracy data are detailed in Table 2 . For both SAP threshold and FDT C-20-5, higher quality studies reported lower sensitivity and specificity when compared with all studies, whereas two FDT C-20-5 studies not meeting the criteria for higher quality reported very high sensitivity (98% and 100%, respectively). For optic disc photography, compared with all studies, the higher quality studies reported similar sensitivity (74% versus 73%) but lower specificity (82% versus 89%). For HRT II, compared with all studies, the higher quality studies reported higher sensitivity (93% versus 86%), but slightly lower specificity (85% versus 89%). 

Seven studies reported test accuracy in different stages of glaucoma. ^{24 51 52 54 55 60 61} Of those reporting the same tests for different stages of glaucoma, Ieong et al. ⁵⁵ reported a sensitivity of 72% for SAP (suprathreshold) for early stage glaucoma, whereas Enger and Sommer ⁵¹ and Katz et al. ²⁴ both reported a sensitivity of 97% for SAP (threshold) for early/moderate stage glaucoma. 

Studies Directly Comparing Tests.

Six studies directly compared two or more of the following tests for detection of OAG: optic disc photography, HRT II, SAP, FDT, and GAT. ^{23 30 34 36 46 55} Table 3shows the common cutoff selected, sensitivity, specificity, DORs, and relative DORs for these studies. In each study SAP (either suprathreshold or threshold) was included as a comparator. DORs for the tests ranged from 4 for SAP threshold ⁴⁶ to 75 for HRT II ³⁰ (Table 3) . In terms of relative DORs, compared with SAP, GAT performed better in one study ³⁶ but worse in another ²³ (statistically significant), HRT II performed better than SAP in one study ³⁰ (statistically significant) but worse in another, ⁵⁵ FDT C-20-5 ³⁰ and FDT C-20 matrix ⁴⁶ performed better than SAP, whereas optic disc photography ³⁴ showed a broadly similar performance. 

Indirect Comparisons in a Single HSROC Model.

The results of the indirect comparisons in a single HSROC model are shown in Table 4 . From the large number of comparisons undertaken, six showed a statistically significant difference between tests (four in terms of sensitivity and two in terms of specificity). There was evidence that, at the common cutoff, FDT C-20-1 was significantly more sensitive than both ophthalmoscopy (30, 95% CrI 0–62) and GAT (45, 95% CrI 17–68), and that both SAP threshold (41, 95% CrI 14–64) and HRT II (39, 95% CrI 3–64) were significantly more sensitive than GAT. There was also evidence that GAT was significantly more specific than both FDT C-20-5 (19, 95% CrI 0–53) and SAP threshold (14, 95% CrI 1–37). Other differences in accuracy between tests may well exist that could not be detected due to the high level of uncertainty. The wide credible intervals reflected the small number of studies reporting each test and the generally high level of heterogeneity. Because of the imprecision in the estimates, no test (or group of tests) was clearly more accurate, based on a 5% significance level. Further analysis, at 10% and 20% levels of significance, identified additional statistically significant comparisons (Table 4) . For example, in terms of sensitivity, at a 10% significance level, FDT C-20-1 was better than SAP suprathreshold and at a 20% level better than optic disc photography, RNFL photography, and FDT C-20-5. OKP was better than GAT at a 10% level and HRT II better than ophthalmoscopy at a 20% level. In terms of specificity, at a 20% level FDT C-20-1 was better than SAP threshold and FDT C-20-5. 

Discussion

To our knowledge, this is the first systematic review of screening and diagnostic tests in glaucoma and includes 40 studies enrolling over 48,000 people and reporting nine tests. Most tests were reported by only a few, mostly heterogeneous, studies. The included studies reported tests of structure (ophthalmoscopy; optic disc photography, RNFL photography, and HRT II), visual function (FDT, OKP, and SAP) and IOP (GAT and NCT). Other tests were considered, including those of structure (GDx VCC, OCT, and RTA), visual function (SWAP and MDP), or using TonoPen to measure IOP. However, no studies using these tests met our inclusion in reporting of test accuracy outcomes. 

