Initiation of SGLT2 Inhibitors Versus DPP‐4 Inhibitors in Metformin Users and Risk of Incident Depression: A Nationwide, Claims‐Based Active‐Comparator New‐User Study Authors: Sangwoo Park, Seogsong Jeong, Hye Jun Kim, Dae Ho Lee, Soo Jung Choi, Sang Min Park Published: 2026 ABSTRACT ABSTRACTAimsTo examine the association between initiation of sodium‐glucose cotransporter‐2 (SGLT2) inhibitors and risk of incident depression compared with dipeptidyl peptidase‐4 (DPP‐4) inhibitors among adults with newly diagnosed Type 2 diabetes receiving metformin.Materials and MethodsWe conducted a nationwide active‐comparator new‐user cohort study with a 90‐day adherent‐survivor landmark design, using the Korean National Health Insurance Service database linked with national health check‐up and mortality data (2014–2022). Adults with newly diagnosed Type 2 diabetes receiving metformin who newly initiated SGLT2 inhibitors or DPP‐4 inhibitors as add‐on therapy were identified. This study was designed as a 90‐day adherent‐survivor landmark analysis. Propensity score matching was performed in a 1:1 ratio, yielding 10 013 matched pairs. The primary outcome was incident depression, defined as ≥ 2 outpatient or inpatient claims with ICD‐10 codes F32–F33. The secondary outcome was suicide mortality. Cox proportional hazards models were used to estimate hazard ratios (HRs) under intention‐to‐treat and as‐treated approaches.ResultsAmong 38 297 eligible participants, 300 depression events occurred among SGLT2 inhibitor users and 391 among DPP‐4 inhibitor users in the matched cohort. Initiation of SGLT2 inhibitors was associated with a lower risk of depression compared with DPP‐4 inhibitors (HR 0.85, 95% CI 0.73–0.99). Similar findings were observed in as‐treated analyses (HR 0.80, 95% CI 0.65–0.98). Associations were generally consistent across clinically relevant subgroups and appeared more pronounced among older adults, females and individuals with greater comorbidity burden. Suicide mortality was infrequent in both groups.ConclusionsAmong 90‐day adherent survivors with Type 2 diabetes receiving metformin, initiation of SGLT2 inhibitors as add‐on therapy was associated with a lower risk of incident depression compared with DPP‐4 inhibitors. Introduction Diabetes is the third leading non‐communicable disease globally based on disability‐adjusted life years (DALYs), following ischemic heart disease and stroke and affects more than 90 million people worldwide [1]. It is a major contributor to premature mortality, particularly among men, with a substantially increased probability of dying before age 70 years [2]. Depression is a common comorbidity in diabetes, with a 1.15‐ to 2.01‐fold greater risk than the non‐diabetic population and the risk increases with disease severity [3]. Depression in diabetes is associated with poorer glycaemic control, higher complication rates and increased mortality, underscoring the importance of understanding factors that may influence mental health outcomes in diabetic patients. Inflammatory reactions and oxidative stress, which alter brain neurotransmission, are recognised contributors to depression in diabetes [4]. Sodium‐glucose cotransporter‐2 inhibitors (SGLT2 inhibitors), such as canagliflozin, dapagliflozin and empagliflozin, are oral antidiabetic medications that improve glycaemic control by lowering the renal glucose excretion threshold and inhibiting renal glucose reabsorption [5]. Recent studies suggest that SGLT2 inhibitors exert systemic benefits beyond glucose lowering, including favourable effects on vascular, cardiac and renal pathways, reductions in pro‐inflammatory markers and improvements in oxidative stress [6, 7]. Moreover, dapagliflozin was found to reverse lipopolysaccharide‐induced depressive‐like behaviours in mice [8]. Clinical findings remain mixed. A recent cross‐sectional study of 229 patients found that SGLT2 inhibitors use was associated with poorer sleep quality but not with depression, whereas dipeptidyl peptidase‐4 inhibitors (DPP‐4 inhibitors) showed no association with psychiatric outcomes [9]. In contrast, a large cohort study of antidiabetic drug monotherapy found that SGLT2 inhibitors use was associated with a lower risk of depression within 5 years after the initial treatment compared with DPP‐4 inhibitors [10]. A recent active‐comparator new‐user cohort study comparing GLP‐1 receptor agonists with SGLT2 inhibitors for incident depression reported that GLP‐1 receptor agonist use was associated with a higher risk of depression compared with SGLT2 inhibitor use, with DPP‐4 inhibitors included in sensitivity analyses [11]. These findings indirectly suggest a differential psychiatric risk profile across major second‐line antidiabetic drug classes. However, whether these associations persist in routine clinical practice, where these drugs are commonly prescribed as add‐on therapy to metformin, remains unclear. Given the biologic plausibility and emerging evidence, further clarification is needed regarding the long‐term impact of SGLT2 inhibitors on depression risk. Therefore, we conducted a nationwide, population‐based cohort study to examine the association between SGLT2 inhibitors use and risk of depression among adults with Type 2 diabetes in South Korea. Methods Study Design and Data Source We conducted a nationwide, population‐based cohort study using a 90‐day adherent‐survivor landmark design, comparing SGLT2 inhibitors with DPP‐4 inhibitors as add‐on therapy to metformin for the prevention of incident depression among adults with Type 2 diabetes [12, 13, 14]. The study used claims data from the Korean National Health Insurance Service (NHIS), which covers nearly the entire Korean population and contains information on demographics, eligibility, inpatient and outpatient encounters, diagnoses, procedures and prescription records. These data were linked to the NHIS health check‐up database, which includes standardised biennial examinations with self‐reported health behaviours and measured clinical parameters (e.g., blood pressure, body mass index [BMI] and laboratory values) [15, 16]. Death information, including cause of death, was obtained from national mortality statistics [17]. The study period spanned from September 2014, when SGLT2 inhibitors were first reimbursed