Fractal analysis of retinal microvasculature and cardiovascular outcome: a systematic review and meta-analysis

Abstract

Background

To access associations between retinal vascular fractal dimension (FD) and the risk of the development of cardiovascular disease (CVD) in this systematic review.

Methods

PubMed, Web of science, and Cochrane Library databases were searched for cohort and cross-sectional studies with data on the association between FD and CVD risk published up to June 2025. The combined risk estimates were estimated using the random-effects model. Subgroup analysis was performed according to pre-defined characteristics. Sensitivity analysis was conducted to examined the robustness of the results. This study is registered on PROSPERO, number CRD42024591961.

Results

Fourteen eligible studies (eight were cohort design and six were cross-sectional design) were included and analyzed. When FD was regarded as a continuous variable, the significant association between decreased FD and the incidence of the CVD (RR: 1.14, 95%CI: 1.05–1.24) was found. When FD was regarded as a categorical variable, the participants in the lower FD group at the baseline had the greater incidence risk of CVD compared to those in higher FD group (RR: 1.51, 95%CI: 1.07–2.14). Furthermore, people with reduced FD may inclined to have the established CVD (RR: 1.70, 95%CI: 1.23–2.37). The sensitivity analysis indicated that the above results are robust. We also found that when arterioles and venules were analyzed separately, the association between decreased arteriolar FD and CVD risk remained significant (cohort studies, RR: 1.08, 95%CI: 1.01–1.17; cross-sectional, RR: 1.90, 95%CI: 1.51–2.37), whereas no such association was observed in venular FD (cohort studies, RR: 0.96, 95%CI: 0.84–1.08; cross-sectional, RR: 1.42, 95%CI: 0.89–2.29).

Conclusion

The decreased retinal microvascular FD was significantly associated with the CVD risk, especially in retinal arteriole. The findings of this study supported that FD may be a potential marker of CVD risk including stroke and coronary heart disease. Noninvasive FD measurement may act as an adjunct to existing CVD risk stratification, facilitating early screening and diagnosis of individuals at elevated CVD risk.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12886-025-04492-z.

Introduction

Cardiovascular disease (CVD), including stroke, coronary heart disease (CHD), hypertensive disease, and peripheral vascular diseases, is one of the most common reasons of mortality and disability around the world [1]. A report from global burden of disease study presented that the incidence of CVD rose by 77.12%, escalating from 31.31 million in 1990 to 55.45 million cases in 2019 [2]. Concurrently, mortality rates increased by 53.81%, from 12.07 million to 18.56 million deaths, from 1990 to 2019. Despite the identification of numerous risk factors, early diagnosis CVD continues to pose a significant challenge. The common diagnostic methods exhibit delays and invasiveness. Retinal microvasculature, as a kind of blood vessel that can be observed from vitro directly, are anatomically and physiologically homologous to the microvasculature of the heart and brain. Tedeschi-Reiner et al. observed the significant association between the retinal arteriolar atherosclerosis and coronary artery severity [3]. These findings endorse the notion that the retinal microvasculature offers a valuable approach for gaining insights into the structure and function of the vessels. Therefore, retinal microvasculature might act as promising biomarkers for evaluating microvascular impairment associated with the progression of CVD.

Fractal analysis serves as a method for assessing the branching pattern and blood flow density of the retinal vasculature. The outcome of this measurement is the fractal dimension (FD), an indicator that reflect the complexity of the vascular network. A pilot study found that the retinal FD was associated with the cerebral blood flow [4]. Recently, several studies have found the relationship between the FD and the CVD outcome, including both cross-sectional and cohort studies [5–13]. For example, Mordi et al. suggested that FD was an independent risk factor of CVD in a 10-year follow-up duration [7]. Another study conducted by Grauslund et al. found that FD was a biomarker of the established CVD [14].

However, the studies reported that relationship between FD and CVD risk have not always been consistent. The contribution of FD to CVD risk stratification whether in terms of incidence or prevalence, lacks robust epidemiological validation. Some research did not observe a significant association between FD and CVD outcome [14–17]. Furthermore, several studies have demonstrated that there are differences in the associations between retinal arteries and veins and CVD risk [7, 9, 15]. Moreover, the association of FD with different CVD outcomes also varied. Sandoval-Garcia et al. observed the significant association between FD and stroke, but not in CHD [13].

