To assess the association between use of sodium–glucose cotransporter 2 (SGLT2) inhibitors and retinal vein occlusion (RVO) using data from the National Health Insurance Service in South Korea.
We used an active comparator, new user design, and nationwide data from 2014 to 2017. Based on a 1:1 propensity score match, we included 47,369 new users of SGLT2 inhibitors and 47,369 users of other glucose-lowering drugs (oGLDs). In the matched sample, we used the Cox proportional hazards model to estimate hazard ratios (HRs) with 95% CIs for developing RVO. Based on the main outcome, exploratory subgroup analyses were undertaken.
During a follow-up of 2.57 years, the incidence rate of RVO was 2.19 and 1.79 per 1,000 person-years in patients treated with SGLT2 inhibitors and oGLDs, respectively. The new use of SGLT2 inhibitors was associated with an increased risk of RVO compared with oGLD use (HR 1.264 [95% CI 1.056, 1.513]). In the subgroup analyses, a significant interaction with SGLT2 inhibitors was observed for age and estimated glomerular filtration rate (eGFR); the HR for RVO was higher in patients aged ≥60 years and those with eGFR <60 mL/min/1.73 m2 than in others.
In a matched cohort study, we found that SGLT2 inhibitors were associated with a significantly increased risk of RVO. Older patients and those with chronic kidney disease were at higher risk for RVO.
Introduction
Retinal vein occlusion (RVO) is the second most common cause of vision loss after diabetic retinopathy (DR) owing to retinal vascular disease and is an important cause of visual morbidity and blindness in older individuals (1,2). RVO incidence increases rapidly with age (3), and risk factors include hypertension, lipid abnormalities, atherosclerosis, diabetes, obesity, smoking, glaucoma, hyperviscosity syndrome, and retinal arteriolar abnormalities (4,5). Although the exact pathogenesis of RVO remains unclear, patients with type 2 diabetes have an increased risk of RVO (6,7). The prevalence of RVO in patients with type 2 diabetes is higher in Asians and West Indians than in Caucasian Europeans (8,9). RVO is diagnosed on the basis of variable painless vision loss and any combination of fundal findings, and treatment is directed at secondary complications of RVO that affect vision (10). If left untreated, RVO can result in permanent vision impairment or blindness because of retinal ischemia. Early recognition and prompt treatment are key to preserving vision and achieving good outcomes. Recently, several anti–vascular endothelial growth factor agents have become the treatment of choice for RVO (11).
Type 2 diabetes is one of the fastest growing and largest global health burdens, with high levels of morbidity and mortality (12). Hyperglycemia is associated with the risk of macrovascular and microvascular complications (13,14). Currently, various glucose-lowering drugs (GLDs) are available to reduce the development or progression of diabetes complications (15). Sodium–glucose cotransporter 2 (SGLT2) inhibitors lower blood glucose levels by increasing urinary glucose excretion and can also reduce body weight and blood pressure (BP) (16,17). These inhibitors show beneficial effects beyond glycemic control on metabolic, cardiovascular, and renal outcomes (18). RVO is a peripheral vascular occlusive disease that shares cardiovascular risk factors (19). Because patients with type 2 diabetes have a markedly higher rate of RVO than those without diabetes, assessing the risk of RVO associated with the use of new drugs for diabetes is important. However, to the best of our knowledge, the association of RVO with the use of SGLT2 inhibitors has not yet been investigated.
Using nationwide data from the National Health Insurance Service (NHIS) in South Korea between 2014 and 2017, we determined whether the use of SGLT2 inhibitors, compared with that of an active comparator, other GLDs (oGLDs), is associated with RVO. In addition, we conducted subgroup analyses to assess the clinical factors that interacted with SGLT2 inhibitors against RVO.