A systematic review of test accuracy is unlikely to identify the best test but can identify more promising ones. It is difficult to rank tests on paired values of sensitivity and specificity, as a highly specific test may be associated with a low sensitivity and vice versa. The choice of test depends on the importance of the tradeoff between missed cases, and false positives. OAG affects an estimated 2% of the adult population. A test of low specificity would be likely to overburden a health service with people who do not have glaucoma and cause unnecessary anxiety for a many individuals, and equally a test of low sensitivity would miss treatable disease which might be unacceptable to society. The DOR, a single measure of test accuracy, is a useful measure for comparing accuracy of several tests in a meta-analysis. ¹⁵ Based on a DOR ≥50, FDT C-20-1, OKP (both tests of visual function), and HRT II (a test of glaucomatous optic neuropathy) merit further evaluation of their performance as screening tests for glaucoma. It should be noted that these findings are relevant to the common cutoff point selected for each test; selection was based on the most frequently reported cutoff and when several cut-offs were reported, the cutoff most likely to represent early glaucoma. Furthermore, these findings are based on heterogeneous data of limited quality and as such are associated with considerable uncertainty. 

Methods of meta-analysis of diagnostic accuracy which combine studies in which both sensitivity and specificity vary have been available since 1990 and are continuing to evolve. ^{13 62 63 64 65 66} These methods are based on the idea of a tradeoff relationship between sensitivity and specificity, as occurs when studies vary in threshold, and seek to estimate the shape and position of the underlying receiver operating curve. From the estimate of this curve, it is possible to identify operating points. The approach adopted in our review identifies the average operating point for each test and makes comparisons between them, based on those studies reporting each test that share a common cutoff point. The Cochrane Collaboration are commencing publication of systematic reviews of diagnostic test accuracy, and the analytical approach we have followed is the one that they are recommending. ⁶⁷ Estimation of a summary point specific to a test being used at a common threshold obtains the best estimate of test accuracy in parameters that are clinically meaningful. The tradeoff between sensitivity and specificity is important in judging the performance of a test and is best depicted by an ROC curve across different cutoff points. However, the included studies did not usually provide information across the whole range of cutoff points to allow such analyses to be undertaken. 

We used a Bayesian Hierarchical SROC model, as standard methods for meta-analysis do not address the threshold effect and are therefore not appropriate. ⁶⁸ Several different levels of analyses were undertaken, including an analysis in which all tests were modeled simultaneously by using this Bayesian approach. This method allowed indirect comparison of sensitivities and specificities to be made, in addition to allowing DORs to be calculated, which is one of the advantages of the Bayesian method adopted. To produce results that are comparable to those from standard methods of meta-analysis we did not use informative priors. 

In addition to providing sensitivity and specificity estimates we also reported the DOR results. Some meta-analysis models can only provide the DOR estimate, and therefore we included this measure for comparability. One strength of the DOR is that it is a mathematically robust measure (like the standard odds ratio) and represents diagnostic accuracy as a single value. However, a disadvantage is that different combinations of sensitivity and specificity can lead to the same DOR. 

To be included, studies had to meet specific inclusion criteria. The validity of indirect comparisons does depend on assumptions regarding the characteristics of the included studies; however, the indirect method is formally performing the comparison that users of the report are likely to make when assessing the pooled results for the individual tests. As such, this method of indirect comparisons serves an important purpose and reaffirms the lack of certainty about which test is indeed the best. 

There are many potential sources of bias in primary diagnostic accuracy studies. Despite the huge volume of literature, no good-quality studies were found that provided a positive response to all questions on the modified QUADAS checklist. Based on limited evidence, of the tests reported by higher quality studies, including the three tests that were considered to merit further evaluation, estimates of sensitivity and specificity varied according to study quality. 

