We conducted a systemic review and meta-analysis to assess the OCTA measurements in AD and MCI, in accordance with the guidelines presented by the Cochrane Collaboration TC (2011) and Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guideline (Stroup et al., 2000).

We included studies that fulfilled all the inclusion criteria listed below:

The exclusion criteria were as follows:

Two independent reviewers (TCY and CTK) conducted a systematic and comprehensive search in PubMed and EMBASE database from 1 January 2014 to 1 May 2021. We used the medical subject headings (MeSH) for the PubMed query with the following (Tomography, Optical Coherence [MeSH] AND (Retinal Diseases[MeSH Terms]) OR (Retinal Vessels[MeSH Terms])) AND (((((Dementia[MeSH Terms]) OR (Alzheimer's Disease[MeSH Terms])) OR (Cognitive Dysfunction[MeSH Terms])) OR (“Cognitive Impairment”)) OR (“Cognitive Decline”)). The literature search was limited to English-written full-text manuscripts that were published in peer-reviewed journals. The list of the detailed search strategy is presented in Figure 1 in accordance with the Preferred Reporting Items for Systematic Reviews and Meta- Analysis (PRISMA). To reduce the chance of missing relevant articles, we also manually searched the reference lists of all gray literature, including relevant articles and reviews. No ethical approval was required for this review study.

Two reviewers (TCY and CTK) independently read each article to ascertain its eligibility and extracted data. Disagreements were resolved by discussions with a senior reviewer (CYB). Extracted information included authors, the title of the study, publication year, country, method of eye selection (i.e., analysis including one eye or both eyes of each subject), number of subjects in each population, mean age, gender ratio, mean MMSE score, OCT model, macular scan size (3x3 or 6x6 mm²), method of VD assessment (whether provided by the device or calculated by software or manually), and minimum SSI accepted for a quality scan. The Newcastle-Ottawa Scale criteria were used to assess the quality of the included case-control studies, with a scale range of 0–9 points (Table 1; Stang, 2010; Wells et al., 2016).

The primary outcome variables of OCTA measurements (in mean and SD) include FAZ, VD in SCP, DCP, and peripapillary. The value of each sub-sector of the measured area, if any, was also extracted. We calculated the average values of multiple subsectors using Review Manager Software (version 5.3; Cochrane Collaboration, Oxford, United Kingdom).

We performed meta-analyses, heterogeneity analysis, subgroup analyses, assessment of publication bias, and meta-regression by Stata software (StataCorp, Texas; version 16.0). Means and standard deviations were used to calculate the standard mean difference (SMD) with respective 95% confidence interval (CI) for continuous variables. We applied random-effect models in all meta-analyses with the restricted maximum likelihood (REML) method. The REML estimation has previously been advocated for continuous outcomes (Veroniki et al., 2016). We tested between-group heterogeneity using the I² test. The I² statistic ranges from 0 to 100%, with 25, 50, and 75% indicating low, moderate, and high between-study heterogeneity. In addition, we assessed potential publication bias using Egger's Tests.

Random-effects meta-regression was performed to assess the potential confounders on the effect sizes. The included variables were mean MMSE scores, mean age differences between study groups, gender ratio differences, OCT model, and macular scan size. We performed subgroup analyses according to the types of OCTA model and macular scan size. Additionally, we performed sensitivity analyses to account for unmeasured confounding, as random-effects meta-analyses of observational studies can result in biased estimates if the synthesized studies contain unmeasured confounding (Mathur and VanderWeele, 2020). We determined the minimum strength for an unmeasured confounder on the risk ratio scale (E-value) required to eliminate the SMDs between case and control groups. The E-value was defined as the minimal magnitude of association that an unmeasured confounder needs to have with both the exposure and the outcome to fully explain a specific exposure-outcome association (Mathur and VanderWeele, 2020). P-values < 0.05 were indicated as statistically significant, and all statistical analyses were two-tailed.

A risk of bias assessment for each study was performed by applying a standardized QUADAS-2 assessment (Figure 2). TCY and CTK independently assessed the risk of bias, and disagreements were resolved by discussions with a senior reviewer (CYB). Egger's test was used to evaluate publication bias.

