Epidemiology of diabetic retinopathy, diabetic macular edema and related vision loss

Prevalence of DR

In general, patients with type 2 diabetes in Western communities have a higher prevalence of DR than their Asian counterparts. In the USA, studies estimate that 28.5–40.3 % of patients with type 2 diabetes had DR, and 4.4–8.2 % of them had VTDR [17, 18]. In contrast, most Asian countries report DR prevalence to be between 12.1–23.0 %, and VTDR prevalence to be between 4.3–4.6 % [19–22].

Singapore is a notable exception to this trend. Despite being an Asian country, paralleling the rapid urbanization, industrialization and internal migration that took place over the past five decades in Singapore, DR prevalence in Singapore is reported to be higher (33.9 %) than other Asian countries but comparable to that seen in the Western world [23]. Within the three major ethnic groups in Singapore, the Malays and Indians were reported to have higher a prevalence of DR (33.4 % in Malays, 33.0 % in Indians) compared to the Chinese (25.4 %) [23]. In addition to ethnic differences, a study conducted in Singapore also highlighted geographic heterogeneity in the prevalence of DR within ethnic Indian groups living in Singapore (30.4 %) [24] and in urban India (18 %) [21, 25]. It has been speculated that increased acculturation to a westernized lifestyle associated with increased prevalence of obesity and diabetes, and increased awareness among Indians living in Singapore has led to a higher prevalence, while selective mortality of those with DR in the urban Indian cohorts led to a lower prevalence. In the Middle-East, Saudi Arabia [26] and Iran [27] both report prevalence that are similar to Western communities (36.8 and 29.6 %, respectively). Of concern is that a large proportion of diagnosed DR is vision threatening, with VTDR prevalence estimated to be higher (10.6–17.5 %) than that observed in the Western world. These observations imply that most of these cases of DR have been detected late, when it has already progressed to a vision-threatening stage, or that these populations are particularly susceptible to severe DR due to ethnic predisposition. Other developed Asian countries such as Hong Kong [19] and South Korea [20] report DR prevalence that is much lower than the global average (12.1 and 15.8 %, respectively).

Apart from the east–west divide, rapidly developing economies in Asia such as China and India are observing urban–rural divides in terms of DR disease burden. In China, prevalence of DR was reported to be higher among adults with type 2 diabetes living in rural regions (29.1–43.1 %) [22, 28], compared to their urban counterparts (18.1 %) [22]. Conversely, in a study conducted in Chennai, India, DR prevalence was reported to be higher in urban (18.0 %) [21] compared to rural areas (10.8 %) [29], possibly due to the increasing affluence accompanied by changes in diet in the urban regions and selective mortality of those with diabetes-related complications in rural regions because of poor access to healthcare. The reason why this urban–rural relationship is reversed in China may represent a case of ethnic predisposition, but this is an area that requires further study. In the past two years, reports on DR prevalence from many developing countries in Asia and Africa have been published [30–35]. Prevalence of DR in Sri Lanka, Bangladesh, Nepal, Tunisia, Kenya and Ethiopia ranged from 21.6–41.4 %. While the sample sizes of these studies tend to be smaller, they still provide insight into the burden of DR in these communities.

Although duration of diabetes is a major risk factor for DR, a few studies reported DR prevalence in newly diagnosed diabetes. Prevalence found in these studies ranged from 2.8 % in South Korea to 28.6 % in Singapore [20, 27, 32, 36–39]. Surprisingly, a large percentage (19.2 %) of newly diagnosed patients with diabetes have DR in Scotland, UK, where there is universal healthcare. This prevalence is even higher than in Nepal (13.0 %) [32], where access to healthcare is presumably more limited. However, the prevalence of advanced stages of DR or DME was found to be lower among those with newly diagnosed diabetes suggesting diagnosis of DR early in the course of the disease [40].

Incidence of DR

In the UK, population studies involving patients with type 2 diabetes estimated cumulative incidence of DR to be 26.0 % at 4 years [49] 38.1–41.0 % at 6 years [50, 51], and 66 % at 10 years [52]. These findings seem comparable to that found in US population studies, which estimated cumulative incidence of DR to be 22.5–34.0 % at 4 years [53, 54] and 72.3 % at 14 years [55], despite differences in ethnicity and age of the cohorts at the time of diabetes diagnosis. Cohorts in Australia [56], Barbados [57] and Mauritius [58] report cumulative incidence that is similar to the UK and US studies. In contrast, the 4-year cumulative incidence of DR in a Spanish cohort is much lower, estimated at 8.1 % [59]. Age and duration of diabetes are comparable between the US, UK and Spanish studies, and this significant difference in incidence is attributed to unusually good glycemic control within the Spanish cohort, with mean HbA1c at 7 %, with 55 % of the cohort achieving HbA1c of less than 7 %. In contrast, patients in one of the US cohorts [53] had HbA1c of 9.9 % on average.