in Korea, through 31 December 2022 [18]. Study Population The target population consisted of adults with newly diagnosed Type 2 diabetes who initiated metformin and subsequently added either SGLT2 inhibitors or DPP‐4 inhibitors. First, we identified individuals with incident Type 2 diabetes based on diagnostic codes and antidiabetic drug prescriptions and then selected those who started metformin. Among them, we defined new users of SGLT2 inhibitors or DPP‐4 inhibitors [19, 20] as patients who received their first prescription of SGLT2 inhibitors or DPP‐4 inhibitors after metformin initiation, with no prior use of either class. This study was conducted using a 90‐day adherent‐survivor landmark design. The treatment group was determined according to the add‐on drug class prescribed at first dispensing. Eligibility for the landmark cohort additionally required patients to remain alive, enrolled, free of the study outcome and adherent to the initiated treatment, defined as an MPR of at least 0.8 [21] during the first 90 days [22]. The index date was the Day‐90 landmark and outcome follow‐up began only after this date. We applied the following exclusion criteria before matching (Figure 1): age < 19 years at index date; prior use of three or more oral anti‐diabetic drug (OAD) classes before index date; ≥ 90 days of insulin use before index date any recorded diagnosis of depression before index date; and absence of a health check‐up within 2 years before the index date. After these exclusions, 38 297 patients remained, comprising 10 029 SGLT2 inhibitors initiators and 28 268 DPP‐4 inhibitors initiators [23]. Study flow for the study. Exposure Definition The exposure of interest was initiation of SGLT2 inhibitors versus DPP‐4 inhibitors as add‐on therapy to metformin. SGLT2 inhibitors included dapagliflozin, canagliflozin, empagliflozin, ertugliflozin and ipragliflozin; DPP‐4 inhibitors included sitagliptin, vildagliptin, saxagliptin, alogliptin, linagliptin, gemigliptin, evogliptin, teneligliptin, anagliptin and their fixed‐dose combinations. Drugs were identified using Anatomical Therapeutic Chemical (ATC) codes and all OAD classes considered in this study are summarised in Table S4. Combination products were classified according to their constituent ATC codes; for example, fixed‐dose combinations of metformin with SGLT2 inhibitors or DPP‐4 inhibitors were counted as exposure to the respective index class. Outcome Definitions The primary outcome was incident depression. Incident depression was defined using International Classification of Diseases, 10th Revision (ICD‐10) codes F32–F33, consistent with World Health Organization definitions. To improve diagnostic specificity in claims data, we required at least two outpatient or inpatient encounters with these codes, an approach that has been shown to have acceptable validity in the Korean NHIS database [24, 25]. We did not require antidepressant prescriptions in the definition to avoid misclassification related to off‐label use [26]. The secondary outcome was death from suicide, defined using underlying cause‐of‐death codes (X60–X84) for intentional self‐harm in the national death registry [17, 27]. For both outcomes, patients with corresponding diagnoses before the index date were excluded from the cohort. Covariates Baseline covariates were assessed using the NHIS health check‐up closest to and within 2 years before, the index date and claims data from the same look‐back period [15]. Demographic and lifestyle variables included age, sex, household income quintile (collapsed into low vs. high income in subgroup analyses), smoking status (current smoker vs. non‐smoker), alcohol consumption (drinker vs. non‐drinker) and physical activity (low vs. high, based on frequency of moderate‐to‐vigorous exercise). Clinical measurements included BMI, systolic and diastolic blood pressure, fasting plasma glucose and total cholesterol [28]. We also captured diabetes‐related complications (retinopathy, nephropathy, neuropathy and diabetic foot), prior dementia (Alzheimer's disease, vascular dementia and other dementias), cardiovascular comorbidities (angina, atrial fibrillation, myocardial infarction, stroke, heart failure, hypertension, venous thromboembolism and peripheral vascular disease) and other comorbid conditions (chronic kidney disease, hyperlipidaemia, liver disease, chronic obstructive pulmonary disease, asthma, obesity, sleep apnoea, alcohol use disorders, thyroid disease, osteoporosis, osteoarthritis, rheumatoid arthritis, gout, Parkinson disease, head injury, fracture and malignancy) [14]. Comorbidities were defined using ICD‐10 codes and required at least one inpatient or two outpatient claims. The Charlson comorbidity index (CCI) was calculated as a summary measure of multimorbidity. Baseline concomitant medications, including statins (atorvastatin, rosuvastatin and others), renin‐angiotensin system blockers, antiplatelets, anticoagulants, β‐blockers, nitrates, nonsteroidal anti‐inflammatory drugs, opioids and dementia drugs, were identified from prescription claims in the year before index date [14]. All covariates listed in Table 1 were considered potential confounders. TABLE 1Baseline characteristics before and after propensity score matching.CharacteristicsBefore propensity score matchingAfter propensity score matchingDPP‐4 inhibitors (n = 28 268)SGLT‐2 inhibitors (n = 10 029)Standardised differenceDPP‐4 inhibitors (n = 10 013)SGLT‐2 inhibitors (n = 10 013)Standardised differenceAge (years), mean (SD) a 55.2 (10.9)50.0 (10.3)0.49350.0 (10.1)50.0 (10.3)0.005Sex (male) a 62.863.10.00562.563.00.011Body mass index, mean (SD) a 26.2 (3.8)27.8 (4.2)0.39527.6 (4.3)27.8 (4.2)0.035Systolic blood pressure, mean (SD) a 128.2 (14.7)127.5 (14.7)0.048127.4 (14.3)127.5 (14.7)0.009Diastolic blood pressure, mean (SD)78.3 (9.8)79.2 (10.3)0.09279.2 (9.9)79.2 (10.3)0.004Fasting blood sugar, mean (SD) a 139.5 (45.3)137.3 (44.6)0.050138.1 (41.8)137.3 (44.6)0.017Total cholesterol, mean (SD) a 184.6 (48.6)186.9 (50.9)0.047187.2 (51.5)186.9 (50.9)0.005Physical activity, n (%) a None17 563 (62.1)6335 (63.2)0.0516249 (62.4)6326 (63.2)0.0171–2 times/week6703 (23.7)2459 (24.5)2527 (25.2)2454 (24.5)2–3 times/week3170 (11.2)970 (9.7)976 (9.8)968 (9.7)≥ 4 times/week832 (2.9)265 (2.6)261 (2.6)265 (2.7)Alcohol consumption, n (%) a None17 422 (61.63)5967 (59.5)0.0585918 (59.1)5958 (59.5)0.0081–2 times/week7184 (25.41)2803 (27.95)2827 (28.2)2796 (27.9)2–3 times/week2542 (8.99)933 (9.3)941 (9.4)933 (9.3)≥ 4 times/week1120 (3.96)326 (3.25)327 (3.3)326 (3.3)Smoking status, n (%) a Never14 804 (52.4)5059 (50.4)0.0445076 (50.7)5052 (50.5)0.005Past6413 (22.7)2271 (22.6)2258 (22.6)2268 (22.7)Current7051 (24.9)2699 (26.9)2679 (26.8)2693 (26.9)Income level a First quarter (highest)9132 (32.3)3294 (32.8)0.029—3310 (33.06)3291 (32.87)0.010Second quarter7367 (26.1)2741 (27.3)2720 (27.16)2735 (27.31)Third quarter5859 (20.7)2016 (20.1)2039 (20.36)2010 (20.07)Fourth quarter5440 (19.2)1844 (18.4)1805 (18.03)1843 (18.41)Basic beneficiary (lowest)470 (1.7)134 (1.3)139 (1.39)134 (1.34)Dementia a Alzheimer's dementia172 (0.61)22 (0.22)0.06120 (0.2)22 (0.22)0.003Vascular dementia57 (0.2)8 (0.08)0.0337 (0.07)8 (0.08)0.003Other dementia64 (0.23)12 (0.12)0.0268 (0.08)12 (0.12)0.010Diabetes complications a Retinopathy1932 (6.83)756 (7.54)0.027755 (7.54)755 (7.54)0.000Nephropathy1880 (6.65)901 (8.98)0.087896 (8.95)893 (8.92)0.001Neuropathy1907 (6.75)671 (6.69)0.002678 (6.77)669 (6.68)0.004Diabetic foot1285 (4.55)478 (4.77)0.010462 (4.61)478 (4.77)0.008Cardiovascular comorbidities a Angina pectoris1798 (6.36)939 (9.36)0.112912 (9.11)935 (9.34)0.009Atrial fibrillation412 (1.46)178 (1.77)0.025178 (1.78)177 (1.77)0.001Myocardial infarction337 (1.19)270 (2.69)0.109243 (2.43)269 (2.69)0.019Stroke1065 (3.77)254 (2.53)0.071229 (2.29)254 (2.54)0.014Heart failure807 (2.85)424 (4.23)0.074390 (3.89)423 (4.22)0.018Hypertension15 085 (53.36)5467 (54.51)0.0235418 (54.11)5456 (54.49)0.008Venous thromboembolism218 (0.77)91 (0.91)0.01586 (0.86)91 (0.91)0.005Peripheral vascular disease3138 (11.1)1003 (10)0.0361012 (10.11)1001 (10)0.004Other comorbidities a CKD768 (2.72)375 (3.74)0.058362 (3.62)371 (3.71)0.005Hyperlipidaemia21 950 (77.65)8336 (83.12)0.1388364 (83.53)8320 (83.09)0.011Liver disease13 677 (48.38)5255 (52.4)0.0805273 (52.66)5243 (52.36)0.006COPD4292 (15.18)1337 (13.33)0.0531349 (13.47)1335 (13.33)0.004Asthma2415 (8.54)792 (7.9)0.024764 (7.63)792 (7.91)0.010Obesity50 (0.18)63 (0.63)0.07144 (0.44)61 (0.61)0.027Sleep apnoea79 (0.28)57 (0.57)0.04452 (0.52)55 (0.55)0.005Alcohol use and related disorders1060 (3.75)362 (3.61)0.007373 (3.73)362 (3.62)0.006Thyroid disease5029 (17.79)2025 (20.19)0.0612024 (20.21)2018 (20.15)0.002Osteoporosis2047 (7.24)509 (5.08)0.090475 (4.74)509 (5.08)0.014Osteoarthritis6845 (24.21)2003 (19.97)0.1022004 (20.01)2003 (20)0.000Rheumatoid arthritis735 (2.6)233 (2.32)0.018229 (2.29)233 (2.33)0.003Gout1687 (5.97)730 (7.28)0.053775 (7.74)726 (7.25)0.020Parkinson disease60 (0.21)8 (0.08)0.0357 (0.07)8 (0.08)0.003Head injury930 (3.29)292 (2.91)0.022272 (2.72)291 (2.91)0.011Fracture1291 (4.57)389 (3.88)0.034359 (3.59)387 (3.86)0.014Malignancy1903 (6.73)566 (5.64)0.045537 (5.36)565 (5.64)0.012Drugs a ACEi371 (1.31)193 (1.92)0.049173 (1.73)192 (1.92)0.015ARB6064 (21.45)1951 (19.45)0.0501919 (19.17)1949 (19.46)0.007Atorvastatin7451 (26.36)2355 (23.48)0.0672411 (24.08)2353 (23.5)0.013CCB2804 (9.92)797 (7.95)0.069763 (7.62)795 (7.94)0.011Fluvastatin48 (0.17)17 (0.17)0.00021 (0.21)17 (0.17)0.010Lovastatin16 (0.06)2 (0.02)0.0193 (0.03)2 (0.02)0.005Pitavastatin1617 (5.72)686 (6.84)0.046719 (7.18)686 (6.85)0.014Pravastatin413 (1.46)203 (2.02)0.043183 (1.83)199 (1.99)0.012Rosuvastatin3745 (13.25)1874 (18.69)0.1491830 (18.28)1868 (18.66)0.010Simvastatin775 (2.74)242 (2.41)0.021241 (2.41)242 (2.42)0.001Anticoagulant851 (3.01)467 (4.66)0.086440 (4.39)464 (4.63)0.013Antiplatelet4851 (17.16)1710 (17.05)0.0031648 (16.46)1706 (17.04)0.015Aspirin4117 (14.56)1408 (14.04)0.0151360 (13.58)1405 (14.03)0.013β‐blocker1210 (4.28)558 (5.56)0.059537 (5.36)557 (5.56)0.009Dementia drug138 (0.49)12 (0.12)0.0675 (0.05)12 (0.12)0.013Nitrate962 (3.4)600 (5.98)0.122562 (5.61)597 (5.96)0.017NSAIDs14 409 (50.97)4933 (49.19)0.0364926 (49.2)4926 (49.2)0.000Opioid222 (0.79)102 (1.02)0.02584 (0.84)101 (1.01)0.018Others statin (ulinastatin, somatostatin)123 (0.44)37 (0.37)0.01041 (0.41)37 (0.37)0.006Thrombolytics11 188 (39.58)3555 (35.45)0.0853553 (35.48)3551 (35.46)0.000 Note: Values are number (percentage) unless stated otherwise. For categorical variables with multiple levels (physical activity, smoking status, alcohol consumption and income), the standardised mean difference represents the maximum absolute difference across categories. a Variables included in the propensity score model. Baseline characteristics before and after propensity score matching. Note: Values are number (percentage) unless stated otherwise. For categorical variables with multiple levels (physical activity, smoking status, alcohol consumption and income), the standardised mean difference represents the maximum absolute difference across categories. Variables included in the propensity score model. Propensity Score Estimation and Matching To reduce confounding by indication, we used an active‐comparator new‐user design [19, 20] and estimated each patient's propensity to initiate an SGLT2 inhibitor versus a DPP‐4 inhibitor using a multivariable logistic regression model. The propensity score model incorporated a broad set of prespecified baseline covariates capturing sociodemographic factors, health behaviours measured at the national health check‐up, anthropometric and laboratory measurements, diabetes‐related complications, cardiovascular and other chronic comorbidities and baseline concomitant medications [29]. SGLT2 inhibitor initiators were then matched 1:1 to DPP‐4 inhibitor initiators using nearest‐neighbour matching without replacement and a calliper width of 0.2 of the standard deviation of the logit of the propensity score. After matching, 10 013 SGLT2 inhibitor initiators were matched to 10 013 DPP‐4 inhibitor initiators, yielding a well‐balanced propensity score‐matched cohort (Figure 1). Covariate balance before and after matching was assessed using standardised mean differences, with absolute values < 0.10 considered indicative of adequate balance. Intention‐to‐Treat (ITT) Analysis The primary analysis followed an ITT approach. In the propensity score‐matched cohort, follow‐up began on the index date and continued until the earliest of outcome occurrence (depression or