The inconsistencies in the aforementioned findings leave uncertainty regarding how FD alterations relate to elevated CVD risk and which specific CVD types are involved. However, few studies have systematically reviewed this relationship. We therefore conducted a meta-analysis of relevant observational studies to evaluate the association between FD and CVD risk. In the present study, we not only innovatively explored cohort and cross-sectional studies, but also analyzed aforementioned association when FD as a continuous variable and a categorical variable separately. Furthermore, in subgroup analyses, we summarized changes in that association when investigating retinal arterioles and venules individually, as well as across different CVD outcome types. We hypothesized that FD is associated with CVD, and that this relationship varies by vessel types and specific CVD outcomes.

Methods

Literature search

We conducted a meta-analysis following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The study protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO) under registration number CRD42024591961. A comprehensive search of PubMed, Web of science and Cochrane library was performed for article published up to June 2025. The following relevant search terms were used including “fractal dimension”, “fractal analysis”, “fractals” and “cardiovascular diseases”, “cerebrovascular disease”, “cerebrovascular disorders”, “stroke”, “coronary heart disease”, “coronary artery disease”, “hypertension”, “peripheral arterial”, “heart failure”, “atherosclerosis”. The detailed search strategy was presented in Table S1. We restricted our search to human species and English paper.

Selection criteria

Studies were included if they met the following inclusion criteria: (1) had a cohort or cross-sectional design; (2) the imaging source was fundus photographs; (3) the exposure is the decreasing retinal microvascular FD at baseline; (4) the outcome was CVD, including stroke, coronary heart disease, hypertension, peripheral arterial disease, heart failure and CVD-related death. (5) provided the quantitative multivariate-adjusted risk estimates with the corresponding 95% confidence interval (CI), including risk ratio (RR), hazard ratio (HR) and odds ratio (OR). The following were the exclusion criteria: (1) unadjusted or only age or sex adjusted risk estimates were reported; (2) case reports, noncontrolled case series, conference abstract, review articles and incomplete reports; (3) lack of necessary data. The selected literatures are imported into Endnote 21. After removing duplicate literatures, two reviewers screen the literatures separately. When there is a disagreement, a third reviewer will intervene to resolve it.

Definition of fractal dimension

FD was calculated through the box-counting method. This method involved segmenting the retinal fundus image into equal-sized squares and counting the number of squares that contained vascular segments [4]. This process was repeated for squares of different sizes, and FD was calculated as the logarithmic gradient of the number of boxes containing blood vessels and the size of the boxes (Figure S1). Most studies defined the field of view (FOV) as the peripheral region centered on the optic disc, spanning either 3.5 optic disc radii or distances of 0.5 and 2.0disc diameters from the disc center. The corresponding retinal vascular analysis systems are provided in Table S1. Most semi-automatic fundus image-based vascular measurement tools distinguish retinal arterioles from venules by combining manual intervention with features such as color and anatomical characteristics, while fully automatic tools primarily utilize machine learning algorithms to analyze features including color, intensity, and vascular crossings for automatic arteriovenous classification [18].

Identification of CVD

The CVD outcomes were ascertained through medical records, clinical examinations or self-reports. Medical records, primarily retrieved from data registered in healthcare institutions, include documentation of typical symptoms, examination results, diagnostic findings, and records of surgical interventions or pharmacological therapies. These records are further coded in accordance with the International Classification of Diseases, 9th Revision (ICD-9; 390–459) or 10th Revision (ICD-10; I00–I99). In some studies, participants underwent computed tomography, magnetic resonance imaging, coronary angiography, or other relevant examinations to confirm CVD outcomes. Self-reported sources were also obtained through standardized questionnaires (Table S2).

