Rapid Reduction of HbA_1c and Early Worsening of Diabetic Retinopathy: A Real-world Population-Based Study in Subjects With Type 2 Diabetes

Diabetic retinopathy (DR) is one of the most frequent complications of diabetes and remains as the first cause of preventable blindness in the working-age population (1,2). Currently, the global prevalence of this complication is still high, affecting 2 in every 10 patients with diabetes, and it is expected to remain high through 2045, especially among countries in the Middle East, North Africa, and the Western Pacific (3).

Poor glycemic control and hypertension are the main modifiable risk factors for reducing the development and progression of DR (4). However, the rapid improvement of hyperglycemia can also be a risk factor for progression of DR. An early worsening of DR (EWDR) was observed in 13.1% of 711 patients with type 1 diabetes assigned to intensive treatment in Diabetes Control And Complications Trial (DCCT) at 6 and/or 12 months in comparison with 7.6% of 728 patients assigned to conventional treatment (odds ratio [OR], 2.06; P < 0.001) (5). A similar phenomenon was reported in subjects with type 2 diabetes after a rapid improvement of blood glucose levels when they were changed from oral agents or diet alone to insulin therapy (6,7).

The main variables accounting for the EWDR are the magnitude of the reduction of HbA_1c and the presence of pre-existent DR (5,8,9). Regarding the former, it should be particularly relevant when HbA_1c is reduced by >1.5% within 3 months or >2% within 6 months (10,11). Therefore, the higher the reduction of HbA_1c, the higher the risk of EWDR. Notably, although this reduction generally occurs after treatment with a highly effective antidiabetic drug treatment, EWDR has also been reported after bariatric surgery (12). This observation points to the reduction of blood glucose levels rather than any particular antidiabetic agent as the main reason involved in EWDR. Nevertheless, the potential association between glucagon like peptide-1 receptor agonists (GLP-1RA) (in particular, semaglutide) and EWDR is still under debate (13,14). The presence of DR is also an essential risk factor of EWDR (10,11). However, the influence of the degree of DR remains to be elucidated. This is because most of the studies have included subjects without data regarding staging of DR.

On this basis, the aim of the current study is to evaluate, in a nested case-control study, whether EWDR is also an issue in patients with type 2 diabetes with mild or moderate nonproliferative DR (NPDR), which represent the vast majority of patients attended in primary care (15,16). In addition, the relationship between EWDR and the different antidiabetic drugs was also analyzed.

We conducted a retrospective nested case-control study in a cohort of subjects with type 2 diabetes with a previous history of DR. The primary care SIDIAP (“Sistema d'informació pel Desenvolupament de la Recerca a Atenció Primària”) database was used as the data source. The SIDIAP database routinely collects pseudoanonymized health care data from users who attended the primary health care centers of Catalonian Health Institute (Institut Català de la Salut). Institut Català de la Salut is the largest health care provider in Catalonia (Spain), covering about 80% (5,564,292 persons) of the Catalonian population. The SIDIAP database contains different patient data, such as visits with health care professionals, diagnostic codes, demographic information, clinical variables, laboratory tests results, prescriptions, referrals to specialists and hospitals, and medication dispensed in pharmacies (17). For this analysis, data from an 11-year period were extracted (2010–2020 inclusive).

As inclusion criteria, all subjects were 30 years or older, with a diagnosis of type 2 diabetes (defined as the presence of International Classification of Diseases, 10th Revision diagnostic codes: E11 (and their subcodes) in clinical history. Additional inclusion criteria were having at least two screening procedures of fundus photography during the observational period and at least two HbA_1c determinations before the inclusion date (screening date when case and control subjects were identified). We excluded subjects with other types of diabetes (type 1 diabetes, gestational diabetes, secondary, or other), and those in whom severe NPDR, PDR, or clinically significant diabetic macular edema (CSDME) was diagnosed.

DR screening was performed at primary care health centers by trained general practitioners using eye fundus photographs (FP) taken with digital 45° color cameras. Two FP from each eye were taken (macula centered, and between macula and the optic nerve). When DR was suspected, the diagnosis was confirmed by ophthalmologists. The diagnosis of DR was taken from the worst affected eye and the most recent FP in case of more than one evaluation during the study period.