suicide, analysed separately), death, loss of NHIS eligibility or 31 December 2022, regardless of subsequent changes in antidiabetic therapy, treatment discontinuation or switching between SGLT2 inhibitors and DPP‐4 inhibitors. We calculated incidence rates per 1000 person‐years and estimated adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs) for SGLT2 inhibitors versus DPP‐4 inhibitors using Cox proportional hazards models. Model 1 adjusted for age and sex; Model 2 further adjusted for income, smoking status, alcohol consumption, physical activity, BMI, systolic blood pressure, fasting glucose, total cholesterol and CCI. Analogous models were fit in the pre‐matching cohort for comparison. As‐Treated (AT) Analysis To approximate the per‐protocol effect under sustained treatment, we performed AT analyses in the propensity score‐matched cohort. For these analyses, follow‐up began at the index date and continued until the earliest of outcome occurrence, treatment discontinuation, treatment switching, death, loss of eligibility or end of follow‐up [30]. Treatment discontinuation was defined as a gap of > 30 days between the end of the last prescription's days of supply and the next refill of the index agent. Treatment switching was defined as initiation of the alternative comparator class between SGLT2 inhibitors and DPP‐4 inhibitors [14]. Person‐time after discontinuation or switching was censored and not attributed to the original exposure group. This analysis restricted attributed person‐time to periods before recorded discontinuation or switching but may remain susceptible to informative censoring. Cox models with the same covariate adjustment as in the ITT analysis were used to estimate aHRs and 95% CIs for depression and suicide. Subgroup Analyses We conducted prespecified subgroup analyses to evaluate potential effect modification of the association between SGLT2 inhibitors use and depression by clinically relevant baseline characteristics. Subgroups were defined by age (< 65 vs. ≥ 65 years), sex, BMI (≥ 25 vs. < 25 kg/m2), alcohol consumption (drinker vs. non‐drinker), smoking status (current smoker vs. non‐smoker), income (low vs. high), physical activity (low vs. high) and CCI (≤ 2 vs. > 2). For each subgroup, we refit the fully adjusted Cox models in the propensity score‐matched cohort and reported aHRs and 95% CIs (Table S1 for ITT and Table S2 for AT analyses). We also evaluated multiplicative interaction by including product terms between exposure group and subgroup indicators in the models; p values for interaction were calculated to assess statistical heterogeneity, although the study was not powered specifically for interaction testing. Analytical Framework The design, eligibility criteria, treatment strategies, follow‐up and outcomes of this observational study were mapped onto a hypothetical randomised controlled trial following established principles of active‐comparator new‐user design. A detailed comparison of the protocol components of the hypothetical trial and the observational study is provided in Table S3. Briefly, the ITT analysis was interpreted as an estimate of the ITT effect, whereas the AT analysis corresponded to a secondary, exposure‐restricted analysis and should not be interpreted as a fully adjusted per‐protocol effect estimate. Statistical Software All statistical analyses were performed using SAS software, version 9.4 (SAS Institute Inc). Two‐sided p values < 0.05 were considered statistically significant. This study was reported in accordance with the TARGET (Transparent Reporting of observational studies Emulating a Target trial GuidEline) reporting guideline (Table S3). To quantify the robustness of the primary result against unmeasured confounding, in Table S9, the E‐value was calculated using the method proposed by VanderWeele and Ding. The fragility index was calculated using a 2 × 2 contingency table approach with Fisher's exact test. To evaluate the potential influence of unmeasured confounding, negative control outcome analyses were performed using lung cancer (ICD‐10: C34) and fracture (ICD‐10: S02, S12, S22, S32, S42, S52, S62, S72, S82, S92, T02, T08, T10, T12) as negative control outcomes, each requiring ≥ 2 outpatient or ≥ 1 inpatient claims. Individuals with a prior diagnosis of the respective negative control outcome before the index date were excluded. Fracture was selected as the second negative control outcome in preference to hypothyroidism, given emerging evidence of potential associations between SGLT2 inhibitor use and thyroid function. An additional outcome analysis was performed for incident anxiety disorders (ICD‐10: F40–F48), defined as ≥ 2 outpatient or ≥ 1 inpatient claims, excluding individuals with a prior diagnosis of anxiety disorders before the index date. Sensitivity Analysis To assess the potential influence of the difference in follow‐up duration between treatment arms, a sensitivity analysis was performed restricting follow‐up to 1 year for both groups. Additional sensitivity analyses included: a Fine‐Gray subdistribution hazard model treating all‐cause death as a competing event; AT analyses using alternative grace periods of 60 and 90 days; and an analysis with additional adjustment for annual outpatient visit frequency and inpatient admission frequency in the year prior to the index date. Kaplan–Meier survival curves were generated for the propensity score–matched cohort. Results Baseline Characteristics Before propensity score matching, 38 297 eligible patients initiating add‐on therapy with either an SGLT2 inhibitor (n = 10 029) or a DPP‐4 inhibitor (n = 28 268) were identified. Compared with DPP‐4 inhibitor initiators, patients receiving SGLT2 inhibitors were younger, had a higher BMI and differed with respect to several comorbidities and concomitant medications (Table 1). After 1:1 propensity score matching, 10 013 matched pairs were retained and baseline characteristics, including health behaviours, anthropometric and clinical measurements, comorbidities and medication use, were well‐balanced between groups, with all absolute standardised mean differences below 0.10 (Table 1). Incidence of Depression and Suicide by Intention‐to‐Treat Analysis In the unmatched cohort, median follow‐up was 3.83 years (IQR, 2.11–5.64) for DPP‐4 