Risk of diabetic ketoacidosis with sodium-glucose cotransporter-2 inhibitors compared with dipeptidyl peptidase-4 inhibitors in type 2 diabetes: A nationwide cohort study

Abstract

Background:

Sodium-glucose cotransporter-2 inhibitors (SGLT2is) have significantly improved glycemic control and reduced cardiovascular and renal risks in type 2 diabetes mellitus (T2DM). However, diabetic ketoacidosis (DKA), though rare, remains a serious safety concern. While previous studies have established an elevated DKA risk with SGLT2is, real-world evidence in Asian populations remains limited.

Objectives:

This study evaluated the real-world risk of DKA associated with SGLT2i use compared to dipeptidyl peptidase-4 inhibitors (DPP4is) in patients with T2DM in Taiwan.

Design:

A nationwide population-based cohort study using Taiwan’s National Health Insurance Research Database.

Methods:

This study included over 700,000 patients with T2DM who initiated SGLT2is or DPP4is between 2016 and 2020. A prevalent new-user design was applied, and inverse probability treatment weighting was used to balance baseline characteristics. DKA cases were identified using validated diagnostic codes from hospital admissions and emergency visits.

Results:

The incidence of DKA was higher in SGLT2i users (0.34 per 1000 person-years (PY)) than in DPP4i users (0.19 per 1000 PY), with an adjusted hazard ratio of 2.04 (95% CI: 1.64–2.54). Subgroup analyses revealed a consistently elevated DKA risk, particularly among men, older adults, and patients with advanced diabetes complications. Sensitivity analyses, including an E-value of 3.72, suggested minimal impact from unmeasured confounders.

Conclusion:

SGLT2i use was associated with a twofold increased DKA risk compared to DPP4is in this Asian cohort. Although the absolute incidence was low, targeted monitoring is warranted in high-risk populations. The lower DKA incidence observed in Taiwan compared to Western studies suggests potential regional influences, necessitating further investigation. These findings contribute to the understanding of SGLT2i safety in Asian populations and inform clinical risk management strategies.

Plain language summary

Risk of a serious diabetes-related condition (ketosis) with common type 2 diabetes medications: A nationwide study from Taiwan

Sodium-glucose cotransporter-2 inhibitors (SGLT2is) are widely used to treat type 2 diabetes. In addition to lowering blood sugar, they provide heart and kidney benefits. However, they carry a small risk of a rare but serious condition called diabetic ketoacidosis (DKA), where excess ketone production causes acid buildup in the blood.

To better understand this risk, we analyzed health records from over 700,000 people with type 2 diabetes in Taiwan. We compared patients who started treatment with SGLT2is to those who used a different group of diabetes medications called dipeptidyl peptidase-4 inhibitors (DPP4is), which have a well-established safety profile.

Our study found that although DKA was uncommon overall, people using SGLT2is had about twice the risk of developing DKA compared to those using DPP4is. The risk was especially higher in men, older adults, and individuals with more advanced diabetes complications. Importantly, the overall rate of DKA in Taiwan was lower than what has been reported in Western countries, suggesting that regional differences in care or patient characteristics may influence risk. These results emphasize the need for careful monitoring when prescribing SGLT2is, particularly for patients with additional risk factors. It is also important for patients and caregivers to be aware of early signs of DKA, even when blood sugar levels are not very high.

By using real-world data from a large Asian population, this study adds important insights into the safety of SGLT2is. It supports informed treatment decisions and helps improve safe diabetes care in both Asian and global settings.

Keywords

type 2 diabetes mellitus (T2DM)diabetic ketoacidosis (DKA)sodium-glucose cotransporter-2 inhibitors (SGLT2i)dipeptidyl peptidase 4 inhibitors (DPP4i)prevalent new-user design inverse probability of treatment weighting (IPTW)

Introduction

Sodium-glucose cotransporter-2 inhibitors (SGLT2is) represent a significant advancement in the management of type 2 diabetes mellitus (T2DM) since their approval by the U.S. Food and Drug Administration (FDA) in 2013. Initially designed to lower blood glucose by promoting urinary glucose excretion, SGLT2is have demonstrated remarkable benefits beyond glycemic control. Landmark trials, such as EMPA-REG OUTCOME (Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes) and CREDENCE (Canagliflozin and Renal Events in Diabetes with Established Nephropathy Clinical Evaluation), revealed the protective effect of SGLT2i in reducing major cardiovascular events and slowing the progression of chronic kidney disease (CKD) in T2DM patients.^1–4 These findings make SGLT2i a preferred choice for diabetes management, particularly in patients at high cardiovascular or renal risk.

SGLT2is have been rapidly adopted worldwide. In the United States, annual prescriptions doubled from 2015 to 2020, and by 2017, they accounted for 22% of new anti-diabetic prescriptions in the United Kingdom.^5,6 This increasing utilization reflects their clinical benefits. However, concerns about their safety profile emerged as early as 2015, when the FDA issued a warning about an elevated risk of diabetic ketoacidosis (DKA), a rare but potentially severe complication.^7–9 Euglycemic DKA, characterized by ketoacidosis with relatively mild hyperglycemia (blood glucose <200 mg/dl), presents diagnostic challenges, often delaying treatment and increasing the risk of adverse outcomes.

Several randomized controlled trials (RCTs) have reported a higher incidence of DKA with SGLT2is compared to placebo.^1,3,10 However, the rarity of this event complicates drawing definitive conclusions. Observational studies and meta-analyses have yielded mixed findings,^9,11–16 particularly when comparing the risk of DKA between SGLT2is and other glucose-lowering agents, such as dipeptidyl peptidase-4 inhibitors (DPP4is) or sulfonylureas (SUs).