Data extraction

We extracted the data from the published articles into the Microsoft Excel sheet. Two independent reviewers extracted the following items: first author, study design, publication year, study region, population size, number of CVD event, demographic characteristics of study participants, follow-up period, assessment and analysis systems of FD, ascertainment of CVD outcome, risk estimates and 95% CI and adjusted covariates (See Supplement Method for details). When necessary, the study authors were contacted for supplement information.

Assessment of quality

The quality of all studies that met the inclusion criteria was assessed. Two people independently assessed study quality using the Newcastle–Ottawa scale (NOS) for cohort studies and Agency for Healthcare Research and Quality (AHRQ) for cross-sectional studies. The NOS consist of three sections, with the grade from 0 to 9. The studies scoring 0–3, 4–6 and 7–9 were considered low, moderate, and high quality, respectively. The AHRQ contains 11 items, and when more items were met, the quality of the study was higher.

Statistical analysis

Data from cohort studies and cross-sectional studies were pooled separately. HRs and ORs, considered equivalent to RRs. The RR of the CVD risk was analyzed when FD was used as a continuous and categorical variable, respectively. We employed a random-effects model to synthesize the RR estimates which considers within-study and between-study variabilities. The heterogeneity was evaluated by the tau-squared (τ²). The amount of the between-study variance attributed to heterogeneity was defined by I². Cochran’s Q statistical test was performed to assess the statistical significance of heterogeneity. Subgroup analyses by CVD outcome, follow-up time, population, retinal fundus analysis systems, retinal vessels (arteriole or venule), baseline CVD excluded or not, sample size were conducted. Sensitivity analyses were performed by removing single study to evaluate the impact that each study has on the overall estimates. Studies with only self-reported outcomes, non-optic disc-centered fields, non-mainstream analysis systems, or those conducted solely in the diabetes population were excluded to assess result robustness. Publication bias was assessed by symmetry of the funnel plot and Egger’s test. If publication bias was indicated, the “nonparametric trim-and-fill” method was used to measure the RR corrected for bias. P < 0.05 was considered statistically significant for the two-tailed test. All analysis process were performed using R software, version 4.4.1 (Figure S2).

Results

Characteristics of the studies

The process and results of the literature selecting is presented (Fig. 1). Following the inclusion criteria, fourteen studies exploring the association between the FD and CVD risk [5–11, 13–15, 19]. Prior to this, two studies were excluded because they only explored the association between FD and lacunar stroke subtypes in the stroke population [20, 21]. Six of them only included FD as a continuous variable to explore the association, two studies only included FD as categorical variables (lower FD compared to higher FD), and six studies included FD as both continuous and categorical variables. Eight of the studies were cohort design and six were cross-sectional design. Four studies addressed multiple CVD outcomes. Two studies discriminated the CVD outcome, including coronary events, stroke events, and hypertension [13, 15], whereas two studies did not [7, 14]. The remaining studies focused on single outcome: stroke in five studies [5, 6, 9, 10, 19], coronary events in three studies [8, 12, 16], hypertension in one study [11], and heart failure in one study [17]. Table S2 shows the characteristics of studies. These studies were published between 2010 and 2023, covering populations from Asia (6 studies, 52.1%), Australia (5 studies, 34.7%), and Europe (3 studies, 13.2%). Most of them had moderate to large sample sizes. The following-up duration of cohort studies ranged from 5 to 13.1 years. Study quality assessed by the NOS and AHRQ presented that included studies had a high quality (Table S3-S4).

Fig. 1.

Flowchart of study selection

Association between retinal fractal dimension and the risk of cardiovascular disease

These included studies were sufficiently similar to allow meta-analysis of the relationship between FD and CVD risk. The multivariate-adjusted RRs with 95% CI of FD associated with CVD from individual studies and the combined RRs are presented (Figs. 2 and 3). When FD was regarded as a continuous variable, the results showed that the decreased FD were significantly associated with the increased incidence of CVD (RR: 1.14, 95% CI: 1.05–1.24, per unit decrease, Fig. 2). However, we did not find the significant association between decreased FD and prevalence of established CVD (RR: 1.25, 95% CI: 0.92–1.70, per unit decrease, Fig. 2). When FD was regarded as a categorical variable, participants with lower FD at baseline also presented a greater incidence and prevalence of CVD (RR: 1.51, 95% CI: 1.07–2.14; RR: 1.70, 95% CI: 1.23–2.37, low versus high FD, Fig. 3). The heterogeneity ranged from low to moderate.