inhibitor initiators and 3.27 years (IQR, 1.79–4.84) for SGLT2 inhibitor initiators. The crude incidence rate of depression was 11.42 per 1000 person‐years among DPP‐4 inhibitor users and 8.83 among SGLT2 inhibitor users. The corresponding aHR, accounting for age and sex, was 0.86 (95% CI 0.76–0.98; p = 0.023). In the fully adjusted model, the aHR was 0.87 (0.77–0.99; p = 0.039) (Table 2). TABLE 2Comparative risk of depression/suicide between initiators of SGLT‐2 inhibitors and DPP‐4 inhibitors in before/after propensity score‐matched cohort.Before propensity score matchingAfter propensity score matchingDPP‐4 inhibitors (n = 28 268)SGLT‐2 inhibitors (n = 10 029)DPP‐4 inhibitors (n = 10 013)SGLT‐2 inhibitors (n = 10 013)DepressionEvents1265300391300Person years110 723.533 987.1638 366.9533 935.07Incidence rate per 1000 person years11.428.8310.198.84aHR (95% CI) a 1.00 (Ref.)0.86 (0.76–0.98)* 1.00 (Ref.)0.86 (0.74–1.00)* aHR (95% CI) b 1.00 (Ref.)0.87 (0.77–0.99)* 1.00 (Ref.)0.85 (0.73–0.99)* SuicideEvents339139Person years116 045.334 855.4739 806.6634 803.39Incidence rate per 1000 person years0.280.260.330.26aHR (95% CI) a 1.00 (Ref.)1.01 (0.47–2.15)1.00 (Ref.)0.81 (0.34–1.90)aHR (95% CI) b 1.00 (Ref.)0.95 (0.44–2.05)1.00 (Ref).0.80 (0.34–1.89)Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval; DPP‐4 = dipeptidyl peptidase‐4; Ref. = reference; SGLT‐2 = sodium‐glucose cotransporter‐2. a Adjusted for age and sex. b Adjusted for age, sex, household income, smoking status, alcohol consumption, physical activity, body mass index, systolic blood pressure, fasting serum glucose, total cholesterol, Charlson comorbidity index.* p < 0.05. Comparative risk of depression/suicide between initiators of SGLT‐2 inhibitors and DPP‐4 inhibitors in before/after propensity score‐matched cohort. Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval; DPP‐4 = dipeptidyl peptidase‐4; Ref. = reference; SGLT‐2 = sodium‐glucose cotransporter‐2. Adjusted for age and sex. Adjusted for age, sex, household income, smoking status, alcohol consumption, physical activity, body mass index, systolic blood pressure, fasting serum glucose, total cholesterol, Charlson comorbidity index. p < 0.05. In the propensity score–matched cohort, median follow‐up was 3.73 years (IQR, 2.05–5.54) for DPP‐4 inhibitor initiators and 3.27 years (IQR, 1.79–4.84) for SGLT2 inhibitor initiators. Three hundred and ninety‐one depression events occurred among DPP‐4 inhibitor users and 300 among SGLT2 inhibitor users, yielding incidence rates of 10.19 and 8.84 per 1000 person‐years, respectively. The risk of depression remained lower in SGLT2 inhibitor initiators (aHR 0.86 [0.74–1.00]; p = 0.046 in the age‐sex model and 0.85 [0.73–0.99]; p = 0.031 in the fully adjusted model). Suicide mortality was rare in both groups. In the matched cohort, the incidence rate was 0.33 per 1000 person‐years for DPP‐4 inhibitors and 0.26 for SGLT2 inhibitors. No statistically significant differences were observed (fully adjusted aHR 0.80, 95% CI 0.34–1.89). Kaplan–Meier survival curves for the propensity score–matched cohort are presented in Figure 2. Graphical assessment of the survival plots showed no meaningful evidence of violation of the proportional hazards assumption. The cumulative incidence function curves from the Fine‐Gray competing risk model are presented in Figure S1. In Table S9, the E‐value for the point estimate of the primary result was 1.63 (E‐value for the upper confidence limit: 1.11), indicating that an unmeasured confounder would need to be associated with both treatment initiation and incident depression by a risk ratio of at least 1.63‐fold to fully explain away the observed association. The fragility index was 39, indicating that a shift of 39 additional events in the SGLT2 inhibitor group would have rendered the result statistically non‐significant. Kaplan–Meier curve. Kaplan‐Meier survival curves for depression‐free survival among propensity score–matched initiators of SGLT2 inhibitors versus DPP‐4 inhibitors. Tick marks indicate censored observations. Numbers at risk are shown at yearly intervals. CI, confidence interval; DPP‐4i, DPP‐4 inhibitor; HR, hazard ratio; SGLT2i, SGLT2 inhibitor. Subgroup Analyses Across most clinically relevant subgroups, including age, sex, obesity, smoking status, alcohol consumption, household income, CCI and physical activity, the direction of association consistently favoured a lower risk of depression among SGLT2 inhibitor users (Table S1). Significant interactions were observed in older adults (≥ 65 years), where SGLT2 inhibitor initiation was associated with a notably reduced risk of depression (aHR 0.52 [0.33–0.81] after matching). No subgroup showed an increased risk associated with SGLT2 inhibitors. Sensitivity Analysis In a sensitivity analysis restricting follow‐up to 1 year for both treatment groups, the fully adjusted aHR was 0.82 (95% CI 0.63–1.08), consistent with the primary analysis (Table S6). In the Fine‐Gray competing risk analysis treating all‐cause death as a competing event, the results were consistent with the primary Cox proportional hazards analysis (age‐ and sex‐adjusted subdistribution HR 0.86, 95% CI 0.74–1.00; fully adjusted subdistribution HR 0.85, 95% CI 0.73–0.99). The cumulative incidence function curves are presented in Figure S1 and full results are provided in Table S5. In additional sensitivity analyses, the Fine‐Gray competing risk model yielded results consistent with the primary analysis (fully adjusted subdistribution HR 0.85, 95% CI 0.73–0.99; Table S5). AT analyses using grace periods of 60 days (aHR 0.81, 95% CI 0.66–0.98) and 90 days (aHR 0.81, 95% CI 0.67–0.97) were also consistent with the primary AT analysis. Additional adjustment for healthcare utilisation yielded a fully adjusted aHR of 0.88 (95% CI 0.76–1.03). Full results are presented in Table S6. In negative control outcome analyses, the aHRs were 1.08 (95% CI 0.64–1.83) for lung cancer and 0.95 (95% CI 0.85–1.07) for fracture, both close to the null and statistically non‐significant, providing reassurance that the primary finding is unlikely to be entirely attributable to unmeasured confounding (Table S7). In the additional outcome analysis, SGLT2 inhibitor initiation was associated with a directionally consistent though