Subjects

The villages of Tailu, Beijiao, and Xiubang in the Tailu township in Lienchiang county and Kungtung, Kunghsi, Yantai, and Wenwo in Haidao township in Xiapu county in Fujian Province were randomly selected for this cross-sectional investigation, which took place from July, 2011 to November, 2011. These participants were selected because it has been shown that coaster villagers are at higher risk for hypertension [26].

Inclusion criteria were voluntary participants aged 30 years or more living in Tailu and Haidao townships for more than 5 years. Subjects were excluded if they were diagnosed with or were suspected to have secondary hypertension, acute myocardial infarction, acute stroke, infection, or an inflammatory disease. Individuals who had persistent arrhythmias (e.g., atrial fibrillation), chronic heart dysfunction, pulmonary hypertension, acute and chronic infectious diseases, malignancy, hematuria or urinary tract infection, peripheral vascular disease, ankle-brachial index (ABI) <0.6, or paralysis were also excluded. In addition, individual with a history of serious eye diseases such as corneal diseases, glaucoma, macular degeneration, or eye injuries, pregnant women, and anyone who could not cooperate with study requirements were excluded.

All participants provided signed informed consent form. This study was conducted in accordance with the Declaration of Helsinki, and was approved by the Institutional Review Board of the Fujian Provincial Hospital.

Data collection

Age, educational background, smoking history, alcohol history, and relevant medical history (e.g. hypertension, chronic heart failure and arrhythmia) were obtained using questionnaires. Physical examinations included height (cm), weight (kg), heart rate (HR, beats/minute) and BP. Examinees could not smoke or drink coffee within 30 minutes before measuring blood pressure. After emptying his/her bladder and resting in a quiet environment for 5 to 10 minutes, the examinee’s BP was measured by using a standard vertical mercury sphygmomanometer on the right upper arm. The average of three BP measurements obtained 1–2 minutes apart was calculated and recorded.

Classification definitions

BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²). According to the BMI, subjects were divided in three categories: normal or underweight (BMI<24 kg/m²), overweight (24–28 kg/m²) and obese (>28 kg/m²) [27]. Hypertension was defined as SBP≥140 mm Hg and diastolic BP (DBP) ≥90 mm Hg in the presence or absence of anti-hypertension medications [28]. In the absence of anti-hypertensive medications, a systolic BP of 120–139 mm Hg and diastolic BP of 80–89 mm Hg was defined as “prehypertension” and a systolic BP of <120 mm Hg and diastolic BP of <80 mm Hg was defined as normal [29].

Laboratory testing

After an 8-hour overnight fast, blood samples were obtained measurement of triglyceride (TG), total cholesterol (TC), HDL-C, low-density lipoprotein cholesterol (LDL-C), fasting plasma glucose (FPG), HbA_1c, high-sensitivity C-reactive protein (hs-CRP), and serum uric acid. Participants diagnosed with diabetes, those currently taking anti-diabetic medications, or with an HbA_1c≥6.5% were defined as diabetic [29].

Data collection and quantitative analysis of retinal vessels

Digital fundus photographs were obtained using non-mydriatic fundus photography. Binocular digital photographs were taken using a 45° high-resolution fundus camera (Topcon NW-8, Topcon Corp, Tokyo, Japan; Nikon D90, Nikon Corp, Tokyo, Japan). Retinal fundus images were centered on the optic disc. The side of fundus photograph with a better quality image was analyzed, and double-blinded analysis was performed by two professionally trained ophthalmologists.

Semi-automated software (Singapore I Vessel Assessment, Version 3.0 [SIVA]; Exploit Technologies Private Limited, Singapore) was used for the quantitative analysis. SIVA software was been developed in collaboration with the Singapore Eye Research Institute, and has the capability of measuring a number of retinal vessel parameters including arteriolar/venular caliber, tortuosity, branching angles, and fractals.