As with prevalence, incidence data from Asia is restricted only to that of type 2 diabetes. A population-based study in urban Shanghai, China, found the 5-year cumulative incidence to be much higher than in the US and UK, at 46.9 %, of which more than a third of it is VTDR. This may just be due to differences in known diabetes duration of the cohorts; the Chinese cohort has diabetes duration of 11 years on average at baseline assessment, while studies in the US and UK report diabetes duration to be 4 to 7 years on average. More prospective studies are warranted to compare the incidence of DR in Asia with that observed in Europe or the US.

Progression and regression of DR

In general, progression was much more common than regression. Two Asian cohort studies, both hospital-based and carried out in Hong Kong, investigated the regression of DR. One of the studies found 4-year progression of DR to be 34.7 % and 4-year regression to be 13.2 % [60], which is similar to that seen in the population-based US cohorts. However, the other study found 4-year regression to be substantially higher (45.8 %) and progression to be lower (6.6 %) [61]. This study defined progression or regression by 1-step change in severity, while most of the other studies defined progression or regression by 2-step changes in severity. Moreover, this study was based in a community optometry clinic. Hence, the population sample may be biased towards patients with mild baseline severity of DR, as patients with more severe disease may have been referred to tertiary hospitals for follow-up. Indeed, 91.7 % of patients with DR at baseline in this study had only mild NPDR, and the 1-step regression of mild NPDR to no DR accounted for the majority of the regression observed in this study. The results of this study are hence not directly comparable with that of the other cohorts, but it highlights the high probability of disease regression in patients with only mild NPDR. The absence of data on population-based cohorts in Asia also precludes direct comparison of progression and regression rates between Asian and Western populations.

Prevalence of DME

In the Diabetic Retinopathy Screening Service for Wales, a high prevalence of DR (56.0 % in type 1 diabetes, 30.3 % in type 2 diabetes) was reported, but the prevalence of DME was not found to be higher than other studies (4.2 % in type 1 diabetes, 1.4 % in type 2 diabetes) [10].

There were a few outliers among the studies that reported exceptionally high prevalence of DME. In Kenya, a population-based study found a prevalence of DME of 33.3 % among participants with diabetes [34], while a Canadian study found DME prevalence to be 15.7 %. It is difficult to ascertain if this abnormally - high observed prevalence is due to genuinely high prevalence in these populations or a difference in methodology. Of note, clinical stereoscopic fundus examination by an ophthalmologist was carried out in both of these studies and factored in the diagnosis of DME whereas most studies relied on non-stereoscopic fundus photographs alone, thus raising the question if prevalence studies using non-stereoscopic fundus photographs may be severely underdiagnosing DME. In patients with newly diagnosed diabetes, observed prevalence of DME was almost non-existent, with studies reporting it to be within 0 to 0.8 % [21, 39]. A Cochrane review of prevalence of DME assessed by optical coherence tomography (OCT) has found a large range of prevalence rates (19–65 %) [72]. Of note, none of the studies included in the review were population-based studies. OCT-detected DME was found to have a great degree of disagreement with the clinical definition of CSME, and not all patients who had macular thickening detected on OCT progressed to have clinical DME, hence its validity as a diagnostic tool in epidemiologic studies is questionable.

Incidence of DME

Risk factors for DR and DME

Duration of diabetes

Cohort studies with the longest follow-up times found that almost all patients with type 1 diabetes develop some degree of retinopathy if duration of disease exposure is long enough [44, 45]. This relationship is not as clear in cohort studies on type 2 diabetes, probably due to the competing risk of mortality in patients with type 2 diabetes, who are older and may have more age-related comorbidities. Nevertheless, many studies, both in type 1 and type 2 diabetes [49, 52, 59, 75–77], found disease duration to be a significant risk factor for DR, and this is independent of adequacy of glycemic control.

Puberty and pregnancy

Puberty is a well-known risk factor for DR in type 1 diabetes. Pre-pubertal years of diabetes exposure contributes to added risk of DR [78, 79], but it seems that it is disease exposure during puberty itself, when the body is undergoing rapid development and maturation, that has the greater impact on the risk of DR. In Finland, the FinnDiane Study Group found that onset of diabetes during pubertal or post-pubertal age increases risk of developing severe retinopathy requiring laser treatment when compared to patients with pre-pubertal onset of diabetes [80]. This was particularly significant among the male participants. Biological pathways that may contribute to this phenomenon include the transforming growth factor beta (TGF-β) signaling pathway, which is an important mediator of renal microvascular damage [81]. Androgens promote and accelerate TGF-β transcriptional activity, which can explain the male preponderance. However, evidence of activation of similar pathways in retinal vessels is lacking.

