The Association of Retinal age gap with metabolic syndrome and inflammation

2.1 Study population

The UK Biobank is a population-based cohort study with over 500 000 participants recruited in the United Kingdom between 2006 and 2010. A total of 22 assessment centers were set across England, Wales, and Scotland. Approximately 9.2 million residents aged 40–69 years within 25 miles of a nearest assessment center were invited to participant in this study, among whom 5.5% were enrolled in the baseline assessment. All participants completed comprehensive health care questionnaires, detailed physical measurements, and biological sample collections. The overall study protocols and data are available elsewhere.¹⁶ This study was restricted to a subset of participants with metabolic syndrome and/or inflammation data available at the initial assessment (March 2006 to December 2010).

2.2 Ethical approval

The UK Biobank Study received approval from the National Information Governance Board for Health and Social Care and the National Health Service North West Multicentre Research Ethics Committee (11/NW/0382). Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.

2.3 Fundus photography

Ophthalmic examinations were introduced from 2010. LogMAR visual acuity, autorefraction, and keratometry (Tomey RC5000, Tomey GmbH, Nuremberg, Germany); intraocular pressure (Ocular Response Analyzer, Reichert, New York, USA); and paired retinal fundus and optical coherence tomography imaging (OCT, Topcon 3D OCT 1000 Mk2, Topcon Corp, Tokyo, Japan) examinations were performed. A 45-degree nonmydriatic and nonstereo fundus image centered on the macula with the optic disc included was taken for each eye. A total of 131 238 images from 66 500 participants were obtained from the UK Biobank study. A total of 80 170 images from 46 970 participants who passed the image quality check were used for the following analyses.

2.4 Estimation of retina age and retinal age gap

Retinal age was predicted according to the methods described by Zhu et al.¹³ Briefly, only color fundus images were fed into a DL model using an Xception architecture to predict biological age. DL is the advanced subset of machine learning with multiple neural networks and could self-learn complex representations without requiring human engineering and domain expertise to design feature extractors or calculation methods.

The difference between the retinal age predicted by the DL model and chronological age was defined as the retinal age gap. A positive retinal age gap indicated an “older” appearing retina, whereas a negative one indicated a “younger” appearing retina.

2.5 Metabolic syndrome and inflammation index

MetS was defined as the presence of three or more of the following: central obesity, hypertension, dyslipidemia, hypertriglyceridemia, and hyperglycemia.^{1, 17} Waist circumference was measured at the smallest part of the trunk using a 200-cm tape measure (SECA). Central obesity was defined as a waist circumference of 88 cm or above for women and 102 cm or above for men. Systolic and diastolic blood pressure were measured twice using an IntelliSense blood pressure monitor model HEM-907XL (Omron) after participants rested for at least 5 min and the averages were taken as the final results.¹⁸ Hypertension was defined as a systolic blood pressure of ≥130 mm Hg and/or a diastolic blood pressure of ≥80 mm Hg or taking antihypertensive drugs. Dyslipidemia was defined as levels of high-density lipoprotein (HDL) cholesterol <50 mg/dL in women and <40 mg/dL in men. A triglyceride level ≥150 mg/dL was defined as hypertriglyceridemia. Hyperglycemia was defined as fasting blood glucose levels >110 mg/dL or taking hypoglycemic medication or using insulin. Inflammation index was defined as a high-sensitivity C-reactive protein (CRP) level >3.0 mg/L.

2.6 Demographic and health variables

Demographic factors and health variables included baseline age, sex, ethnicity (recorded as white and nonwhite), Townsend deprivation indices (an area-based proxy measure for socioeconomic status), education attainment (recorded as college or university degree, and others), smoking status (recorded as current/previous and never), drinking status (recorded as current/previous and never), physical activity level (recorded as above moderate/vigorous/walking recommendation and not), history of heart disease (angina or heart attack), history of stroke, and general health status (recorded as excellent/good and fair/poor).

2.7 Statistical analyses

Data are presented as mean (SD) or median (interquartile range, IQR) for continuous variables and numbers (percentages) for categorical variables. The differences of the baseline characteristics of the study population among retinal age gap quartiles were compared using analysis of variance and chi-square test. The differences of the baseline characteristics between the MetS group and non-MetS group were compared using chi-square test and/or t test. Logistic regression models were applied to investigate the association between retinal age gap (independent variable) and MetS and inflammation (outcome), respectively. We then investigated associations between retinal age gaps at different quartiles with metabolic syndrome and inflammation. The associations between retina age gap and abdominal obesity, hypertension, elevated serum HDL, elevated serum triglycerides, and hyperglycemia were also examined respectively by logistic regression models. All logistic regression models were adjusted for baseline age, sex, ethnicity, Townsend deprivation indices, educational level, smoking status, drinking status, physical activity level, general health status, history of heart disease, and history of stroke. Odds ratios (OR) with their 95% confidence intervals (CIs) were reported. Variance inflation factors (VIF) procedure was used to test collinearity for all variables and all covariables' VIF were <2. For all the analyses, complete data were used. A two-sided p value of <.05 indicated statistical significance. Analyses were performed using Stata (version 13, StataCorp, Texas, USA).