Fig. 2.

Association of decreased retinal microvascular fractal dimensions with cardiovascular risk

Fig. 3.

Association of lower retinal microvascular fractal dimensions group with cardiovascular risk compared with higher fractal dimensions group

There are four studies explore the relationship between the FD and incidence of stroke individually [5, 6, 13, 19]. Overall, combined results showed that the participants with lower FD presented significantly increased incidence of stroke (RR: 1.23, 95% CI: 1.03–1.47, per unit decrease; RR: 1.85, 95% CI: 1.02–3.34, low versus high FD, Fig. 4). In addition, similar associations were observed when combining the two CHD-related studies (RR: 1.25, 95% CI: 1.02–1.53, low versus high FD, Fig. 4). Furthermore, two cross-sectional studies revealed that decreased FD was associated with an increased prevalence of established stroke [9, 10], and the combined results remained significant (RR: 1.90, 95% CI: 1.51–2.37, per unit decrease, Table S5). While only one study exhibited the association between the decreased FD and slightly greater prevalence of established CHD (RR: 1.01, 95% CI: 1.00-1.02, per unit decrease, Table S5) [16].

Fig. 4.

Forest plot of the association of decreased retinal microvascular fractal dimensions with the incidence of the stroke and coronary heart disease

Subgroup analysis

The subgroup analyses of cohort studies were shown in Table 1. When FD was regarded as a continuous variable: reduced FD was significantly associated with CVD risk in the period exceeding 10 years, but not in the period of less than 10 years. An increased incidence of CVD was observed with reduced FD in retinal arterioles, but not in venule. Furthermore, subgroup analysis by the population showed the significant association in Australian and European people, while not in Asian. Subgroup analyses of cross-sectional studies were shown in Table S5. When FD was regarded as a continuous variable, the significant association between decreased arteriolar FD and established CVD remains valid, with no significant association observed in venules. Subgroup analysis focusing on FD as a categorical variable was not carried out in cross-sectional studies, largely attributable to the insufficient number of investigations addressing this variable classification.

Table 1.

Subgroup analyses of relative risk of cardiovascular outcome incidence associated with decreasing retinal microvascular fractal dimension

Subgroup	Number	Decreased FD#				Number	Lower FD group vs. higher FD group$
		RR (95%CI)	P value	I2, %	P value		RR (95%CI)	P value	I2, %	P value
All study	7	1.14(1.05, 1.24)	0.001	27	0.221	6	1.51(1.07, 2.14)	0.018	65	0.014
CVD outcome
Stroke	4	1.23(1.03, 1.47)	0.023	43	0.154	4	1.85(1.02, 3.34)	0.043	54	0.088
CHD	2	1.13(0.90, 1.41)	0.302	65	0.092	2	1.25(1.02, 1.53)	0.030	65	0.091
HTN	1	0.97(0.84, 1.20)	0.678	/	/	0	/	/	/	/
Follow-up time (year)
< 10	4	1.12(0.97, 1.31)	0.130	27	0.248	3	1.64(0.76, 3.52)	0.206	64	0.060
>= 10	3	1.16(1.04, 1.29)	0.007	51	0.128	3	1.40(1.00, 1.98)	0.051	69	0.040
Population
Australian	3	1.30(1.12, 1.51)	< 0.001	0	0.503	3	1.69(1.21, 2.35)	0.002	29	0.245
Asian	2	1.03(0.83, 1.26)	0.808	1	0.315	1	0.74(0.33, 1.67)	0.467	/	/
European	3	1.11(1.03, 1.19)	0.006	8	0.339	2	1.52(0.77, 3.02)	0.227	76	0.043
Analysis system
IRIS	1	1.26(1.06, 1.51)	0.010	/	/	2	1.65(1.06, 2.58)	0.028	41	0.192
SIVA	2	1.03(0.83, 1.26)	0.808	1	0.315	1	0.74(0.33, 1.67)	0.467	/	/
Other	1	1.42(1.05, 1.92)	0.023	/	/	1	2.43(1.04, 5.68)	0.040	/	/
VAMPIRE	2	1.08(1.02, 1.16)	0.017	0	0.825	1	2.38(1.19, 4.76)	0.014	/	/
RMHAS	1	1.27(1.04, 1.56)	0.021	/	/	1	1.16(1.03, 1.18)	< 0.001	/	/
Retinal vessels
Arteriole	2	1.08(1.01, 1.17)	0.029	16	0.091	0	/	/	/	/
Venule	2	0.96(0.84, 1.08)	0.482	18	0.269	0	/	/	/	/
Baseline CVD exclude
No	3	1.16(1.04, 1.29)	0.006	0	0.508	3	1.73(1.23, 2.44)	0.002	37	0.204
Yes	4	1.15(0.98, 1.35)	0.092	52	0.099	3	1.24(0.69, 2.24)	0.477	51	0.129
Sample size
< 5000	5	1.16(1.03, 1.31)	0.017	26	0.246	5	1.68(1.11, 2.54)	0.014	47	0.112
>=5000	2	1.14(0.98, 1.32)	0.088	54	0.165	1	1.16(1.03, 1.18)	< 0.001	/	/

