Abstract
Aims
To identify predictors of diabetic ketoacidosis (DKA) in patients with an insulin‐deficient phenotype initiating sodium‐glucose cotransporter 2 inhibitor (SGLT2i) therapy.
Materials and Methods
This retrospective cohort study analysed data from 31 900 patients with diabetes aged 18–70 identified as having an insulin‐deficient phenotype. After applying inclusion and exclusion criteria, patients were matched and divided into SGLT2i users (n = 6572) and non‐users (n = 6382). The primary endpoint was the first DKA event in patients with no prior history of DKA. Independent risk factors for DKA were assessed using Cox regression.
Results
Over a median follow‐up of 4.4 years, 239 patients experienced DKA (143 [2.22%] SGLT2i users vs. 96 [1.54%] non‐users; HR [95% confidence interval, CI] 1.39 [1.07–1.79]; p = 0.014). The adjusted model confirmed an increased DKA risk with SGLT2i use (adjusted hazard ratio, aHR [95% CI] 1.50 [1.15–1.95]; p = 0.003). Baseline HbA1c >9% was associated with a 53% higher risk (aHR [95% CI] 1.53 [1.18–1.99]; p = 0.0016), while body mass index (BMI) ≤25 kg/m2 was linked to a 61% increased risk (aHR [95% CI] 1.61 [1.24–2.09]; p = 0.0003). Insulin use further heightened risk (aHR [95% CI] 2.35 [1.71–3.23]; p < 0.0001).
Conclusions
SGLT2i use in patients with an insulin‐deficient phenotype is associated with increased DKA risk, particularly in those with HbA1c >9% and BMI ≤25 kg/m2. Clinicians should exercise caution in these patients, carefully assessing risks and implementing mitigation strategies to ensure safe use.
Keywords: antidiabetic drug, observational study, SGLT2 inhibitor, type 1 diabetes
1. INTRODUCTION
Sodium‐glucose cotransporter 2 inhibitors (SGLT2i) have transformed the management of type 2 diabetes mellitus (T2DM), chronic kidney disease (CKD) and heart failure (HF) providing significant kidney and cardiovascular benefits. 1 As glucose‐lowering agents, they provide a novel therapeutic approach through an insulin‐independent mechanism, offering important benefits to diabetes management strategies. Numerous randomized controlled trials (RCTs) have demonstrated the efficacy of SGLT2i in improving glycaemic control, slowing kidney disease progression and reducing cardiovascular risk in patients with T2DM. 2 , 3 , 4 , 5 , 6 , 7 Despite their well‐established benefits in T2DM, the use of SGLT2i in type 1 diabetes mellitus (T1DM) has been approached with caution due to concerns over an increased risk of diabetic ketoacidosis (DKA), a rare but potentially life‐threatening complication. Moreover, the risk of DKA exists in patients with T2DM as well, often in the context of acute illness in subjects with severe insulin deficiency.
Clinical trials evaluating the efficacy and safety of SGLT2i added to insulin therapy in T1DM have demonstrated moderate reductions in HbA1c, improving glycaemic variability and lowering insulin requirements while conferring benefits in terms of weight, blood pressure and kidney function. 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 However, the incidence of DKA was higher in the SGLT2i‐treated groups. 16 This heightened risk has limited the broader adoption of SGLT2i in T1DM despite their promising clinical outcomes. Furthermore, a population‐based cohort study of adults with T2DM initiating SGLT2i therapy found that HbA1c levels above 10% and a prior history of DKA, both indicative of an insulin‐deficient phenotype, were significant risk factors for developing DKA. 17 These findings highlight the importance of patient‐specific factors in determining the safety profile of SGLT2i.
While DKA remains a critical safety concern, the impact of SGLT2i on DKA risk in insulin‐deficient individuals with no prior history of ketoacidosis is still unclear. To address this gap, our study aims to investigate the risk of DKA in patients with an insulin‐deficient phenotype, including, but not limited to, those with T1DM, who are initiating SGLT2i therapy and have no history of DKA. Our definition of an insulin‐deficient phenotype is based on simple clinical characteristics—low body mass index (BMI), elevated HbA1c and/or insulin use—which have been shown to be reliable proxies for underlying insulin deficiency. 18 In fact, recent analyses from clinical trials and real‐world data suggest that such continuous clinical measures can outperform complex cluster‐based diabetes classifications for predicting disease progression and therapy response. Additionally, these features align with the severe insulin‐deficient diabetes cluster identified in data‐driven models of diabetes subtypes, which is associated with higher risks of complications such as retinopathy and poor glycaemic control. 19 Additionally, we aim to determine whether BMI or HbA1c levels modulate DKA risk in this population.
