Introduction

Globally, it is estimated that ~2.2 billion people live with some form of vision impairment and/or eye disease, with at least 1 billion of these having preventable visual impairment [1, 2]. The leading causes of visual impairment include several eye diseases, including cataract, glaucoma, and diabetic retinopathy [3], with prevalence rates accelerating over the last 10 years due to population growth and ageing. There are also large differences in eye disease prevalence depending on geographic location, with the greatest prevalence being in low income countries [3].

A large body of literature reports that those with eye disease may be at a higher risk of physical and mental health complications when compared to those who are normally sighted (e.g. mobility limitations [4], chronic kidney disease [5], gout [6], obstructive sleep apnoea [7], depression [8], lower cognitive function [9], and suicidal behaviour [10]) and, importantly, increased risk of cardiovascular disease mortality [11, 12].

Given the incidence, morbidity, and mortality rates associated with eye disease, numerous systematic reviews and meta-analyses have attempted to quantify this disparate literature. To date, most systematic reviews have focused on a single eye disease end point and there has not been a systematic evaluation of the relationships between eye disease and diverse physical and mental health outcomes. Moreover, the strength and reliability of the relationships reported in the literature is unclear. In order to address the breadth of the literature of complex conditions and comorbid outcomes, an increasing number of studies have used an ‘umbrella review’ approach (i.e., the syntheses of existing systematic reviews with meta-analyses, to capture the breadth of outcomes associated with a given exposure) [13, 14].

Therefore, the aim of the present study is to assess the strength and credibility of the evidence on eye disease and associated health outcomes derived from meta-analyses of observational studies using an umbrella review approach, aiming to the answer the following questions:

  1. 1.

    Which comorbid outcomes are associated with eye diseases?

  2. 2.

    What is the epidemiological credibility of the relationships between eye diseases and comorbid outcomes?

Methods

An umbrella review was carried out following standardised procedures [13, 15]. The protocol for the present umbrella review was preregistered with PROSPERO (registration number CRD42018093358).

Search strategy and selection criteria

We searched PsycINFO, Medline, CINAHL, and Embase databases (from inception to 15/03/2021) to identify systematic reviews with meta-analyses, pooling observational (cross-sectional, case-control, cohort) studies to examine any association between eye disease and any comorbidity/medical condition. The following search key was used:

“(meta-analysis or meta-anal* or systematic review) AND (vision OR visual* impair* OR eyesight OR blindness OR macular degeneration OR retinopathy OR cataract OR glaucoma OR corneal opacit* OR trachoma OR onchocerciasis)”.

Two independent reviewers (MT, DP) searched titles/abstracts for eligibility, and then evaluated the full text of those articles surviving title/abstract phase. A third reviewer resolved any potential conflict (LS). When more than one meta-analysis assessed the same risk factor or the same outcome, we only included the one with the greatest number of included studies [16,17,18]. Exclusion criteria were: 1) meta-analyses of randomised controlled trials (RCTs); 2) studies published in languages other than English, 3) meta-analyses reporting only one study for an outcome, since no meta-analysis was possible.

Data extraction

Data was independently extracted by two investigators (MT, DP) into a pre-prepared spreadsheet. For each meta-analysis, we extracted PMID/DOI, first author, publication year, population included in the study, study design, number of included studies, the total sample size and number of cases, i.e. people having the outcome of interest. The methodological quality of each included meta-analysis was assessed with the Assessment of multiple systematic reviews (AMSTAR) 2 tool (available at https://amstar.ca/Amstar-2.php), which is a recent update of AMSTAR [19], by two independent investigators (MT, DP). The AMSTAR2 tool was chosen because it has been used in several similar umbrella reviews [20,21,22].

Data analysis

For each association of meta-analyses providing individual study data, we extracted effect sizes (ESs) of individual studies and re-performed the meta-analysis calculating the pooled effect size and the 95% confidence intervals (CIs), with random-effects models [23]. Heterogeneity was assessed with the I2 statistic [24]. Additionally, we calculated the 95% prediction intervals (PIs) for the summary random ESs providing the possible range in which the ESs of future studies is expected to fall [25].

We also tested the presence of small-study effect bias [16, 26,27,28], which is deemed to be present in case of both pooled estimates larger than the individual largest study, and publication bias (Egger’s regression asymmetry test p < 0.10). We then assessed the existence of excess significance bias by evaluating whether the observed number of studies with nominally statistically significant results (p < 0.05) was different from the expected number of studies with statistically significant results (significance threshold set at p < 0.10) [28, 29], a test designed to assess whether the published meta-analyses comprise an over-representation of false positive findings [28].

