Study design and population
This hospital-based, cross-sectional case control study was conducted from January 2016 to June 2017. This research adhered to the Declaration of Helsinki and was approved by the local ethics committee of the Institutional Review Board of Peking University First Hospital. We obtained informed consent for participation from all subjects. Newly confirmed NTG cases and healthy age-matched controls were recruited during the same period from Peking University First Hospital in Beijing, China, and all diagnoses were made by glaucoma experts.
We extracted basic demographic features (age, sex, race, and preexisting medical conditions); performed complete routine ophthalmic examination; measured central corneal thickness (CCT) by A-scan ultrasound pachymetry (SW-2100, Sowei, Tianjin); and obtained stereoscopic optic disc photographs centered on the disc with a digital color retinography (EOS 50D CMOS, Canon CR-2, Tokyo, Japan) and a 3.5-mm-diameter circle centered on the optic disc with spectral-domain optical coherence tomography (Optovue, Fremont, CA, USA); and 24–2 Humphrey Swedish Interactive Threshold Algorithm (SITA) standard VF testing (Humphrey Field Analyzer, Carl Zeiss Meditec, Dublin, CA, USA) for all participants.
Eligibility criteria
The patients with NTG demonstrated reproducible visual field loss consistent with neuroretinal rim narrowing or excavation and nerve fiber layer defects. The diurnal IOP curve showed the highest intraocular pressure ≤ 21 mm Hg. No subjects had abnormal anterior chamber angles on gonioscopy or secondary causes of optic nerve damage on posterior segment examination.
Controls had IOP ≤ 21 mm Hg, cup disc ratio ≤ 0.6, and cup disc ratio asymmetry ≤ 0.2. Normal slit lamp examinations, normal optic nerves with no rim or RNFL changes and normal visual field were also confirmed in controls. Only age-matched control subjects without a family history of glaucoma were included.
Reliable perimetry was defined as a fixation loss rate ≤15% and false positive and negative rates ≤15%. A signal strength index > 50 was used to define good quality OCT images. An important inclusion criterion for cases and controls was subjects with refractive error between -6 D and +3 D and a cylinder correction within ±3 D, and axial length < 26 mm, because over refraction cause discrepancies in photographic magnification and affect the accuracy of retinal vessel measurement[4, 5]. We excluded subjects with diabetes, hypertension, hyperlipemia, cardiac-cerebral abnormalities, migraine, any autoimmune connective tissue disease, history of anti-glaucoma medications, history of intraocular or laser surgery, eye trauma, and older than 70 years because these events may contribute to vascular structural changes[6]. This research only included mild and moderate (visual field MD ≥ -12 dB) NTG eyes with single-hemifield involvement, which means superior or inferior visual field loss consistent with nerve fiber layer defects on fundus photography.
Retinal arteriole caliber measurements
Professional technicians used a digital non-mydriatic retinal camera to acquire images centered on the optic disc and covering the optic cup per patient and stored them at 2592*1728 pixels in size (the actual size of the image is 22.3*14.9 mm) on a computer. Then, these images were forwarded to a drive where they were analyzed by 2 glaucoma-trained masked ophthalmologists.
Photographs with poor quality or any one of four main retinal arteriole branches within the disc were excluded. The retinal arteries are bright red, with obvious reflection, and the branches are inclined at an acute angle. While, retinal veins are dark red, the reflection is not obvious, and the branches are obtuse. Normal arteriovenous ratio is 2:3[7]. We plotted superior temporal, inferior temporal, superior nasal, and inferior nasal retinal arteries and then measured their calibers at the optic disc border using ImageJ (National Institutes of Health, Bethesda, MD), as shown in Fig.1.
The measurement protocol was as follows:
- The original full size version of a received image was opened using ImageJ.
- A scale bar was set for this image. Since the amplification of a Canon CX-1 fundus camera is 1.1, 2592 pixels equals 20273 microns. We entered 2592 for the distance in pixels, 20273 for the known distance, and micron for the unit of length.
- The image was magnified 200 times. Two ophthalmologists masked to diagnosis evaluated four parapapillary retinal arteriole calibers (PRACs) per eye, which were measured five times on each artery with the average recorded. Both observers measured all images two times to assess interrater reliability.
Optical coherence tomography measurements
RTVue-OCT was used according to the standard glaucoma protocol, and each eye underwent a three-dimensional optic disc scan. The main parameters of the ONH scan used in this research were the average RNFL thickness (RNFLT), superior RNFL thickness, inferior RNFL thickness, cup area, rim area, and cup/disc area ratio.
Statistical methods
All statistical analyses were performed in SPSS version 20.0. Differences in NTG vs controls and cases with superior RNFL defects vs inferior RNFL defects were tested by using independent two-tailed Student’s t-tests for continuous variables and χ2 tests for categorical variables. We used repetitive measurement and analysis of variance to compare four PRACs for each group.
In case-only analyses, univariate and multiple linear regression analyses were used to assess associated factors for PRACs. Pearson’s correlation test was performed using factors of age, sex, CCT, cup area, rim area, cup/disc area ratio, RNFL thickness and visual field global indices for four PRACs. Multivariate analyses were executed in an enter approach, and standardized beta coefficient (β) and P values are reported. The area under the receiver operating characteristic curve (AUROC) was used to compare the powers to detect NTG in the parameters of the PRAC, RNFL thickness and mean deviation (MD) read in Humphrey perimetry.
Interclass correlation coefficients obtained by the two-way mixed-effect model showed good interrater and intrarater reliability (intrarater ICC ≥ 0.993 and interrater ICC ≥ 0.973; Table 1). All statistical tests were two sided, and a P value < 0.05 was considered statistically significant.