Diagnostic accuracy of imaging devices in glaucoma: an updated meta-analysis

Author: Moradi Y, Moradkhani A, Pourazizi M, Rezaei L, Azami M.

Geographical coverage: United States China, Republic of Korea, Spain, India, France, Hungary, Colombia, Canada, Japan, Turkey, Italy, and Britain.

Sector: Service delivery

Sub–sector: Screening, diagnostic accuracy

Equity focus: Not explicitly stated

Study population: Adult patients with glaucoma

Review type: Effectiveness review

Quantitative synthesis method: Meta-analysis

Qualitative synthesis method: Not applicable

Background

Glaucoma is the leading cause of irreversible blindness globally, particularly in high‑income regions. Its prevalence rises with age, making early diagnosis essential to prevent vision loss. Conventional clinical examinations often miss disease at its earliest stage. In the past two decades, advanced imaging technologies—optical coherence tomography (OCT), Heidelberg retinal tomography (HRT), optical coherence tomography angiography (OCTA) and scanning laser polarimetry with the GDx device (SLP/GDx)—have become central to screening and diagnosis because they quantify the optic nerve head and macular structures. However, studies report variable sensitivity and specificity for these modalities, raising concerns about their reliability.

Objectives

To compare the diagnostic accuracy of OCT, HRT, OCTA and SLP/GDx and provide ophthalmologists and policy‑makers with evidence to support earlier detection and improved management of glaucoma.

Main findings

Twenty‑three cross‑sectional studies met the inclusion criteria: four from the United States, three from China, three from the Republic of Korea, two each from Spain, India, France and Hungary, and one each from Colombia, Canada, Japan, Turkey, Italy and Britain.

Imaging devices were analysed in two anatomical groups: optic nerve head (ONH) and macular. For devices assessing the ONH (27 data sets), pooled sensitivity was 77 % (95 % confidence interval [CI] 70 to 83) and pooled specificity 89 % (95 % CI 84 to 92). For the macular region (16 data sets), pooled sensitivity rose to 87 % (95 % CI 80 to 92) and specificity to 90 % (95 % CI 84 to 94).

By modality, pooled estimates were: OCT—sensitivity 85 % (95 % CI 81 to 89), specificity 89 % (95 % CI 85 to 92); HRT—sensitivity 72 % (95 % CI 57 to 83), specificity 79 % (95 % CI 62 to 90); OCTA—sensitivity 82 % (95 % CI 66 to 91), specificity 93 % (95 % CI 87 to 96). Overall, the macular region showed higher diagnostic performance than the ONH, and OCT was the most sensitive modality, whereas OCTA was the most specific.

Methodology

The authors searched PubMed, Scopus and Web of Science for descriptive, cross‑sectional or diagnostic‑accuracy studies published from 2004 to 2022 that evaluated the above imaging devices against a recognised reference standard and reported sensitivity, specificity or predictive values. Two reviewers independently screened records. Methodological quality was assessed with the revised Quality Assessment of Diagnostic Accuracy Studies (QUADAS‑2) checklist. Data were synthesised using a hierarchical summary receiver operating characteristic (HSROC) model; pairwise multiplicity corrections were applied. Summary receiver operating characteristic (SROC) curves and forest plots illustrated diagnostic performance. Publication bias was assessed with a Deeks funnel plot.

Applicability / external validity

The meta‑analysis pooled evidence from multiple regions and platforms, supporting generalisability across diverse clinical settings. Nevertheless, heterogeneity in study methods, glaucoma definitions and imaging parameters may limit the direct transferability of the pooled estimates to individual practices.

Geographic focus

Included studies were conducted in the United States, China, Republic of Korea, Spain, India, France, Hungary, Colombia, Canada, Japan, Turkey, Italy and Britain.

Summary of quality assessment

Confidence in the pooled estimates is reduced by several methodological limitations. The review did not report searching the grey literature, provide a list of excluded studies or specify whether data extraction was performed in duplicate. Although QUADAS‑2 was applied, results were not stratified by risk of bias, and considerable heterogeneity remained unexplained. These factors suggest the conclusions should be interpreted with caution.