The global burden of diabetes has risen significantly, from 4.6% in 2000 to 10.5% in 2021, with projections estimating 643 million cases by 2030 and 783 million by 2045.¹⁷ This growing prevalence, alongside expanding indications for SGLT2is,^18–22 underscores the need to better understand their safety profile in real-world settings. Traditional clinical trials, while informative, often fail to capture the complexities of clinical practice, including patient comorbidities, concomitant medication use, and long-term adherence. Furthermore, most existing studies have predominantly focused on Western populations, with limited data available for Asian patients, whose diabetes characteristics differ significantly—such as more significant beta-cell dysfunction, lower body mass index, and a higher propensity for visceral fat accumulation.^23–25 These factors may also influence susceptibility to DKA, emphasizing the importance of region-specific research. To address this gap, we conducted a large-scale, population-based cohort study using real-world data from Taiwan’s National Health Insurance Research Database (NHIRD), providing a comprehensive assessment of SGLT2i-associated DKA risk in an Asian population compared to DPP4is.

Materials and methods

Ethics statement

This study was approved by the Institutional Review Board (IRB) of National Yang Ming Chiao Tung University (NYCU113050AE). Confidentiality and data protection were ensured according to the Health and Welfare Data Science Center regulations, Ministry of Health and Welfare, Taiwan. As the data were anonymized and encrypted prior to release, informed consent was waived by the IRB. All procedures complied with ethical standards set by the IRB and applicable governmental regulations.

Data sources

We utilized the NHIRD, a nationwide claims-based database managed by the National Health Insurance Administration (NHIA) of Taiwan, a mandatory insurance program that has provided most of the healthcare services in Taiwan since 1995.²⁶ The NHIRD encompasses medical claims, diagnoses, treatment records, prescribed medications, and healthcare utilization across outpatient, inpatient, and emergency settings for over 99% of Taiwan’s population. Data from 2011 to 2021 were collected, including disease diagnoses (according to the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) before 2016 and ICD-10-CM afterward), medications (classified using the Anatomical Therapeutic Chemical (ATC) Classification System), demographic information, reimbursement amounts, and healthcare provider identifiers for analysis. Additionally, individuals in our study were linked to data from the National Death Registry to confirm the date of death.

Study design

This study employed a prevalent new-user cohort design, a method well-suited for comparative drug effects study of newer medications like SGLT2is versus established treatments.²⁷ This design allows for a more comprehensive evaluation of drug safety, particularly by including patients who switch from older glucose-lowering agents to newer therapies, thus reflecting real-world clinical practice. DPP4is were chosen as the comparator for several reasons. First, DPP4is have been widely prescribed as one of the most common second-line oral anti-diabetic drugs (OADs) in Taiwan since their market introduction before the availability of SGLT2is. Many patients prescribed SGLT2is may have previously used DPP4is or other OADs like SUs or thiazolidinediones. Second, DPP4is are known for their favorable safety profile compared to other second-line agents, making them a suitable comparator. Furthermore, before 2019, NHIA prohibited the concurrent use of SGLT2is and DPP4is, further supporting the choice of DPP4is as an ideal comparator for evaluating the transition to SGLT2i therapy.

Study cohort

We identified patients with T2DM who initiated either SGLT2is (ATC: A10BK) or DPP4is (ATC: A10BH) between 2016 and 2020. Figure 1 illustrates the index date assignment for SGLT2i and DPP4i users. The available SGLT2is during the study period included dapagliflozin and empagliflozin, which were reimbursed starting in May 2016, and canagliflozin, reimbursed from January 2019.

Figure 1.

The diagram illustrates four groups and the index date of follow-up based on the prevalent new user design.

The study cohort was divided into four groups: (A) incident new users of SGLT2is, (B) incident new users of DPP4is, (C) prevalent new users of SGLT2is, and (D) prevalent users of DPP4is, as illustrated in Figure 1. The index date was defined as the first prescription date for SGLT2is (groups A and C) or DPP4is (group B). For prevalent users of DPP4is (group D), a pseudo index date was assigned based on the index date of the matched SGLT2i prevalent new users (group C) to ensure comparability in cohort entry timing.

Patients who discontinued DPP4is prior to their index date were excluded from the study. Further exclusions were applied for individuals who used both SGLT2is and DPP4is concurrently at the index date, had missing data for sex or age, were younger than 18 years, had a history of type 1 diabetes mellitus, end-stage renal disease, or kidney transplant, or experienced DKA within 1 year before the index date. The final cohort was used to compare the risk of DKA among T2DM patients using either SGLT2is or DPP4is. Detailed information on ICD codes used for disease diagnoses is available in Supplemental Table S1.

Inverse probability treatment weighting

We employed inverse probability treatment weighting (IPTW) to balance baseline characteristics between SGLT2i and DPP4i users, optimizing comparability between the two groups. IPTW was selected over other propensity score (PS) methods, such as PS matching (PSM),²⁸ because it retains most of the sample population, thus increasing statistical power by minimizing data loss. In contrast, PSM typically discards unmatched individuals, which can reduce sample sizes. However, IPTW can result in large weights, increasing variance, and potentially biasing effect estimates. To mitigate this, we applied 1% and 5% trimming to remove extreme weights and capped weights exceeding 10 and 20.^29,30 The PS weights were calculated based on several key variables: age, sex, year of cohort entry, diabetes duration, diabetes-related complications, and severity scores, including the adapted Diabetes Complications Severity Index (aDCSI),^31–33 Charlson Comorbidity Index (CCI),³⁴ comorbidities, and prior medication use (Figure 2 and detailed in Table 1). Both aDCSI and CCI are recognized as effective markers for capturing the severity of diabetes and associated comorbidities in claims-based studies.^31–33,35

Figure 2.