^#Decreased FD means FD as a continuous variable; ^$Lower FD group versus higher FD group means FD as a categorical variable; FD, fractal dimension; CVD, cardiovascular disease; CHD, coronary heart disease; HTN, hypertension; RR, relative risk. SIVA, Singapore I Vessel Assessment; IRIS, International Retinal Imaging Software; VAMPIRE, Vessel Assessment and Measurement Platform for Images of the Retina; RMHAS, Retina-based Microvascular Health Assessment System

Sensitivity analysis and publication bias assessment

We conducted sensitivity analysis to assess the robustness of the results. The combined RRs remained largely unchanged when studies were excluded one at a time (Figure S3). Studies with only self-reported outcomes, non-optic disc-centered fields, non-mainstream analysis systems, or those conducted solely in the diabetes population were excluded [5, 7, 13–15], the combined results were still almost robust respectively (Table S6). Publication bias was evaluated by observing the symmetry of the funnel plot. The funnel plots were visualized to be symmetrical, and the Egger’s tests found no significant publication bias (Figure S4). The certainty of evidence for the critical associations of this study was assessed as moderate (See Supplement Result for details).

Discussion

This meta-analysis has reviewed the relationship between the reduced FD and the risk of the CVD. Fourteen eligible studies (eight were cohort design and six were cross-sectional design) were included and analyzed, which comprising 7,389 CVD cases. Furthermore, the association remained significant in analyses only for stroke and CHD.

In the past decade, a number of studies have investigated the role of FD in cardiovascular and cerebrovascular diseases. Some studies provided effect sizes for the combined risk of CVD. Mordi et al. found that the reduced FD can be the independent predictor of 10-year major adverse cardiovascular events, including CVD death, myocardial infarction, and stroke, with an 8% increase in event risk for 1 SD decrease in FD [7]. Sandoval-Garcia et al. observed a negative association between FD and the incidence of CVD in a cohort followed for eight years, although this association was not statistically significant. However, the study found that the risk of cerebrovascular disease increased by 40% as per SD decrease in FD after multivariable adjustment, and the risk of cerebrovascular disease was reduced by 58% in the highest tertile of FD compared with the lowest tertile. In addition, the inclusion of FD in the basic model with traditional vascular risk factors improved its discrimination efficiency for cerebrovascular events [13]. In addition, Betzler et al. and Grauslund et al. assessed the correlation between FD and combined risk of CVD, but found no marked association [14, 15].

