We screened patients with T2DM who were treated at the outpatient diabetes clinic of Severance Hospital, Yonsei University College of Medicine, Republic of Korea, between April 2014 and August 2017. Patients aged 18 to 75 years who had taken an SGLT2 inhibitor (empagliflozin, 10 mg once daily; dapagliflozin, 10 mg once daily; or ipragliflozin, 50 mg once daily) as monotherapy or an add-on therapy for at least 90 consecutive days were included. We excluded patients with type 1 diabetes mellitus, gestational diabetes, insulin use, or significant renal impairment (eGFR <45 mL/min/1.73 m²). Patients who changed the class of SGLT2 inhibitors and those with missing data regarding body weight, HbA1c level, or eGFR were also excluded. Finally, a total of 804 subjects (128 empagliflozin users, 500 dapagliflozin users, and 176 ipragliflozin users) were enrolled and followed for a median of 192 days (interquartile range [IQR], 168 to 323 days). This study was approved by the Institutional Review Board of the Yonsei University Health System, Severance Hospital (No. 4-2018-0175). Informed consent was waived due to the retrospective nature of the study.

Demographic information including age, gender, time since diabetes diagnosis, other anti-diabetic drug use, and comorbidities (hypertension, hyperlipidemia, coronary artery occlusive disease, peripheral artery occlusive disease, stroke, and transient ischemic attack) was collected through an examination of electronic medical records. The subjects regularly visited the outpatient clinic every 3 to 6 months according to the degree of glycemic control. At each visit, the body weight and body mass index (BMI) were calculated as the body weight divided by the height squared (kg/m²). Blood samples were collected after an overnight fasting. Fasting plasma glucose (FPG) levels were assessed by enzymatic colorimetric assay using a Cobas c 702 (Roche, Basel, Switzerland) and Hitachi 7600-200-DDP (Hitachi, Tokyo, Japan). HbA1c levels were determined by high-performance liquid chromatography on a Cobas Integra 800 device (Roche). Lipid panels (total cholesterol, triglyceride, and high density lipoprotein cholesterol [HDL-C]), serum aspartate aminotransferase, alanine aminotransferase, blood urea nitrogen, and creatinine levels were measured using a Hitachi 7600 chemistry analyzer (Hitachi). We calculated low density lipoprotein cholesterol (LDL-C) levels using the following formula: LDL-C=(total cholesterol−HDL-C)−(triglyceride/5). The eGFR was determined using the 4-variable modification of diet in renal disease formula [15].

We assessed the efficacy of SGLT2 inhibitor according to baseline anti-diabetic drug use: baseline metformin, sulfonylurea (SU), dipeptidyl peptidase 4 (DPP4) inhibitor, or thiazolidinedione (TZD) users and non-users. Glycemic response was evaluated using the change from baseline HbA1c (ΔHbA1c) and the percentage of subjects with baseline HbA1c ≥7% who achieved HbA1c <7% after follow-up. Body weight changes (Δweight loss) were adjusted by the baseline body weight as follows: [(baseline weight−follow-up weight)/baseline weight] ×100.

Normally distributed data are expressed as mean±standard deviations and non-normally distributed data are expressed as median (IQR). Independent t-test or Mann-Whitney U test for continuous variables and chi-square test for categorical variables were used to assess the differences in baseline characteristics. Changes in clinic-laboratory values between baseline and follow-up were analyzed by paired t-test or Wilcoxon signed-rank test. To compare the response according to the baseline anti-diabetic agent use, Mann-Whitney U test were used. Subgroups based on initial HbA1c and BMI categories were compared by Kruskal-Wallis test. We used linear regression analyses to determine the factors responsible for the changes in HbA1c. Multivariate model was adjusted for age, sex, initial BMI, diabetes duration, duration of SGLT2 inhibitor use, baseline HbA1c and eGFR levels, and anti-diabetic agent use (metformin, SU, DPP4 inhibitor, and TZD). IBM SPSS Statistics for Windows, version 20.0 (IBM Corp., Armonk, NY, USA) was used for the statistical analyses and P<0.05 was considered significant.

The baseline characteristics of the study participants are described in Table 1. The median age was 57 years (IQR, 49 to 64 years), 54.4% were male, and the median diabetes duration was 7.2 years. The baseline median BMI was 27.3 kg/m² and the initial HbA1c and FPG were 7.7% and 152 mg/dL, respectively. Of the 804 subjects, 767 (95.4%) were baseline metformin users, 364 (45.1%) were SU users, 217 (26.9%) were DPP4 inhibitor users, and 52 (6.5%) were TZD users.

When classified by concurrently prescribed anti-diabetic drugs, the DPP4 inhibitor users showed a similar age and sex compared to non-users, but the diabetes duration was longer (Table 1). Fasting and postprandial plasma glucose levels were higher but there was no difference in initial median HbA1c levels (7.7% vs. 7.8%). When classified by metformin use, most of the subjects were on metformin therapy at baseline and the characteristics were balanced except for diabetes duration (Supplementary Table 1). The SU users were older, had a longer diabetes duration, and had the least controlled diabetes compared to non-users (Supplementary Table 2). A few patients were on TZD treatment and there was no difference in baseline glycemic status between users and non-users (Supplementary Table 3).