Patient selection flowchart.

Table 1.

Baseline characteristics of SGLT2i and DPP4i users before and after IPTW.

Characteristics	Before IPTW			After IPTW
	SGLT2i	DPP4i	ASMD^a	SGLT2i	DPP4i	ASMD^a
	%	%	ASMD^a	%	%	ASMD^a
Sample, N	166,156	609,635		723,101	777,957
Age (year), mean (SD)	56.8 (12.4)	63.8 (12.7)	0.558	60.9 (27.7)	62.4 (14.4)	0.068
Male	58%	52%	0.103	54%	54%	0.006
Year of cohort entry
2016	15%	23%	0.208	33%	22%	0.254
2017	22%	22%	<0.001	17%	22%	0.112
2018	21%	19%	0.030	17%	19%	0.075
2019	21%	18%	0.066	17%	19%	0.058
2020	21%	17%	0.111	16%	18%	0.046
Diabetes duration
0 year	5%	38%	0.858	50%	39%	0.233
0–1 year	14%	21%	0.195	17%	21%	0.091
2–3 years	42%	17%	0.584	13%	16%	0.085
4–5 years	39%	25%	0.310	19%	24%	0.120
DM-related comorbidities
Diabetic nephropathy	28%	33%	0.113	31%	32%	0.022
Diabetic neuropathy	10%	13%	0.078	12%	12%	0.002
Diabetic retinopathy	13%	14%	0.043	14%	14%	0.006
Hyperglycemia	1%	1%	0.036	1%	1%	0.004
Hypoglycemia	1%	2%	0.135	2%	2%	0.001
aDCSI, mean (SD)	0.2 (0.6)	0.3 (0.6)	0.167	0.2 (0.6)	0.3 (0.6)	0.167
Other comorbidities
Hypertension	62%	66%	0.095	64%	65%	0.032
Hyperlipidemia	34%	43%	0.202	40%	41%	0.029
Ischemic heart disease	16%	16%	0.007	15%	16%	0.020
Cardiac arrhythmia	1%	2%	0.159	2%	2%	0.046
HF	5%	6%	0.059	6%	6%	0.012
PVD	5%	6%	0.048	6%	5%	0.003
Ischemic stroke	4%	6%	0.122	6%	6%	0.010
CKD	8%	15%	0.231	12%	13%	0.038
Chronic lung disease	11%	44%	0.811	31%	37%	0.124
Cancer	4%	7%	0.105	7%	6%	0.003
Dementia	1%	4%	0.182	3%	3%	0.007
Hypovolemia	0%	1%	0.042	1%	1%	0.002
CCI, mean (SD)	2.4 (1.6)	2.7 (2.0)	0.166	2.7 (1.8)	2.7 (1.9)	<0.001
Anti-diabetic drugs
Metformin	73%	64%	0.195	69%	67%	0.046
SU	43%	47%	0.072	47%	47%	0.001
GLP1-RA	1%	0%	0.131	1%	1%	0.104
Meglitinides	10%	12%	0.087	14%	14%	0.003
TZD	11%	9%	0.081	14%	14%	0.003
α glucosidase inhibitors	47%	47%	0.001	47%	47%	0.001
Insulin	18%	17%	0.027	17%	17%	0.003
Other drugs
ACEI	7%	8%	0.048	8%	8%	0.011
ARB	54%	54%	0.005	53%	54%	0.026
α blockers	74%	64%	0.218	69%	67%	0.046
b blockers	31%	32%	0.022	31%	32%	0.021
CCB	34%	43%	0.185	40%	41%	0.021
Loop diuretics	8%	14%	0.209	12%	13%	0.014
Thiazide	6%	9%	0.100	8%	8%	0.010
Potassium-sparing agents	4%	5%	0.049	5%	5%	0.005
Antiplatelets	30%	35%	0.110	33%	34%	0.023
Statins	71%	62%	0.195	64%	64%	0.004
NSAIDs	30%	33%	0.056	32%	33%	0.004
Health care utilization
Any hospitalization before 90 days	7%	12%	0.169	11%	11%	0.016
Any hospitalization before 91–365 days	11%	15%	0.107	13%	14%	0.013
Number of office visits, mean (SD)	28.7 (18.7)	32.4 (21.2)	0.185	31.1 (20.0)	31.6 (21.0)	0.024

aDCSI: adopted diabetic comorbidity severity index; ACEI: angiotensin-converting enzyme inhibitor; ARB: angiotensin receptor blocker; ASMD: absolute standardized mean difference; CCB: calcium channel blocker; CCI: Charlson comorbidity index; CKD: chronic kidney disease; DPP4i: dipeptidyl peptidase-4 inhibitor; GLP1-RA: glucagon-like peptide-1 receptor agonists; IPTW: inverse probability treatment weighting; HF: heart failure; NSAID: non-steroidal anti-inflammatory drugs; PVD: peripheral vascular disease; SD: standard deviation; SU: sulfonylurea; SGLT2i: sodium-glucose cotransporter-2 inhibitor; TZD: thiazolidinedione.

ASMD <0.1 denotes balance in the covariate between two groups.