In the SIDIAP database, the results of FP were categorized into five possibilities following ICDR (International Clinical Diabetic Retinopathy) classification (18): normal FP, mild NPDR, moderate NPDR, severe NPDR, and PDR. Diabetic macular edema was not graded and was only considered when it was clinically significant (CSDME), and required a specific intervention by an ophthalmologist (i.e., intravitreal injections of anti-VEGF agents and/or corticosteroids, or photocoagulation). From a practical point of view, according to the degree of decrease in vision (vision-threatening DR [VTDR]) (19), these six categories were regrouped into three functional categories: normal FP (absence of visible DR), non-VTDR (comprising mild and moderate NPDR), and VTDR (comprising severe NPDR, proliferative DR, and CSDME). For the purpose of the study, only non-VTDR were considered in the analysis. The family physicians made an initial evaluation of the fundus, and pathological images were sent to a reference ophthalmologist. For data validation, 230 retinographies out of 2,300 subjects included in the current study (10% of the total sample) were randomly selected and evaluated by an expert ophthalmologist. Interrater agreement between the ophthalmologist and the SIDIAP database (κ coefficient) was 0.83, thus indicating a good level of agreement. Considering the ophthalmologist readout as the gold standard, we obtained a sensitivity of 89% and a specificity of 94%. These data revealed a high reliability on the classification of DR provided by the SIDIAP database. These findings were consistent with the results of the validation of the protocol for screening in our primary health care organization (20).

We considered as case subjects those subjects in whom DR progressed (from mild or moderate NPDR to VTDR). Candidate control subjects were subjects with type 2 diabetes without worsening of DR during the observation period in two or more consecutive screening visits.

Groups (case subjects and control subjects) were paired (ratio = 1:1) by age, diabetes duration, baseline HbA_1c (nearest method), year of inclusion (exact method), and sex (exact method), using the propensity score matching using MatchIt R package (21). Supplementary Fig. 1 shows the flowchart of the study.

At inclusion date, we collected sociodemographic variables (i.e., age, gender), toxic habits (i.e., current tobacco use), and clinical variables related to diabetes (i.e., diabetes duration, and HbA_1c levels). Data on cardiovascular risk factors such as BMI, total cholesterol, LDL cholesterol, HDL cholesterol, non-HDL cholesterol, and blood pressure were also extracted from the database. Additional data on medication were obtained. The estimated glomerular filtration ratio (eGFR) was calculated according to the Chronic Kidney Disease (CKD) Epidemiology Collaboration equation (22).

Main exposure indicators on HbA_1c reduction were computed as maximum absolute difference between two previous HbA_1c (%) in 12 months, monthly reduction of HbA_1c calculated as the maximum absolute difference between two HbA_1c/time in months between the two assessments, “rapid reduction” of HbA_1c if the difference absolute between two HbA_1c was > 1.5% in less than 12 months, and “very rapid reduction” of HbA_1c if the monthly change between previous HbA_1c corresponds to a reduction greater than 2% in 6 months.

First, we carried out a descriptive statistical analysis, summarizing the quantitative variables with the mean and its SD, and the qualitative variables with the frequency and its percentage. Description and comparison by groups was performed with compareGroups R package (23) using t test, and Fisher exact test for quantitative and categorical variables.

To assess the association of the principal exposure variables and worsening of DR, unadjusted and adjusted OR with 95% CIs were estimated and summarized. Different logistic regression models were developed to analyze the relationship among previous expositions, and case occurrence. Models were adjusted by clinical variables considering potentially confounding such as age, sex, diabetes duration, BMI, systolic blood pressure, tobacco, baseline HbA_1c, albumin/creatinine ratio, history of coronary heart disease, and antidiabetic drugs, from unadjusted model to fully adjusted model. P values less than 0.05 were considered statistically significant. Statistical analysis was performed with R statistical software, version 4.1.1. The R packages used were MatchIt (version 4.2.0) (21), compareGroups (version 4.6.0) (23), and stats (version 4.1.1) (https://www.r-project.org) (24).

This study was approved by the Institutional Review Board (or Ethics Committee) of IDIAP Jordi Gol i Gurina Foundation, Barcelona, Spain (protocol code 21/049-P and date of approval 24 March 2021).

After applying the study eligibility criteria, 19,902 subjects were identified during the observation period. From those, 1,150 met the criteria for case subjects, and 18,537 were identified as potential control subjects.

The elapsed time between the two photographic screenings was a median of 1.77 (interquartile range [IQR] 1.09–3.15) years for case subjects and 1.95 [IQR 1.15–3.46] years for the control group (P = NS). The second set of retinographies were performed at 2.8 [IQR 1.2–5.2] months after the last HbA_1c in the case subjects and at 3 [1.2–5.2] months in the control subjects (P = NS).

The baseline characteristics of case and control subjects are shown in Table 1. Both groups were well balanced for matching variables (age, sex, duration of diabetes, baseline HbA_1c, and date of inclusion). Patients in whom EWDR was observed presented significantly higher values of systolic blood pressure and albumin excretion rate, higher frequency of coronary heart disease, more use of antiplatelet drugs, and more advanced stages of NPDR.

