Risk of diabetic retinopathy and diabetic macular oedema with sodium–glucose cotransporter 2 inhibitors and glucagon-like peptide 1 receptor agonists in type 2 diabetes: a real-world data study from a global federated database

Diabetes mellitus affects 463 million people worldwide and this number is projected to rise to 700 million by 2045, making it a major global epidemic [1]. In the UK, 4.9 million individuals have diabetes, with £10 billion annually allocated by the National Health Service [2, 3]. Diabetic retinopathy and diabetic macular oedema (DMO) are leading causes of vision loss in the diabetic population [4]. The Wisconsin Epidemiologic Study of Diabetic Retinopathy XV notes a concerning progression in two-thirds of individuals who had diabetic retinopathy at baseline to more severe stages of diabetic retinopathy, and 20% to proliferative diabetic retinopathy and DMO [5]. While treatments for proliferative diabetic retinopathy and vision-threatening DMO such as intravitreal anti-vascular endothelial growth factor (VEGF) injections exist, their invasiveness, cost and complications pose challenges [6]. Despite the current focus on glycaemic control, the multifactorial nature of diabetic retinopathy and DMO suggests unidentified risk factors and potential new therapeutic targets [7].

Glucagon-like peptide 1 receptor agonists (GLP1-ra) have been used for over a decade in the management of type 2 diabetes [8]. Other than improving glycaemic control, GLP1-ra have positive effects on body weight, and a number of GLP1-ra agents have shown cardiovascular (CV), reno-protective and mortality benefits. Similarly, sodium–glucose cotransporter 2 inhibitors (SGLT2is) are a more recent class of medication now firmly established in diabetes management. Similar to GLP1-ra, the therapeutic effect of SGLT2is has been re-appraised with dedicated CV and renal outcome trials, with notable benefits even in the absence of type 2 diabetes [9]. Additionally, recent studies have suggested a possible role for both GLP1-ra and SGLT2is in the management of diabetic retinopathy and DMO that extends beyond glycaemic control [10, 11]. Possible mechanisms include the reduction of retinal endothelial cell apoptosis and mitigation of gradual central retinal thinning with SGLT2i, a neuroprotective role in retinal ganglion cells with GLP1-ra, as well as a reduction in the oxidative stress and vascular remodelling with both medications [12‐16]. While CV and renal benefits are widely acknowledged, their potential impact on diabetic retinopathy and DMO, particularly with SGLT2i, is promising but not as firmly established.

In this this real-world study using a global federated dataset, we aimed to evaluate the impact of SGLT2i and GLP1-ra therapy on the risk of diabetic retinopathy and DMO in individuals with type 2 diabetes taking insulin. We compared the impact of SGLT2i vs a control cohort, GLP1-ra vs control cohort and SGLT2i vs GLP1-ra.

Diabetic retinopathy and DMO are leading causes of blindness worldwide. In this study, we explore the impact of SGLT2i and GLP1-ra therapy on the risk for developing diabetic retinopathy and DMO, through a real-world study from a global federated database, including around two million people with type 2 diabetes receiving insulin treatment. Our principal findings are as follows: (1) SGLT2i reduced the incidence of DMO, while GLP1-ra did not significantly affect DMO incidence; (2) SGLT2i resulted in a minimal but statistically significant increase in the risk of diabetic retinopathy; however, it is unlikely to be clinically significant and it should be interpreted in the context of multiple testing; GLP1-ra increased the incidence for diabetic retinopathy compared with a propensity score matched control cohort; and (3) direct comparison of SGLT2i and GLP1-ra therapies demonstrated that the GLP1-ra cohort had an increased risk for the development of both diabetic retinopathy and DMO. In keeping with well-established literature, SGLT2i and GLP1-ra were associated with a reduced risk of HF, hospitalisation and all-cause mortality [8, 9]. To the best of our knowledge, this is the largest real-world study to evaluate the benefits of SGLT2i and GLP1-ra on diabetic retinopathy and DMO in individuals with type 2 diabetes receiving insulin.