Outcome measurement

The primary outcome was DKA, identified using ICD-10-CM code E11.X, occurring during hospital admissions or emergency department visits. The positive predictive value (PPV) for DKA diagnosis was validated at 89%.³⁶ A secondary outcome, hyperosmolar hyperglycemic syndrome (HHS, ICD-10-CM E87.0), was used as a negative control to assess potential bias.³⁷ Both DKA and HHS events were recorded during the study period.

Statistical analysis

Baseline characteristics of SGLT2i and DPP4i users were compared before and after applying IPTW using absolute standardized mean differences (ASMDs). An ASMD greater than 0.1 indicated an imbalance between groups. To estimate adjusted hazard ratios (HRs) for the risk of DKA, Cox proportional hazard models were employed, incorporating IPTW to account for baseline confounders.

The time-to-event analysis began on the index date and continued until the first occurrence of DKA or any of the following censoring events: medication discontinuation (with a 60-day grace period for prescription refills), death, treatment switch or add-on, or the completion of 1 year of follow-up.

Given differences in treatment discontinuation, switching, and add-on therapy rates between the groups, an as-treated analysis was conducted to more accurately reflect real-world treatment patterns (see Supplemental Table S3). This approach accounts for modifications in treatment over time, offering a more representative depiction of the therapeutic experience. All statistical analyses were performed using SAS/STAT 9.4 (SAS Institute Inc.) and R (version 3.2.5 for Windows). A p-value of <0.05 was considered statistically significant.

Subgroup and sensitivity analyses

Preplanned subgroup analyses were conducted based on age, sex, prior insulin use, CKD history, HF history, CCI, and aDCSI scores. Sensitivity analyses tested the robustness of the results using various PS methods (e.g., trimming and stratification), intention to treat analysis, varying grace periods for discontinuation, incident new user design, and different DKA definitions (e.g., hospital admissions only). To evaluate the potential impact of unmeasured confounders, we calculated the E-value, which estimates the strength an unmeasured confounder would need to have to fully explain the observed association between SGLT2i use and DKA.

Results

Baseline characteristics of the study population

After the application of exclusion criteria, a total of 166,156 patients were identified as SGLT2i users and 609,635 as DPP4i users between 2016 and 2020. Prior to adjusting for confounders, notable differences in baseline characteristics between the two groups were observed (Table 1). Patients using SGLT2is were younger (mean age 58.3 vs 64.7 years), included fewer women (42% vs 48%), and had longer diabetes duration (4–5 years: 39% vs 25%) compared to DPP4i users. Comorbidities such as CKD (8% vs 15%), diabetic nephropathy (10% vs 13%), chronic lung disease (11% vs 44%), and dementia (1% vs 4%) were less prevalent among SGLT2i users compared to DPP4i users. Additionally, SGLT2i users were less likely to have a history of loop diuretic or antiplatelet use, had a lower CCI, and experienced fewer outpatient visits and hospitalizations than those in the DPP4i group.

After applying IPTW, the final weighted cohort included 723,101 individuals in the SGLT2i group and 777,957 in the DPP4i group, resulting in well-balanced covariates across groups, as indicated by the reduction of ASMDs for all variables to below 0.1 (Table 1), suggesting successful balancing of baseline characteristics.

Incidence and HRs for DKA

During the follow-up period, the SGLT2i group exhibited a higher incidence of DKA compared to the DPP4i group. A total of 208 patients in the SGLT2i group developed DKA, compared to 130 patients in the DPP4i group. The incidence rate was 0.34 per 1000 person-years (PY) (95% confidence interval (CI), 0.28–0.39) among SGLT2i users, compared to 0.19 per 1000 PY (95% CI, 0.14–0.24) among DPP4i users (Table 2). This indicates a higher frequency of DKA events among patients treated with SGLT2is.

Table 2.

The incidence (1000 PY) and hazard risk of DKA and HHS of SGLT2i users compared to DPP4i users.

Outcome	Exposure	Events	PYs	IR (95% CI)	HR^a (95% CI)
DKA	SGLT2i	208	612,159	0.34 (0.28–0.39)	2.04 (1.64–2.54)
DKA	DPP4i	130	684,176	0.19 (0.14–0.24)	Reference
HHS	SGLT2i	2,005	571,225	3.51 (3.20–3.83)	1.02 (0.96–1.08)
HHS	DPP4i	2,501	696,657	3.59 (3.30–3.81)	Reference

CI: confidence interval; DPP4i: dipeptidyl peptidase-4 inhibitor; DKA: diabetic ketoacidosis; HHS: hyperosmolar hyperglycemic syndrome; HR: hazard ratio; IR: incidence rate; PY: person-year; SGLT2i: sodium-glucose cotransporter-2 inhibitor.

HR was estimated based on conditional Cox regression analysis.

The HR for DKA in SGLT2i users versus DPP4i users was 2.04 (95% CI, 1.64–2.54) (Table 2), suggesting a more than twofold increase in the risk of DKA with SGLT2i use. In contrast, the HR of HHS, used as a negative control outcome, was 1.02 (95% CI, 0.96–1.08), showing no significant difference between the two groups. This suggests that the elevated risk is specific to DKA rather than other hyperglycemic emergencies, further confirming the strong association between SGLT2i use and DKA.

Subgroup analyses of the risk of DKA associated with SGLT2i use

Subgroup analyses were performed to assess the consistency of the association between SGLT2i use and DKA across different patient characteristics (Figure 3). The association between SGLT2i and increased DKA risk was observed consistently across most subgroups, including age, insulin use, and CKD history. However, the association was not significant in females (HR 1.42, 95% CI, 0.89–2.26) or in patients with an aDCSI score <1 (HR 1.25, 95% CI, 0.77–2.04). Significant interaction effects were identified for sex (p for interaction = 0.02) and CKD history (p for interaction = 0.01), suggesting that these factors may modify the risk of DKA with SGLT2i use.

