Patients

Subjects were enrolled consecutively into a diagnostic evaluation study using area under the curve analysis (AUC) recruited from 2 centers, one in Sydney, Australia, and the other in London, United Kingdom. The protocol and study design in Australia were approved prospectively by the South Eastern Sydney Area Health Network Human Research Ethics Committee (reference number 16/055). In the United Kingdom, the study was registered retrospectively as an audit of practice (institutional review board reference CA18/NO/01). Informed consent was obtained from all patients. All testing was conducted according to the tenets of the Declaration of Helsinki.

Inclusion and Exclusion C riteria

NTG patients had their diagnoses confirmed by a fellowship-trained glaucoma subspecialist based on characteristic optic disc appearance with matching glaucomatous visual field loss. IOP had never to have been recorded above 21 mm Hg, and all patients had open drainage angles with dark room gonioscopy. Secondary glaucomas such as pseudoexfoliation and pigment dispersion syndrome were excluded.

Other exclusion criteria included age younger than 18 years, significant media opacity, clinical evidence of diabetic retinopathy, and macular degeneration or any other retinal disease. Any participants unable or unwilling to undergo magnetic resonance imaging (MRI) of the brain and orbits with contrast were excluded.

NGON was diagnosed by a fellowship-trained neuro-ophthalmologist. Non-arteritic anterior ischemic optic neuropathy required documented painless disc swelling with visual field defects that either improved or stabilized over a 6-week period. A “disc at risk” was required in the contralateral eye and, if the patient was older than 50 years of age, an erythrocyte sedimentation rate and a C-reactive protein assay with normal results were also required. Arteritic anterior ischemic optic neuropathy or posterior ischemic optic neuropathy required a positive temporal artery biopsy. Prior optic neuritis was diagnosed based upon a clinical history consistent with optic neuritis and a contrast-enhanced MRI demonstrating optic nerve enhancement at the time the patient was symptomatic. Patient conditions were diagnosed with drusen on the basis of the International Drusen Collaboration recommendations. All hereditary optic neuropathies were included only if genetic testing confirmed a known mutation for the condition. Any NGON patient with IOP >21 mm Hg, narrow drainage angles, or a known family history of glaucoma were excluded.

All subjects underwent a complete ophthalmic examination including best-corrected visual acuity, near vision, color vision (Ishihara plates), IOP with Goldmann applanation tonometry, Humphrey visual field ([Humphrey Field Anaylser 2 [HFA2], Carl Zeiss Meditec, Dublin, CA] 24-2; SITA-standard, white on white), optical coherence tomography (OCT) (Spectralis OCT [Heidelberg Engineering, Heidelberg, Germany] MRW-BMO, retinal nerve fibre layer [RNFL] thickness [RNFLT], ganglion cell complex), and optic nerve head photos [Zeiss Visucam, Carl Zeiss Meditec, Jena, Germany]).

All NTG patients underwent blood tests for full blood count, electrolytes, urea, creatinine, vitamin B ₁₂ , and folate and an MRI of the brain and orbits with gadolinium enhancement using a standard protocol.

The OCT was performed by a trained technician at each facility. Only OCT scans with a mean quality score >15 were included for analysis. The scanning protocols for the Spectralis OCT followed previously published protocols. Briefly, the MRW-BMO scanning pattern is a radial scan consisting of 24 equally distributed high-resolution 15-degree B-scans which compute the neuroretinal rim measurements. B-scans were averaged from 20 to 30 individual B-scans, with 768 A-scans per B-scan, centered on the optic nerve head. The RNFLT is a circumpapillary 12-degree circular scan with 1,536 A-scans centered on the optical nerve head. The data is averaged from 16 individual B-scans.

The MRW-BMO scan has a global value which is segmented into 6 sectors (temporal, nasal, superotemporal [ST], inferotemporal [IT], superonasal [SN], and inferonasal [IN]). The BMO was identified automatically by the software but was manually checked by a single operator and corrected in case of any segmentation or BMO location errors.

Only 1 eye for each patient was included in the study. If both eyes were eligible, the worst affected eye was included for analysis.

Statistical Analysis

Univariate logistic regression models were used to assess whether the MRW-BMO measurement was significantly associated with the presence of GON. This was analysis was followed by 5-fold cross-validated logistic regression to assess the model’s predictive power to classify GON versus NGON. AUC, sensitivity, specificity, positive predictive values (PPV), negative predictive values (NPV), and likelihood ratios were calculated from the cross-validated model. The positive likelihood ratio (PLR) was calculated as sensitivity/(1-specificity), and the negative likelihood ratio (NLR) was calculated as the (1-sensitivity)/specificity. A higher PLR (>1) gives an increased probability of having a disease following a positive test result, whereas a lower NLR (between 0 and 1) gives a lower probability of having a disease following a negative test result. For reference, a PLR of 5 represents an approximate increase of 30% in the post-test probability of having a disease, whereas an NLR of 0.2 represents an approximate decrease of 30% in the post-test probability of a patient having a disease. Four pre-specified models were analyzed using different combinations of MRW-BMO measurements: 1) all 6 sectors; 2) of only the global measurement; 3) of only the ST and IT sectors; and 4) of only the ST, IT, and temporal sectors. For each of the sectors, both the MRW-BMO alone, to distinguish GON from NGON, and the MRW-BMO-to-RNFL ratio were analyzed. The rationale for the latter was that it would anatomically match the sector of cupping with any associated RNFL loss in that particular sector.

All analyses were conducted using R version 3.5.1 software (R Project, Vienna, Austria) with the caret application (version 6.0-80) for cross-validation of logistic regression models and the pROC application (version 1.13.0) for plotting receiver operating characteristic (ROC) curves.