Participants

The Australian Heart and Eye Study (AHES) is a cross-sectional study encompassing 1680 patients who presented to Westmead Hospital (Sydney, Australia) for diagnosis of coronary artery disease by coronary angiogram (S1 and S2 Files). All eligible consecutive patients between June 2009 and January 2012 were included. No absolute contraindications to coronary angiography were enforced and the procedure was performed where the risk-benefit ratio was favourable. Relative contraindications included contrast allergy and renal impairment. Most patients were recruited post-angiogram and investigations/data were collected as part of routine care.

Data collected included: detailed medical history, visual acuity, biochemistry, angiographic and peripheral arteriolar wave form analysis, pulse wave form, ankle-brachial index, peripheral and invasive blood pressure measurements, transthoracic echocardiogram, electrocardiogram, blood analysis and digital retinal photography. Patients were excluded if information regarding retinal vessel calibre, diabetic retinopathy or age-related macular degeneration status were incomplete.

Medical history

A 252-item questionnaire (S3 File) was administered encompassing medical history and cardiovascular risk factors including family history. Data were collected and collated by two study personnel creating a single entry. Questionnaire categories included: cardiac rhythm, angina history, previous myocardial infarction, previous angiograms, previous cardiac intervention (open surgical bypass or percutaneous coronary stenting), hypertension (defined as participants with a formal diagnosis of hypertension or using anti-hypertensive agents), hypercholestrolemia, diabetes mellitus (defined as patients with formal diagnosis or using hypoglycemic agents or insulin) or impaired fasting glucose state, chronic medical conditions, current medication, smoking status, alcohol consumption, previous stroke/transient ischemic attack.

Retinal vessel calibre assessment

Digital retinal photography was used to assess retinal vessel calibre. A pre-calibrated Canon 60^o fundus camera (Model CF-60DSi, Canon Inc., Tokyo, Japan) with an attached digital camera (Model 1DSmklll, Canon Inc., Tokyo, Japan) was utilised to capture dilated images of the optic disc and macula bilaterally. Retinal vessel calibre measurements from the right eye of each participant were used preferentially unless ungradable. One masked study personnel measured retinal vessel caliber using a computer-assisted program (IVAN, University of Wisconsin, Madison) as previously described [5, 12] and these studies reported high reproducibility with this methodology. For the present study intra-grader reliability of this method was high with quadratic weighted kappa values of 0.85 (CRAE) and 0.90 (CRVE). The diameter of all retinal vessels coursing through a pre-selected retinal region was measured. Values were combined with the Parr-Hubbard formula as modified by Knudtson, Lee [26] to obtain an average representing the central retinal arteriolar/venular equivalent (CRAE, CRVE).

Assessment of covariates

Peripheral blood pressure was measured using the Intellisense^TM OMRON digital automatic blood pressure monitor (Model HEM-907; OMRON Healthcare, Singapore) as a single measurement in the right arm in supine position. Invasive blood pressure measurements were achieved via a fluid filled catheter in the central aorta attached to the Mac-Lab hemodynamic system (GE Healthcare Milwaukee, WI). Diabetic status was determined from self-report, reported use of diabetes medications and blood glucose levels. Where the status was unknown or undetermined the patient was excluded from the study. Data collected from patient records included: renal function, full blood count, cardiac enzymes (creatinine kinase, troponin T), fasting blood glucose, HbA1c, fasting lipids and thyroid function tests.

Assessment of coronary artery disease

Routine diagnostic coronary angiography was performed after six hours fasting via either the femoral or radial artery using a catheter of known dimension (5Fr to 7Fr). Selective coronary injections of Ultravist (Schering) were filmed in standard projections on a Siemens Bi-Plane radiographic unit (Siemens Healthcare, Germany).

All angiograms were analyzed offline by a trained cardiologist masked to the results of the adjunctive investigations and retinal grading. The coronary artery segments were defined using the Syntax system, which divides the arterial tree into 16 segments, based on the modified American Heart Association (AHA) classification[27]. For each segment, the severity of obstruction was documented using several grades: normal, 1–25%, 25–50%, 50–74%, 75–99% and 100% (occluded). Each lesion that was visually scored as greater than 50% luminal obstruction in a vessel that was ≥1.5mm diameter was further analyzed using quantitative coronary analysis (QCA). QCA was performed using validated computerized edge-detection software (QCAPLUS, Sanders data Systems, Palo Alto, California, USA).

Coronary angiograms were scored according to two methods to document both the severity and extent of CAD:

1. (i). Gensini score (severity score): This has been described previously[28]. Briefly, the coronary arterial tree was divided into segments with multiplying factors according to the functional importance of any given segment (5 for the left main trunk to 0.5 for the most distal segments) and the percentage reduction in luminal diameter of each narrowing was assigned a score (0, 1, 2, 4, 8, 16 or 32), according to the degree of stenosis. The sum of the scores of all segments provides the Gensini score, which places emphasis on the severity of the disease[29].
2. (ii). Extent score: The extent score was proposed by Sullivan et.al to define the proportion of the coronary arterial tree with angiographically detectable coronary atheroma[30]. The proportion of each vessel involved by atheroma, identified by lumen irregularity, was multiplied by a factor for each vessel, which is related to the length of that vessel. The scores for each vessel were added to give a total score out of 100. This percentage represents the proportion of the coronary intimal surface area containing coronary atheroma[29].