Although this is a retrospective study and thus randomisation and controlling of the confounding variables was not possible, we propensity score matched for a significant number of the key known confounding factors, which were well balanced amongst the comparison groups. Despite propensity score matching, unmeasured confounding variables may bias our findings. However, E-values of greater than 1.5 (for SGLT2i vs control and GLP1-ra vs control analyses) suggest that confounders with such magnitude of association with both exposure and outcome are unlikely. Propensity score matching could not account for duration/severity of diabetes in this platform, leading to the use of insulin treatment as an active comparator design to reduce the extent of confounding by indication. In our propensity score matching for HbA_1c, BMI and eGFR, we opted for categorical instead of continuous values, as imputation for missing values was not computationally possible. Employing continuous values in complete case analyses would result in a reduction of sample size and statistical power. However, categorising data comes with limitation of residual confounding. This being a real-world federated database study, raw data were inaccessible due to confidentiality agreements. We obtained access to summary-level data relying on clinical ICD-10 coding, introducing some limitations. Due to lack of individual data, multivariant analysis is not possible. Data completeness cannot be assured, and the availability of 5 year data for all participants from the index event remains uncertain. Nonetheless, censoring was used to account for participant attrition from the cohort during the analysis, ensuring their exclusion from the statistical analysis. The diagnostic modality and severity of the main and positive control outcomes could not be established due to these constraints on data availability. The sequence of initiating insulin relative to GLP1-ra or SGLT2i is not known. Moreover, the length of treatment with SGLT2i, GLP1-ra and/or insulin or adherence to treatment and therefore the time of exposure to those medications are not known. All participants were taking insulin, a medication usually used in more advanced type 2 diabetes, and the results may not be generalisable to other patient populations. The dataset's majority being from the USA, with its insurance-based healthcare system, may affect generalisability when considering early insulin use due to cost considerations.

In a meta-analysis of RCTs, SGLT2i demonstrated a protective effect against retinal disease when directly compared with glucose-lowering agents [22]. A further systematic review and network meta-analysis suggested that canagliflozin increased diabetic retinopathy, although this signal was not present for two other SGLT2is (empagliflozin and dapagliflozin) [23]. Lin et al demonstrated the rate of developing diabetic retinopathy was comparable between individuals receiving SGLT2i and GLP1-ra, although individuals receiving SGLT2i had a lower risk of proliferative diabetic retinopathy [14]. Similar to our study, a post hoc analysis of the EMPA-REG OUTCOME trial showed no increased risk of retinopathy with empagliflozin vs placebo [15]. Other retrospective studies have shown favourable outcomes in individuals taking SGLT2i, either by slowing the progression of diabetic retinopathy or by reducing the risk for developing diabetic retinopathy, or both [15, 24]. In a multi-institutional cohort study, for example, Su et al observed a 25% reduced risk of developing DMO in individuals receiving SGLT2i compared with individuals receiving GLP1-ra [25]. Also, Tatsumi et al demonstrated that SGLT2i can reduce the risk of DMO (measured through central retinal thickness) in individuals with diabetic retinopathy [26].

Studies investigating the effect of GLP1-ra on diabetic ocular disease have also been heterogenous in nature. The SUSTAIN-6 RCT demonstrated that the rates of retinopathy complications, including vitreous haemorrhage, blindness or photocoagulation, were higher with semaglutide compared with placebo [27], although this trial was a CV outcome trial rather than a diabetic retinopathy-orientated trial. A systematic review and meta-analysis also suggested that subcutaneous semaglutide increased the risk of diabetic retinopathy, but this finding was not demonstrated with oral semaglutide [23]. The LEADER trial found a non-significant increased incidence of retinopathy in the GLP1-ra (liraglutide) group compared with placebo [28]. However, in nationwide cohort and Mendelian randomisation studies, both showed lower risk of diabetic retinopathy for individuals treated with GLP1-ra [29]. The AngioSafe Type 2 Diabetes Study demonstrated no association between GLP1-ra and severe diabetic retinopathy, as diagnosed with fundus imaging [30]. This has been further supported by a systematic review and meta-analysis which showed no increase in diabetic retinopathy risk in the GLP1-ra groups [31].