statistically non‐significant lower risk of anxiety disorders compared with DPP‐4 inhibitor initiation (age‐ and sex‐adjusted aHR 0.88, 95% CI 0.76–1.01; fully adjusted aHR 0.87, 95% CI 0.75–1.00; Table S8). AT Analyses In the AT analysis, median follow‐up was 1.51 years (IQR, 0.67–3.05) for DPP‐4 inhibitor initiators and 1.44 years (IQR, 0.65–2.88) for SGLT2 inhibitor initiators. When restricting follow‐up to periods of continuous treatment (censoring at discontinuation or switching), the comparative risk reduction associated with SGLT2 inhibitors became more pronounced. In the matched cohort, depression incidence rates were 10.15 and 8.13 per 1000 person‐years for DPP‐4 inhibitors and SGLT2 inhibitors, respectively. The age‐ and sex‐adjusted aHR was 0.81 (95% CI 0.66–0.99; p = 0.043). The fully adjusted aHR was 0.80 (95% CI 0.65–0.98; p = 0.031), consistent with a lower risk among adherent SGLT2 inhibitor users (Table 3). TABLE 3Sensitivity analysis: As‐treated estimates for depression risk in SGLT‐2i versus DPP‐4i initiators.After propensity score matchingDPP‐4 inhibitors (n = 10 013)SGLT‐2 inhibitors (n = 10 013)DepressionEvents214159Person years21 073.0219 556.91Incidence rate per 1000 person years10.158.13aHR (95% CI) a 1.00 (Ref.)0.81 (0.66–0.99)* aHR (95% CI) b 1.00 (Ref.)0.80 (0.65–0.98)* Note: As‐treated (ATT) analyses were conducted in the propensity score–matched cohort by following patients from the index date (90 days after initiation of the add‐on SGLT‐2 inhibitor or DPP‐4 inhibitor with ≥ 80% medication possession ratio during the initial 90‐day treatment window) until the earliest of incident outcome, treatment discontinuation, treatment switch, death or end of follow‐up. Treatment discontinuation was defined as a gap of more than 30 days after the end of the last prescription of the index drug and treatment switch was defined as initiation of the alternative comparator class (SGLT‐2 inhibitor ↔ DPP‐4 inhibitor). Person‐time after discontinuation or switching was censored and not attributed to the original exposure group. This analysis restricted attributed person‐time to periods before recorded discontinuation or switching but may remain susceptible to informative censoring.Suicide mortality results are not presented in tabular format owing to the very small number of events in the as‐treated analysis (SGLT2 inhibitor group: 1 event; DPP‐4 inhibitor group: 4 events), which precluded reliable statistical estimation.Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval; DPP‐4 = dipeptidyl peptidase‐4; Ref. = reference; SGLT‐2 = sodium‐glucose cotransporter‐2. a Adjusted for age and sex. b Adjusted for age, sex, household income, smoking status, alcohol consumption, physical activity, body mass index, systolic blood pressure, fasting serum glucose, total cholesterol, Charlson comorbidity index.* p < 0.05. Sensitivity analysis: As‐treated estimates for depression risk in SGLT‐2i versus DPP‐4i initiators. Note: As‐treated (ATT) analyses were conducted in the propensity score–matched cohort by following patients from the index date (90 days after initiation of the add‐on SGLT‐2 inhibitor or DPP‐4 inhibitor with ≥ 80% medication possession ratio during the initial 90‐day treatment window) until the earliest of incident outcome, treatment discontinuation, treatment switch, death or end of follow‐up. Treatment discontinuation was defined as a gap of more than 30 days after the end of the last prescription of the index drug and treatment switch was defined as initiation of the alternative comparator class (SGLT‐2 inhibitor ↔ DPP‐4 inhibitor). Person‐time after discontinuation or switching was censored and not attributed to the original exposure group. This analysis restricted attributed person‐time to periods before recorded discontinuation or switching but may remain susceptible to informative censoring. Suicide mortality results are not presented in tabular format owing to the very small number of events in the as‐treated analysis (SGLT2 inhibitor group: 1 event; DPP‐4 inhibitor group: 4 events), which precluded reliable statistical estimation. Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval; DPP‐4 = dipeptidyl peptidase‐4; Ref. = reference; SGLT‐2 = sodium‐glucose cotransporter‐2. Adjusted for age and sex. Adjusted for age, sex, household income, smoking status, alcohol consumption, physical activity, body mass index, systolic blood pressure, fasting serum glucose, total cholesterol, Charlson comorbidity index. p < 0.05. Suicide events were extremely rare under the AT approach (1 event in the SGLT2 inhibitor group and 4 events in the DPP‐4 inhibitor group), precluding reliable statistical estimation. Given the very small number of events and the resulting statistical instability, these results are not presented in tabular format. Subgroup findings in the AT analyses were broadly aligned with the ITT results, with lower risk estimates among females, patients with BMI < 25 kg/m2, non‐drinkers, non‐smokers and those with higher comorbidity burden (Table S2). Discussion In this nationwide, propensity score‐matched cohort study using a 90‐day adherent‐survivor landmark design, among adults with newly diagnosed Type 2 diabetes who initiated an SGLT2 inhibitor or a DPP‐4 inhibitor as add‐on therapy to metformin, we found that SGLT2 inhibitors' use was associated with a modest but consistently lower risk of incident depression. This inverse association was observed in both ITT and AT analyses, despite different censoring strategies and remained robust across multiple sensitivity and subgroup analyses. Suicide mortality events were infrequent and effect estimates were imprecise. Confounding by indication was systematically minimised through the active‐comparator new‐user design, which compared two drug classes with similar glycemic indications and through propensity score matching on a comprehensive set of baseline covariates achieving standardised mean differences below 0.10 for all variables after matching. Taken together, these findings suggest that the choice of add‐on glucose‐lowering therapy after metformin may have implications for mental health and warrants confirmation in randomised or pragmatic studies. Patients with Type 2 diabetes are at substantially increased risk of depression, with reported