Research Design and Methods
Data Source and Study Population
We conducted an active comparator and new user design cohort study using data from the NHIS, which is a representative population-based sample cohort in South Korea (20). The cohort can be used by public health researchers and policymakers to evaluate the effects of medical practice on health outcomes. The NHIS database covers almost the entire South Korean population and includes information on participants’ insurance eligibility, medical treatment history, health care provider’s institution, and general health examination. Claims are accompanied by diagnostic data collected on ICD-10 codes, procedures, prescription drugs, patient personal information, hospital information, direct medical costs of both inpatient and outpatient care, and dental services. This study adhered to the tenets of the Declaration of Helsinki and was approved by the institutional review board of The Catholic University of Korea (study protocol no. SC20ZISE0098). The need for written informed consent was waived by the institutional review board.
From this cohort, we identified 146,435 new users of SGLT2 inhibitors and 1,776,333 new users of oGLDs between January 2014 and December 2017, from the date of availability of the first SGLT2 inhibitor in the country until the last available data. New use was defined as initiation of either drug class among patients with no previous use of either drug class within the past year. The first prescribed drug was defined as the index drug, and the first prescription date was designated as the index date. Patients aged ≥20 years who were diagnosed with type 2 diabetes were included. We also included patients who had undergone a health checkup within 4 years from the index date and were dispensed more than one drug within the past year. We excluded patients who had already been diagnosed with RVO before the index date and those with missing data. We estimated propensity scores and matched new users of SGLT2 inhibitors and oGLDs at a ratio of 1:1. The study group comprised 47,369 participants with an SGLT2 inhibitor and a control group of 47,369 participants. The first claim date of the SGLT2 inhibitor or oGLD was defined as the drug index date. A flowchart of the study population is shown in Supplementary Fig. 1.
Measurements and Definitions of Covariates
Height, weight, waist circumference, and BP were measured during regular medical checkups according to standardized methods. BP was measured three times, and the mean value of the second and third measurements was used for the analysis. BMI was calculated as weight in kilograms divided by height in meters squared. Blood samples for the measurement of fasting plasma glucose, hemoglobin, total cholesterol, HDL cholesterol, and triglyceride levels were obtained after an overnight fast before each examination. Estimated glomerular filtration rate (eGFR) was calculated using the abbreviated MDRD equation. Detailed histories of smoking status, alcohol consumption, and physical activity were obtained using a self-administered questionnaire. Individuals were categorized as nonsmokers, ex-smokers, or current smokers, and heavy alcohol drinkers were defined as individuals who drank >30 g of alcohol per day. Regular exercise was defined as either strenuous physical activity more than three times per week or moderate-intensity activity at least five times per week. Low income was defined as the lowest household income quintile. Comorbidities were defined by a combination of ICD-10 codes and self-reported medication history. The presence of hypertension was defined according to ICD-10 codes I10, I13, or I15; prescription of an antihypertensive agent; or systolic/diastolic BP ≥140/90 mmHg. The presence of dyslipidemia was defined according to the prescription of antidyslipidemic medication under ICD-10 code E78. DR diagnosis was defined according to ICD-10 code H36.0. Hospitals where these health examinations were performed are certified by the NHIS and subjected to regular quality control measures.
Study Outcomes and Follow-up
The main outcome of this study was the development of RVO. The classification of RVO is based on the site of retinal vein involvement: Central RVO (CRVO) occurs when a proximal retinal vein is occluded, leading to involvement of the entire retina, and branch RVO (BRVO) occurs when a distal retinal vein is occluded, leading to fluid leakage and hemorrhages on the retina (10). The RVO group included all patients who received inpatient and outpatient care for an initial diagnosis of RVO (Korean Standard Classification of Diseases code H34.8). Here, we used a composite of BRVO and CRVO and excluded subjects who had been treated for RVO. These cases were regarded as new incident cases of RVO. We followed patients from their first prescription until the occurrence of RVO, death, or the end of the study period. Patients were censored at the time of the first outcome event in the analyses of the primary outcome. The mean ± SD follow-up time for the primary intention-to-treat analysis was 2.57 ± 0.9 years.