There is no universally agreed on optimal reference standard for the diagnosis of OAG, although progressive structural optic neuropathy has been proposed as the best possible reference standard. ^{69 70} In this review either of two reference standards was considered. There was no obvious pattern in the sensitivity and specificity of the tests in the seven studies ^{18 34 40 43 44 48 58} in which the first and best reference standard of OAG was confirmed on longitudinal follow-up compared with the remainder in which ophthalmologist-diagnosed OAG was used. Although the latter is suboptimal compared with the former, it is the accepted reference standard in clinical practice. However, establishing a reference standard in glaucoma is problematic, as in some people, optic disc damage precedes visual field loss, whereas in others the reverse is the case. 

The accuracy of a test may vary according to the population in which it is performed. Samples with higher prevalence often arise through preferential inclusion of suspect cases, which shifts the disease severity to include more moderate and severe disease, and since it is easier to differentiate between severely diseased and nondiseased persons, a test would be expected to report improved (apparent) sensitivity and specificity. Therefore, studies with a significantly higher prevalence than expected in a screening population should be interpreted with this limitation in mind. ^{21 28 29 30 36 39} These studies, including two that met the criteria for higher quality studies, ^{30 39} tended to recruit their participants through media advertising rather than contacting individuals in a predefined population and can be considered to be more representative of screening in higher risk populations. 

Twenty of the 40 studies included were hospital based, which by nature, is an enriched population likely to include a disproportionate number of participants with high IOP and with previous experience taking tests, potentially leading to overly optimistic performance estimates. ^{71 72 73 74} Most of the case–control studies identified applied stringent criteria for inclusion, such as visual acuity of 6/9 or no other ocular disease and as such were highly prone to bias. ⁷⁵ To minimize this spectrum bias, case–control studies (n = 57) in which the participants were considered unrepresentative of a case mix found in a general population where OAG screening would be performed were excluded from the review. 

In the meta-analysis models for the individual tests, statistical heterogeneity was evident across most studies. Empirically, there was no obvious single cause of the heterogeneity, but potential contributory factors include differences in populations, study design, setting, prevalence, and severity of glaucoma within studies. Other factors include differences in reference standard, and in tests included within the same category (e.g., different types of perimetry and ophthalmoscopy have a large number of variants, potentially leading to heterogeneity in discriminatory power across studies reporting those tests), and the extent to which studies were affected by other potential biases (e.g., partial and differential verification bias, incorporation bias, test, and diagnostic review bias). 

Limitations

Relatively few studies were identified for each test, and it was not possible to perform sensitivity analysis based on study design. The common cutoff chosen for each test was the one most frequently reported across the included studies for that test, although this may not be the most appropriate. Most of the studies were poorly reported, an issue that has been highlighted in recent literature. ^{76 77 78 79} Only 6 of the 40 studies directly compared two or more tests. It was not possible to provide summary results of studies that directly compared tests because of small numbers. Studies not providing sufficient information to allow the calculation of 2 × 2 tables were excluded, although they may have contributed information on sensitivity and specificity. 

Systematic reviews provide a robust and rigorous evaluation of the available evidence, but according to their nature, as new studies are published, the review requires updating. Since the completion of our meta-analysis, further studies have been published on the performance of the tests included in this review. These include population-based studies in the United States, United Kingdom, Hungary, Japan, and China. These studies provide additional information on the performance of FDT perimetry alone, ^{80 81 82} in combination with GDx VCC, ⁸³ and combined with an IOP measurement ⁸⁴ and data on the performance of HRT II in an elderly population in the United Kingdom ⁸⁵ and in a community screening program in Japan, ⁸⁶ comparing HRT II with nonmydriatic fundus photography. Although systematic reviews rapidly become out of date, which is a limitation, one strength of a systematic review is that the methods are transparent and reproducible such that the review can be updated as further data become available in the future. Priorities for future research and optimal study designs can also be identified. 

Implications for Practice and Recommendations for Research

Ideally, a screening test for OAG should be safe, easy to administer and interpret, portable, quick, and acceptable to the people who are to be tested and should be sufficiently valid to distinguish between those who do and do not have OAG. Many potential screening tests for glaucoma are available. Of the many candidate tests, no one test or group of tests was clearly more accurate. Based on limited data, relatively poorly performing tests, ophthalmoscopy, standard automated perimetry, retinal photography, and Goldmann applanation tonometry were identified. 