Figure 3.

Subgroup analysis for the risk of DKA stratified by patient characteristics.

Sensitivity analyses for DKA risk of SGLT2i users compared to DPP4i users

We conducted a series of sensitivity analyses to confirm the robustness of the association between SGLT2i use and the risk of DKA (Figure 4). Different PS methods and analytic techniques were applied, including trimming at 99% and 95%, truncation at weights of 10 and 20, and PS stratification (with 5 and 10 strata). These adjustments consistently supported the elevated HRs for DKA among SGLT2i users. For example, trimming at the 95th percentile yielded an HR of 1.92 (95% CI: 1.29–2.87), and truncation with a weight cap of 10 resulted in an HR of 1.80 (95% CI: 1.13–2.89), affirming the increased DKA risk associated with SGLT2i.

Figure 4.

Sensitivity analysis for the risk of DKA among SGLT2i users compared to DPP4i users.

Further analyses, such as the intention to treat approach and varying grace periods for medication discontinuation (90 and 45 days), reinforced the association. Although some analyses, such as 1:1 PSM (HR, 1.56; 95% CI: 0.98–2.48), the new-user study design (HR, 1.14; 95% CI: 0.68–1.91), and restricting DKA outcomes to hospital admissions (HR, 1.61; 95% CI: 0.93–2.32), did not reach statistical significance, the overall trend remained consistent, indicating an elevated DKA risk among SGLT2i users compared to DPP4i users.

The calculated E-value for the association between SGLT2i use and DKA was 3.52, indicating that an unmeasured confounder would need to be strongly associated with both the exposure and outcome to nullify the observed association. This high E-value reinforces the robustness of our findings, suggesting that the relationship between SGLT2i use and DKA is unlikely to be explained by unmeasured confounding factors.

Agent-specific risk of DKA among SGLT2is

Among SGLT2i users, dapagliflozin was the most commonly prescribed agent (52%), followed by empagliflozin (39%) and canagliflozin (9%). The incidence rate of DKA varied across different SGLT2i agents. Dapagliflozin had the highest incidence rate at 0.35 per 1000 PY (95% CI: 0.31–0.39), followed by empagliflozin at 0.22 per 1000 PY (95% CI: 0.19–0.25), and canagliflozin at 0.22 per 1000 PY (95% CI: 0.20–0.24). Compared to DPP4is, the HR for DKA was significantly elevated for dapagliflozin (HR 1.41, 95% CI: 1.07–1.84) and empagliflozin (HR 2.43, 95% CI: 1.89–3.11), whereas canagliflozin did not show a significant increase in risk (HR 0.88, 95% CI: 0.42–1.86) (Table S4).

Discussion

This study aimed to evaluate the risk of DKA associated with the use of SGLT2is compared to DPP4is in patients with T2DM using data from Taiwan’s NHIRD. While the absolute incidence rates were low, SGLT2i use was significantly associated with an elevated risk of DKA compared to DPP4i use. This association remained consistent across most subgroups, including age, insulin use, and CKD history, and was supported by sensitivity analyses. The risk varied among specific SGLT2i agents, with dapagliflozin and empagliflozin associated with a higher incidence of DKA compared to canagliflozin.

The elevated DKA risk associated with SGLT2i use observed in our large Taiwanese cohort aligns with previous findings from both RCTs and real-world observational studies. RCTs consistently report a twofold increase in DKA risk among SGLT2i users compared to placebo, including empagliflozin in the EMPA-REG OUTCOME trial (HR: 2.18, 95% CI: 1.10–4.30),¹ canagliflozin in the Canagliflozin Cardiovascular Assessment Study (CANVAS) Program (HR: 2.33, 95% CI: 0.76–7.17),³⁸ and dapagliflozin in the Dapagliflozin Effect on Cardiovascular Events-Thrombolysis in Myocardial Infarction 58 (DECLARE-TIMI) 58 trial (HR: 2.18, 95% CI: 1.10–4.30).² Real-world evidence further supports these findings, including analyses from the FDA Adverse Event Reporting System (FAERS)³⁹ and a large multicenter cohort study in Canada and the UK,¹³ both of which identified an elevated DKA risk with SGLT2i use. While some observational studies have reported conflicting results^11,12,40—likely due to variations in sample size, study design, and matching methodologies—a comprehensive meta-analysis reinforced the increased DKA risk, reporting a higher incidence in both observational studies (HR: 1.7) and RCTs (HR: 2.5).⁴¹ These findings collectively demonstrate the consistency of evidence across different study designs.

Although an increased DKA risk with SGLT2i use has been well established, reported incidence rates vary widely. In our Taiwanese cohort, the incidence of DKA among SGLT2i users was 0.34 per 1000 PY (95% CI: 0.28–0.39), placing it at the lower end of the range reported in observational studies (0.2–6.3 events per 1000 PY) and below most RCTs (0.6–2.2 per 1000 PY).^{3,9,13,14,38,41} This lower incidence may reflect differences in patient demographics, treatment practices, or healthcare systems, but further studies are required to clarify the underlying factors.