Statistical Analyses
We used an active comparator, new user design to mitigate the risk of confounding. The 1:1 nearest neighbor propensity score matching (PSM) method (caliper width 0.2 of the SD of the logit propensity score) was applied to estimate the treatment effect (21). Variables that could potentially affect treatment assignment or outcomes were selected, including sociodemographic characteristics, comorbidities, and concomitant drugs (baseline characteristics listed in Table 1). The standardized difference was calculated to assess the relative imbalance between the two groups, with values of <0.100 considered to be adequately balanced. We used an intention-to-treat exposure principle in which patients were defined as being exposed to the study drug from cohort entry throughout follow-up. The disease-free probability of the primary outcome was calculated using Kaplan-Meier curves, while a log-rank test was used to analyze intergroup differences. The Cox proportional hazards regression model was used to compare the risk of developing RVO in the matched sample. Hazard ratios (HRs) with 95% CIs were estimated for developing RVO. We further performed exploratory subgroup analyses to assess effect modification of treatment in the PSM cohort. We used an interaction term between treatment status and subgroup to assess the heterogeneity of treatment differences according to subgroup status; in these analyses, we considered P < 0.05 to be statistically significant. We performed several sensitivity analyses to adjust for the differences in the follow-up duration between the two groups. Statistical significance was defined at a two-sided P < 0.05. All analyses were performed using the SAS 9.4 statistical software (SAS Institute, Cary, NC).
. | SGLT2 inhibitors . | oGLDs . | ASD . |
---|---|---|---|
Participants, n | 47,369 | 47,369 | |
Male sex | 26,719 (56.41) | 26,767 (56.51) | 0.0020 |
Age (years) | 57.23 ± 11.03 | 57.22 ± 10.57 | 0.0011 |
BMI (kg/m2) | 26.83 ± 3.99 | 26.92 ± 4.19 | 0.0236 |
Waist circumference (cm) | 88.86 ± 12.09 | 88.93 ± 9.94 | 0.0061 |
Systolic BP (mmHg) | 128.11 ± 14.85 | 128.06 ± 14.84 | 0.0035 |
Diastolic BP (mmHg) | 78.92 ± 10.06 | 78.87 ± 9.98 | 0.0052 |
Fasting plasma glucose (mg/dL) | 154.09 ± 54.41 | 155.13 ± 56.03 | 0.0188 |
eGFR (mL/min/1.73 m2) | 89.79 ± 26.74 | 89.51 ± 25.75 | 0.0104 |
Total cholesterol (mg/dL) | 190.1 ± 51.06 | 187.12 ± 47.35 | 0.0606 |
Triglycerides (mg/dL) | 152.54 (151.74–153.34) | 148.95 (148.17–149.73) | 0.0409 |
HDL cholesterol (mg/dL) | 49.73 ± 13.49 | 49.62 ± 13.75 | 0.0075 |
LDL cholesterol (mg/dL) | 105.81 ± 44.51 | 103.78 ± 43.02 | 0.0464 |
Hemoglobin (mg/dL) | 14.37 ± 1.63 | 14.25 ± 1.67 | 0.0701 |
Duration of diabetes (years) | 6.12 ± 3.84 | 6.07 ± 3.6 | 0.0137 |
Hypertension | 29,612 (62.51) | 29,703 (62.71) | 0.0039 |
Dyslipidemia | 35,958 (75.91) | 36,178 (76.37) | 0.0108 |
DR | 6,453 (13.62) | 6,860 (14.48) | 0.0247 |
Low income | 9,307 (19.65) | 9,252 (19.53) | 0.0029 |
Current smoker | 1,1701 (24.7) | 11,779 (24.87) | 0.0054 |
Heavy alcohol drinker | 4,277 (9.03) | 4,253 (8.98) | 0.0058 |
Regular physical activity | 9,759 (20.6) | 9,685 (20.45) | 0.0038 |
Hypoglycemic medications, previous 1 year | |||