We found 311 case subjects (27.0%) and 307 control subjects (26.7%) with good baseline glycemic control (HbA_1c < 7%). Both case and control subjects with HbA_1c ≥ 7% at baseline had a higher percentage of moderate NPDR than those with good control, without reaching statistical significance (50.4% vs. 44.4%, P = 0.069, and 25.1% vs. 22.8%, P = 0.413).

We did not observe any statistical difference in the absolute reduction of HbA_1c between case and control subjects (0.13 ± 1.21 vs. 0.21 ± 1.18; P = 0.12). The final HbA_1c achieved in cases and control subjects was 7.94 ± 1.58 vs. 7.87 ± 1.50, respectively (P = NS). The elapsed time to evaluate HbA_1c improvement was also similar between case and control subjects (median 6.4 [IQR 5.0–8.3] vs. 6.4 [IQR 4.8–8.3] months; P = NS). In addition, the monthly reduction of HbA_1c was also similar between case and control subjects (0.04 ± 0.30 vs. 0.05 ± 0.31; P = 0.41). Notably, the percentage of patients according the criteria used to define EWDR (rapid or very rapid reduction of HbA_1c) was similar between case and control subjects (Fig. 1). All case subjects with mild NPDR at baseline (n = 589; 51.2%) progressed to VTDR (severe NPDR, PDR, and CSDME). This means that in this group, all subjects experienced an increase of more than two steps in the Early Treatment Diabetic Retinopathy Study (ETDRS) severity level. In addition, those patients with moderate NPDR at baseline progressed to severe NPDR (n = 339; 60.4%), PDR (n = 109; 19.4%), and CSDME (n = 113; 20.1%). In this group, those subjects who progressed to PDR and CSDME can be considered as having experienced an increase of two or more steps in the ETDRS scale. However, those patients who progressed from moderate to severe NPDR in the ICDR scale experienced an increase of one step (from level 47 to level 53) or two steps (from level 35 or 43 to level 53) in the ETDRS scale. Therefore, we can state that at least 811 out of 1,150 case subjects (70.5%) progressed two steps in the ETDRS scale.

The results of different logistic regression models adjusted by potential confounding factors are shown in Fig. 2. We did not find any significant association between the reduction of HbA_1c (“rapid” or “very rapid”) and the EWDR either in the unadjusted model or after progressive adjustments. When stratification by baseline HbA_1c was performed, we did not find that those patients with higher levels of HbA_1c (>9.5%) in whom a “rapid” or “very rapid” reduction of HbA_1c was achieved presented a higher risk of EWDR (Fig. 3).

When the analysis was performed stratifying the population by antidiabetic drugs used, we also confirm the lack of relationship with EWDR (Supplementary Table 1).

Our results suggest that the reduction of HbA_1c is not associated with progression of mild or moderate NPDR. This is important because the vast majority of patients with type 2 diabetes have no DR or only present early stages of the disease. Therefore, in these patients, the optimization of metabolic control should not be delayed. In addition, we did not find any association between EWDR and the use of any particular antidiabetic drug.

Although there is no doubt regarding the beneficial impact of glycemic control on long-term development and progression of DR, EWDR has been reported as a consequence of rapid improvement of hyperglycemia. This potential complication in the management of patients with DR is generally well known by ophthalmologists and diabetologists, but it has been less recognized by nonspecialized physicians and health care providers. However, after the publication of the SUSTAIN-6 clinical trial (25), the EWDR has been emphasized and revisited. This study demonstrated significant beneficial results in cardiovascular outcomes (CVOT) in those patients treated with semaglutide, but also identified a significant increase in the rate of severe “DR complications” (i.e., vitreous hemorrhage, blindness, or conditions requiring treatment with an intravitreal agent or photocoagulation) in comparison with placebo (3% vs. 1.8%) (25).

Although the percentage of DR complications was low, they were clinically relevant. The main reason accounting for these sight-threatening complications was the rapid lowering in blood glucose levels and the inclusion of patients with advanced stages of DR. It should be noted that in the PIONEER 6 study, in which semaglutide was also used, but patients with advanced stages of DR (i.e., proliferative DR or maculopathy requiring acute treatment) were excluded, EWDR was not detected (26). The results herein reported also support this concept. Taken together, it seems clear that the degree of DR is crucial for the potential deleterious effect of a rapid decrease in glucose levels, and, therefore, an appropriate grading of DR at study entry and follow-up is needed in any randomized controlled trial aimed at evaluating DR outcomes.