The SIVA software was per the developer’s protocol. The software employs automatic optic disc detection and measures caliber and geometry of peripheral vessels up to 2 DD from the optic disc margin. Measurements were taken within a concentric grid (0.5–2.0 DD) centered on the optic disc [30]. The program automatically traced and identified all vessels (artery or vein) within the concentric circular grid, thus generating a skeleton image of the retinal microvasculature. The spatial resolution of each image was 3216×2136 pixels, and the images were stored without compression before analysis. The size of the scaling window was 1366×731 pixels. With the box counting method, the digitized retinal image was divided into a large number of equally sized square boxes, and the number of boxes containing a section of the refined skeletonized line tracing is counted; the process is then repeated with a different sized box. The fractal dimension of the skeletonized line tracing is the slope of the line obtained via plotting the logarithm of the number of boxes through which the tracing passes against the logarithm of the size of the boxes. Trained graders then correctly identified all vessels traced and removed all artifacts from the fundus photographs that may have been incorrectly drawn as vessels from the tracing. The reproducibility of this method has been shown in several studies [9], [15], [30].

CRAE and CRVE measurements, performed 0.5- to 1.0-disc diameter away from the optic disc margin, were calculated by the revised Knudtson-Parr-Hubbard formula (Fig. 1) [31]. Fractal geometry can be used to quantify a branching pattern that exhibits self-similarity, and the retinal vasculature has a fractal-like architecture. D(f), which is a measure of a fractal structure characterizing the distribution of the branching vascular system in two-dimensional space, was used to quantify the branching pattern of the retinal blood vessels. Retinal vascular D(f) was calculated from a skeletonized line tracing using the box-counting method, which divides each digital photograph into a series of squares of various side lengths, and the number of boxes is counted [8], [15]. D(f) was defined as the gradient of logarithms of the number of boxes and the size of the boxes [9]. Larger values indicate a more complex branching pattern, and two-dimensional fractal dimensions are between 1 and 2. D(f) measurements were performed 0.5- to 2.0-disc diameters away from the optic disc margin, as described by Cheung et al. [8].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Fundus photographs processed by SIVA version 3.0 software.

Test zone was 0.5- to 2.0-disc diameters away from the optic disc margin. A red line represents an artery and a blue line represents a vein. Fractal dimension, D(f), is a complexity indicator determined by reactive vessels obtained from vascular tracking through the box counting dimension method.

https://doi.org/10.1371/journal.pone.0106551.g001

Statistical analysis

Continuous variables were presented as mean ± standard deviation, and categorical variables by count and percentage. To investigate the potential confounders associated with the retinal vascular parameters, simple and multiple linear regressions were performed. In univariate analyses, all significant factors were selected for entry into the multivariate models, except for those that had collinearity with other variables. Statistical analyses were performed with SAS software version 9.2 (SAS Institute Inc., Cary, NC). A two-sided P value<0.05 was considered to be statistically significant.

Relationship of retinal vessel parameters with cardiovascular risk factors

A previous study reported a correlation between CRAE and plasmas TG [32]. The current study found that a narrow CRAE was associated with obesity and gender, while a large CRAE was associated age, smoking, and elevated HbA_1c, consistent with previous reports [2]–[5], [33], [34]. The current study also showed that a decreased CRVE was associated with an increase of LDL-C; however, this correlation was not significant after adjusting for other cardiovascular risk factors. Therefore, we could not conclude that CRVE was associated with lipid levels. Our study did not find an association between CRVE and hs-CRP, consistent with the Cardiovascular Health Study [35], but inconsistent with the Beaver Dam [31] and multi-ethnic study of atherosclerosis (MESA) studies [5]. We suggest that differences between the current study and the Beaver Dam study including the participants’ age, angle of fundus photographs, software application for data analyses, and racial composition of the study cohorts contributed to the different findings. Compared to the MESA study, the current study subjects were ethnically homogeneous and the proportion of males was low (37.4% vs 53.6%, respectively). Additionally, our study subjects had lower educational levels, which might also have influenced the results.

A decrease in D(f) was associated with an increase of age in the current study. This finding was also shown in studies using a box counting dimension method to measure D(f) in whites and Malaysians [9], [15]. Another study using Fourier fractal dimension (FFD) techniques to express vascular network complexity also had the same results [36].