2. MATERIALS AND METHODS
2.1. Study design and data source
This retrospective observational cohort study evaluated the incidence of DKA events in adult patients with an insulin‐deficient phenotype treated with or without SGLT2 inhibitors. Anonymized data were obtained from Clalit Health Services (CHS), Israel's largest integrated payer/provider healthcare organization, covering approximately 4.5 million members. Data extraction utilized the Clalit Research Data Sharing platform, powered by MDClone (https://www.mdclone.com). The CHS database contains comprehensive clinical and administrative data for inpatient and outpatient care including all patients' interactions—diagnoses, medical visits, hospitalizations, prescription and purchased medications and laboratory tests.
The study included data from 1 January 2015 (when SGLT2i were first introduced in Israel) to 31 December 2022, for initiation of SGLT2i, with follow‐up until 31 December 2023. The study protocol was approved by the CHS Community Helsinki and Data Utilization Committees. Individual informed consent was not required due to the study's retrospective nature and the de‐identified patient data.
2.2. Study population
The study population consisted of all patients with diabetes aged 18–70 in 2019 who met the following criteria: age at diabetes diagnosis ≤50 years, BMI ≤27 kg/m2, HbA1c ≥8.5% (for insulin‐naïve patients) or any HbA1c value for patients previously treated with insulin. These criteria were designed to identify patients with a high likelihood of having an insulin‐deficient phenotype. Our approach, based on younger age at diagnosis, lower BMI and elevated HbA1c in insulin‐naïve individuals, is supported by findings from prior clinical and epidemiological studies. Dennis et al. demonstrated that models using these simple, continuous clinical features rather than discrete data‐driven clusters were superior for stratifying disease progression and tailoring therapy in type 2 diabetes. 18 Moreover, Ahlqvist et al. previously identified a ‘severe insulin‐deficient diabetes’ cluster using similar parameters, with elevated risk of complications. 19 Our definition thus aligns with both pragmatic clinical prediction and established data‐driven phenotypes.
Patients were excluded if they met any of the following criteria: (1) treatment with SGLT2i prior to 1 January 2015, (2) baseline estimated glomerular filtration rate (eGFR) <30 mL/min/1.73 m2, (3) a history of kidney transplant or diagnosis of polycystic kidney disease, (4) membership in CHS for less than 6 months prior to study entry, or (5) lack of creatinine lab results prior to the index date.
2.3. Cohort building process
The cohort was constructed in two steps. First, SGLT2i users and non‐users were 1:1 matched using a mixed approach: exact matching for gender, eGFR category, urinary albumin‐to‐creatinine ratio (UACR) category, duration of diabetes and the use of short‐acting insulin, and propensity score matching for age, BMI, diagnosis of heart failure (HF), history of cardiovascular disease (CVD) (defined as a medical diagnosis of cerebrovascular accident [CVA], ischaemic heart disease [IHD], myocardial infarction [MI] or peripheral vascular disease [PVD]), use of statins, GLP‐1 receptor agonists (RA) and sulphonylurea/repaglinide. The expected median date of the first SGLT2i purchase (30 June 2019) was used as the reference date for extracting the baseline characteristics of both SGLT2i users and non‐users. The baseline characteristics closest to this reference date were used for matching the two cohorts. Second, each SGLT2i user was assigned an index date corresponding to their actual first purchase. The matched patient from the control group was then assigned the same index date, with the baseline characteristics updated accordingly. Patients who did not meet the eligibility criteria when applying the updated index dates based on the actual date of drug prescription were excluded, along with their matched pair, from further analysis. Finally, patients with a previous episode of DKA were excluded from the analysis.