Among the limited studies on SGLT2i-related DKA in Asian populations, two large-scale Korean studies reported no significant increase in DKA risk with SGLT2is compared to DPP4is, a finding that contrasts with our results.^40,42 Differences in study design may partly account for these discrepancies, as one study focused only on hospitalized DKA cases, potentially underestimating overall incidence. Additionally, reported DKA rates varied, with one study observing 0.26 per 1000 PY among older adults,⁴² while another reported 2.51 per 1000 PY during the first 30 days of SGLT2i use, which later declined to 0.614 per 1000 PY over 3 years.⁴⁰ These variations highlight the importance of harmonized methodologies to enhance comparability and accurately assess DKA risk across different populations.

Our study also provides molecule-specific insights, demonstrating an increased risk of DKA with dapagliflozin and empagliflozin, while canagliflozin did not show a statistically significant risk increase. The lower observed risk for canagliflozin may be attributed to its later introduction to the Taiwanese market in 2019, leading to a smaller patient population and fewer recorded DKA events. Consequently, statistical power may have been insufficient to detect a significant association, and further large-scale studies are necessary to confirm these findings.

Patients with an aDCSI score ⩾1 faced a higher risk, likely due to the impact of diabetes-related complications on metabolic stability. Although direct evidence linking aDCSI to DKA is limited, more severe complication scores are generally associated with poor glycemic control and increased hospitalization rates, both of which may contribute to DKA susceptibility.^35,43 The elevated DKA risk with SGLT2is compared to DPP4is persisted regardless of prior insulin therapy, consistent with findings from Douros et al., who reported a 2.85-fold increased risk, particularly among those without prior insulin use. This trend has been attributed to SGLT2i-induced ketogenesis driven by reductions in the insulin-to-glucagon ratio.⁴⁴ These results emphasize the need for careful risk assessment in patients with advanced diabetes complications when considering SGLT2i therapy.

This large, population-based cohort study provides one of the most comprehensive evaluations of SGLT2i-associated DKA risk in an Asian population. Leveraging NHIRD data, we applied a prevalent new-user design and advanced statistical methods, including IPTW, to minimize bias. Sensitivity analyses (E-value: 3.72) suggested that unmeasured confounders were unlikely to have significantly impacted our findings, enhancing their reliability.

However, several limitations should be considered. As an observational study, causality cannot be definitively established. Clinical variables such as HbA1c and creatinine were not available, though validated composite indices (CCI and aDCSI) were used to approximate diabetes severity. The reliance on administrative data may introduce misclassification bias, but incorporating diagnostic codes from emergency visits and hospitalizations helped improve case identification. Additionally, prescription refill records were used to indirectly measure medication adherence, though they do not account for actual drug intake. As with all observational studies using administrative claims, residual confounding from unmeasured factors such as lifestyle, undiagnosed comorbidities, and diabetes duration cannot be fully eliminated. Although DKA was identified using validated ICD-10 codes with high PPV, some degree of outcome misclassification may still be present. Because an E10 diagnosis in Taiwan is closely tied to eligibility for the National Health Insurance catastrophic illness certification program, the likelihood of miscoding is considered low. Individuals with any prior E10 coding were, therefore, excluded to prevent inclusion of true type 1 diabetes. While a very small number of miscoded cases may still occur, such patients would only be included if they initiated SGLT2 inhibitors or DPP4 inhibitors. Any resulting misclassification would be rare and non-differential between exposure groups, thereby biasing risk estimates toward the null. Thus, the potential impact on DKA ascertainment is expected to be minimal. Dose-specific analyses were not feasible because DKA events were rare for individual SGLT2 inhibitors, dose variability during the study period was minimal when these agents were used primarily for glycemic management in type 2 diabetes, and reimbursement-based dose records do not capture titration or actual daily exposure. Although a higher rate of switching from SGLT2 inhibitors to the comparator could reduce the accumulation of DKA events in the exposed group, such bias would be expected to attenuate the observed association. The finding that SGLT2 inhibitor users still exhibited a significantly higher DKA risk, therefore, suggests that our estimate is conservative and that the increased risk is unlikely to be explained by differential switching. Because an as-treated approach was used, follow-up time varied according to treatment switching or discontinuation. However, the Cox model directly incorporates person-time at risk, ensuring that such variation in follow-up duration does not bias the hazard ratio. The absolute incidence of DKA in Taiwan was relatively low, which may reduce statistical power in rare subgroup analyses and may limit the generalizability of our findings to populations with a higher baseline risk of DKA. Finally, while this study provides valuable insights into SGLT2i safety in Taiwan, further research is needed to assess its applicability to other populations and healthcare systems.

Conclusion

This study provides real-world evidence that SGLT2i use is associated with a higher risk of DKA compared to DPP4i, particularly among patients with advanced diabetes complications, men, and older adults. Although the absolute incidence remained low, these findings indicate the importance of patient education and vigilant monitoring to mitigate risk. The lower incidence observed in Taiwan compared to other studies suggests potential regional differences that may influence risk, though further research is needed to clarify the contributing factors. By addressing gaps in Asian-specific data, this study advances the understanding of SGLT2i safety and informs risk management strategies for their clinical use.

Supplemental Material

sj-docx-1-taj-10.1177_27558428261421697 – Supplemental material for Risk of diabetic ketoacidosis with sodium-glucose cotransporter-2 inhibitors compared with dipeptidyl peptidase-4 inhibitors in type 2 diabetes: A nationwide cohort study

Supplemental material, sj-docx-1-taj-10.1177_27558428261421697 for Risk of diabetic ketoacidosis with sodium-glucose cotransporter-2 inhibitors compared with dipeptidyl peptidase-4 inhibitors in type 2 diabetes: A nationwide cohort study by Hsiao-Yun Yeh, Hong-Fu Lu, Liang-Yu Lin and Li-Nien Chien in Sage Open Chronic Disease

Footnotes

Acknowledgements

We thank Taiwan’s National Health Insurance Administration (NHIA) and Health and Welfare Data Science Center (HWDC) for making the databases used in this study available; however, the content of this article does not represent any official position of the NHIA or HWDC.