The underlying mechanisms involved in EWDR remain unclear (27). The most robust hypothesis is based on the concept of the autoregulation of capillary blood flow at the retinal level (28). Since hyperglycemia is an important factor involved in capillary blood flow and shear stress, the rapid reduction of blood glucose levels could lead to changes in blood flow, thus precipitating the progression from severe NPDR to PDR or triggering hemorrhages in those patients with PDR. However, our results suggest that the clinical impact of this eventual pathogenic mechanism in early stages of DR is questionable. It is worth mentioning that, apart from a more advanced stage of DR, we have observed higher blood pressure, coronary artery disease, and microalbuminuria in the EWDR group. This finding suggests a more compromised endothelial system in these patients, and, therefore, it is possible that those patients with more cardiovascular burden are more prone to EWDR. However, specific studies addressed to investigating this issue are needed.

The strong relationship between the optimization of HbA_1c and the beneficial effects on DR has obscured the necessity of performing clinical trials to investigate the potential direct effect of antidiabetic drugs on the development or progression of DR, independently of their effectiveness in lowering blood glucose levels. The lack of studies addressed to answering this specific question means that clear information on this issue is missing. Nevertheless, several meta-analyses and real-world population-based studies did not find that dipeptidyl peptidase 4 inhibitors (DPP-4i), GLP-1RA, and sodium–glucose transporter 2 (SGLT2) inhibitors have a higher risk of DR than placebo (29–31). By contrast, some evidence suggests that sulfonylureas may be associated with increased risk of DR (30). In the current study, we did not find any relationship between antidiabetic drugs and EWDR in a real-world population-based sample.

As a result of the SUSTAIN-6 study, several cohort studies (32,33), meta-analyses (30,34), and scholar reviews (13) have been performed. These studies have shown that GLP-1RA did not increase the risk of development and progression of DR. Furthermore, two independent reports of the Food and Drug Administration Adverse Reporting System on this issue supported this conclusion (33,35). More recently, specific meta-analysis and metaregression studies examining the effect of GLP-1RA on DR including only randomized controlled trials on CVOTs have been performed (36,37). These studies have shown an association between GLP1-RA (i.e., liraglutide, semaglutide, and dulaglutide) and EWDR, but, again, the rapid reduction of HbA_1c rather than any eventual retinal adverse effect of these drugs seemed to be the main mechanism involved. It is worth mentioning that none of the included CVOTs were designed or powered to assess retinopathy outcomes.

Overall, it seems that the intensity and velocity in the reduction of HbA_1c rather than the antidiabetic agent used to exert this action is the crucial factor for the EWDR. In this regard, EWDR has been reported after bariatric surgery (12), thus reinforcing this concept that the risk of EWDR is inherent not to any drug but to their capacity of lowering HbA_1c. Nevertheless, the ongoing FOCUS trial (ClinicalTrials.gov identifier NCT03811561) will examine long-term effects of semaglutide compared with placebo on DR using validated and standard ophthalmic assessments and will clarify whether or not semaglutide has a direct retinal effect in DR progression.

Our study has several limiting factors. First, patients with advanced DR were not included, and, therefore, our results can only be applied to patients with type 2 diabetes with mild or moderate NPDR. It should be noted that subjects with advanced stages of DR (i.e., NPDR, PDR, and CSDME) would need interventional treatment (i.e., photocoagulation and/or intravitreal injections of anti-VEGF agents or corticosteroids) in a relatively short period by a specialized ophthalmologist. Because of the setting and nature of our study, we preferred not to include this sight-threatened population. The second limiting factor is that the relatively fair metabolic control observed in our cohort might minimize the effect of rapidly lowering the blood glucose levels. Therefore, further studies including more subjects with poor glycemic control would be needed. Finally, the ICDR classification is not as accurate as ETDRS severity levels, but it is feasible in current clinical practice. Considering the equivalences between the two classification systems (38), we found a progression of two steps in ETDRS scale in at least 70.5% of case subjects, a result that is clinically meaningful.

In summary, clinicians should not be afraid to optimize blood glucose in a relatively short time period in those subjects with mild or moderate NPDR. However, this strategy cannot be extended to sight-threatening DR, in which a careful decrease in HbA_1c should be planned. Therefore, presence and degree of DR is relevant information that should be available before initiating an intensive glucose-lowering approach.

Acknowledgments. The authors thank Dr. Pere Romero (Ophthalmology Service, University Hospital Sant Joan, Institut d’Investigació Sanitaria Pere Virgili, University of Rovira i Virgili, Reus, Spain) for his time spent evaluating retinographies from a subset of subjects included in the study.

Funding. This study was supported by an unrestricted grant from Novo Nordisk Pharma S.A.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. R.S., J.F.-N., J.R., and D.M. conceived and designed the study. J.R. and J.F. analyzed data. All authors interpreted the data. R.S. and J.F. wrote the first draft of the manuscript. All authors participated in its critical review with important intellectual contributions, and approved the final version of the manuscript. J.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.