2.4. Definition of variables
All data were captured from the CHS computerized database. Diabetes duration was considered as a categorical variable. Laboratory data were taken as the value most prior to the index date (up to 3 years prior to the index date). Estimated GFR (eGFR) value was calculated by the Chronic Kidney Disease Epidemiology Collaboration (CKD‐EPI) equation. Baseline medications were defined as active medications in the 6‐month period prior to the index date. A prior diagnosis of DKA was identified according to diagnosis codes from hospital, emergency room (ER) or community medical records. For the outcome analysis, DKA event was defined as any urgent ER visit or hospital admission due to DKA. Patient outcomes were censored at end of membership at CHS, death or December 31, 2023. Definitions of covariates based on diagnosis codes are provided in Table S1.
2.5. Statistical analyses
Outcomes were analysed based on SGLT2i exposure. The primary endpoint was the occurrence of the first DKA event and was analysed by subgroups of interest. Baseline characteristics are reported as frequencies and percentages for categorical variables, and as mean and standard deviation (SD) or median and interquartile range (IQR) for continuous variables. Differences between groups were calculated by t test and by chi‐square as appropriate. p‐Value <0.05 and standard mean difference (SMD) ≥0.2 were considered significant.
Time‐to‐DKA was analysed using Kaplan–Meier curves and compared between groups with bivariate and multivariate Cox proportional hazards models. During propensity score–matched analyses, we identified a significant imbalance in the prevalence of prior DKA events, with a higher proportion among patients who did not initiate SGLT2i. Given the strong association between prior DKA and future risk, we excluded all individuals with documented DKA prior to the index date. Subsequent analyses were restricted to individuals without prior DKA. Inverse probability of treatment weighting (IPTW) was applied to both the bivariate and multivariate Cox models to balance the groups based on baseline characteristics. Propensity scores were estimated using logistic regression, incorporating relevant demographic and clinical covariates. Variables that remained imbalanced (p < 0.05 and SMD ≥ 0.2) and others with modest imbalance (SMD ≥ 0.1) but clinical relevance—such as insulin use, BMI, HbA1C and Charlson comorbidity index—were included in the IPTW model. Covariate balance after weighting was assessed using SMDs.
Multivariable Cox models were constructed to evaluate the association between SGLT2i use and incident DKA, with variable selection guided by clinical relevance and prior literature. Subgroup analyses were conducted for age, sex, BMI, diabetes duration, eGFR, UACR categories and baseline cardiovascular disease. To test for heterogeneity of effect, an interaction term was included in the Cox regression model.
Statistical analyses were conducted SAS Enterprise Guide 8.3.
3. RESULTS
3.1. Baseline characteristics
After applying inclusion and exclusion criteria and performing matching, 6732 patients remained in each of the two cohorts of SGLT2i users and non‐users. Exclusion of patients with prior episodes of DKA resulted in 6572 patients in the SGLT2i group and 6382 patients in the non‐user group (Figure S1).
Baseline characteristics are presented in Table 1. The mean age of the cohort was 53.3 ± 8.3 years, BMI 26.0 ± 2.8 kg/m2 and HbA1c 9.0% ± 1.6%. SGLT2i users were more likely to have a history of previous MI; accordingly, anti‐platelet drugs, statins and other cardiovascular and blood pressure lowering agents were more prevalent in the SGLT2i group. Additionally, SGLT2i users were more likely to be using GLP1‐RA and DPP‐4 inhibitors.
TABLE 1.
Baseline characteristics.