By extension, our results indicated that participants with higher FD have an 85% increased risk of developing a stroke. Two previous systematic reviews have also qualitatively explored the association between reduced FD and increased risk of stroke [22, 23]. One study included assessed the circular FD, a method distinct from the box counting FD employed in this meta-analysis. Their findings indicated that the complexity of the retinal vasculature was significantly reduced in stroke patients as compared to the control group [24]. This result was similar to our findings. Stroke is categorized into ischemic and hemorrhagic types. Ong et al. reported a significant association between reduced FD and ischemic stroke, as well as its principal etiologic subtypes [9]. Nevertheless, there remains a scarcity of research examining the relationship between FD and hemorrhagic stroke. Furthermore, several studies have also revealed differences in retinal characteristics between stroke type. Doubal et al. reported that decreased microvascular complexity in lacunar stroke compared to mild cortical stroke [21]. In contrast, another study observed an increased FD in patients with lacunar stroke compared to patients with other types of acute ischemic stroke [20]. In the present meta-analysis, we only included studies that measured the FD of retinal photographs, and there were previous studies that measured FD by optical coherence tomography angiography (OCTA). Tao et al. employed FD to assess the complexity of the superficial and deep retinal vasculature among aging adults (aged ≥ 50 years). They observed a significant correlation between reduced FD in the deep retinal layer and the presence of lacunes [25]. A separate study demonstrated a marginal association between decreased FD in the deep capillary plexus and the occurrence of ischemic stroke. However, no significant association was identified between FD in either the deep or superficial capillary plexus and lacunar infarction [26].

Our review of articles examining the association between FD and coronary events revealed that three were cohort studies and one was cross-sectional. In present results, patients with lower FD have a 25% increased risk of CHD. In addition to the included studies, Dinesen et al. found that eyes in the fourth quartile with higher FD were less likely to have coronary artery bypass grafting compared to patients in the first and second quartiles after adjusted for gender and age [27]. In another study, Fu et al. assessed retinal vascular complexity in the fundus by employing not only the FD but also parameters such as vessel segment number, vascular skeleton density, and vessel area density. Their findings indicated that reduced vascular complexity and density within the retinal vascular network could be indicative of a heightened risk for CHD [8]. However, only one study has investigated the association between FD and heart failure prevalence, and no significant association was found [17]. In recent years, some studies have explored the association between FD and the risk of hypertension. Liew et al. observed that retinal vessels were significantly sparser and the FD was lower in hypertensive patients, with this difference being statistically significant [28]. MacGillivray et al. also found the consistent results in 40 fundus images [29]. Xu et al. employed OCTA to assess the vascular complexity of the fundus in 137 hypertensive patients and 79 healthy individuals. They discovered that the FD of both the superficial and deep retinal layers was significantly reduced in hypertensive eyes compared to those in healthy eyes [30]. In addition, several studies have also shown a negative association between blood pressure level and FD [31, 32].

Although many studies have provided evidence for the association between FD and CVD risk, the underlying mechanism of this association remains unclear. Impaired endothelial function, a prevalent risk factor for both retinal blood perfusion anomalies and CVD, impedes the preservation of vascular network integrity [33]. Atherosclerosis, caused by endothelial dysfunction, is a chronic inflammatory condition that affects blood vessels throughout the body. This condition triggers blood vessel wall hardening, narrowing, and a loss of elasticity, culminating in secondary thrombosis [34]. In cases of vessel occlusion, perfusion to the retinal microvasculature is compromised, causing tissue hypoxia and reduced collateral circulation. This may account for the observed the thinning of the retinal microcirculation and decreased complexity of retinal blood vessels. Retinal vessels have similar anatomical and physiological structures to those of the brain and heart [6]. As part of the systemic circulation vessels, retinal vessels represent the changes of peripheral vessels. Therefore, deviations in FD suggest widespread microvascular damage. In this light, the aforementioned pathological mechanisms may explain the diminished complexity of the retinal vessel network associated with increased CVD-related risk.