Figure 1 illustrates the selection process for the 16 studies included in the meta-analysis. A total of 1041 relevant articles were initially identified across all databases. After removal of duplicates, the remaining 844 studies were screened based on their titles and abstracts. A total of 828 studies were then excluded for various reasons including those not related to AD, MCI, and OCTA, yielding 16 case-control studies involving 444 AD subjects, 391 MCI subjects and 770 controls ultimately eligible for the quantitative analysis. The study by van de Kreeke et al. (2020) was not included in the meta-analysis as it did not report average vessel density between groups. In addition, although Jiang et al. (2018b) measured SCP and DCP in both AD and MCI patients, the original data was not provided to conduct meta-analyses. Despite the possible duplicate populations of authors such as O'Bryhim et al. (2018, 2021), we extracted different outcome variables in these studies. The relevant studies excluded due to insufficient data to perform meta-analyses were still included in the systemic review, and the main characteristics of all studies are presented in Table 2.

In all studies, there were no significant differences in mean age between the AD, MCI, and control groups. The mean MMSE scores in the control group ranged from 24.8 to 30.0, whereas scores in the MCI group ranged from 22.6 to 28.0, and scores in the AD group ranged from 16.9 to 23.0. Regarding the OCTA models, nine studies used the RTVue XR Avanti (Optovue, Inc., Fremont, CA, USA; Lahme et al., 2018; O'Bryhim et al., 2018, 2021; Zabel et al., 2019; Zhang et al., 2019; Criscuolo et al., 2020; Wu et al., 2020; Wang et al., 2021), six studies used the Zeiss Cirrus HD-5000 with AngioPlex (Carl Zeiss Meditec, Dublin, CA; Haan et al., 2019; Querques et al., 2019; Yoon et al., 2019; Chua et al., 2020; Robbins et al., 2021; Shin et al., 2021), and one study used Heidelberg Spectralis (Salobrar-Garcia et al., 2020).

Vessel density of SCP significantly lower in the AD and MCI groups compared to controls (Figures 3, 4). Microvascular densities of SCP were significantly lower in AD vs. controls (SMD, −0.48; 95% CI, −0.70 to −0.27; p = 0.04; I² = 48.21%), and MCI vs. controls (SMD, −0.42; 95% CI, −0.81 to −0.03; p = 0.03; I² = 79.29%). To nullify the above pooled SMD estimates of vessel density in SCP, an unmeasured confounder would need to be associated with a risk ratio of 2.47 and 2.29 for AD or MCI risk and OCTA retinal measurements, respectively.

The results of subsector and quadrantal analyses revealed a significantly superficial microvascular loss between AD and healthy controls in the fovea (<1 mm from the fovea; SMD, −0.38; 95% CI, −0.66 to −0.10; p = 0.01; I² = 36.57%), and the parafovea region (1.0–3.0 mm from the fovea; SMD, −0.48; 95% CI, −0.71 to −0.25; p < 0.001; I² = 41.55%) (Supplementary Figures 1–3). The E-values would be 2.18 and 2.47 for AD risk and OCTA measurements to nullify the above pooled SMDs. There was no significant difference in superficial microvascular density in other regions, including superior, inferior, nasal, and temporal.

Vessel density of DCP was significantly decreased in the AD and MCI groups compared to controls (Figures 3, 4). Microvascular densities of DCP were significantly decreased in AD vs. controls (SMD, −1.19; 95% CI, −2.00 to −0.38; p < 0.001; I² = 92.38%), and MCI vs. controls (SMD, −0.53; 95% CI, −0.85 to −0.22; p < 0.001; I² = 59.68%). To nullify the above pooled SMD estimates of vessel density in DCP, an unmeasured confounder would need to be associated with a risk ratio of 5.35 and 2.62 for AD or MCI risk and OCTA measurements, respectively. The results of subsector and quadrantal analyses did not show significant differences in deep microvascular loss between AD and healthy controls, which may be due to the limited numbers of studies reporting the subsector DCP in subjects with AD or MCI.