Insulin | 5,520 (11.65) | 5,274 (11.13) | 0.0163 |
Sulfonylurea | 19,671 (41.53) | 18,680 (39.44) | 0.0426 |
Metformin | 43,478 (91.79) | 44,503 (93.95) | 0.0842 |
Thiazolidinedione | 4,904 (10.35) | 4,684 (9.89) | 0.0154 |
Dipeptidyl peptidase 4 inhibitors | 17,356 (36.64) | 16,814 (35.5) | 0.0238 |
Number of concomitant oral hypoglycemic agents, previous 1 year | |||
1 | 18,433 (38.91) | 18,386 (38.81) | 0.0163 |
2 | 19,416 (40.99) | 1,868 (39.44) | 0.0426 |
≥3 | 9,520 (20.1) | 10,301 (21.75) | 0.0842 |
Index year | |||
2014 | 3,553 (7.5) | 3,553 (7.5) | 0.0001 |
2015 | 9,663 (20.4) | 9,663 (20.4) | 0.0001 |
2016 | 15,033 (31.74) | 15,033 (31.74) | 0.0001 |
2017 | 19,120 (40.36) | 19,120 (40.36) | 0.0001 |
Follow-up duration (years) | 2.39 ± 0.93 | 2.74 ± 0.91 | 0.3822 |
. | SGLT2 inhibitors . | oGLDs . | ASD . |
---|---|---|---|
Participants, n | 47,369 | 47,369 | |
Male sex | 26,719 (56.41) | 26,767 (56.51) | 0.0020 |
Age (years) | 57.23 ± 11.03 | 57.22 ± 10.57 | 0.0011 |
BMI (kg/m2) | 26.83 ± 3.99 | 26.92 ± 4.19 | 0.0236 |
Waist circumference (cm) | 88.86 ± 12.09 | 88.93 ± 9.94 | 0.0061 |
Systolic BP (mmHg) | 128.11 ± 14.85 | 128.06 ± 14.84 | 0.0035 |
Diastolic BP (mmHg) | 78.92 ± 10.06 | 78.87 ± 9.98 | 0.0052 |
Fasting plasma glucose (mg/dL) | 154.09 ± 54.41 | 155.13 ± 56.03 | 0.0188 |
eGFR (mL/min/1.73 m2) | 89.79 ± 26.74 | 89.51 ± 25.75 | 0.0104 |
Total cholesterol (mg/dL) | 190.1 ± 51.06 | 187.12 ± 47.35 | 0.0606 |
Triglycerides (mg/dL) | 152.54 (151.74–153.34) | 148.95 (148.17–149.73) | 0.0409 |
HDL cholesterol (mg/dL) | 49.73 ± 13.49 | 49.62 ± 13.75 | 0.0075 |
LDL cholesterol (mg/dL) | 105.81 ± 44.51 | 103.78 ± 43.02 | 0.0464 |
Hemoglobin (mg/dL) | 14.37 ± 1.63 | 14.25 ± 1.67 | 0.0701 |
Duration of diabetes (years) | 6.12 ± 3.84 | 6.07 ± 3.6 | 0.0137 |
Hypertension | 29,612 (62.51) | 29,703 (62.71) | 0.0039 |
Dyslipidemia | 35,958 (75.91) | 36,178 (76.37) | 0.0108 |
DR | 6,453 (13.62) | 6,860 (14.48) | 0.0247 |
Low income | 9,307 (19.65) | 9,252 (19.53) | 0.0029 |
Current smoker | 1,1701 (24.7) | 11,779 (24.87) | 0.0054 |
Heavy alcohol drinker | 4,277 (9.03) | 4,253 (8.98) | 0.0058 |
Regular physical activity | 9,759 (20.6) | 9,685 (20.45) | 0.0038 |
Hypoglycemic medications, previous 1 year | |||
Insulin | 5,520 (11.65) | 5,274 (11.13) | 0.0163 |
Sulfonylurea | 19,671 (41.53) | 18,680 (39.44) | 0.0426 |
Metformin | 43,478 (91.79) | 44,503 (93.95) | 0.0842 |
Thiazolidinedione | 4,904 (10.35) | 4,684 (9.89) | 0.0154 |
Dipeptidyl peptidase 4 inhibitors | 17,356 (36.64) | 16,814 (35.5) | 0.0238 |
Number of concomitant oral hypoglycemic agents, previous 1 year | |||
1 | 18,433 (38.91) | 18,386 (38.81) | 0.0163 |
2 | 19,416 (40.99) | 1,868 (39.44) | 0.0426 |
≥3 | 9,520 (20.1) | 10,301 (21.75) | 0.0842 |
Index year | |||
2014 | 3,553 (7.5) | 3,553 (7.5) | 0.0001 |
2015 | 9,663 (20.4) | 9,663 (20.4) | 0.0001 |
2016 | 15,033 (31.74) | 15,033 (31.74) | 0.0001 |
2017 | 19,120 (40.36) | 19,120 (40.36) | 0.0001 |
Follow-up duration (years) | 2.39 ± 0.93 | 2.74 ± 0.91 | 0.3822 |
Data are mean ± SD or n (%), with the exception of data for triglycerides, which are presented as median (interquartile range). ASD, absolute standardized difference.
