Biguanides Beat SGLT2 Inhibitors on Blood Sugar in Japanese Database Study

When Japanese doctors choose between two of the most important drug classes for type 2 diabetes, the decision appears to hinge less on body weight than current guidelines might suggest. A retrospective analysis of 1925 patients from a nationwide clinical database found that the use of diuretics and other cardiovascular medications was the strongest factor distinguishing who received a biguanide and who received a sodium-glucose cotransporter-2 (SGLT2) inhibitor. Body mass index did influence prescribing, particularly among patients adding a second glucose-lowering drug, but it was not the dominant driver of treatment choice in routine practice.

The study, published in the journal Drugs – Real World Outcomes, was led by Ryo Mishima and Daisuke Koide and drew on the Millennium Medical Record Database, a repository that integrates electronic health records, diagnosis procedure combination data and insurance claims from 54 medical institutions across Japan. The database held approximately 1.79 million patient records as of April 2023, and about 70 percent of those records come from large hospitals with more than 300 beds. Because the database collects identifiable information, researchers can link multiple data sources and track patients across participating institutions, which mitigates a common limitation of electronic health record databases regarding patient transfers.

The research team focused on adults aged 18 years or older who received a first prescription of either a biguanide or an SGLT2 inhibitor between 1 July 2019 and 30 April 2023. Patients who received both drug classes at the same time were excluded, as were those with type 1 or gestational diabetes. The investigators also excluded patients with an estimated glomerular filtration rate below 30 mL/min/1.73 m2, for whom metformin is contraindicated, along with those on renal replacement therapy and those diagnosed with renal failure or heart failure, because these conditions could substantially restrict treatment selection. Each patient needed at least 90 days of data before the index prescription and at least 28 days of follow-up afterward.

Patients were divided into two cohorts. The first-line cohort included 1440 people with no previous antidiabetic medication before the index prescription, of whom 861 started a biguanide and 579 started an SGLT2 inhibitor. The second-line cohort included 485 people with prior antidiabetic medication history, of whom 302 started a biguanide and 183 started an SGLT2 inhibitor. The median follow-up time in the full analysis set was 315 days. Patients remained in their initially assigned treatment groups even if they switched medications during follow-up, and no formal sample size calculation was performed because the analysis was exploratory and included all available data.

To identify the factors behind prescribing decisions, the researchers used classification and regression tree analysis built on propensity scores estimated from a logistic regression model. The candidate variables spanned demographics, diabetes complications, comorbidities, concomitant medications, laboratory values including HbA1c and kidney function, and measurements such as body mass index and blood pressure. For first-line therapy, the most influential variable for classification was diuretic use, which accounted for 65.7 percent of relative importance, followed by lipid-modifying agents at 15.6 percent and baseline HbA1c at 13.1 percent. For second-line therapy, baseline HbA1c ranked first at 35.9 percent, followed by body mass index at 26.5 percent and diuretics at 17.9 percent.

The baseline characteristics reinforced this picture of differential cardiovascular risk. In the first-line group, patients who received SGLT2 inhibitors had roughly 8 percentage points more males, approximately 15 percentage points higher rates of cardiovascular risk management medication, and a median estimated glomerular filtration rate 7 mL/min/1.73 m2 lower than the biguanide group. Median HbA1c was 0.5 percentage points lower in the SGLT2 inhibitor group, while body mass index differed by only 0.2 kg/m2. Similar patterns appeared in the second-line group, where the SGLT2 inhibitor group had a 0.4 point lower median HbA1c and a 0.9 kg/m2 higher median BMI. After matching, covariate balance improved in both cohorts, with absolute standardized mean differences below 0.25.

The researchers addressed confounding with a coarsened exact matching procedure that used the terminal nodes of the classification trees as propensity score strata, configured as a full-matching approach so that no patients were algorithmically excluded. Missing covariate data were handled through multiple imputation by chained equations, generating 20 imputed datasets. For glycemic outcomes, the team applied a mixed model for repeated measures with matching weights, which included fixed effects for treatment group, time point, their interaction and baseline HbA1c. A total of 1378 patients entered the HbA1c analysis after exclusions for missing follow-up values.

Both drug classes produced clinically meaningful glucose control. In the first-line cohort, weighted analysis showed that HbA1c fell by 1.37 percentage points (95% CI -1.48 to -1.26) with biguanides and by 1.16 percentage points (95% CI -1.29 to -1.02) with SGLT2 inhibitors at 24 weeks, a between-group difference of 0.21 percentage points favoring biguanides (95% CI -0.38 to -0.04). In the second-line cohort, the reductions were larger: 2.26 percentage points (95% CI -2.47 to -2.05) for biguanides and 1.79 percentage points (95% CI -2.04 to -1.54) for SGLT2 inhibitors, with a between-group difference of 0.47 percentage points favoring biguanides (95% CI -0.80 to -0.15). The authors noted that the difference was modest in the first-line setting but numerically larger in the second-line setting.

Treatment persistence, defined as the duration from initiation of the index drug to discontinuation with a 60-day grace period after the last prescription, told a different story. In intention-to-treat analyses using weighted Kaplan-Meier estimation, persistence at 48 weeks was 47.0 percent (95% CI 42.7-51.2) for biguanides and 56.2 percent (95% CI 51.2-61.2) for SGLT2 inhibitors in the first-line cohort. In the second-line cohort, persistence was comparable between groups at 51.8 percent (95% CI 45.2-58.4) for biguanides and 49.4 percent (95% CI 40.3-58.6) for SGLT2 inhibitors. The authors reported that one-year continuation rates were slightly lower than those in previous studies, possibly because the database primarily captures large hospitals where clinicians may adjust medications more actively, and because definitions of persistence vary across studies.

The authors cautioned that several limitations affect interpretation. The database predominantly captures care in large hospitals, so the findings may not generalize to primary care clinics, and eligibility rules may have preferentially retained patients with more stable follow-up. Qualitative proteinuria data were unavailable and albumin-to-creatinine ratios were measured in only a limited subset, and because SGLT2 inhibitors are more likely to be prescribed to patients with proteinuria, this unmeasured factor could bias the comparative estimates toward smaller HbA1c reductions in the SGLT2 inhibitor group. Insulin initiation remained slightly more common in the biguanide group even after matching, which may have contributed to the larger early HbA1c reduction observed at 12 weeks. Even so, the authors concluded that treatment selection in Japanese practice was influenced primarily by concomitant cardiovascular medication use, that body mass index played a role especially in second-line therapy without predominating, and that biguanides were associated with larger HbA1c reductions while persistence over 48 weeks was higher with SGLT2 inhibitors only among first-line patients.

Subject of Research: Factors influencing the choice between biguanides and SGLT2 inhibitors and their glycemic efficacy and persistence in type 2 diabetes in Japan

Article Title: Treatment Selection Patterns and Associated Outcomes for Biguanides and SGLT2 Inhibitors in Type 2 Diabetes: A Retrospective Database Study in Japan

Article References: Mishima, R., & Koide, D. (2026). Treatment Selection Patterns and Associated Outcomes for Biguanides and SGLT2 Inhibitors in Type 2 Diabetes: A Retrospective Database Study in Japan. Drugs – Real World Outcomes, 13(3), 395-408. https://doi.org/10.1007/s40801-026-00567-5

Image Credits: AI Generated

DOI: 10.1007/s40801-026-00567-5

Keywords: type 2 diabetes, biguanides, SGLT2 inhibitors, HbA1c, medication persistence, Japan, retrospective cohort study, treatment selection, BMI, diuretics, real-world evidence, coarsened exact matching