Relationship of retinal vessel parameters with BP

Hypertensive retinopathy is the most common ophthalmologic manifestation of hypertension [37]. Our study results indicated that CRAE was highly correlated with MAP, consistent with numerous studies in different populations including Singaporeans [23], Japanese [38], Caucasians [2], [3], [33], [39], African-Americans [4], and multiethnic populations [5]. MAP showed a stronger association with CRAE than PP, indicating that elevated BP has a stronger impact on central retinal artery stenosis than BP fluctuations. Similar to other studies, this study could not show causality; however, CRAE can predict stroke, mortality of cardiovascular diseases, and coronary heart disease independently of BP levels [39]–[42].

A number of studies [23], [24], [35], [37], [43] have shown that CRVE is not associated with hypertension. The current study also demonstrated no significant association between CRVE and hypertension, but found that a narrow CRVE was not related to an increased PP difference, which is inconsistent with the Rotterdam Study [34]. Increased BP and PP were associated with significant reductions in the retinal arteriolar and venular diameters and the arteriolar-to-venular ratio in the Rotterdam Study. In addition, a study with participants from southwestern Singapore found that retinal arteriolar tortuosity was correlated with elevated BP [44]. PP difference not only reflects a high SBP, but also implies different levels of arterial stiffness. We hypothesize that changes in venules may be due to the same pathological processes resulting in atherosclerosis, i.e., endothelial damage, oxidative stress, and lipid overload.

The current study showed that CRVE was independently associated with uric acid. Uric acid is an important pro-inflammatory factor [45], and some studies have found that uric acid plays a role similar to CRP. Elevated uric acid may suggest occult vasculitis or remodeling of pericapillary spaces [46]. Although the present study could not find the association between CRVE and CRP, other studies have found that increased CRVE was related to specific inflammatory factors. Despite the fact that uric acid is an antioxidant, high uric acid concentrations are associated with microvascular functional reserve or microvascular dysfunction [46]–[48]. This may be related to a series of downstream reactions caused by free radicals generated in the degradation process. Yuan et al. [49] studied a population at high-risk for diabates and found that uric acid concentrations were associated with decreased CRAE and increased CRVE. Our study was a population-based cross-sectional survey and there were methodological differences between the current study and that of Yuan et al.

Studies conducted in whites and Malaysians [9], [15] showed that D(f) was independently associated with SBP, DBP, MABP, and PP. Compared to the association between CRAE and SBP, D(f) showed a stronger association with SBP. The current study found D(f) to be highly correlated with BP classification (prehypertension, hypertension). We speculate that D(f) may have greater sensitivity than CRAE for hypertension morbidity and severity.

Relationship of retinal vessel parameters and FPG, Hba_1c, and diabetes

A study in northwest Shanghai reported that participants with narrower retinal arteriolar calibers were diagnosed with metabolic syndrome nearly twice as often as their normal counterparts (odds ratio 1.78) [50]. A relationship of larger CRVE with high HbA_1c has been demonstrated [2]–[5], [33]. In addition, Tsai et al. [51] reported that in diabetic patients a wider CRAE was correlated with increasing glucose (P<0.001) and HbA_1c levels (P<0.001). Other studies have found correlations between increasing Hba_1c levels, plasma glucose levels, and retinal venular caliber measurements [52]. It is likely that we did not find an association with these measurements due to the differences in our study population compared with previous studies.

To the best of our knowledge, D(f) independently associated with HbA_1c has not been reported in other studies: however, this finding might help to explain the association between D(f) and diabetic retinopathy [16], [17], [53]. We suggest that the incidence of low D(f) and diabetic retinopathy is increased in diabetic patients with long-term poor control of blood sugar. Thus, we hypothesize that D(f) might be considered as a predictor of diabetic microangiopathy and target organ damage in diabetic patients.

Study strengths and limitations

Strengths of the current study are the large number of patients and the fact that it was restricted to the Chinese Han population in coastal regions. Limitations are the study design, which cannot determine causality, selection bias because a relatively large number of subjects were excluded, and patients with ophthalmic diseases such as refractive errors and lens opacity were excluded, which may have influenced D(f). Genetic factors, hormone levels, cell factors, and environmental influences which may have affected retinal vascular parameters were not evaluated. Some bias may be present because retinal vessel diameters are affected by the cardiac cycle.