Results
Study Population
After 1:1 PSM, a matched cohort with 94,738 new users of SGLT2 inhibitors or oGLDs was generated (Supplementary Fig. 1). Of the total follow-up for SGLT2 inhibitors, the proportion by drug started at cohort entry was 70.1% for dapagliflozin, 22.3% for empagliflozin, and 7.6% for ipragliflozin. Before PSM, SGLT2 inhibitor users were younger, more obese, and had a higher prevalence of dyslipidemia than oGLD users. SGLT2 users had a higher rate of prescriptions for metformin, sulfonylurea, and thiazolidinedione (Supplementary Table 1). The baseline characteristics were well balanced betw-een the groups after PSM (all absolute standardized differences <0.1) (Table 1). Overall, the mean ± ΣΔ age was 57.2 ± 10.8) years, and 43.5% were female, 52.6% had hypertension, and 76.1% had dyslipidemia. Overall, the fasting plasma glucose level was 154.4 ± 55.1 mg/dL and the mean eGFR value was 89.44 ± 29.18 mL/min/1.73 m2. For RVO, the mean follow-up time was 2.39 ± 0.93) years among the SGLT2 inhibitor users and 2.74 ± 0.91) years among the oGLD users.
The Risk of RVO: SGLT2 Inhibitor or oGLD Treatment
Figure 1 shows the cumulative incidence of the primary outcome RVO. Kaplan-Meier curves show a substantial increase in RVO among SGLT2 inhibitor users compared with oGLD users (log-rank P = 0.0104). During follow-up, there were 480 events of incident RVO of which 248 occurred in the SGLT2 inhibitor group (incidence rate 2.19 per 1,000 person-years) and 232 occurred in the oGLD group (1.79 per 1,000 person-years) (Table 2). Use of SGLT2 inhibitors, compared with oGLDs, was associated with a significantly increased risk of RVO (HR 1.264 [95% CI 1.056, 1.513]) (Table 2).
. | n . | RVO cases, n . | Person-years . | Incidence rate (per 1,000 person-years) . | HR (95% CI) . |
---|---|---|---|---|---|
oGLDs | 47,369 | 232 | 129,820.82 | 1.78708 | 1 (reference) |
SGLT2 inhibitors | 47,369 | 248 | 113,177.26 | 2.19125 | 1.264 (1.056, 1.513) |
. | n . | RVO cases, n . | Person-years . | Incidence rate (per 1,000 person-years) . | HR (95% CI) . |
---|---|---|---|---|---|
oGLDs | 47,369 | 232 | 129,820.82 | 1.78708 | 1 (reference) |
SGLT2 inhibitors | 47,369 | 248 | 113,177.26 | 2.19125 | 1.264 (1.056, 1.513) |
HR, hazard ratio; CI, confidence interval; oGLD, other glucose-lowering drugs; SGLT2, sodium–glucose cotransporter 2.
Subgroup Analyses
Figure 2 shows the results of subgroup analyses based on baseline characteristics. No significant interactions were observed between SGLT2 inhibitor users and most subgroups with respect to the primary outcome. Effect estimates of SGLT2 inhibitors on RVO varied according to age (P for interaction = 0.0135) and eGFR (P for interaction = 0.0083). Patients aged ≥60 years had a significantly increased risk for RVO (HR 1.523 [95% CI 1.198, 1.936]), whereas those aged <60 years did not (HR 0.987 [95% CI 0.75, 1.301]). The effect estimate in patients who had an eGFR <60 mL/min/1.73 m2 was different from that in those with ≥60 mL/min/1.73 m2 (HR 3.134 [95% CI 1.554, 6.318] vs. HR 1.174 [95% CI 0.973, 1.416]).