DR and DME can progress rapidly during pregnancy, especially in patients with type 1 diabetes. A recent study found progression of DR in pregnancy to be almost 3 times as likely to occur in mothers with type 1 diabetes as mothers with type 2 diabetes (31.3 vs. 11.7 %, p = 0.001) [82]. This progression is often transient and accompanied by rapid regression of DR in the postpartum period. At the end of 6.5 years of follow-up on average, prevalence and severity of retinopathy was comparable between women with pregnancies and women without pregnancies [83]. Possible mechanisms behind the progression of DR in pregnancy include both hormonal and immune theories [84, 85].

Hyperglycemia

Hyperglycemia is one of the most important risk factors for DR and DME. A meta-analysis of three large population-based studies found a graded relationship between the level of glycemia and frequency of retinopathy signs [86]. The United Kingdom Prospective Diabetes Study (UKPDS) and the Diabetes Control and Complications Trial (DCCT) provided strong evidence that tight control of glycemia (HbA1c <7 %) reduces the risk of development and progression of DR in both type 1 and type 2 diabetes [87]. The DCCT showed that intensive glycemic control reduced the incidence of retinopathy by 76 % and progression from early to advanced retinopathy by 54 % [88]. This highlights that strict glycemic control is much more effective in preventing or delaying the onset of DR in patients with diabetes without DR, rather than limiting the severity of DR after it has occurred. In the case of DME, intensive glycemic control was associated with 46 % reduction in the incidence of DME at the end of the trial and a 58 % reduction 4 years later compared with those in the conventional group [89]. The burden of primary prevention of DR and DME hence falls heavily on primary care physicians, who are in the best position to achieve good glycemic control in patients who have not developed complications. In everyday clinical care however, it is difficult to replicate the intensity of glycemic control seen in these studies that were achieved under trial conditions. From the findings reported by the DCCT, intensive glycemic control actually increases risk of progression of existing DR in the first year of treatment [90]. However, this should not deter achieving tight glycemic control in patients with existing DR, as the long-term progression risk reduction outweighs that of the increased risk in the first year alone.

Glycemic control should be achieved early in the disease course and maintained for as long as possible, since its protective effect is sustained even if tight glycemic control is lost. This is the metabolic memory effect observed after the DCCT. Within a year after the end of DCCT, the glycemic control in the conventional group and intensive control group had converged, but the participants in the intensive control group still had lower prevalence of DR and DME than the participants in the conventional control group at 10 years after DCCT [91]. Risk reduction in the intensive control group was 52 % between years 1 to 10 after DCCT, but dwindled to 12 % between years 11 to 18 [92]. This implies that the metabolic memory effect fades with time, but this is confounded by improved glycemic control and risk reduction in the conventional control group since the end of DCCT. Besides implications for clinical treatment, metabolic memory also has implications on methodology of diabetes research, seeing that obtaining mean HbA1c of the entire course of diabetes may be needed to control for the effect of metabolic memory [93].

Apart from the absolute value of glycemia alone, the short-term variability of glycemia, such as spikes in post-prandial glucose, is found to be associated with increased risk of microvascular complications [94]. However, there is insufficient data at this point to conclude that fluctuations in blood sugar levels is a causative factor in microvascular complications considering increased glycemic fluctuation can be due to a multitude of correlated factors that may all contribute to microvascular injury, such as severity of disease or poor compliance.

The benefits of achieving euglycemia should be balanced with the risk of hypoglycemia, especially in the elderly. In both the Action in Diabetes and Vascular Disease (ADVANCE) [95] and Action to Control Cardiovascular Risk in Diabetes (ACCORD) [96] trials, aggressive glycemic control (HbA1c <6.5 %) did not significantly reduce risk of retinopathy development or progression in type 2 diabetes. In ACCORD, it was found that such an aggressive manner of glycemic control may in fact be associated with increased mortality, but it was not ascertained whether this was directly due to metabolic complications of treatment, such as hypoglycemia. Current institution guidelines state that treatment goals of hyperglycemia are to be anywhere between <6.5 to <7.5 % of HbA1c. According to a recently published Cochrane review [97] however, there is no concrete evidence on any specific treatment target. Instead, the authors recommend that clinicians set individualized treatment goals based on age, disease progression, risk of hypoglycemic episodes, and psychological factors of the patient.