After administration of SGLT2 inhibitors (empagliflozin, dapagliflozin, or ipragliflozin) for a median of 192 days, the median HbA1c level had decreased by 0.7%, from 7.7% (IQR, 7.0% to 8.6%) to 7.0% (IQR, 6.5 to 7.8%; P<0.001) and fasting and postprandial glucose levels were significantly reduced (Table 2). The weight loss was about 3 kg, from 75 kg (IQR, 67 to 85 kg) to 72 kg (IQR, 63 to 82 kg; P<0.001). Serum triglyceride, HDL-C, and liver function test also improved and modest increases in serum hemoglobin and hematocrit levels were also observed. There were no significant changes in total cholesterol, LDL-C, and serum creatinine levels or eGFR. In addition, when patients were classified according to the kind of SGLT2 inhibitors, there were no significant differences in the changes in HbA1c and body weight (P=0.629 and P=0.125, respectively) (Supplementary Table 4).

The effects of SGLT2 inhibitors on the HbA1c level and body weight were compared according to baseline HbA1c (cut-off, 7%) and BMI (cut-off, 25 kg/m²). In well-controlled diabetes (left panel, Fig. 1A), the change from baseline HbA1c was minimal. In patients with inadequately controlled diabetes (right panel, Fig. 1A), the degree of HbA1c reduction was significantly larger and obese patients tended to have a better response (ΔHbA1c, −0.6% vs. −0.8%). In Fig. 1B, the adjusted weight loss did not differ in well-controlled diabetes (left panel, −4.6% vs. −4.0%), whereas significant weight reduction was observed in obese subjects with inadequately controlled diabetes (right panel, −2.9% vs. −4.3%).

Glycemic responses according to other baseline characteristics are shown in Supplementary Fig. 1. There was no difference in glycemic response to SGLT2 inhibitors according to age and gender. However, a significantly larger HbA1c reduction following SGLT2 inhibitor therapy was shown in patients with shorter diabetes duration (<5 years), preserved eGFR (≥90 mL/min/1.73 m²), and poorly controlled diabetes (HbA1c ≥9%).

Next, the effect of SGLT2 inhibitors was observed according to concomitant anti-diabetic medications. Baseline metformin or TZD use with SGLT2 inhibitor treatment did not impact the change from baseline HbA1c or the proportions of subjects achieving HbA1c <7% (Fig. 2A and B). SU users experienced a greater reduction in HbA1c after the addition of SGLT2 inhibitor versus non-users (−0.6% vs. −0.4%, P=0.001) (Fig. 2A), but the percentage of patients who reached the glycemic target was higher in SU non-users (29.1% vs. 48.5%, P<0.001) (Fig. 2B). As the SU users had inadequately controlled diabetes at baseline (Supplementary Table 2), it might be easier to lower the baseline high glucose, but it was not enough to reach the HbA1c <7%. On the other hand, the effect of SGLT2 inhibitor on prevalent DPP4 inhibitor users compared to non-users was significant both in the change from baseline HbA1c (−0.6% vs. −0.4%, P=0.002) and in the proportions of subjects who achieved HbA1c <7% (44.4% vs. 35.6%, P=0.036). The adjusted weight loss was about 4% and was not impacted by baseline metformin, SU, or DPP4 inhibitor use, whereas the combination therapy with TZD offset the weight loss effects of the SGLT2 inhibitor (P=0.015) (Fig. 2C).

When subjects were divided according to respective drug combinations, 18 (2.2%) started SGLT2 inhibitor as monotherapy; 305 (37.9%) were on dual therapy; 353 (43.9%) were on triple combination therapy; and 128 (15.9%) were on quadruple combination therapy with addition of SGLT2 inhibitor to preexisting metformin, SU, TZD, or DPP4 inhibitor therapy. The detailed combinations of anti-diabetic drugs and the reduction of HbA1c are described in Supplementary Table 5. Because the combinations of anti-diabetic drugs were so diverse and those except for the four major combinations were difficult to analyze due to the small number of subjects, we analyzed subjects by baseline user or non-user through the manuscript.

To identify the predictive markers of SGLT2 inhibitors on HbA1c levels, linear regression analyses were conducted (Table 3). After adjusted for age, sex, initial BMI, diabetes duration, duration of SGLT2 inhibitor use, baseline HbA1c, eGFR, and other anti-diabetic agent use (metformin, SU, DPP4 inhibitor, and TZD), shorter diabetes duration, higher baseline HbA1c, and eGFR were the common predictors for a better response following SGLT2 inhibitor treatment in all subjects (all P<0.05). However, the baseline BMI had an inverse effect according to glycemic status. Lower BMI in well-controlled diabetes patients (baseline HbA1c <7%) and higher BMI in inadequately controlled diabetes patients (baseline HbA1c ≥7%) were associated with better response. Preexisting anti-diabetics use did not have an impact on subjects with lower baseline HbA1c <7%, but baseline SU and DPP4 inhibitor use were potential moderators of the SGLT2 inhibitor effects in subjects with baseline HbA1c ≥7%. SU use was associated with a lower response after adjusting for covariates whereas DPP4 inhibitor use was related to a significantly better response after the addition of SGLT2 inhibitor.