Conclusions
In this nationwide cohort study of ∼100,000 PSM adults with type 2 diabetes, we found that new use of SGLT2 inhibitors was significantly associated with a greater risk of RVO than that with oGLD treatments. Exploratory subgroup analyses revealed that the treatment effect of SGLT2 inhibitors on the risk of RVO was mostly consistent with the findings of the main analysis, and a significant difference across age and eGFR was found. Patients aged ≥60 years and those with chronic kidney disease were at an increased risk for RVO after new use of SGLT2 inhibitors. These findings are relevant because of the common prescription of SGLT2 inhibitors and the importance of this outcome for patients. Ongoing monitoring and preventive education regarding the incremental risk of RVO should be provided when initiating SGLT2 inhibitors so that patients may gain optimum benefit from these medicines.
SGLT2 inhibitors are a new class of oral medications that lower glucose levels independent of insulin. These agents are generally well tolerated, but clinical experience and emerging clinical data have revealed several potential adverse effects, including genitourinary tract infections, euglycemic diabetic ketoacidosis, acute kidney injury, lower-extremity amputations, and bone fractures (22). There is a need to further examine the risk profile of SGLT2 inhibitors. RVO is associated with cardiovascular risk factors, such as atherosclerotic and thrombophilic factors (19). RVO is categorized as CRVO and BRVO based on the occlusion site, and the pathogenesis of CRVO or BRVO remains incompletely understood (4,5). Although the similarity of the underlying mechanism between CRVO and BRVO is unclear, their risk factors show some overlap (7,23). The pathogenesis of RVO is multifactorial, and damage to the retinal vessel wall from atherosclerosis and compression affects the adjacent vein, causing thrombogenesis involving stasis, hypercoagulability, and, thus, occlusion (24). We show that new use of SGLT2 inhibitors increases the risk of RVO, although the underlying mechanism is not clear.
SGLT2 inhibitors cause osmotic diuresis, which may lead to adverse reactions related to volume depletion in patients with type 2 diabetes (25–27). We speculate that by promoting glucosuria and intravascular volume contraction, and the subsequent hyperviscosity, in patients treated with SGLT2 inhibitors, the risk of RVO might increase. Increased blood viscosity results in abnormal erythrocyte aggregation, and high vascular resistance predisposes the patient to circulatory venous stasis. This combination makes the retinal vein particularly susceptible to thrombosis in the hyperviscosity state (28–30). Severe dehydration causes a reduction in plasma volume, leading to a state of acute hyperviscosity and resulting in venous thrombosis.
Retinal hyperviscosity is usually bilateral, but in a previous case report, RVO as a result of dehydration was identified in three patients with unilateral CRVO and in one patient with bilateral CRVO (31). The risk of RVO could increase in more vulnerable sites through the interaction of dehydration and concomitant ocular risk factors, including atherosclerosis, glaucoma, and acquired vascular changes (32). However, because of a lack of available data, we could not assess heterogeneity according to the RVO occurrence site. Further studies investigating the RVO site, extent, and severity with SGLT2 inhibitor use are needed.
The current study shows that the risk of RVO after treatment with SGLT2 inhibitors was significantly increased in patients aged ≥60 years and in those who had an eGFR <60 mL/min/1.73 m2. Older people are vulnerable to dehydration because of physiological changes, and dehydration is a risk factor of reduced renal function in older patients. In addition, a study reported that the risk of dehydration is particularly increased among patients aged >75 years with an GFR <60 mL/min/1.73 m2 and receiving diuretics (33). Our findings are consistent with the postulated hypothesis that RVO is an adverse event related to volume depletion. Care should be taken to ensure that adequate measures against dehydration associated with SGLT2 inhibitor use, including patient education, are implemented and that the risk of RVO is closely monitored, particularly in older patients and those with impaired renal function.