estimates ranging from approximately 15% to 70% higher than those without diabetes, depending on disease severity and treatment context [3, 31]. Emerging evidence also suggests that SGLT2 inhibitors may be associated with a lower risk of depression compared with DPP‐4 inhibitors [10, 32, 33]. Furthermore, a recent active‐comparator new‐user study reported that GLP‐1 receptor agonist initiation was associated with a higher depression risk compared with SGLT2 inhibitor initiation [11], suggesting a potential hierarchical pattern in the psychiatric risk profiles of major second‐line antidiabetic drug classes that warrants further investigation. However, most prior studies focused on first‐line initiation or monotherapy and thus may not reflect routine clinical practice, where these medications are commonly prescribed as add‐on therapy to metformin. In this nationwide active‐comparator new‐user cohort study with a 90‐day adherent‐survivor landmark design and propensity score matching, our findings aligned with previous reports and indicated a modest, consistent reduction in incident depression among SGLT2 inhibitors initiators with an approximately 20% lower risk compared with DPP‐4 inhibitors. These results highlight that, in routine metformin‐based care, selecting an add‐on agent may have clinically relevant mental health implications and should be evaluated alongside cardiometabolic benefits in future confirmatory studies. On a metformin background, adding an SGLT2 inhibitor induces a fasting‐like and insulin‐sparing metabolic phenotype leading to glucosuria with caloric loss, modest weight reduction, osmotic diuresis/natriuresis with plasma‐volume contraction and a shift towards lipid oxidation and mild ketone body availability [34, 35]. These changes have been associated with improved glycemic stability and reductions in glycemic variability [36]. Because long‐term glycemic variability itself has been linked to a higher risk of incident depression and anxiety in large population cohorts, improved glycemic stability may plausibly contribute to the lower depression risk observed with SGLT2 inhibitors [37, 38]. Beyond glycemia, experimental and translational literature suggests that SGLT2 inhibitors can attenuate oxidative stress and inflammatory signalling and may exert neuroprotective and anti‐neuroinflammatory effects [39, 40]. In particular, empagliflozin has been shown to suppress pro‐inflammatory activation of primary microglia, a mechanism increasingly recognised as a major component of depression pathophysiology, supporting a plausible neuroimmune pathway linking SGLT2 inhibition to mood‐related outcomes [40, 41]. Future studies should evaluate candidate mediators, such as changes in glycaemic variability, weight, inflammatory markers and cardio‐renal events, to clarify whether these pathways explain the lower depression incidence observed with SGLT2 inhibitors. In the present study, subgroup analyses within the ITT framework showed no strong evidence of effect heterogeneity across most demographic and clinical strata. The inverse association between SGLT2 inhibitor initiation and incident depression was directionally consistent across sociodemographic and clinical subgroups. Notably, the association appeared more pronounced among adults aged ≥ 65 years, women and individuals with a higher comorbidity burden. Rather than indicating a qualitatively different drug effect, these patterns may reflect greater absolute benefit in groups with higher baseline vulnerability, where multimorbidity and cardiometabolic complexity are more common, systemic inflammation is more prominent and depression is more prevalent and burdensome, making treatment‐related reductions in risk easier to detect [42, 43]. These subgroup findings should be interpreted cautiously and evaluated in future studies specifically designed to assess effect modification and to quantify absolute risk differences. We leveraged a nationwide claims database with near‐complete capture of prescriptions, diagnoses and mortality. Using an active‐comparator new‐user design (SGLT2 inhibitors vs. DPP‐4 inhibitors added to metformin) with well‐balanced propensity score matching, we found consistent results across ITT and AT analyses and multiple robustness checks. However, several limitations merit consideration. First, as with all observational studies, residual confounding cannot be excluded, particularly, by psychosocial factors (such as life stressors, social support, family history) and clinical variables not reliably captured in claims data (e.g., baseline depression symptoms and detailed glycaemic indices). Additionally, propensity score matching was used for confounding control; although this approach has been widely adopted in pharmacoepidemiological studies of the same drug classes, alternative methods such as inverse probability weighting may yield different point estimates and should be considered in future studies. Negative control outcome analyses using lung cancer and fracture yielded HRs close to the null, suggesting that the degree of unmeasured confounding is unlikely to fully explain the observed association; however, residual confounding cannot be entirely excluded. Second, depression was identified using diagnostic codes requiring ≥ 2 encounters, which improves specificity but may under‐ascertain milder or untreated cases and does not capture severity, symptom course or treatment response. Third, exposure misclassification is possible because prescription claims do not guarantee medication intake and treatment discontinuation or switching may be related to emerging symptoms or clinical deterioration, potentially introducing informative censoring in AT analyses. The AT analysis censored follow‐up at treatment discontinuation or switching without inverse probability‐of‐censoring weighting. Therefore, informative censoring related to prognosis or time‐varying clinical status may remain and the AT estimate should not be interpreted as a fully adjusted per‐protocol effect. Fourth, although the cohort was nationwide, the magnitude of association may differ in settings with other healthcare systems, ethnic compositions or prescribing patterns. Fifth, the shorter median follow‐up among SGLT2 inhibitor initiators relative to DPP‐4 inhibitor initiators