3.1 Study sample

Of 35 918 included participants, the mean (SD) age was 56.6 (8.04) years, and 20 002 (55.7%) were women. Table 1 shows baseline characteristics of the study participants overall and stratified by quartiles of retinal age gaps. As shown, participants with a higher quartile of retinal age gaps were more likely to be younger, of female gender, of nonwhite ethnicity, nonsmokers, physical inactivity, less healthy in general status, more deprived, better educated, and without a history of chronic heart diseases and stroke (p < .001).

The distribution of retinal age gap of the included participants is shown in Figure 1. The mean (SD) and median (IQR) of the retinal age gap were −1.31 (4.82) and −1.18 (−4.18 to 1.79). The retinal age gap was divided into four groups of equal size, with Q1 as the lowest 25% retinal age gap, Q2 as the lowest 25%–50%, Q3 as 50%–75%, and Q4 as the highest 25% of retinal age gaps. The retinal age gap ranges for the four quartiles are Q1 (−27.9 to −4.18), Q2 (−4.18 to −1.18), Q3 (−1.18 to 1.79), and Q4 (1.79 to 19.0). We also added the fundus photographs of Q1-Q4 stages in the Figure S1. Compared to the fundus images in Q1, the fundus images in Q4 tended to have more retinal vascular changes including arteriolar narrowing and blood vessel tortuosity.

3.2 Metabolic syndrome and inflammation

Table 2 presents baseline characteristics of the study participants stratified by MetS. Compared to the nonmetabolic syndrome group, participants with MetS tend to be older, of male gender, former/current smokers, nondrinkers, physical inactivity and with greater deprivation, poor education, and poor general health status, and with a history of chronic heart diseases and stroke (p < 0.001).

3.3 Retinal age gap and metabolic syndrome and/or inflammation

Associations of retinal age gap with metabolic syndrome and/or inflammation are reported in Table 3. Each year increase in the retinal age gap was associated with a 1% risk increase of MetS (OR: 1.01; 95% CI, 1.00,1.02; p = .016), a 1% risk increase of inflammation (OR: 1.01; 95% CI, 1.00,1.02; p = .021), and a 1% risk increase of MetS and inflammation combined (OR: 1.01; 95% CI, 1.00,1.02; p = .011) in the fully adjusted model. Compared to participants with the lowest quartile of retinal age gaps, the risk of MetS significantly increased by 10% and 14% respectively for participants with retinal age gap in the third and fourth quartiles (OR: 1.10; 95% CI, 1.01,1.21; p = .030; OR: 1.14; 95% CI, 1.03,1.26; p = .012, respectively). A similar trend was identified for the risk of inflammation that participants with retinal age gaps in the third and fourth quartiles had 10% and 25% increased risks (OR: 1.10; 95% CI, 1.01,1.21; p = .048; OR: 1.25; 95% CI, 1.12,1.38; p < .001) compared to those in the first quartile, respectively. A significant association existed between retinal age gaps and a combined condition of inflammation and MetS. Table 4 shows the association between retinal age gaps and the five specific subcomponents of MetS. Per year increase in retinal age gaps was associated with a 2% risk increase in abdominal obesity (OR: 1.02, 95% CI, 1.01,1.02; p < .001), a 1% risk increase in hypertension (OR: 1.01; 95% CI, 1.00,1.02; p = .002), a 6% risk increase in hyperglycemia (OR: 1.06, 95% CI, 1.04,1.07; p < .001) but not associated with elevated serum HDL and elevated serum triglycerides. Compared with the participants with the lowest retinal age gap quartile, participants with higher quartiles tend to have a 10% to 27% risk increase in abdominal obesity and 14% to 19% risk increase in hypertension. Moreover, participants with the higher quartiles of retinal age gaps were associated with a 25% to 104% risk increase in hyperglycemia (OR: 1.25; 95% CI, 1.10,1.42; p < .001; OR: 1.45; 95% CI, 1.27,1.66;, p < .001; OR: 2.04; 95% CI, 1.74,2.38; p < .001, respectively).