This was a retrospective cohort study using data from the JMDC Claims Database (JMDC Inc., Tokyo, Japan; Fig. S1) between January 1, 2015, and September 30, 2022. The JMDC Claims Database is a large administrative database containing inpatient, outpatient, and pharmacy claims and health checkup data provided by health insurance associations in Japan. As of February 2024, cumulative data of approximately 17 million individuals are included. The study protocol was approved by the Medical Affairs Japan-Protocol Review Committee of Astellas Pharma Inc. (protocol no. 1941-MA-3477). The study was conducted in accordance with the protocol, the ethical principles of the Declaration of Helsinki, and all applicable regulations and guidelines governing clinical study conduct. Patient informed consent was not required because data used in this study were retrospectively collected and deidentified.

Patients included in this study had at least one prescription for either an SGLT2i or a DPP-4i on or after January 1, 2015 (see Table S1 for medication codes) and had a confirmed diagnosis of T2DM (International Classification of Diseases, Tenth Revision [ICD-10]: E11.x) or unspecified diabetes mellitus (ICD-10: E14.x) on or before the index date, where the index date was defined as the date of the first receipt of an SGLT2i or a DPP-4i, preceded by a minimum of 1 year of database record, on or after January 1, 2015. Patients were excluded from this study if they met any of the following criteria: aged  < 18 years at the index date; had used an SGLT2i or a DPP-4i, or a fixed-dose combination drug containing both an SGLT2i and a DPP-4i, during the pre-index period, which was defined as the period between the index date and 1 year before the index date; had used both an SGLT2i and a DPP-4i, including a fixed-dose combination drug containing both, on the index date; had a diagnosis of type 1 diabetes mellitus (ICD-10: E10.x) on or before the index date, or gestational diabetes (ICD-10: O24) during the pre-index period or on the index date;  < 1 year of database record before the index date; or had a diagnosis of microvascular or macrovascular disease (see Table S2 for microvascular and macrovascular disease codes), DR, or DME (see Table S3 for DR, DME, and ophthalmoscopy codes) during the pre-index period or on the index date.

The primary endpoint was the first record of DR diagnosis (including DME) preceded by a record of ophthalmic examination  ≤ 1 month prior to the month of the DR diagnosis (see Table S3 for DR, DME, and ophthalmoscopy codes). The secondary endpoint was the first record of ophthalmologic procedures related to DR (including DME) following a DR diagnosis preceded by an ophthalmic examination record as defined for the primary endpoint above (see Table S4 for ophthalmologic procedure codes). Focusing solely on DME, the exploratory endpoint was the first record of DME diagnosis preceded by ophthalmic examination  ≤ 1 month prior to the month of the DME diagnosis (see Table S5 for DME and relevant ophthalmic examination codes). For DR and DME to be considered as events, both ophthalmoscopy and diagnosis had to occur on or after the index date.

All patients meeting the eligibility criteria were included in the analysis and were matched 1:1 (SGLT2i:DPP-4i) using propensity score matching, which was calculated using a multivariate logistic regression model. Baseline variables used as covariates for propensity score matching were sex, age, index year, hospital size (i.e., number of beds), smoking status, BMI, dyslipidemia, hypoglycemia, Charlson Comorbidity Index (CCI), adapted Diabetes Complication Severity Index (aDCSI), coronary revascularization, SBP, HbA1c, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), medications (antidiabetic drugs, antihypertensive drugs, antiplatelet drugs, renin-angiotensin system [RAS] inhibitors, antihyperlipidemic drugs), treatment status of antidiabetic drug (treatment-naïve and treatment non-naïve), and cataract surgery. Code lists for CCI and aDCSI are shown in Tables S6 and S7. Missing data were not imputed; if data for any of the variables used for matching were not recorded during the baseline period, defined as the pre-index period plus the index date, patients were not included for matching or the subsequent analyses. Variables that had  ≥ 25% missing baseline data were excluded completely from the propensity score calculation. The degree of balance was assessed using standardized mean difference (SMD), where an SMD > 0.1 indicated a residual imbalance between the treatment groups.

Baseline characteristics, laboratory measurements, chronic conditions, and relevant medical history were summarized descriptively for the crude population and the matched cohort. These baseline data were extracted during the baseline period, except for BMI, SBP, HbA1c, and laboratory measurements, including LDL-C and HDL-C, which were measured within 180 days before the index date. Patients were followed from the index date to the earliest of the following events: clinical event of interest (DR diagnosis, ophthalmologic procedures related to DR, and DME diagnosis), treatment discontinuation, end of patient data, and death. Treatment discontinuation was defined using a grace period of 60 days; patients were considered to have discontinued the treatment if the difference between the prescription date plus the days of medication administration and the current prescription date was  ≥ 61 days or if they switched or added on the other index drug (SGLT2i or DPP-4i). The incidence of DR was compared between the treatment groups as primary analysis by evaluating the event-free rate. In the primary analysis, DR was defined as having any of the Japan-specific standard disease or procedure codes for DR or ophthalmoscopy, listed in Table S3, excluding retinopathy (standard disease code: 3,621,043), maculopathy (standard disease code: 3,625,036), chronic retinopathy (standard disease code: 8,840,403), circinate retinitis (standard disease code: 8,841,100), and neovascular maculopathy (standard disease code: 8,849,638). The incidence rates (IRs) per 1000 PY were calculated for each group, and cumulative IRs were plotted using the Kaplan-Meier method. Hazard ratio (HR) and its 95% confidence interval (CI) were calculated using Cox proportional hazard models to compare DR risk between the groups; P-values were calculated using log-rank tests, with a two-sided significance level of 0.05. The same analyses were performed to compare the incidence of ophthalmologic procedures related to DR and incidence of DME. Sensitivity analyses were performed for the primary endpoint by calculating the IRs of DR diagnosis and adjusted HRs using a shorter grace period of 30 days and a broader disease definition (i.e., having any of the DR or ophthalmoscopy codes listed in Table S3). Multivariate Cox regression analysis was performed using the crude population to estimate adjusted HRs. Furthermore, subgroup analyses were performed for the primary endpoint. Subgroup analyses were based on BMI (< 25 kg/m2, ≥ 25 kg/m2), SBP (< 130 mmHg, ≥ 130 mmHg), HbA1c (< 7%, 7–8%, > 8%), age (< 65 years, ≥ 65 years), concomitant use of lipid-lowering drugs (yes, no), concomitant use of RAS inhibitors (yes, no), and treatment status of antidiabetic drugs (treatment-naïve, treatment non-naïve). All statistical analyses were conducted using SAS 9.4 or higher (SAS Institute Inc., Cary, NC, USA).