Hypertension

Multiple epidemiologic studies have identified hypertension as a risk factor for DR and DME [51, 53, 76, 77, 98]. In the UKPDS, tight blood pressure control (defined as target blood pressure <150/85 mmHg) in patients with type 2 diabetes reduced the risk of microvascular disease by 37 %, the rate of progression of DR by 34 %, and the risk of deterioration of visual acuity by 47 % [99]. Unlike in the case of hyperglycemia, the protective effect of blood pressure control waned quickly upon stopping intensive control [100]. Anti-hypertensive medications targeting the renin-angiotensin-aldosterone system (RAAS) are now the first line treatment for control of hypertension in patients with nephropathy as it was found that they had additional beneficial effects independent of their absolute hypotensive action. Since retinopathy and diabetic nephropathy are related microvascular complications, clinical trials such as the Diabetic Retinopathy Candesartan Trials (DIRECT) and Renin-Angiotensin System Study (RASS) measured the beneficial effects these classes of anti-hypertensive medications had on DR and DME. Candesartan was found to reduce the incidence of retinopathy by two or more steps in severity on the ETDRS scale by 18 % or by three or more steps by 35 % in type 1 diabetes, and increased regression of retinopathy by 34 % in type 2 diabetes [101, 102]. However, regression only occurred in mild DR, and candesartan had no effect on incidence or progression of DME. In the RASS, enalapril and losartan reduced the risk of retinopathy progression by 65 and 70 %, respectively. Since it was observed that this effect was independent of blood pressure changes across the period of the trial, it was proposed that DR risk reduction was not mediated by an effect on hypertension.

A recently published Cochrane review concluded that intensive blood pressure control had a modest effect in reducing incidence of DR, but does not reduce risk of progression [103]. Insufficient evidence on adverse effects of strict blood pressure control in patients with diabetes made a cost-benefit analysis impossible in the review, and both clinicians and researchers should be aware of this gap in literature. Hence, the overall recommendation is to avoid intensive blood pressure control for the sole purpose of slowing DR progression. Instead, control of hypertension in a patient with diabetes should be focused on preventing or limiting progression of other vascular complications, particularly nephropathy, as well as lowering mortality. There is insufficient evidence for the use of RAS targeting anti-hypertensive medication specifically for preventing or treating retinopathy.

Dyslipidemia

As outlined in a previous review, the evidence for dyslipidemia as a risk factor for DR are inconsistent, and no single lipid measure had been consistently found to be associated with DR or DME [74]. In recent cohort studies, only the Madrid Diabetes Study found an association between low density lipoprotein (LDL) cholesterol and incidence of DR [59]. Moreover, a meta-analysis found that there was a dose-dependent relationship of statin use with increasing risk of diabetes [104]. It was then believed that statins might have effects on glucose homeostasis, such as decreasing insulin production or increasing insulin resistance, or both [105]. Therefore, while the use of statins is first-line treatment for dyslipidemia in the prevention of cardiovascular events in patients with diabetes, the evidence for intensive control by statins for the purposes of treating DR and DME are lacking.

Fenofibrate, a peroxisome proliferator-activated receptor alpha (PPARα) agonist, has gathered interest on its effects on DR and DME. In an ancillary study of the Fenofibrate Intervention and Event-lowering in Diabetes (FIELD) cohort, participants treated with fenofibrate had a 31 % reduced risk of requiring laser treatment for PDR or DME, compared to placebo [106]. However, 2-step progression of retinopathy did not differ significantly between the fenofibrate and placebo group, except for the subgroup with pre-existing DR. In this subgroup, risk of 2-step progression was almost a fifth of that compared to placebo. Moreover, in a more recent trial by the ACCORD group, adjunct fenofibrate with simvastatin compared to simvastatin alone reduced the rate of progression of DR (6.5 vs. 10.2 %, respectively) by at least 3 steps at 4 years [107]. Fenofibrate treatment may also have beneficial effects on DME, as it was found to have a moderate effect in decreasing macular volume in patients with DME [108]. The sample size of this study however, was relatively small, and more studies are required to study this association. Given the current evidence, it is found that patients with DR benefit most from fibrate therapy if they have hypertriglyceridemia and low serum high density lipoprotein (HDL)-cholesterol, and hence treatment can be justified in this subset of patients, with the hopes of slowing progression to PDR. However, generalization of fibrate treatment to all patients with diabetes at risk of DR is not recommended without stronger evidence [109].

Obesity

The effect obesity has on DR has been relatively well-studied but with inconclusive and conflicting findings [110]. It may be possible that obesity has differing impacts on DR in type 1 diabetes as compared to type 2 diabetes. In the Diabetic Incidence Study in Sweden involving predominantly participants with type 1 diabetes, it was found that risk of developing DR increased by 1.11 (95 % confidence interval (CI) 1.04–1.18) times per kg/m² increase in Body Mass Index (BMI) after 10 years of follow-up [111]. In the EURODIAB Prospective Complications Study, also involving patients with type 1 diabetes, larger waist to hip ratio was associated with incidence of DR after more than 7 years of follow-up [112].

In contrast, many studies in type 2 diabetes, performed primarily in Asia, found an inverse relationship between obesity and DR. In a cross-sectional study of the Shanghai Diabetes Registry Database, participants who were overweight had reduced risk of DR and VTDR [113]. A similar study on the multi-ethnic population in Singapore found the same risk reduction in obese patients for DR, VTDR and CSME [114].