Frequency doubling technology, (C-20-1), HRT II, and OKP were identified as having better diagnostic performance than other candidate tests, although these findings were based on poor-quality evidence. Further investigations should evaluate the most promising tests in directly comparative studies in a relevant population. 

Appendix 1

 Supported by the UK National Institute for Health Research Health Technology Assessment programme Project no. 04/08/02 and by core funding of The Health Services Research Unit and the Health Economics Research Unit by the Chief Scientist Office of the Scottish Government Health Directorates.

 The study was developed from a Health Technology Assessment on the clinical and cost-effectiveness of screening for OAG. The views expressed in this report are those of the authors and not necessarily those of the funding agencies.

 Submitted for publication November 22, 2007; revised February 26, April 18, and June 17, 2008; accepted October 14, 2008.

 Disclosure: G. Mowatt, None; J . M. Burr, None; J . A. Cook, None; M.A .R. Siddiqui, None; C. Ramsa y, None; C. Frase r, None; A. Azuara-Blanco, None; J . J. Deeks, None

 The publication costs of this article were defrayed in part by page charge payment. This article must therefore be marked “advertisement” in accordance with 18 U.S.C. §1734 solely to indicate this fact.

 

Figure 1.

 

Flow of studies through the review process.

Table 1.

 

Characteristics of the Included Studies

Study	Index Test(s)	Test(s) Performers and Interpreters	Reference Standard	Enrolled (n)	Analyzed (n)	Mean Age y (range)	Sex	Country	Period
Population-Based Studies (Cross-sectional)
Anton 2004 16	GAT	Ophthalmologists	Ophthalmic examination	569	510	(40–79)	M: 232; F: 278	Spain (Segovia Study)	N/S
Bengtsson 1980 17	GAT	Ophthalmologists	Ophthalmic examination	1938	1511	(55–69)	N/S	Sweden (Dalby Population Survey)	1977–1978
Bonomi 2001 18	GAT	Ophthalmologists	Follow-up confirmation	5816	4297 eyes of 4297 people	(40–80+)	M: 1882; F: 2415	Italy (Egna-Neumarkt Study)	N/S
Detry-Morel 2004 20	FDT C-20-5	Residents in training, paramedical staff	Ophthalmic examination	1802	3211 eyes of 1620 people	63 (22–97)	M: 680; F: 940	Belgium	Oct 1999
Harasymowycz 2005 21	HRT II	Ophthalmic photographer	Ophthalmic examination	303	264 right eyes, 265 left eyes of 271 people	62.2 (SD 11.6)	M: 90; F: 179	Canada	Aug 2003–Feb 2004
Hollows 1966 22	GAT	Ophthalmologists	Ophthalmic examination	4608	4231	55 (40–74)	Approx: M: 3639; F: 592	UK (Rhondda Valley Study)	Summer 1963
Ivers 2001 23	SAP suprathreshold; GAT	N/S	Ophthalmic examination	4433	3654 (both tests)	(49–97)	M: 1582; F: 2072	Australia (Blue Mountains Eye Study)	1992–1994
Katz 1991 24	SAP threshold	N/S	Ophthalmic examination	355	355 eyes of 355 people	Cases: 61; controls: 53	N/S	USA (Glaucoma Screening Study)	1981–1992
Katz 1993 25	SAP suprathreshold	N/S	Ophthalmic examination	5308	4733	(40–80+)	M: 2109; F: 3199	USA (Baltimore Eye Survey)	Jan 1985–Nov 1988
Kozobolis 2000 27	GAT	Uncertain	Ophthalmic examination	1300	1107	(40–80+)	M: 463; F: 644	Greece (Crete, Greece Glaucoma Study)	Feb 1993–June 1998
Mansberger 2005 28	FDT C-20-5	N/S	Ophthalmic examination	296	251 eyes of 251 people	45 (30–65)	M: 117; F: 174	India	N/S
Mundorf 1989 29	SAP suprathreshold	N/S	Ophthalmic examination	145	145	71	M: 40; F: 105	USA	N/S