ORCID iD

Hsiao-Yun Yeh

Ethical considerations

This study was approved by the IRB of National Yang Ming Chiao Tung University (NYCU113050AE). Data confidentiality was maintained per Health and Welfare Data Science Center (HWDC) regulations, and informed consent was waived due to prior anonymization and encryption.

Consent to participate

Informed consent was waived due to prior anonymization and encryption.

Author contributions

Hsiao-Yun Yeh: Conceptualization; Data curation; Validation; Writing – original draft; Writing – review & editing.

Hong-Fu Lu: Conceptualization; Formal analysis; Methodology; Software; Visualization; Writing – original draft.

Liang-Yu Lin: Conceptualization; Methodology; Writing – review & editing.

Li-Nien Chien: Conceptualization; Data curation; Funding acquisition; Methodology; Project administration; Writing – original draft; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the National Science and Technology Council of Taiwan (NSTC-112-2314-B-A49-080). The funders had no role in the study design, the collection, analysis, and interpretation of data, the writing of the report, or the decision to submit the article for publication.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The data used in this study are from the Taiwan National Health Insurance Research Database (NHIRD), maintained by the Health and Welfare Data Science Center (HWDC), Ministry of Health and Welfare, Taiwan. Due to legal and ethical regulations, the data are not publicly available. Access requires formal application and approval from both the HWDC and the Institutional Review Board (IRB) of National Yang Ming Chiao Tung University.

Supplemental material

Supplemental Material for this article is available online.

References

Zinman

Wanner

Lachin

, et al. Empagliflozin, cardiovascular outcomes, and mortality in type 2 diabetes. N Engl J Med 2015; 373(22): 2117–2128.

Wiviott

Raz

Bonaca

, et al. Dapagliflozin and cardiovascular outcomes in type 2 diabetes. N Engl J Med 2019; 380(4): 347–357.

Perkovic

Jardine

Neal

, et al. Canagliflozin and renal outcomes in type 2 diabetes and nephropathy. N Engl J Med 2019; 380(24): 2295–2306.

Zelniker

Wiviott

Raz

, et al. SGLT2 inhibitors for primary and secondary prevention of cardiovascular and renal outcomes in type 2 diabetes: a systematic review and meta-analysis of cardiovascular outcome trials. Lancet 2019; 393(10166): 31–39.

Adhikari

Jha

Dardari

, et al. National trends in use of sodium-glucose cotransporter-2 inhibitors and glucagon-like peptide-1 receptor agonists by cardiologists and other specialties, 2015 to 2020. J Am Heart Assoc 2022; 11(9): e023811.

Wilkinson

Douglas

Stirnadel-Farrant

, et al. Changing use of antidiabetic drugs in the UK: trends in prescribing 2000–2017. BMJ Open 2018; 8(7): e022768.

FDA Drug Safety Communication: FDA warns that SGLT2 inhibitors for diabetes may result in a serious condition of too much acid in the blood, https://wayback.archive-it.org/7993/20170112031553/http://www.fda.gov/Drugs/DrugSafety/ucm446845.htm (2015).

Ogawa

Sakaguchi

Euglycemic diabetic ketoacidosis induced by SGLT2 inhibitors: possible mechanism and contributing factors. J Diabetes Investig 2016; 7(2): 135–138.

Fralick

Schneeweiss

Patorno

Risk of diabetic ketoacidosis after initiation of an SGLT2 inhibitor. N Engl J Med 2017; 376(23): 2300–2302.

10.

Heerspink

HJL

Stefansson

Correa-Rotter

, et al. Dapagliflozin in patients with chronic kidney disease. N Engl J Med 2020; 383(15): 1436–1446.

11.

Wang

Desai

Ryan

, et al. Incidence of diabetic ketoacidosis among patients with type 2 diabetes mellitus treated with SGLT2 inhibitors and other antihyperglycemic agents. Diabetes Res Clin Pract 2017; 128: 83–90.

12.

Wang

Voss

Weaver

, et al. Diabetic ketoacidosis in patients with type 2 diabetes treated with sodium glucose co-transporter 2 inhibitors versus other antihyperglycemic agents: an observational study of four US administrative claims databases. Pharmacoepidemiol Drug Saf 2019; 28(12): 1620–1628.

13.

Douros

Lix

Fralick

, et al. Sodium-glucose cotransporter-2 inhibitors and the risk for diabetic ketoacidosis: a multicenter cohort study. Ann Intern Med 2020; 173(6): 417–425.

14.

Dawwas

Flory

Hennessy

, et al. Comparative safety of sodium-glucose cotransporter 2 inhibitors versus dipeptidyl peptidase 4 inhibitors and sulfonylureas on the risk of diabetic ketoacidosis. Diabetes Care 2022; 45(4): 919–927.

15.

Patorno

Pawar

Franklin

, et al. Comparative effectiveness and safety of empagliflozin: an interim analysis from the empagliflozin comparative effectiveness and safety (Emprise) study. J Am Coll Cardiol 2019; 73(9): 683–683.

16.

Ueda

Svanstrom

Melbye

, et al. Sodium glucose cotransporter 2 inhibitors and risk of serious adverse events: nationwide register based cohort study. BMJ 2018; 363: k4365.

17.