Assessment of the credibility of the evidence

Credibility of meta-analyses providing individual study data was assessed according to stringent criteria based on previously published umbrella reviews [18, 20, 26, 27, 30, 31]. In brief, associations that presented nominally significant random-effects summary effect sizes (p < 0.05) were ranked as convincing, highly suggestive, suggestive, and weak evidence based on number of events, strength of the association, and the presence of several biases (criteria available in Supplementary Table 1).

Results

Search

The flow diagram of search, selection and inclusion process is fully reported in Supplementary Fig. 1. Out of 9239 hits initially identified, after duplicate removal, 4508 were assessed at title/abstract level. Finally, 34 systematic reviews and meta-analyses were included examining a total of 58 independent outcomes [5,6,7, 32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62].

Findings from the case-control and cross-sectional studies

Overall, 41 outcomes were assessed by case-control or cross-sectional studies. The most common outcome examined was modifiable risk factors (n = 14), followed by mental health/cognition outcomes (n = 12), disease outcomes (n = 11), pregnancy related condition (n = 2), and visual impairment (n = 2). The median number of studies was 7 and the median number of participants was 3865. Full information can be found in Table 1 and Fig. 1.

Table 1 Main findings of the case-control and cross-sectional studies.
Full size table
Fig. 1: Significant associations between various eye diseases and health outcomes.
figure 1

a odds ratios; b standard mean difference; c risk ratio; d hazard ratio.

Full size image

The p value for effect-size, under a random effects model, was <0.05 in 24/41 outcomes, and three reported a p value < 1*10−6. Among the 41 outcomes, 18 reported low heterogeneity (I2 < 50%), 11 moderate heterogeneity (I2 between 50 and 75%) and 12 high heterogeneity. Small study effect affected 10/41 outcomes, whilst 6/41 had excess significance bias (see Table 1). The largest study, in terms of participants, for each outcome was significant in 19 associations. For five outcomes, the PIs excluded the null value.

Using the criteria to grade the quality of the evidence, no outcome reached a convincing evidence (class I), three outcomes reached highly suggestive evidence (class II), six reached suggestive evidence (class III), 15 a weak strength of evidence (class IV), and 17 outcomes had no statistical significance. Regarding the class II evidence, open-angle glaucoma was associated with a myopia (n = 11 studies; OR = 1.92; 95% CI: 1.54–2.38) and with diabetes (n = 13 studies; OR = 1.46; 95% CI: 1.27–1.68); and any cataract was associated with a higher presence of type 2 diabetes (OR = 1.64; 95% CI:1.42–1.88) (see Table 1).

Findings from cohort studies

Overall, 17 outcomes were explored in prospective and retrospective designs. Mortality was the most explored outcome (n = 9), followed by pregnancy conditions (n = 4), disease outcomes (n = 3), and modifiable risk factors (n = 1). The median number of studies was 10, and the median number of participants was 30,118. Full information can be found in Table 2 and Fig. 1.

Table 2 Main findings of the prospective and retrospective studies.
Full size table

Almost half (8/17) of the associations included were statistically significant under a random-effects model, with three outcomes having a p value < 1*10−6. Among the 17 outcomes included, six were of low heterogeneity (I2 < 50%), three were of moderate heterogeneity (I2 between 50 and 75%) and eight were of high heterogeneity. Small study effects were present in five outcomes, and three outcomes showed excess significance bias (see Table 2). The largest study, in terms of participants, for each outcome was significant in 10/17 outcomes.

Using the criteria to grade the quality of the evidence, no outcome reached a convincing evidence (class I), three reached highly suggestive evidence (class II), two reached suggestive evidence (class III) and three showed weak strength of evidence (class IV). Regarding class II evidence, retinopathy of prematurity was associated with a higher incidence of chorioamnionitis (n = 71 studies; OR = 1.38; 95% CI: 1.3–1.57) and a higher risk of sepsis (n = 42; OR = 1.98; 95% CI: 1.69–2.33), and diabetic retinopathy was positively associated with incident cardiovascular disease (n = 12; OR = 2.42; 95% CI: 1.77–3.32).

Study quality

The majority of meta-analyses scored critically low (n = 31/34) on AMSTAR2, and three scored low (see Table 3). The main reasons for the critically low scoring was that most studies failed to report an explicit statement that the review methods were established prior to the conduct of the review (AMSTAR2 question 2; 3/34 studies satisfied this criteria) and failed to provide a list of excluded studies and justify the exclusions (AMSTAR2 question 7; 1/34 studies satisfied this criteria).