Robin 2005 30	Ophthalmoscopy; HRT II; SAP threshold; FDT C-20-5	Appropriately trained staff	Ophthalmic examination	704	261 eyes of 261 people (all tests)	65	M: 281; F: 378	Australia	Nov 2001
Weih 2001 37	Ophthalmoscopy	N/S	Consensus by panel of ophthalmologists, based on results of ophthalmic examination	4744	4636	59 (SD 12)	M: 2230; F: 2514	Australia (Visual Impairment Project)	1992–1996
Wolfs 1999 38	Optic disc photography	Technicians	Ophthalmic examination	6777	5143 eyes of 5143 people	(55 and over)	N/S	Netherlands (Rotterdam Eye Study)	N/S
Yamada 1999 39	OKP; FDT C-20-1	Technicians	Decision of glaucoma specialists, based on ophthalmic records	259	175 eyes of 175 people (OKP); 240 eyes of 240 people (FDT)	FDT: 59.6 (SD 14.7); OKP: 58.8 (SD 15.6)	M: 108; F: 135	USA	N/S
Population-Based Studies (Cohort)
Christoffersen 1995 19 (Patient source; general practice)	OKP	GPs, medical secretaries	Ophthalmic examination	195	187	57 (40–84)	M: 51; F: 136	Norway	N/S
Vernon 1990 32 (Patient source: general practice)	Ophthalmoscopy; SAP suprathreshold; NCT	Ophthalmoscopy: experienced ophthalmol-ogists; NCT/SAP: non-ophthalmically trained staff	Ophthalmic examination	988	854 (ophth); 855 (SAP); 874 (NCT)	65	M: 374; F: 500	UK	N/S
Wang 1998 36 (Patient source: general practice)	Ophthalmoscopy; SAP suprathreshold; GAT (RNFL photography)	N/S	Ophthalmic examination	530 from primary care clinic)	400 (ophth); 214 (SAP); 357 (GAT) [136 (RNFL photo)]	(40–65+)	M: 111; F: 294	USA	Jul 1991–Feb 1992
Population-Based Studies (Case–Control)
Vitale 2000 34 (Patient source: Cases and controls: sample of patients with and without glaucoma from the Baltimore Eye Study Follow-up Study)	Optic disc photography; SAP suprathreshold	Experienced technicians	Follow-up confirmation	249	182 (disc photo); 228 (SAP);	68	M: 100; F: 149	USA (Baltimore Eye Study Follow-up)	1994
Already-Suspect Population (Cohort Studies)
Ekstrom 1993 40 (Patient source: people previously examined in a population-based glaucoma survey)	GAT	N/S	Follow-up confirmation	760	413	(65–74)	M: 364; F: 396	Sweden (Tierp Glaucoma Survey)	Mar 1984–Mar 1986
Hammond 1979 41 (Patient source: eye clinic)	Ophthalmoscopy	Nurses skilled in use of the ophthalmoscope	Ophthalmic examination	219	188	(21 and over)	N/S	USA	N/S
Khong 2001 42 (Patient source: eye clinics)	FDT C-20-5	N/S	Ophthalmic examination	228	113	68.5 (23–91)	M: 104; F: 119	Australia	Dec 1999–Jan 2000
Leibowitz 1980 43 (Patient source: Framingham Eye Study)	GAT	Generally performed by 2nd or 3rd year residents in ophthalmology	Follow-up confirmation	2631	574	(<65–75+)	M: 272; F: 302	USA (Framingham Eye Study)	Feb 1973–Feb 1975
Marraffa 1989 44 (Patient source: eye clinic)	SAP suprathreshold	Ophthalmologists	Follow-up confirmation	104	182 eyes of 104 people	54.3 (18–76)	M: 45; F: 59	Italy	N/S
Schultz 1995 45 (Patient source: clinical practices of glaucoma specialist, cataract surgeon, and general ophthalmol-ogist)	Optic disc photography	Carried out: N/S Interpreted: 3rd-year ophthalmology residents	Ophthalmic examination	258	365 eyes of? people	(<40–>70)	M: 112; F: 144; Unknown: 2	USA	N/S
Spry 2005 46 (Patient source: hospital eye service)	SAP threshold; FDT C-20 matrix	SAP: clinic staff trained in visual field testing; FDT: N/S	Ophthalmic examination	48	48 (both tests)	67.3 (SD 13.5)	M: 24; F: 24	UK	Oct 2003–Jan 2004