IDF. IDF diabetes atlas. 10th ed. International Diabetes Federation. Accessed June 12, 2023. https://diabetesatlas.org/atlas/tenth-edition/

18.

Solomon

McMurray

JJV

Claggett

, et al. Dapagliflozin in heart failure with mildly reduced or preserved ejection fraction. N Engl J Med 2022; 387(12): 1089–1098.

19.

Packer

Anker

Butler

, et al. Cardiovascular and renal outcomes with empagliflozin in heart failure. N Engl J Med 2020; 383(15): 1413–1424.

20.

Gao

Y-M

Feng

S-T

Wen

, et al. Cardiorenal protection of SGLT2 inhibitors—perspectives from metabolic reprogramming. eBioMedicine 2022; 83: 104215.

21.

The

E-KCG

Herrington

Staplin

, et al. Empagliflozin in patients with chronic kidney disease. N Engl J Med 2023; 388(2): 117–127.

22.

Sharma

Razaghizad

Joury

, et al. Primary and secondary cardiovascular and kidney prevention with canagliflozin: insights from the CANVAS program and CREDENCE trial. J Am Heart Assoc 2024; 13(3):e031586.

23.

Rhee

EJ.

Diabetes in Asians. Endocrinol Metab (Seoul) 2015; 30(3): 263–269.

24.

Chan

Malik

Jia

, et al. Diabetes in Asia: epidemiology, risk factors, and pathophysiology. JAMA 2009; 301(20): 2129–2140.

25.

Deurenberg

Deurenberg-Yap

Guricci

Asians are different from Caucasians and from each other in their body mass index/body fat per cent relationship. Obes Rev 2002; 3(3): 141–146.

26.

Hsieh

Shao

, et al. Taiwan’s National Health Insurance Research Database: past and future. Clin Epidemiol 2019; 11: 349–358.

27.

Suissa

Moodie

Dell’Aniello

Prevalent new-user cohort designs for comparative drug effect studies by time-conditional propensity scores. Pharmacoepidemiol Drug Saf 2017; 26(4): 459–468.

28.

Austin

Stuart

EA.

Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015; 34(28): 3661–3679.

29.

Sturmer

Webster-Clark

Lund

, et al. Propensity score weighting and trimming strategies for reducing variance and bias of treatment effect estimates: a simulation study. Am J Epidemiol 2021; 190(8): 1659–1670.

30.

Chesnaye

Stel

Tripepi

, et al. An introduction to inverse probability of treatment weighting in observational research. Clin Kidney J 2021; 15(1): 14–20.

31.

Wicke

Glushan

Schubert

, et al. Performance of the adapted Diabetes Complications Severity Index translated to ICD-10. Am J Manag Care 2019; 25(2):e45–e49.

32.

Glasheen

Renda

Dong

Diabetes Complications Severity Index (DCSI)—update and ICD-10 translation. J Diabetes Complications 2017; 31(6): 1007–1013.

33.

Chang

Weiner

Richards

, et al. Validating the adapted Diabetes Complications Severity Index in claims data. Am J Manag Care 2012; 18(11): 721–726.

34.

Quan

Sundararajan

Halfon

, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005; 43(11): 1130–1139.

35.

Yeh

Liu

, et al. Adapted Diabetes Complications Severity Index and Charlson Comorbidity Index in predicting all-cause and cause-specific mortality among patients with type 2 diabetes. BMJ Open Diabetes Res Care 2023; 11(2): e003262.

36.

Bobo

Cooper

Epstein

Jr , et al. Positive predictive value of automated database records for diabetic ketoacidosis (DKA) in children and youth exposed to antipsychotic drugs or control medications: a Tennessee Medicaid Study. BMC Med Res Methodol 2011; 11: 157.

37.

Pasquel

Umpierrez

GE.

Hyperosmolar hyperglycemic state: a historic review of the clinical presentation, diagnosis, and treatment. Diabetes Care 2014; 37(11): 3124–3131.

38.

Neal

Perkovic

Mahaffey

, et al. Canagliflozin and cardiovascular and renal events in type 2 diabetes. N Engl J Med 2017; 377(7): 644–657.

39.

Blau

Tella

Taylor

, et al. Ketoacidosis associated with SGLT2 inhibitor treatment: analysis of FAERS data. Diabetes Metab Res Rev 2017; 33(8): 10.1002/dmrr.2924.

40.

Kim

Jeon

Han

, et al. Sodium-glucose co-transporter-2 inhibitors and the risk of ketoacidosis in patients with type 2 diabetes mellitus: a nationwide population-based cohort study. Diabetes Obes Metab 2018; 20(8): 1852–1858.

41.

Colacci

Fralick

Odutayo

, et al. Sodium-glucose cotransporter-2 inhibitors and risk of diabetic ketoacidosis among adults with type 2 diabetes: a systematic review and meta-analysis. Can J Diabetes 2022; 46(1): 10–15.e2.

42.

Han

Lee

, et al. Effectiveness and safety of sodium-glucose co-transporter-2 inhibitors compared with dipeptidyl peptidase-4 inhibitors in older adults with type 2 diabetes: a nationwide population-based study. Diabetes Obes Metab 2021; 23(3): 682–691.

43.

Chen

H-L

Hsiao

F-Y.

Risk of hospitalization and healthcare cost associated with Diabetes Complication Severity Index in Taiwan’s National Health Insurance Research Database. J Diabetes Complications 2014; 28(5): 612–616.

44.

Perry

Shulman

GI.

Sodium-glucose cotransporter-2 inhibitors: understanding the mechanisms for therapeutic promise and persisting risks. J Biol Chem 2020; 295(42): 14379–14390.