Vessel density of RPC revealed no significant difference in the AD and MCI groups compared to controls (Figures 3, 4).

The area of the FAZ was significantly larger in AD subjects compared to controls (SMD, 0.54; 95% CI, 0.09 to 0.99; p = 0.02; I² = 84.07%) (Figure 3). To nullify the pooled SMD estimates of FAZ, an unmeasured confounder would need to be associated with a risk ratio of 2.65 for AD risk and OCTA measurements. There was no statistical difference in the FAZ area between either AD subjects and MCI subjects or MCI subjects and controls (Figures 4, 5).

There was no statistical difference in choroidal thickness between either AD subjects and MCI subjects compared to controls (Figures 3, 4).

We performed meta-regression on studies comparing mean MMSE scores, mean age, gender ratio, OCT model, and macular scan size macular thickness between AD, MCI, and controls (Table 3). Our results showed significant associations between the type of OCTA model and the effect sizes of the foveal avascular zone between AD and controls (β = 0.923; p = 0.008; r² = 53.72%). Furthermore, there were also significant associations between the macular scan size and the effect sizes of the area of foveal avascular zone between MCI and controls (β = 0.543, p = 0.030; r² = 71.57%), the effect sizes of the SCP differences between AD and controls (β = 0.422, p = 0.008; r² = 100%), and effect sizes of the DCP differences between AD and controls (β = −1.438, p = 0.031; r² = 47.34%). All other factors had no significant impact (p > 0.05) on the effect sizes of the differences of VD in SCP and DCP or FAZ in AD, MCI groups.

Subgroup analyses were performed to compare the effect sizes between studies using RTVue XR Avanti and Cirrus 5000 Angioplex, and studies with a macular scan size of 3x3 and 6x6 mm. We observed that the differences of FAZ area between AD and controls were only statistically significant (p < 0.05) among studies using RTVue XR Avanti and a macular scan size of 6x6 mm. Likewise, the differences in FAZ area between MCI and controls were only statistically significant among studies with a macular scan size of 6x6 mm.

Regarding VD, the differences of SCP between MCI and controls was only statistically significant among studies with a macular scan size of 6x6 mm (Supplementary Figures 4, 5).

Egger's tests showed no publication bias in most of the analyses (P ≥ 0.05) (Table 4), except for the analyses of the FAZ area and DCP vessel density between AD and healthy controls. Most of the eligible studies were of low risk of bias (Table 5, Figure 2).

The strengths of this meta-analysis lie in the comprehensive search and review, including the latest related studies, which helped us strengthen the findings of each study and obtain stronger evidence. Furthermore, this is to date the largest meta-analysis evaluating the role of OCTA in detecting early retinal microvasculature changes, including not only AD but MCI individuals. As recent studies suggest a critical window to early initiation of therapy, including MCI subjects may be essential to identify early pathologic biomarkers (Chalermpalanupap et al., 2013). Our study also provides a more precise estimate of the effect size and increases the generalizability of individual studies by performing stringent meta-regression and subgroup analyses. Even though heterogeneity remains among the eligible studies, we attempted to adjust for possible confounding bias in the meta-regression analysis to address the heterogeneity.

Our study has some limitations, and we have highlighted several knowledge gaps and priority needs for future research. First, there was a limited number of prospective studies included in our study. More prospective and longitudinal studies are required to provide comprehensive insights into the disease mechanism. Second, the number of studies may not be sufficient to achieve adequate statistical power in some sub-sector analyses, although the consistent trend of microvascular changes observed in AD and MCI subjects was still evident. Third, as previously discussed, pre-clinical AD and MCI subjects in different studies were not well-defined due to lack of disease-specific biomarkers and uniformed diagnostic criteria. Finally, parameters for image acquisition among different OCTA models should be documented in a standard format to compare these data across institutions reliably.

The retina provides a unique opportunity and correlation with AD. With the rise of OCTA, the potential biomarkers for early detection of AD are brighter than ever. Our findings in this meta-analysis further emphasize the role, contribution, and importance of the retinal microvasculature in the pathogenesis of AD, and may provide a roadmap for other neurodegenerative diseases.