For this study, pseudonymized real-world dataset was provided by Allied Medical, K.K. The dataset is based on medical practice data, comprising 58,682 patients from 34 primary care clinics across Japan. The dataset is owned by Allied Medical, K.K., and the research team conducted the analysis with permission granted by the company.

In this study, the initiation of treatment is defined as the first visit accompanied by a prescription for diabetes medication, and we included patients who initiated treatment after January 2015. The first 90 days from the first visit were considered the baseline period, with the time afterward defined as the observation period. Patients were included based on the following criteria: (1) having at least one measurement of HbA1c and estimated glomerular filtration rate (eGFR) during the baseline period, (2) having a prescription proportion of days covered (PDC) [15] of 0.75 or higher for the medication of interest (BGs or DPP-4is) during the baseline period. Patients who met the following criteria were excluded: (1) having PDC of 0.75 or higher for both BGs and DPP-4is during the baseline period, (2) experiencing endpoint during the baseline period or prior to the first clinic visit at other clinics (self-report). Patients were divided into the BG arm and the DPP-4i group based on their prescription patterns during the baseline period, and analyses were conducted regardless of any changes in prescription patterns during the observation period.

The study was approved by the ethics committee of Health Outcome Research Institute (submission ID 2024-06). Informed consent was waived by the ethics committee.

Complications related to T2D consisting of renal, neurological, vascular, retinal, and skin complications were comprehensively listed and adopted as a composite endpoint. In this study, we used this composite endpoint as the primary outcome for the incidence of diabetes-related complications. As indicators of the drug safety, hepatic events and cardiac events were included as the secondary outcome. All relevant event names and corresponding ICD codes are shown in Supplementary Table S1.

In addition to health-related outcomes, we calculated and compared the medical costs and their breakdown from the start of treatment until the incidence of an event or loss to follow-up. We adopted the healthcare payers’ perspective, under which all medical costs were included regardless of copayment. Costs were calculated by multiplying the actual usage of resources by the unit prices defined in Japan's National Medical Care Fee Schedule and the National Drug Tariff as of April 2021. For discontinued items, we used the latest available unit price. Under Japan's public health insurance system, all medical practices, except for hospitalization in tertiary medical facilities, are charged on a fee-for-service system. The government strictly regulates the unit prices of medical services to ensure they are uniform across all clinics [16].

After the patients were split into the BG group and the DPP-4i group, one-to-one nearest-neighbor propensity score (PS) matching using logistic regression was conducted to mitigate influences of confounding factors between the groups. The following variables were incorporated to calculate propensity scores: age, sex, disease history (hepatic and cardiac), baseline renal function (eGFR), and baseline severity (HbA1c). The list of event names included in the disease history can be found in Supplementary Tables S2 and S3.

Considering the relatively small population size and number of events, we examined the impact of the type of medication and the prescription intervals during the initial treatment phase (baseline period) on the incidence of T2D-related complications using univariate survival analysis with the log-rank test. The safety of the medication, a secondary outcome, was also evaluated using the log-rank test based on the type of medication.

To further explore the relationship between T2D-related complications and various factors, we conducted a more advanced survival analysis: (1) bootstrap sampling (80%) from the dataset, (2) fit a cross-validated LASSO–Cox proportional hazards model incorporating the following explanatory variables: type of initial medication, initial prescription intervals, age, sex, HbA1c, eGFR, and medical history (3) repeat steps 1 and 2 a hundred times, count the number of times a non-zero coefficient was observed for each variable, and retain variables that were observed more than ten times, (4) fit a Cox proportional hazards model to the entire dataset using the variables retained from step 3 to investigate the impact of each variable on the outcome. Additionally, considering the nature of T2D, we conducted a sensitivity analysis by excluding events occurring or censoring within 180 days from the start of the baseline period, focusing on patients who had a relatively stable course.

Regarding the costs, we divided the patients into four groups based on the type of initial medication (BG or DPP-4i) and the length of the prescription interval (less than 40 days or 40 days and above). We then calculated and compared the annual cost per person for each group.

All statistical analyses were performed using Python version 3.7.9 and R version 4.2.0. Analysis items with a two-tailed P value of < 0.05 were considered statistically significant.