| SGLT2i users (N = 6572) | SGLT2i non‐users (N = 6382) | SMD | SMD post‐IPTW | |
|---|---|---|---|---|
| Demographic characteristics | ||||
| Age, years | 53.1 ± 8.3 | 53.5 ± 8.3 | −0.04 | −0.13 |
| Sex | ||||
| Female | 2354 (35.8) | 2263 (35.5) | 0.01 | −0.03 |
| Male | 4218 (64.2) | 4119 (64.5) | ||
| Socioeconomic level | ||||
| High | 707 (10.8) | 560 (8.8) | 0.21 | 0.2 |
| Medium | 3701 (56.3) | 3085 (48.3) | ||
| Low | 1662 (25.3) | 2130 (33.4) | ||
| No data | 502 (7.6) | 607 (9.5) | ||
| BMI, kg/m2 | 26.2 ± 2.8 | 25.6 ± 2.9 | 0.19 | 0.01 |
| BMI ≤25 kg/m2 | 2121 (32.3) | 2522 (39.5) | 0.15 | −0.01 |
| BMI ≤23 kg/m2 | 792 (12.1) | 1089 (17.1) | 0.14 | 0.00 |
| Smoking | ||||
| Current | 1881 (28.6) | 2014 (31.6) | 0.18 | 0.06 |
| Never | 3456 (52.6) | 3382 (53.0) | ||
| Past | 1226 (18.7) | 947 (14.8) | ||
| No data | 9 (0.1) | 39 (0.6) | ||
| Medical history | ||||
| Age at diabetes diagnosis, years | 41.3 ± 6.7 | 41.8 ± 6.7 | −0.07 | −0.11 |
| Diabetes duration, years | ||||
| <5 | 768 (11.7) | 714 (11.2) | 0.2 | 0.10 |
| ≥5–10 | 1453 (22.1) | 1425 (22.3) | ||
| ≥10–20 | 4124 (62.8) | 3934 (61.6) | ||
| ≥20 | 214 (3.3) | 206 (3.2) | ||
| No data | 13 (0.2) | 103 (1.6) | ||
| Charlson comorbidities score | 3.4 ± 2.1 | 3.3 ± 2.1 | 0.05 | −0.01 |
| Hypertension | 6664 (50.7) | 5506 (43.1) | 0.15 | −0.03 |
| History of cerebrovascular disease | 783 (11.9) | 857 (13.4) | −0.05 | −0.06 |
| History of congestive heart failure | 416 (6.3) | 253 (4.0) | 0.11 | 0.00 |
| History of myocardial infarction | 908 (13.8) | 426 (6.7) | 0.24 | 0.16 |
| History of peripheral vascular disease | 309 (4.7) | 268 (4.2) | 0.02 | 0.03 |
| Cardiovascular drugs used | ||||
| Anti‐platelet drugs | 3086 (47.0) | 2093 (32.8) | 0.29 | 0.10 |
| Statins | 4401 (67.0) | 3421 (53.6) | 0.28 | 0.1 |
| ACE inhibitors or ARBs | 3526 (53.7) | 2543 (39.8) | 0.28 | −0.02 |
| Beta blockers | 1667 (25.4) | 870 (13.6) | 0.3 | −0.04 |
| Calcium channel blockers | 934 (14.2) | 655 (10.3) | 0.12 | −0.01 |
| Thiazides | 119 (1.8) | 106 (1.7) | 0.01 | −0.03 |
| Loop diuretics | 714 (10.9) | 507 (7.9) | 0.1 | 0.05 |
| Glucose‐lowering drugs used | ||||
| Any insulin | 3846 (58.5) | 3549 (55.6) | 0.06 | −0.01 |
| Long‐acting insulin | 3711 (56.5) | 3202 (50.2) | 0.13 | 0.03 |
| Short‐acting insulin | 1661 (25.3) | 1682 (26.4) | −0.02 | 0.02 |
| Sulphonylureas | 1606 (24.4) | 1396 (21.9) | 0.06 | 0.00 |
| GLP‐1 receptor agonists | 1017 (15.5) | 423 (6.6) | 0.29 | −0.03 |
| DPP‐4 inhibitors | 3011 (45.8) | 1020 (16.0) | 0.68 | −0.03 |
| Laboratory and clinical measurements | ||||
| HbA1c (%) | 9.0 ± 1.6 | 8.8 ± 1.7 | 0.15 | 0.00 |
| HbA1c >9 (%) | 2837 (43.2) | 2396 (37.5) | 0.1 | −0.02 |
| eGFR, mL/min/1.73 m2 | 99.9 ± 16.1 | 100.3 ± 16.4 | −0.02 | 0.02 |
| eGFR categories | ||||
| ≥90 | 5248 (79.9) | 5188 (81.3) | 0.2 | 0.00 |
| ≥60–90 | 1147 (17.5) | 995 (15.6) | ||
| ≥45–60 | 145 (2.2) | 133 (2.1) | ||
| ≥30–45 | 32 (0.5) | 63 (1.0) | ||
| Urinary albumin creatinine ratio, mg/g | 110.8 ± 385.4 | 98.0 ± 365.6 | 0.03 | 0.00 |
| Urinary albumin creatinine ratio categories | ||||
| <30 | 4191 (63.8) | 4175 (65.4) | 0.2 | 0.04 |
| ≥30–300 | 1651 (25.1) | 1357 (21.3) | ||
| ≥300 | 514 (7.8) | 416 (6.5) | ||
| No data | 216 (3.3) | 434 (6.8) | ||
| Lipids (mg/dL) | ||||
| HDL cholesterol | 41.9 ± 11.3 | 44.5 ± 12.7 | −0.22 | −0.11 |