RVO is commonly associated with hypertension (34,35). Of the patients included in our study, 63% had a diagnosis of hypertension. In subgroup analyses, the incidence rate of RVO showed an increasing trend in patients with hypertension; however, the effect of SGLT2 inhibitors on RVO was not different in patients with and without hypertension. In addition, patients with DR had a higher incidence rate of RVO than those without; however, there was no difference in RVO risk with SGLT2 inhibitors according to DR status. Although the exact mechanism remains unclear, metformin use showed a protective effect against RVO in a population-based cohort study (36). In the specific GLD analyses of our study, the incidence rate of RVO did not decrease in patients treated with metformin. The risk of RVO was also no different when SGLT2 inhibitors were combined with metformin. The results across combinations of other drugs were consistent with the main outcome.
The pharmacological properties and adverse effects specific to individual SGLT2 inhibitors have been noted (37). In our study, we found no significant differences in the risk of RVO among different SGLT2 inhibitors (data not shown). While uncertainty regarding the potential effects specific to individual SGLT2 inhibitors remains, class-wide effects of volume depletion have been suggested (38). The current study confirmed the class effects of SGLT2 inhibitors on RVO. Further observational studies are needed to support our findings.
The strengths of the current study are the inclusion of a large sample size, which encompassed the entire South Korean population, and multiple subgroup analyses to inform clinical decision-making. There is clinical relevance attributed to a large sample cohort from a generalized population database comprising individuals with variations in age, comorbidities, and lifestyle. We applied the PSM method to estimate treatment effect; approximately one-half of the included patients were not selected in the PSM cohort. To address the cost of the data loss, we analyzed the data before PSM. Our conclusion was strengthened by the fact that we obtained similar results before and after PSM (Supplementary Table 2).
Our study also has some limitations. First, although we used an active comparator design and a new user design to address confounding by indication, our observational study could not completely rule out residual and unmeasured confounding factors. Second, the definition of exposure was based on filled prescriptions; low adherence might bias the results toward the null. Data on covariates and outcomes may be incomplete or misclassified; such misclassification would lead us to underestimate the true magnitude of the association between SGLT2 inhibitors and RVO. However, it is not likely to be different in patients receiving SGLT2 inhibitors from that in those receiving oGLDs. Third, the study was conducted in South Korea, and its potential for generalization to other populations and health care systems is unknown. Finally, the event rates in many subgroup analyses were low, resulting in wide CIs. The diagnosis of RVO was based on the Korean modification of ICD-10 codes, and information on the frequency of retinal screening was unavailable. Underreporting of asymptomatic RVO or misclassification was also possible, leading to underestimation of the incidence of RVO. Additional clinical data on SGLT2 inhibitor use are required for further well-powered analyses.
In a Korean nationwide population-based matched cohort study, we conclude that new use of SGLT2 inhibitors is associated with a significantly increased risk of RVO compared with the use of oGLDs. Older patients and those with chronic kidney disease have a greater risk for RVO than others. We emphasize that SGLT2 inhibitors should be used with caution in consideration of the risk of RVO, especially in older patients with impaired renal function.
This article contains supplementary material online at https://doi.org/10.2337/figshare.14824089.
Article Information
Funding. This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (2021R1A2C2013890) and a faculty grant of Myongji Hospital.
Duality of Interest.No potential conflicts of interest relevant to this article were reported.
Author Contributions. M.-K.L. drafted and edited the manuscript. M.-K.L., B.K., K.H., J.-H.L., M.K., M.K.K., K.-H.B., K.-H.S., H.-S.K., and Y.-J.R. commented on the submitted manuscript. M.-K.L., K.H., H.-S.K., and Y.-J.R. contributed to the study design, analysis, and interpretation of data. B.K. performed the statistical analysis of the data. H.-S.K., and Y.-J.R. provided supervision and revised the manuscript. All authors have read and approved the final manuscript. H.-S.K. and Y.-J.R. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.