Table 3 AMSTAR2 results.
Full size table

Discussion

The present review, including 34 studies and 58 outcomes associated with varying eye diseases, no convincing (Class I) evidence for any comorbidity across all eye diseases was found. Highly suggestive levels of evidence (Class II) for cohort, case-control and cross-sectional studies showed that people with diabetic retinopathy were nearly 2.5 times more likely to suffer from cardiovascular diseases, and 1.8 times more likely to suffer CVD mortality. Diabetic retinopathy is a microvascular disease and it is not surprising that cardiovascular diseases will have a significant effect on the eye, with sepsis and chorioamnionitis being significant risk factors for retinopathy of prematurity [63]. Furthermore, babies with retinopathy of prematurity are nearly twice as likely to suffer from sepsis [53]. Retinopathy of prematurity is a vasoproliferative disease that affects the retinal vascular system in premature babies. As infection is a significant risk factor for neonatal brain damage, and sepsis is the key cause of neonatal inflammation, this could be the reason why the strong association with retinopathy of prematurity has been found. The foetal inflammatory response induced by chorioamnionitis [64], leads to proinflammatory cytokines having a substantial effect on retinal angiogenesis and subsequent development of the retina [65, 66], which could lead to retinopathy of prematurity.

Our analysis shows people suffering from open angle glaucoma are twice as likely to suffer from diabetes. Diabetes is a serious condition and its effects on macrovascular and micro vascular structures are well documented [67, 68]. While the strong association of diabetes and cataract is well known, the link with open angle glaucoma has been open to debate. Our analysis shows highly suggestive evidence of the link between diabetes and open angle glaucoma. One possible mechanism could be because long standing hyperglycaemia increases the risk of neural injury and the reduced capacity for auto-regulation of blood in diabetes could have an effect on the optic nerve and nerves in the eye. Furthermore, diabetes affects nerves in the body (neuropathy) and research has shown diabetes having a negative effect on ganglion cells in the eye [69].

Myopia also yielded a highly suggestive (Class II) association with open angle glaucoma. One possible mechanism is the biomechanical stress induced by increased axial length and oxidative stress, although this needs further investigation. The increasing global prevalence of myopia would have significant consequences on the global burden of eye diseases beyond just refractive error, and may explain, to a certain extent, the increasing prevalence of open angle glaucoma worldwide.

Suggestive levels of evidence (Class III) include cataract (including nuclear sclerosis) being associated with all-cause mortality and gout, diabetic retinopathy with depression, and open angle glaucoma with hypertension and migraine. Weaker strength of evidence (Class IV) links AMD with cognitive function, and glaucoma with sleep apnoea. Further studies need to be carried out to strengthen and confirm possible association between these conditions and the eye diseases.

Umbrella reviews provide top-tier evidence and important insights, however there are a number of limitations. Although we measured for heterogeneity, the meta-analyses included in this study included differing study designs, methods of measuring VI and eye diseases and populations. Furthermore, meta-analyses have inherent limitations [70]: their findings are dependent on estimates that are selected from each primary study and how they are applied in the meta-analysis. Finally, almost all of the studies included scored ‘critically low’ in quality control. Some studies were scored low as they had missed quality indicators such as confirming review methods or details about excluded studies. It is important that all the quality indicators are included in order to assure confidence in the data presented.

Conclusion

Our results show highly suggestive evidence for associations between diabetic retinopathy and cardiovascular disease, open angle glaucoma and diabetes, myopia and open angle glaucoma. Furthermore, we found suggestive evidence for associations between cataract and all-cause mortality and gout, depression and diabetic retinopathy, and hypertension and migraine for open angle glaucoma. Clinicians should take note of these and consider these associations in the delivery of care. Furthermore, public health policies should reflect and accommodate these associations in healthcare policies, practices and guidelines.

Summary table

What this study adds

  • This is the first study to examine the credibility of evidence against strict statistical criteria of eye disease and all types of health outcomes.

  • Six significant associations were classified as ‘highly suggestive’, including cataract and type 2 diabetes; open-angled glaucoma, myopia and diabetes; diabetic retinopathy, cardiovascular disease, and cardiovascular mortality; and retinopathy of prematurity and chorioamnionitis.

  • Eight significant associations were classified as ‘suggestive’, including diabetic retinopathy, all-cause mortality, and depression; diabetic macular oedema and dyslipidaemia; cataract and gout; nuclear sclerosis and all-cause mortality; open angled glaucoma, migraine, and hypertension; age-related macular degeneration and diabetes.

  • 18 significant associations were classified as ‘weak’.

Study limitations

  • The risk of bias of included meta-analyses was high.

  • This study included only meta-analyses of observation studies, which carry inherent limitations.