The exact mechanisms underlying this discrepancy between type 1 and type 2 diabetes are not well understood. It was postulated that unintentional weight loss is a sign of advanced and severe type 2 diabetes, hence the observation of non-obese patients with type 2 diabetes being at higher risk of DR. In contrast, obesity and metabolic syndrome do not contribute to the etiology of type 1 diabetes, which is autoimmune in nature, and obese patients with type 1 diabetes may simply have more difficulties achieving good glycemic control. It should be noted that there are no prospective population-based studies in Asia on DR incidence, and the protective effect of obesity in Asians with type 2 diabetesis yet to be confirmed by a cohort study.

Closely related to obesity is the study of obstructive sleep apnea (OSA) as a potential risk factor for DR and DME. A cross-sectional study in patients with type 2 diabetes found that OSA was associated with DR severity, but not DME [115]. A separate study on patients with CSME found high prevalence of sleep-disordered breathing in these patients, but severity of sleep-disordered breathing was not correlated with severity of DR or DME in this study [116]. However, the sample sizes of these studies were too small to draw any concrete conclusions.

Bariatric surgery is a highly effective treatment for morbid obesity that achieves glycemic control of diabetes rapidly. However, much like how intensive glucose control with medications or insulin increases risk of DR progression in the short-run, this rapid improvement in glycemic control post-bariatric surgery has been associated with progression of DR. Most studies presented in this area are case series, and a recent meta-analysis of these studies found that patients with pre-existing DR are 2.77 times (95 % CI 1.10–6.99) more likely to have adverse outcomes in DR post-operatively than patients without pre-existing DR [117]. As mentioned earlier, increased risk of progression with intensive glycemic control occurred only in the first year of follow-up, with subsequent risk reduction with longer-term control [90]. It remains to be seen if this is the case with bariatric surgery as well, as no studies had sufficient follow-up time to determine if bariatric surgery has long-term benefits on DR.

Inflammation

Retinal and vitreous inflammation was observed in subjects with diabetes, both in animal models and human studies. The role of inflammation in DR and DME is therefore an area of extensive study, and has been reviewed previously [118]. As pointed out in the review however, current data suggests systemic inflammation cannot account for the characteristic lesions seen in DR and DME. Many conditions can lead to systemic inflammation (e.g. sepsis, autoimmune disease), but DR-like lesions and DME are not seen in these diseases. Hence, it seems that the local retinal inflammation seen in subjects with diabetes is not related to systemic inflammation. This challenges the validity of investigating systemic inflammatory markers such as serum C-reactive protein (CRP), interleukin-6 (IL-6) and tumor necrosis factor-α (TNF-α) as risk factors for DR or DME. Indeed, inconsistencies in the association between systemic inflammatory markers and risk of DR and DME exist in the current literature. The EURODIAB Prospective Complications Study found an association between CRP, IL-6, TNF-α and presence of DR in subjects with type 1 diabetes via a cross-sectional study [119]. Other cross-sectional studies found no such association. The Multi-ethnic Study of Atherosclerosis did not find an association between CRP and DR or VTDR (which includes DME), but found an association between fibrinogen, an acute-phase reactant in systemic inflammation, and DR and VTDR [120]. The Singapore Malay Eye Study even found that raised CRP was associated with a lower prevalence of DR [121]. None of the studies found an association between systemic inflammatory markers and DME specifically.

Local retinal inflammation forms the basis of intravenous administration of corticosteroids. The Diabetic Retinopathy Clinical Research Network (DRCR.net) compared intravitreal triamcinolone versus focal/grid laser photocoagulation in patients with DME. The findings showed that the triamcinolone group had better visual acuity at the 4-month interval, but equivalent visual acuity at the 1-year interval. At the 2-year [122] and 3-year interval [123], mean visual acuity was better in the photocoagulation than the triamcinolone groups. Hence, corticosteroid treatment for DME is effective, but the effect is transient. Clinicians also have to be cautious with adverse effects such as elevated intraocular pressure and cataract formation.

Vascular endothelial growth factor (VEGF) is a key modulator of angiogenesis and vascular permeability, and is upregulated by inflammatory cytokines [124]. Anti-VEGF agents have been used successfully for the treatment of both PDR and DME [125, 126]. Ranibizumab, an anti-VEGF agent, was more effective than laser therapy in restoring vision for DME [127], although just like with corticosteroids, ranibizumab is associated with elevations in intraocular pressure [128]. In recent reports, the DRCR.net compared outcomes in DME treated by aflibercept, bevacizumab or ranibizumab, and found that aflibercept provided superior visual recovery if baseline visual acuity was poorer than 69 ETDRS letters (approximately 6/15 Snellen) when compared to the other anti-VEGF agents, but there was no significant difference between aflibercept and the other anti-VEGF agents if baseline visual acuity was better than 69 letters [129].