Furthermore, to explore the potential sources of heterogeneity, we performed subgroup analysis. Duration of follow-up, type of retinal vessel (arteriole or venule), sample size, and status of having or not having a prior CVD may account for the source of the heterogeneity. Previous studies have shown that compared with retinal venule, arteriolar FD are more significantly associated with CVD incidence and disease risk [7, 9, 10, 15]. These were consistent with the results of our subgroup analyses. However, most current studies have focused on the overall vascular network of the fundus, and few studies have focused on the measurement of retinal arteriolar parameters alone. McGrory et al. observed a significant correlation between reduced arteriolar FD and deep white matter hyperintensity scores [35]. It also provided new evidence that retinal arterioles may be more suitable as a marker of CVD risk. Based on previous literature reports, age and sex were also important influencing factors. A meta-analysis showed that the associations between retinal microvascular changes and CHD were more evident in relatively young age and female patients [36]. This phenomenon may be related to the poor regulation of retinal vessels in the elderly and the sparse retinal vessels in women [16, 37]. Nevertheless, a definitive explanation for this occurrence remains elusive. There is a need for further well-structured, prospective research that incorporates stratification by age and sex groups.

Measurement heterogeneity in retinal vascular FD assessment existed across studies, involving imaging devices, preprocessing protocols and analysis software (Table S2). Most used 45° cameras (Topcon TRC-NW8 and Canon CR-DGi), while 2 Australian studies adopted 30° Zeiss FF3 cameras. Smaller 30° view may underestimate FD, weakening FD-CVD risk associations. Excluding peripheral vessels of FOV (e.g., 0.5–2.0disc diameter) may also contain sparser vasculature, underestimating FD-CVD associations. All included studies incorporated preprocessing steps such as image calibration or noise reduction. The absence of these steps may potentially either attenuate (reducing signal clarity) or inflate (introducing systematic bias) the observed effects. Four systems (VAMPIRE, SIVA, IRIS, RMHAS) were applied. Semi-automated analysis software may need manual intervention for partial vascular segmentation and reference point marking, introducing inter-observer variability that could impact pooled results across studies. However, no marked differences between fully and semi-automated system studies were observed herein. Our study further showed that the observed association effects varied across different ethnicities, which may be attributed to the underlying heterogeneity in retinal vascular characteristics among distinct populations [38, 39]. Significant associations were observed only in the European and Australian populations herein. The generalizability of the finding in non-western populations is restricted. Additionally, the mean age of the study population in most studies ranged from 55 to 65 years, which may limit the generalizability of the observed associations to younger populations. Therefore, future studies should refine FD measurement consistency, expand ethnic and age diversity, and include participants with varied comorbidities to boost result generalizability.

There were some limitations in our meta-analysis. First, the results of meta-analyses of observational studies may be affected by residual confounding or unmeasured variables, such as unmeasured lifestyle-related factors. Furthermore, we acknowledge that the evidence from this study does not establish a causal relationship. Second, the moderate number of studies on different CVD outcomes included in this study made the number of studies included in each subgroup relatively small, which may affect the accuracy of the combined results. Third, studies from the same research team may have overlapping populations, inflating estimates of combined risk. Despite these studies examined different CVD outcomes, and we endeavored to contact authors to identify overlapping cohorts, some overlap possibly remained. Forth, the studies we searched were only in English, and some other articles published in non-English that were related to the topic may have been missed.

In conclusion, we summarized the relationship between FD and CVD risk. The results showed that reduced FD was significantly associated with the incidence and prevalence of CVD outcomes. As an adjunct to existing CVD risk stratification rather than a full replacement, noninvasive FD measurement may provide potential incremental value to current CVD risk stratification tools and facilitate early screening of high-risk individuals for conditions such as stroke and CHD.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Y.L., G.L. and H.L. searched, screened and collected the included literature. Y.L. pooled and analyzed the data. Y.L. prepared all the figures and tables. Y.L. and Q.Z. wrote the main manuscript text. G.L. and H.L. provided review comments for the writing of the manuscript. Q.Z. designed and provided financial support for the study.

Funding

The publication fee (APC) will be covered by our new project supported by the National Natural Science Foundation of China (grant numbers 82571212), as the previous project (grant numbers 82070960) has been completed and closed.

Data availability

The datasets, analytic code, and other supporting materials generated or analyzed in this study are available to qualified researchers upon reasonable request. The author responsible for material sharing is Qing Zhang, affiliated with Beijing Institute of Ophthalmology, Beijing Tongren Hospital, Capital Medical University, Beijing, China, who can be contacted via email at qingzhang810@vip.sina.com.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.