Statistical analysis

Statistical analysis was performed using SAS (version 9.2, SAS Institute Inc., Cary, North Carolina, USA) and statistical significance was defined as P < 0.05. Multivariate logical regression analysis involved initially adjusting for age and sex, and then further adjusted for hypertension, BMI, smoking, diabetes history, cholesterol level and prior statin use. Confounders were selected based on previous studies reporting significant association with retinal vessel caliber [9–12].

Retinal arteriolar calibre in IFG/DM

A number of large studies have reported mixed results when exploring the relationship between retinal arteriolar calibre and hyperglycemia. Cross-sectional results reporting positive association include the Australian Diabetes Obesity and Lifestyle study [15], the Multi-Ethnic Study of Atherosclerosis (MESA) study [16], where significance was sustained only in the white population, and a study by Jeganathan, Sabanayagam (17), where the relationship was particularly strong in subjects of Indian background. In agreement a prospective study in the Beaver Dam population reported an association between larger arteriolar calibre and 15 year incident diabetes [31]. Contrastingly, Wong, Shankar [23] and Nguyen, Wang [21] reported retinal arteriolar narrowing, rather than widening, to be associated with incident diabetes at 10 year follow-up in the Beaver Dam and at 5 year follow-up of the AusDiab Study, respectively.

A recent meta-analysis by Sabanayagam, Lye [32] pooling 5 population-based prospective studies confirmed no association between retinal arteriolar calibre and diabetes or impaired fasting glucose. This is in agreement with the findings reported in this current study.

Retinal venular calibre in IFG/DM

Our finding of a positive association between retinal venule widening and DM is consistent with that reported in the literature. Cross-sectional data from the Wisconsin Epidemiological Study of Diabetic Retinopathy (WESDR) [33], MESA [16] and a Singaporean study [17] reported a greater magnitude of retinal venular widening association with presence of DM. Prospective studies later reported this relationship may be predictive with patients having wider baseline venular calibre being more likely to develop DM, as in the Beaver Dam Eye Study [20], or impaired fasting glucose, as in the Blue Mountains Eye Study among patients aged under 70 years [19]. The Wisconsin study further showed that wider retinal venular calibre was related to retinopathy severity, duration of diabetes and higher levels of glycosylated haemoglobin [14]. The association between larger retinal venular calibre and DM was also confirmed in a meta-analysis of 5 prospective population studies [32].

The mechanism of association between retinal venular calibre and DM is currently unknown. Physiological studies have highlighted the role of insulin in increasing skeletal muscle microvascular perfusion [34]. Changes in insulin status have subsequently been shown to alter microvascular recruitment where hyperinsulinemia stimulates total blood flow [35] and with insulin resistance this mechanism is impaired [34]. Whilst studies have been largely confined to skeletal muscle, a similar phenomenon may be extrapolated to the retinal microvasculature. While the selectivity for retinal venules is unclear, wider retinal venules have previously been noted as a marker of endothelial dysfunction and inflammation [36]. Thus, it is possible the selective widening of retinal venules represents the combined effect of inflammation, insulin resistance and endothelial dysfunction. Further studies are required to elucidate the pathological mechanisms underlying this relationship.

After stratification by Extent score, our data suggest this positive association was more apparent in patients with less diffuse CAD (i.e. Extent scores below the median), thus present in the absence of macrovascular disease. To our knowledge this association between retinal venular calibre and diabetes, in the presence of less diffuse CAD, has not been previously described. The loss of statistical significance with multivariate analysis in this case may be due to small sample size and further adequately-powered studies would be helpful to explore the reproducibility and magnitude of this association.

Interestingly, we report a gender difference in the association between retinal venular calibre and DM where significance was only apparent in men. This is in-line with a meta-analysis of pooled data where the association was stronger in men [32]. Benitez-Aguirre, Craig [37] previously reported sex differences in retinal vascular geometry in pre-pubertal Type 1 diabetics where young females had wider baseline retinal venular calibre and greater propensity to develop retinopathy earlier. The influence of sex hormones on the retinal vasculature and its links with hyperglycemia has not been previously explored and may be of interest in future studies.

Limitations

In interpreting the results of our study it is important to consider its limitations. Firstly, the cross-sectional design of our study only allows description of associations and not causal relationships. However, several large prospective studies with similar findings suggest causality [19, 20]. Secondly a number of other baseline cardiovascular risk factors, in particular hypertension [38], have been shown to affect retinal vessel calibre. Multivariate adjustment for key cardiovascular risk factors facilitated evaluation of DM as an independent factor in retina vessel calibre, however, we cannot disregard the potential influence of residual confounding on observed associations. Thirdly, our retinal vessel measurements were not cardiac-gated and variation in measurements of 2–11% in venules have been reported [39]. However randomisation of measurements is likely to minimise this effect and our methods are in line with previous large studies. Fourthly, the interobserver and intraobserver variability in the cardiac scores (Extent, Gensini) have not been calculated. However, these scores have previously been validated by dedicated studies and thus used as such [40, 41]. Another interesting observation was a large number of men in the present cross-sectional study compared to females, which is likely due to bias from the cardiovascular nature of the cohort. We cannot also completely exclude the risk of selection bias, given the cross-sectional nature of this study, and as such, the present findings require confirmation in prospective longitudinal studies.