Anti-VEGF agents appear superior to corticosteroids in terms of efficacy. DRCR.net compared ranibizumab and concurrent photocoagulation against triamcinolone with photocoagulation in patients with DME, and found that ranibizumab achieved better visual outcome at 1-year follow-up than triamcinolone, except in a subset of patients with pseudophakic eyes [130]. In this subset of participants, triamcinolone achieved comparable visual outcome when compared with ranibizumab, possibly because of the removed effect of steroid-induced cataract formation in pseudophakic eyes. Consistent results were obtained at 2-year follow-up [131].

Metabolic hormones

Hormones involved in metabolism have been hypothesized to play key roles in the pathogenesis of microvascular complications in diabetes, due to their roles in both metabolic and inflammatory pathways [132]. In particular, leptin and adiponectin, which are actively secreted by adipocytes to regulate energy balance in the body, have been implicated as potential risk factors.

Leptin may play a role in inciting inflammation. Leptin was found to cause upregulation of VEGF in retinal pericytes [133], hence stimulating angiogenesis in the ischemic retina [134], and possibly contributing to the neovascularization seen in PDR. Elevated serum and vitreous leptin was observed in patients with diabetes, and vitreous leptin was especially elevated in patients with PDR [135]. However, cross-sectional studies could not find an association between elevated serum leptin and DR [136, 137], though it should be noted that the sample sizes of these studies were relatively small and they may be underpowered.

Adiponectin has been found to induce dilation of retinal arterioles via upregulation of endothelial cell nitric oxide production, in animal studies [138]. Studies by the same group in human subjects with mild DR found that serum adiponectin was positively correlated with retinal blood flow velocity and negatively correlated with retinal arterial resistance [139]. Hence, adiponectin may have a role in countering ischemia by promoting reperfusion in the ischemic retina. In vitro studies also found that it downregulates VEGF and thus may have anti-angiogenic properties [140]. Large population-based cross-sectional studies found that elevated serum adiponectin in patients with DR correlated with severity of DR when compared to patients without DR [141, 142]. However, there are inconsistencies in literature, with one study finding decreased serum adiponectin in participants with PDR [143]. Given that basic science suggests adiponectin as mainly protective against the development of microvascular complications, the observation that serum adiponectin is elevated in patients with severe DR appears contradictory. It may be that upregulation of adiponectin secretion can be attributed to a natural response that ameliorates the effects of severe microvascular disease, but prospective cohort studies are needed to establish the temporal link between adiponectin levels and the development and progression of DR. Overall, it appears research in adiponectin has produced more promising and consistent results than leptin. The association between these hormones and DME has yet to be studied.

Oxidative stress

Oxidative stress is the accumulation of free radicals in the form of reactive oxygen species (ROS). Highly efficient physiological mechanisms consisting of endogenous free radical scavengers usually keep oxidative stress low. However, under pathological conditions, ROS production may be increased such that the defensive mechanisms are overwhelmed, or the protective mechanisms themselves may be impaired, or both [144]. Oxidative stress has been linked to the histopathological changes of DR, such as retinal basement membrane thickening [145] and capillary cell loss [146]. Increased ROS and decreased antioxidant potential has also been found in patients with diabetes, especially if they have DR [147]. The effects of oxidative stress are observed early in the course of diabetes, and its effects on microvasculature persist even if hyperglycemia is subsequently corrected. Hence, oxidative stress is likely to be the mechanism behind the “metabolic memory” effect mentioned earlier, where sustained periods of hyperglycemia early in the disease course has long-lasting effects on future microvascular complications [148].

Multiple biochemical pathways involved in DR pathogenesis are linked to oxidative stress. The accumulation of advanced glycation end products (AGE) in retinal pericytes upregulates cellular expression of its receptor (RAGE). AGE-RAGE overexpression produces ROS, activating apoptotic pathways to cause pericyte loss, seen in early DR [149] The polyol pathway is augmented in hyperglycemic conditions, resulting in overconsumption of NADPH, reducing its availability for formation of the key endogenous antioxidant glutathione [150]. ROS has also been found to increase the activity of protein kinase C (PKC), a family of serine-threonine kinases that cause vascular dysfunction by increasing permeability, altering blood flow, and stimulating neovascularization. Vascular dysfunction and neovascularization is potentiated further as PKC induces VEGF [144]. Due to how multiple pathways activate and can be activated by oxidative stress, therapeutic strategies targeting any single pathway is unlikely to be effective, as shown in the multiple randomized-controlled trials [151–153]. Research has since focused on mitochondrial dysfunction as the main upstream source of oxidative stress, but whether research in this area will yield novel treatment strategies remains to be seen [148].