| LDL cholesterol | 92.9 ± 36.9 | 97.7 ± 36.3 | −0.13 | −0.02 |
| Triglycerides | 198.2 ± 188.2 | 171.4 ± 161.5 | 0.15 | 0.09 |
| Non‐HDL cholesterol | 129.6 ± 46.4 | 130.1 ± 44.8 | −0.01 | 0.09 |
| Blood pressure (mmHg) | ||||
| Systolic | 126.6 ± 14.5 | 126.1 ± 14.7 | 0.03 | −0.03 |
| Diastolic | 75.6 ± 9.3 | 74.9 ± 9.1 | 0.07 | 0.03 |
3.2. Study outcomes
The median follow‐up of the whole group was 4.4 years (IQR 2.7–6.4); 4.5 years (IQR 2.8–6.4) for those using SGLT2i and 4.3 years (IQR 2.6–6.3) for non‐users. During the follow‐up period, 239 patients experienced an episode of DKA, including 143 (2.22%) SGLT2i users and 96 (1.54%) non‐users, (HR [95% CI] 1.39 [1.07–1.79]; p = 0.014) (Figure 1A). A higher HbA1c was associated with an increased incidence of DKA in both SGLT2i users and non‐users (Figure 1B).
FIGURE 1.
Cox proportional hazards model showing the risk of developing diabetic ketoacidosis by (A) use of sodium‐glucose cotransporter 2 inhibitor (SGLT2i) and (B) by use of SGLT2i and HbA1c level at baseline.
The increased risk of DKA associated with SGLT2i use was observed across all subgroups compared to non‐users (Figure 2). The frequency of DKA was increased in those with higher HbA1c in both SGLT2 users and non‐users (3.3% and 1.8% in patients with HbA1c >10% using and not using SGLT2i, respectively, vs. 2.1% and 1.3% in patients with HbA1c ≤10% using and not using SGLT2i respectively). Similarly, a lower BMI was associated with a higher proportion of DKA compared to a higher BMI.
FIGURE 2.
Bivariate Cox proportional models showing the effect of sodium‐glucose cotransporter 2 inhibitor (SGLT2i) on the incidence of diabetic ketoacidosis within each subgroup of interest. Hazard ratios and 95% confidence intervals are shown. BMI, body mass index; eGFR, estimated glomerular filtration rate; UACR, urinary albumin creatinine ratio.
3.3. Predictors of DKA
We further studied the independent contribution of SGLT2i, insulin, BMI and HbA1c to the development of DKA. In an IPTW adjusted model, SGLT2i remained independently associated with an increased risk of DKA (HR [95% CI] 1.50 [1.15–1.95]; Table 2). Baseline HbA1c >9% was independently associated with a 53% excess risk of DKA (HR 1.53 [95% CI: 1.18–1.99]; p = 0.0016), while a BMI ≤25 was associated with a 61% increased risk (HR 1.61 [95% CI: 1.24–2.09]; p = 0.0003). Using a BMI cutoff of 23 kg/m2, patients with a BMI ≤23 had twice the risk of DKA compared to those with a BMI >23 (Table S2). When assessed as continuous variables, both lower BMI and higher HbA1c were linearly associated with greater DKA risk (Table S3).
TABLE 2.
Multivariate cox regression analysis of predictors of DKA.
| Variable | aHR (95% CI) | p‐Value |
|---|---|---|
| Use of SGLT2 inhibitors | 1.50 (1.15–1.95) | 0.003 |
| Male sex | 0.86 (0.66–1.13) | 0.29 |
| Diabetes age at diagnosis | 1.00 (0.97–1.03) | 0.86 |
| BMI ≤25 kg/m2 | 1.61 (1.24–2.09) | 0.0003 |
| HbA1c >9% | 1.53 (1.18–1.99) | 0.0016 |
| Age | 0.98 (0.95–1.01) | 0.86 |
| Use of insulin | 2.35 (1.71–3.23) | <0.0001 |
| UACR <30 mg/g | 1.00 | |
| UACR 30–299 mg/g | 0.92 (0.67–1.27) | 0.62 |
| UACR category ≥300 mg/g | 0.94 (0.56–1.60) | 0.82 |
Note: Multivariate Cox proportional hazard model with inverse probability of treatment weighting.
