Sex Differences in Age-Related Changes in Retinal Arteriovenous Area Based on Deep Learning Segmentation Model

The retina is the only part of the human body where various age-related diseases including hypertension and arteriosclerosis can be estimated through its microvasculature.1, 2, 3 Our deep learning–based “Hokkaido University retinal vascular segmentation model” separately identified arterioles and venules on color fundus photographs, and the sum of pixels in the probability maps of arterioles and venules was defined as arteriolar area (AA) and venular area (VA).⁴ Retinal vessels narrow with age, reflecting systemic arteriosclerotic changes, but little is known about sex differences. The present study compared sex differences in correlations between age and the automatically calculated AA and VA.

This retrospective cross-sectional study was conducted according to the guidelines of the Declaration of Helsinki and approved by the institutional review board of Hokkaido University Hospital (# C-T2023-0179). This study involves human participants with retrospective, noninvasive observations. For this type of study, patients were given the opportunity to refuse to participate in the study via the website. The subjects were 9938 color fundus photographs (AFC-330, NIDEK, Co., Ltd) clearly imaged from both eyes of 4969 Japanese examinees who visited the Keijinkai Maruyama Clinic for regular health checkups. The average values of both eyes were used for analysis of AA and VA in each individual. Correlations between age and vessel areas were examined for each sex. To investigate an association with menopause, female eyes were divided into 2 groups, younger and older, at arbitrary ages. The differences in the slope of the regression line of AA on age were analyzed for each of the 2 groups at each age. Systemic factors related to AA and VA were examined. Statistical tests were performed with the R statistical package (version 3.6.1; R Foundation for Statistical Computing). P < 0.05 was considered statistically significant. The clinical background is shown in Table S1 (available at www.ophthalmologyscience.org).

Both AA and VA showed significant negative correlations with age in each sex. Arteriolar area showed a more steeply declining regression on age in males than in females (Fig 1A, B). Venular area showed no sex difference in the steepness (Fig 1C, D). Both male and female AAs showed steeper declines with older age (Fig 1E, F). For AA in females, the difference in the slope of the regression line was the greatest when divided into those <54 years of age and ≥54 years old (−38.3 vs. −170.0 pixels/year, Fig 1F). The slope of the regression line of AA on age for the older women was comparable to that for men of the same age distribution, and the bending slope of the regression line before and after 54 years of age appeared to be actually less angled in men (−53.5 vs. −156.3 pixels/year, Fig 1E) than in women.

Figure 1.

The total retinal arteriovenous areas in males and females. A, Arteriolar area in males. B, Arteriolar area in females. Arteriolar area showed a negative correlation with age in each sex (P < 0.001, Pearson product–moment correlation), together with steeper regression on age in males than in females (P = 0.033, F test). C, Venular area in males. D, Venular area in females. Venular area showed a negative correlation with age in each sex (P < 0.001, Pearson product–moment correlation), together with no sex difference in the steepness of regression on age (P = 0.195, F test). E, Comparison of AA in younger and older male groups. F, Comparison of AA in younger and older female groups. The difference in the slope of the regression line was the greatest (P < 0.001, F test) when divided into the following 2 groups: those <54 years of age (orange dots and line) and ≥54 years old (purple dots and line). The bending slope of the regression line before and after 54 years of age was less angled in males than in females. Red dots in (A-D) represent the value of arteriolar area and the blue dots in (C and D) venular area.

Multiple stepwise linear regression revealed that AA in men was negatively correlated with age, systolic and diastolic blood pressure (SBP and DBP), and hypertension but positively with body mass index (BMI), arrhythmia, and smoking. Arteriolar area in women was negatively correlated with age, SBP, DBP, and diabetes mellitus but positively with BMI and smoking (Table S2, available at www.ophthalmologyscience.org). The β (standard regression coefficient) value was strongest for age in each sex but weaker in men than in women, likely because the absolute β values for SBP, DBP, and smoking were higher in men than in women, thus attenuating the relative correlation with age. Venular area in men was negatively correlated with age, SBP, DBP, and diabetes mellitus but positively with BMI and smoking. Venular area in women was negatively correlated with age, DBP, and diabetes mellitus but positively with BMI. The associations of high BMI with large AA and VA in both sexes would reasonably be explained by incremental blood volume in the systemic circulation. Moreover, current smoking also resulted in large AA and VA, potentially reflecting its diverse and often conflicting bioactions to the systemic vasculature via the autonomic nervous system.

Arteriolar area in younger women negatively correlated with DBP but positively with smoking, while AA in older women was negatively correlated with age, SBP, and DBP (Table S3, available at www.ophthalmologyscience.org). Older women had the higher absolute value of β in the order of age, SBP, and DBP. This trend of β values of AA in older women was equivalent to those of AA in men.

This study demonstrated the following findings: (1) AA in men decreased more steeply with age than that in women; (2) VA showed no significant sex difference; and (3) as concerns AA in women, the difference in the slope of the regression line of AA on age was statistically the greatest when divided at age 54, and the slope of the regression line of older women was comparable to that of men of the corresponding age distribution.

Previous reports have documented that men have narrower central retinal arteriolar equivalent than women,1, 2, 3 with a steeper age-related decline,² which was consistent with our current results. There are obvious sex differences in the epidemiology of diseases caused by arterio-arteriolosclerosis including lacunar infarct and hypertension, etc., and arteriosclerosis manifests itself in retinal arteriolar changes,⁵ suggesting that there are sex differences in retinal arteriolar changes as well. In contrast, VA showed no sex difference, which was consistent with no sex difference in central retinal venular equivalent.^1,2 The lack of sex differences in VA as well as central retinal venular equivalent is probably due to the anatomy and physiology of venular walls, which is thinner and more distensible than arteriolar walls and does not develop sclerotic changes.⁶

Interestingly, the difference in the slope of the regression line of AA on age between the younger and older women was statistically the greatest at age 54 as the cutoff value. This is supported by a survey on age at menopause reporting 80% of women reached menopause between 45 and 54 years of age (48.3 ± 4.8 years).⁷ Although we did not interview participants about menopausal age, which limited interpretation of our results, this was the first to show a difference in the degree of age-dependent decrease in AA between younger and older women, supporting the acceleration of arteriosclerosis.

In conclusion, AA decreased more steeply with age in men than in women, possibly reflecting the progression of sex-related arterio-arteriolosclerosis. The steep decrease in AA with age in older women reasonably suggested an association with sex hormone changes after menopause. The analysis of AA and VA may serve as a clinically significant platform for better understanding the multifactorial etiology of arterio-arteriolosclerosis.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author, M.M., and M.S. The data are not publicly available due to containing information that could compromise the privacy of research participants.