From an epidemiologic standpoint, given the importance of oxidative stress in the pathogenesis of DR, reliable and accessible markers of oxidative stress are valuable measures of disease severity and prognosis. To date, most studies relating oxidative stress to DR involve in vitro and animal studies, and oxidative stress markers have not been investigated in large epidemiologic studies. Small cross-sectional studies have consistently found elevated markers of oxidative stress such as lipid peroxide (LPO) and malondialdehyde in both vitreous and serum of human subjects with DR [154, 155]. In particular, serum LPO was found to correlate highly with vitreous LPO, and that LPO correlated well with key disease mediators such as VEGF, suggesting that serum LPO may be a suitable proxy measure of DR severity [154]. More studies will be needed to confirm this association.

Vitamin D

On top of its well-known effects on calcium metabolism, Vitamin D has anti-angiogenic and anti-inflammatory effects that have implicated Vitamin D deficiency in the pathogenesis of various types of pathology, such as malignancy, autoimmune disease, cardiovascular disease and diabetes [156].

It is thus intuitive that Vitamin D has a protective effect on DR and DME, since anti-angiogenesis may slow progression to PDR and anti-inflammatory properties may counteract development of both DR and DME. Calcitriol, or 1,25-dihydroxycholecalciferol, is the metabolically active form of Vitamin D, and has been found to be a potent inhibitor of retinal neovascularization in vitro [157], possibly through suppressing TGF-β and VEGF levels [158]. Epidemiologic studies have found vitamin D deficiency to be associated with increased prevalence and severity of diabetic retinopathy, in both type 1 [159, 160] and type 2 diabetes [161–163]. However, all these studies are cross-sectional. No data is available on how Vitamin D influences prevalence of DME.

Genetic factors

As highlighted earlier in this review, certain trends in DR prevalence and incidence cannot be explained by environmental or socioeconomic factors, such as the abnormally high prevalence of DR in rural China, or the large proportion of VTDR in the Middle East. Some patients appear predisposed to severe DR even with adequate risk factor control, while others avoided DR despite poor control and long diabetes duration [164]. Familial aggregation studies and clinical trials including the DCCT have demonstrated a heritable tendency for severe retinopathy in type 1 and type 2 diabetes, independent of shared risk factors [165–168]. Hence, the hypothesis of differential genetic susceptibility to DR has drawn interest. The list of polymorphisms reviewed here is not exhaustive, but focuses on genes affecting the biological pathways mentioned earlier in the review.

Polymorphisms in the adipose most abundant gene transcript-1 (apM-1) gene on chromosome 1q21.3-q23 that codes for adiponectin have been detected to influence serum adiponectin levels and risk of DR [142]. Participants with type 2 diabetes heterozygous for the Tyr111His polymorphism at exon 3 (Tyr/His) had significantly higher serum adiponectin levels than participants who were homozygous for Tyr111His (Tyr/Tyr), but this had no statistically significant effect on the risk of DR. Participants with type 2 diabetes who had the mutant +45TG allele at the Gly15Gly polymorphism had no observable differences in serum adiponectin levels when compared to participants with the wild type +45TT allele, but they had a significantly lower risk of DR. It was unclear why the reduced risk of DR in this study appeared independent of serum adiponectin levels. Multiple VEGF polymorphisms have been investigated for their link to DR. The -2578C/A, +936C/T and -460 T/C polymorphisms of VEGF have been associated with DR in Asians by meta-analysis of cross-sectional studies [169, 170]; The C-634G polymorphism was linked to risk of DME. The CC genotype of this polymorphism was associated with the presence of DME, but was also associated with better treatment response to bevacizumab when compared to the CG and GG genotypes [171]. Recently, single nucleotide polymorphisms in the VEGF-C gene have been associated with DR and DME in both type 1 and type 2 diabetes [172].

Aldose reductase is the rate-limiting enzyme in the polyol pathway that contributes to oxidative stress in patients with diabetes. The C(−106)T polymorphism was found on meta-analysis to be associated with risk of DR in type 1, but not type 2 diabetes [173]. Genes coding for enzymes in antioxidant pathways such as catalase, superoxide dismutase and glutathione peroxidase are downregulated in patients with DR compared to patients with diabetes but without DR, but it is unknown if certain polymorphisms predispose to this observation [174]. Vitamin D receptor gene polymorphisms may also predispose to DR. T to C substitution at the Taq I site of the Vitamin D receptor gene [175], and T to C substitution at the start codon FokI [176], was associated with severe DR in patients with type 1 diabetes.

A few genome-wide studies have identified novel gene loci associated with DR [177–180]. Association of novel genes related to vascular endothelium proliferation and capillary permeability, such as PLXDC2 and ARHGAP22, imply that our understanding of angiogenic and inflammatory pathways is still incomplete [178]. Interestingly, polymorphism of RP1-90 L14.1, a long intergenic non-coding RNA gene adjacent to CEP162 was found to be associated with susceptibility to DR [180]. Since CEP162 is a key protein in cell ciliogenesis [181], it raises the question if dysregulation of ciliary assembly plays a role in DR pathogenesis.