Diabetic retinopathy is strongly associated with other microangiopathies [7]. Understanding the effects of SGLT2i and GLP1-ra on diabetic retinopathy and DMO may delineate phenotypes and thus allow for risk stratification for the development of strategies to tackle wider diabetic (macro- and microvascular) complications. Gibbons and Freeman previously demonstrated that treatment-induced diabetic neuropathy (insulin neuritis) occurred in parallel with diabetic retinopathy and microalbuminuria, suggesting a common underlying pathophysiological mechanism [32]. Our data are consistent with several other studies showing a beneficial effect of SGLT2i and GLP1-ra on HF, hospitalisation and all-cause mortality. The EMPA-REG OUTCOME trial demonstrated that the empagliflozin cohort had a reduced risk for hospitalisation for HF and all-cause mortality compared with the placebo group [33]. Similarly, the DAPA-HF trial showed that dapagliflozin can prevent progression of disease in individuals with pre-existing HF, and reduce the risk of HF hospitalisations and CV death [34]. Conversely, the LEADER trial (GLP1-ra: liraglutide), PIONEER 6 trial (GLP1-ra: oral semaglutide) and EXSCEL trial (GLP1-ra: exenatide) described a significant reduction in all-cause mortality and a non-statistically significant reduction in hospitalisation for HF in the GLP1-ra group compared with placebo [28, 35, 36]. We recently showed that combination therapy (with SGLT2i and GLP1-ra) was associated with the greatest risk reduction in all-cause mortality vs a propensity score matched control cohort [19].

The mechanism through which SGLT2i translates to improved diabetic eye disease outcomes is likely multifactorial, although studies have suggested that it extends beyond glycaemic control. Hyperglycaemia results in oxidative stress, inflammation, vascular damage, capillary ischaemia and retinal tissue hypoxia [37]. SGLT2 transporters are present in retinal cells, including mesangial cells and retinal pericytes [38], which may suggest that SGLT2i agents act in directly reducing glucose levels in the retinal microcirculation and hence reduce glucotoxicity, oxidative stress and inflammation, and restore insulin signalling, halting glucose-induced vascular and endothelial dysfunction, progression of microangiopathy and, importantly, diabetic retinopathy [7, 39]. Moreover, SGLT2is have been demonstrated to reduce the total circulating volume, improve sodium balance and increase haematocrit, which may improve oxygen delivery to tissues [11, 40]. These effects may also influence risk of DMO. Regarding the GLP1-ra pharmacodynamics, a hypothesised mechanism through which GLP1-ra predisposes to diabetic retinopathy is via rapid reduction in HbA_1c which can lead to alterations in VEGF and IGF-1 that can lead to the development or early worsening of diabetic retinopathy [41, 42]. A Phase III clinical trial aiming to evaluate the effect of semaglutide on diabetic ocular disease is underway, with the results projected for 2027 [43].

The results of this study support the hypothesis that SGLT2i and GLP1-ra may have a beneficial role in preventing the development of DMO in individuals with type 2 diabetes taking insulin, but due to the study design they do not establish a causal link. There is a clear need to evaluate the impact of SGLT2i and GLP1-ra on diabetic eye disease in dedicated prospective studies, primarily to determine their ability to prevent and even treat DMO. Moreover, it is not known whether the association of GLP1-ra and diabetic retinopathy may be mitigated over time, as short-term tightening of diabetic control can precipitate worsening of diabetic retinopathy, but over the longer-term improved glycaemic control is beneficial in holistic diabetes outcomes.

Our study demonstrates that SGLT2i therapy is associated with a reduced risk of DMO in individuals with type 2 diabetes taking insulin. However, SGLT2i use was not associated with an altered risk of diabetic retinopathy, while GLP1-ra use was associated with an increased risk for diabetic retinopathy. Direct comparison of SGLT2i and GLP1-ra suggests favourable outcomes with SGLT2i in both the development of DMO and diabetic retinopathy. Dedicated RCTs appear justified to determine whether SGLT2i can reduce the risk of potentially blinding diabetic eye disease, particularly in people who are at high risk of DMO.