Theodossiades 2001 47 (Patient source: glaucoma clinics)	Ophthalmoscopy	Optometrists	Ophthalmic examination	50	50 eyes of 50 people	N/S	N/S	UK	N/S
Already-Suspect Population (Case–Control Studies)
Airaksinen 1984 48 (Patient source: not stated)	RNFL photography	N/S	Follow-up confirmation	142	132 eyes of 132 people	Glaucoma: 62 (SD 20.5) Normal: 54 (SD 16.9); OHT: 57 (SD 12.7)	N/S	Canada + Finland	N/S
Anton 1997 49 (Patient source: cases and controls: glaucoma unit)	SAP threshold	Uncertain	Ophthalmic examination	180	180 eyes of 180 people	Glaucoma: 61 (SD 8); Normal: 59 (SD 9)	N/S	Spain	N/S
Damato 1989 50 (Patient source: Cases: not stated Controls: dermatology ward, hospital staff, relatives/friends of patients, patients with unilateral non-glaucomatous disease affecting the fellow eye)	OKP	Staff experienced in perimetry	Ophthalmic examination	102	102 eyes of 102 people	Glaucoma: 57.3; Normal: 54.4	N/S	UK	N/S
Enger 1987 51 (Patient source: Cases and controls: nerve fiber layer study)	SAP threshold	N/S	Ophthalmic examination	112	170 eyes of 112 people	Glaucoma: 61 (28–80); Normal: 51 (26–75)	N/S	USA	N/S
Harper 1994 52 (Patient source: not stated)	OKP; SAP suprathreshold	Uncertain	Ophthalmic examination	212	193 (OKP); 212 (SAP)	Glaucoma: 67.8 (43–85); Normal: 61.5 (41–85)	N/S	UK	N/S
Heeg 2005 53 (Patient source: Cases: glaucoma outpatient department Controls: old people’s homes, blood bank, other public places)	FDT C-20-1; FDT C-20 full threshold	N/S	Ophthalmic examination	1112	208 (FDT C-20-1); 1112 (FDT C-20 full threshold)	Glaucoma: 65 (13–91); Normal: 63 (33–94)	Eligible: Glaucoma: M: 509; F: 542 Normal: M: 118; F: 119	Netherlands (Groningen Longitudinal Glaucoma Study)	Jul 2000–Jun 2001
Ieong 2003 55 (Patient source: Cases: glaucoma subjects Controls: partners of cases, optometrist practice)	HRT II; SAP suprathreshold	Optometrists	Ophthalmic examination	66	66 eyes of 66 people (both tests)	Glaucoma: 69; Normal: 60	Glaucoma: M: 16; F: 13 Normal: M: 16; F: 21	UK	N/S
Johnson 1999 56 (Patient source: not stated)	FDT C-20-1	N/S	Ophthalmic examination	108	160 eyes of 108 people	Glaucoma: 64 (35–85); Normal: 46 (18–81)		USA	N/S
Quigley 1980 57 (Patient source: Cases and controls: Ophthalmol-ogical institute)	Optic disc photography; RNFL photography	Ophthalmologists	Ophthalmic examination	175	294 eyes of? people (both tests)	Readable photos: Glaucoma: 52.7 (SD 2.78); Glaucoma suspect: 45.2 (SD 1.56); Normal: 37.9 (SD 2.8) Unreadable photos: Glaucoma: 62.5 (SD 4.0); Glaucoma suspect: 59.6 (SD 6.3); Normal: 50 (SD 12.1)	M: 86; F: 89	USA	Jan 1978–Apr 1979
Sommer 1979 58 (Patient source: Cases and controls: glaucoma clinic)	Optic disc photography; RNFL photography	N/S	Follow-up confirmation	Unclear	223 eyes of ? people (both tests)	N/S	N/S	USA	N/S
Wollstein 2000 60 (Patient source: Cases: glaucoma clinic and ocular hypertension clinic Controls: spouses or friends of patients, responders-an advertisement)	Optic disc photography	Photos taken by trained technicians; assessed by glaucoma consultants, glaucoma fellow, clinical glaucoma technician	Ophthalmic examination	123	123 eyes of 123 people	Glaucoma: 65.1 (SD 10.06); Normal: 57.1 (SD 12.52)	N/S	UK	N/S
Wood 1987 61 (Patient source: not stated)	Ophthalmoscopy	Ophthalmologists; junior doctors	Ophthalmic examination	22	43 eyes of 22 people	(32–75)	N/S	UK	N/S