reflects the later market entry of SGLT2 inhibitors in Korea rather than an analytical asymmetry; a sensitivity analysis restricting follow‐up to 1 year for both groups yielded results consistent with the primary analysis. Finally, suicide mortality was rare in our cohort, limiting precision; while point estimates did not suggest an increased risk in either analytic framework, modest harms or benefits cannot be definitively excluded. The borderline statistical significance of the primary estimate (E‐value for the point estimate: 1.63; E‐value for the upper confidence limit: 1.11; fragility index: 39) warrants cautious interpretation and confirmation in larger studies or pragmatic randomised trials is needed. Because eligibility for the landmark cohort required survival, continued enrollment, absence of the outcome and an MPR of at least 0.8 during the first 90 days after treatment initiation, the findings pertain to 90‐day adherent survivors and may not be generalizable to all patients receiving an initial prescription. Although the landmark design avoided attributing pre‐landmark person‐time to either treatment group, selection related to early adherence and survival cannot be excluded. The shorter median follow‐up among SGLT2 inhibitor initiators relative to DPP‐4 inhibitor initiators reflects the later market entry of SGLT2 inhibitors in Korea following national reimbursement approval in September 2014, rather than an analytical asymmetry. Longer follow‐up would be valuable for characterizing the durability of the observed association and the present findings should be interpreted in light of this limitation. In this nationwide, propensity score‐matched cohort study using a 90‐day adherent‐survivor landmark design, initiation of SGLT2 inhibitors as add‐on therapy to metformin was associated with a lower risk of incident depression compared with initiation of DPP‐4 inhibitors among 90‐day adherent survivors. This inverse association persisted across both ITT and AT analyses and was broadly consistent across clinically relevant subgroups. Suicide mortality was rare and we found no signal of increased or decreased risk with SGLT2 inhibitors initiation. These findings suggest that mental health outcomes may warrant greater consideration in the benefit–risk framework for selecting add‐on glucose‐lowering therapy, alongside established cardio‐renal and metabolic benefits. Regarding the absolute magnitude of the observed benefit, the incidence rates in the propensity score–matched cohort (10.19 per 1000 person‐years for DPP‐4 inhibitor initiators versus 8.84 for SGLT2 inhibitor initiators) correspond to an absolute risk reduction of approximately 1.35 per 1000 person‐years and a number needed to treat of approximately 741 per year or 250–370 over a 2–3 year treatment horizon. Although modest at the individual level, this estimate may translate into a clinically meaningful reduction in depression burden at the population level, given the large number of patients eligible for second‐line add‐on therapy worldwide and the well‐established association between depression and worse glycemic control, higher complication rates and increased mortality in patients with Type 2 diabetes. The directional consistency of the anxiety disorder findings with the primary depression result further supports the biological plausibility of broader favourable effects of SGLT2 inhibitor initiation on mental health outcomes in this population. Confirmation in pragmatic randomised trials and mediation‐focused studies is needed to determine causality and clarify whether improvements in glycemic stability, inflammatory burden or cardiometabolic events mediate the observed association. Author Contributions Sangwoo Park and Seogsong Jeong contributed equally to this work as co‐first authors and jointly conducted all aspects of the study. Sangwoo Park was responsible for manuscript drafting, study design, statistical analysis and methodological refinement. Seogsong Jeong contributed to manuscript drafting, study design, methodological refinement and the overall conceptual and narrative direction of the manuscript. Hye Jun Kim and Dae Ho Lee reviewed the manuscript and contributed to manuscript revision. Soo Jung Choi and Sang Min Park served as co‐corresponding authors and were involved in all stages of the manuscript preparation, including study conception and design, critical review of the manuscript and final approval of the content. Funding This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (RS‐2024‐00440371). Also, this work was supported by the Gachon University Gil Medical Center (FRD2023‐20 to Soo Jung Choi). Conflicts of Interest The authors declare no conflicts of interest. Ethics statement The study protocol was reviewed by the Institutional Review Board of Seoul National University Hospital (IRB No. E‐2401‐102‐1505) and was determined to be exempt from review. The requirement for informed consent was waived because only de‐identified administrative data were used. Supporting information Figure S1: Cumulative incidence function (CIF) curve. Table S5: Fine‐Gray competing risk analysis Fine‐Gray subdistribution hazard ratios for incident depression treating all‐cause death as a competing event. Table S6: Sensitivity analyses summary of sensitivity analyses for the association between SGLT2 inhibitor versus DPP‐4 inhibitor initiation and incident depression. Table S7: Negative control outcome analyses negative control outcome analyses: lung cancer and fracture. Table S8: Anxiety disorder (additional outcome) association between SGLT2 inhibitor versus DPP‐4 inhibitor initiation and incident anxiety disorders. Table S9: Robustness measures E‐value, fragility index and proportional hazards assessment for the primary analysis. Table S1: Subgroup analyses of depression risk comparing SGLT‐2 inhibitors with DPP‐4 inhibitors (intention‐to‐treat). Table S2: Subgroup analyses of depression risk comparing SGLT‐2 inhibitors with DPP‐4 inhibitors (as‐treated analysis). Table S3: Study Design Framework; Comparing SGLT2 Inhibitors vs DPP‐4 Inhibitors for Incident Depression in an Active‐Comparator New‐User Cohort with a 90‐Day Adherent‐Survivor Landmark Design. Table S4: Classification of oral antidiabetic drugs and ATC codes used in this study.