Epidemiology of diabetes-related vision loss

While treatment options such as pan-retinal laser photocoagulation can largely control neovascularization and prevent blindness, these treatments cannot restore vision, and in fact have vision-impairing effects of their own. Intravitreal agents such as anti-vascular endothelial growth factor (VEGF) agents do not fully restore vision in all patients, and require frequent and costly doses for effective treatment. Vision loss from DR or DME is hence a significant healthcare burden [1].

A recent systematic review estimated that in 2010, 3.63 million people worldwide suffer from moderate and severe vision loss due to DR and its related sequelae, defined as visual acuity in the better eye being worse than Snellen 6/18 but at least 3/60. An estimated 850 thousand more people suffer from DR-related blindness, defined as visual acuity worse than 3/60 in the better eye [2]. Prevalence of vision impairment and blindness due to DR was found to be on the uptrend, even though total prevalence of vision impairment and blindness was decreasing. Findings from reviews of cross-sectional studies in Europe [182], South-East Asia and Oceania [183], consistently found DR to be the fifth most common cause of moderate and severe vision loss and blindness, behind causes such as uncorrected refractive error, cataracts, macular degeneration and glaucoma. In Africa, DR is the sixth most common cause of visual impairment and blindness, behind the above-listed conditions and trachoma [184]. In the USA, the WESDR investigated visual impairment in patients with type 1 diabetes, and found that 25-year cumulative incidence of visual impairment (defined as poorer than 6/12 best-corrected visual acuity in the better eye) and severe visual impairment (defined as poorer than 6/60 best-corrected visual acuity in the better eye) to be 13 and 3 %, respectively [185].

Recent data in Leeds, UK, found that in 2008 to 2010, DR accounted for 6.1–8.3 % of visual impairment certification. Extrapolated to the total population of the metropolitan area in Leeds, this estimates that 30.0 to 43.2 people per million per year will become severely visually impaired due to DR and its sequelae [186]. In Fife, Scotland, between 2000 and 2009, the mean incidence of blindness (defined as above) was 13.8 per million per year for the total population of the county [187]. In the Sankara Nethralaya Diabetic Retinopathy Epidemiology and Molecular Genetics Study (SN-DREAMS) in type 2 diabetes, the prevalence of visual impairment and blindness was 4 and 0.1 %, respectively [188].

Other eye complications of diabetes

While DR and DME are the most important and well-studied diabetes-related eye complication, many patients with diabetes are at risk of vision loss from other diabetes-related eye conditions that range from mild vision impairment to blindness. Diabetes is associated with early and rapid development of cataracts, and is hence a major cause of visual impairment among patients with diabetes. The Singapore Malay Eye Study (SiMES) found patients with diabetes to be more likely to have cortical and posterior subcapsular cataracts [189]. In the WESDR study and SN-DREAMS study, presence of cataracts were significant factors contributing to visual impairment and blindness in patients with diabetes [185, 188]. Many patients with diabetes require cataract surgery at a relatively younger age. In the WESDR, 10-year cumulative incidence of cataract surgery was 8 % in patients with type 1 diabetes and 25 % in patients with type 2 diabetes [190]. While usually a surgical procedure with good outcomes, cataract surgery is complicated in patients with diabetes as they may develop DME after surgery [191].

Although findings have been inconsistent, diabetes has been found to be a risk factor for developing primary glaucoma in some population-based studies [192]. For instance, SiMES found an association between ocular hypertension and diabetes, but not glaucoma [189]. Neovascular glaucoma, which is both a blinding and painful condition, can also arise from PDR. A recent report found that 7.1 % of patients with PDR requiring vitrectomy developed neovascular glaucoma 1 year after surgery [193]. Epiretinal membranes, which can cause significant visual impairment, were also found to be more common among patients with diabetes that have undergone cataract surgery [189].

Relationship of DR and DME with diabetes related systemic complications

Macrovascular complications

The strength of association between DR and macrovascular complications, such as cardiovascular disease is just as strong as in nephropathy [8]. In the Chennai Urban Rural Epidemiology Study, prevalence of coronary heart disease was higher among patients with DR as compared to those without DR [198]. An eight-year cohort study in Japan found that patients who developed signs of mild DR were already at higher risk of coronary heart disease or stroke [9]. Factoring presence of DR in the assessment of patients with diabetes also improved risk assessment of silent myocardial infarcts [199]. Presence of DR was also associated with mortality from cardiovascular disease, especially if there is concomitant nephropathy [200]. Literature relating DR with peripheral vascular disease is sparse, but a recent cross-sectional study in China found an association between presence of PDR with lower ankle-brachial index and lower toe-brachial index [201].