 N/S, not stated. Numbers analyzed are people unless otherwise stated.

Figure 2.

 

Results of the quality assessment of the 40 included studies.

Figure 3.

 

Summary of sensitivity and specificity of tests included in the HSROC meta-analysis models.

Table 2.

 

HSROC Analysis: All Studies Compared with Higher Quality Studies

	Optic Disc Photography			HRT II			FDT C-20-5			SAP Threshold
		Sensitivity % (95% CrI)	Specificity % (95% CrI)	Sensitivity % (95% CrI)	Specificity % (95% CrI)	Sensitivity % (95% CrI)	Specificity % (95% CrI)	Sensitivity % (95% CrI)	Specificity % (95% CrI)
All studies	73 (61–83)	89 (50–99)	86 (55–97)	89 (66–98)	78 (19–99)	75 (57–87)	88 (65–97)	80 (55–93)
Higher quality	74 (30–95)	82 (45–97)	93 (58–99)	85 (47–97)	72 (26–96)	60 (17–92)	73 (28–95)	64 (22–92)

 Optic disc photography (all studies n = 6, higher quality studies n = 3); HRT II (all studies n = 3, higher quality studies n = 2); FDT C-20-5 (all studies n = 5, higher quality studies n = 2); SAP threshold (all studies n = 5, higher quality studies n = 2).

Table 3.

 

Sensitivity, Specificity, DOR, and Relative DOR at the Common Cutoff for Studies Directly Comparing Tests

Study	Test	Common Cutoff	Sensitivity % (95% CrI)	Specificity % (95% CrI)	DOR (95% CrI)	RDOR (95% CrI)
Vitale 2000 34	SAP supra	Three adjacent points missed	50 (37–63)	83 (76–88)	5 (3–9)	1
	Optic disc photo	VCDR >0.6	77 (62–89)	59 (50–67)	5 (2–11)	0.99 (0.36–2.75)
Ieong 2003 55	SAP supra	Optometrist’s judgment	72 (53–87)	95 (82–99)	46 (9–237)	1
	HRT II	Global/one of six segments abnormal	69 (49–85)	95 (82–99)	39 (8–198)	0.85 (0.08–8.54)
Robin 2005 30	SAP threshold	AGIS score ≥3 (common cutoff)	63 (38–84)	74 (68–80)	5 (2–13)	1
	HRT II	≥1 Borderline or 1 severe abnormality	95 (74–100)	81 (75–85)	75 (10–574)	15.01 (1.57–143.82)
	FDT C-20-5	One abnormal point	84 (60–97)	55 (49–61)	7 (2–23)	1.31 (0.27–6.43)
Spry 2005 46	SAP threshold	GHT outside normal limit and/or P < 0.05 with the PSD global index in one/both eyes	80 (52–96)	52 (34–69)	4 (1–18)	1
	FDT C-20 matrix		100 (78–100)	27 (13–46)	12 (1–222)	2.83 (0.11–72.91)
Ivers 2001 23	SAP supra	Three or more points missing	89 (80–94)	73 (71–74)	20 (10–39)	1
	GAT	IOP >22 mm Hg	14 (7–23)	98 (97–98)	6 (3–12)	0.31 (0.12–0.78)
Wang 1998 36	SAP supra	Absolute or relative defects ≥ 17	70 (57–80)	67 (59–74)	5 (2–9)	1
	GAT	IOP > 21 mm Hg	28 (17–40)	96 (93–98)	9 (4–19)	1.89 (0.70–5.13)

 RDOR = Relative DOR = index test DOR/SAP DOR. RDOR calculated as all direct studies had SAP as one of the tests. RDOR > 1 indicates that the test performed better than SAP in the study and <1 indicates that the test performed worse than SAP. AGIS, Advanced Glaucoma Intervention Study; GHT, Glaucoma Hemifield Test; PSD, pattern standard deviation.

Figure 4.

Table 4.

 

Summary of Sensitivity, Specificity and DOR for Tests Included in the HSROC Meta-analysis Models

Test	Number of Studies	Common Cutoff	Sensitivity % (95% CrI)	Specificity % (95% CrI)	DOR (95% CrI)
Ophthalmoscopy	5	VCDR ≥0.7	60 (34–82)	94 (76–99)	26 (6–110)
Optic disc photography	6	VCDR ≥0.6	73 (61–83)	89 (50–99)	22 (3–148)
RNFL photography	4	Diffuse and/or localized defect	75 (46–92)	88 (53–98)	23 (4–124)
HRT II	3	≥1 Borderline or outside normal limits	86 (55–97)	89 (66–98)	51 (11–246)
FDT
C-20-1	3	1 Abnormal point	92 (65–99)	94 (73–99)	181 (25–2139)
C-20-5	5	1 Abnormal point	78 (19–99)	75 (57–87)	10 (0.7–249)
OKP	4	1 Abnormal point	86 (29–100)	90 (79–96)	58 (4–1585)
SAP suprathreshold	9	≥3 Points missing	71 (51–86)	85 (73–93)	14 (6–34)
SAP threshold	5	AGIS score ≥3	88 (65–97)	80 (55–93)	30 (6–159)
GAT	9	IOP >21 mm Hg	46 (22–71)	95 (89–97)	15 (4–49)

 The common cutoff was considered to include the following cutoffs: Ophthalmoscopy (discs graded as normal or suspect, subjective criteria); Optic disc photography (VCDR ≥0.7, normal/glaucomatous disc based on majority opinion of observers); RNFL photography (NFL lost); HRT II (global or I of the 6 segments flagged abnormal); OKP (1 or more points missing, if ≥ 1 chart numbers consistently made the black stimulus disappear); SAP suprathreshold (≥17 relative or absolute defects and/or cluster of 8 in any one quadrant, ≥4 abnormal points in any single quadrant, sufficient points to drop the indicator into the suspicious zone or below, 3 abnormal adjacent points, ≥1 missed point, optometrist judgment, at least 1 absolute defect associated with 1 relative defect or 3 adjacent relative defects or 4 non-adjacent relative defects or sure nasal step); SAP threshold (cross meridional, GHT abnormal/borderline, LDA 59 points, mirror image method, GHT outside normal limit and/or PSD P < 0.05 in one or both eyes); GAT (IOP ≥ 21 mm Hg, IOP 21–22 mm Hg, IOP >22 mm Hg). NFL, nerve fiber layer; AGIS, Advanced Glaucoma Intervention Study; GHT, Glaucoma Hemifield Test; LDA, Logistic Discriminant Analysis.

The authors thank the members of the OAG project group for assistance and comments at all stages of the review. Other members of the OAG Screening Project Group were: Rodolfo Hernandez, Luke Vale, Tania Lourenco, Cynthia Fraser, John Cairns, Richard Wormald, Stephen McPherson, David Wright, Kannaiyan Rabindranath, Rod Taylor, Norman Waugh, and Adrian Grant.