Abbreviations: aHR, adjusted hazard ratio; BMI, body mass index; DKA, diabetic ketoacidosis; HbA1c, haemoglobin A1c; SGLT2, sodium‐glucose cotransporter 2; UACR, urine albumin‐creatinine ratio.
4. DISCUSSION
Our study demonstrates an increased risk of DKA in patients with an insulin‐deficient phenotype treated with SGLT2i. This increased risk was observed regardless of age, sex, BMI, HbA1c, diabetes duration, insulin use, kidney function or cardiovascular disease. Notably, DKA frequency was higher among both SGLT2i users and non‐users with lower BMI and higher HbA1c levels. After adjusting for these and other covariates, SGLT2i use remained an independent predictor of DKA. Additionally, low BMI, elevated HbA1c and insulin use emerged as independent risk factors, with a 62% increased risk for patients with a BMI below 25 kg/m2 and a 54% higher risk for those with an HbA1c above 9%.
The increased risk of DKA associated with SGLT2i can be attributed to their unique mechanism of action. By promoting urinary glucose excretion, SGLT2i reduce plasma glucose levels independently of insulin. However, this insulin‐independent glycaemic control is accompanied by increased glucagon secretion and enhanced free fatty acid mobilization and oxidation, which predispose patients to ketogenesis. 20 , 21 Additionally, SGLT2i‐induced glycosuria may reduce the need for exogenous insulin, leading to lower insulin levels and further tipping the balance towards ketone body production. This mechanism is particularly concerning as it can result in euglycaemic DKA, a condition in which patients present with significant ketosis and acidosis but without the extreme hyperglycaemia traditionally associated with DKA. This atypical presentation may delay recognition and treatment by patients and clinicians alike.
Our findings are consistent with a meta‐analysis of phase three RCTs, which reported a 3.5‐fold increased risk of DKA with SGLT2i use compared to placebo in patients with T1DM (11). While these trials established a clear association, they provided limited insight into specific risk factors that could guide clinical decision‐making. Our study addresses this gap by identifying higher HbA1c and lower BMI as key independent risk factors for DKA. We specifically examined individuals with an insulin‐deficient phenotype and found that all are at increased risk for DKA with SGLT2i use. However, within this insulin‐deficient population, the absolute risk was highest among patients treated with insulin and those with elevated HbA1c or low BMI.
To mitigate the recognized DKA risk with SGLT2i, several protocols have been developed to guide their safe use. 22 , 23 , 24 These recommendations emphasize individualized care, including the use of lower SGLT2i doses to minimize ketogenesis, avoiding substantial reductions in total daily insulin dose to prevent ketosis and refraining from excessive alcohol consumption or low carbohydrate/ketogenic diets that may exacerbate ketone body production. The protocol also highlights the importance of routine blood ketone monitoring for early detection of ketosis and includes a sick‐day management strategy to address potential DKA triggers during periods of illness or stress. While promising, these strategies remain largely untested in real‐world settings. Prospective studies are needed to validate their effectiveness. Moreover, educating patients about euglycaemic DKA and the importance of ketone monitoring is essential to ensure timely intervention and prevent severe outcomes.
The DECLARE‐TIMI 58 study included 17 160 patients with type 2 diabetes and established or high risk for cardiovascular disease randomized to dapagliflozin or placebo and followed for a median of 4.2 years. During follow‐up, 39 patients suffered of DKA, 27 in the dapagliflozin group and 12 in the placebo group. Long standing diabetes, higher HbA1c and baseline insulin use were risk factors for the development of DKA. Interestingly, C‐peptide levels at baseline were a strong predictor of DKA in both treatment arms with no DKA events occurring in patients with baseline C‐peptide >0.97 nmol/L. Thus, testing C‐peptide in patients with a phenotype of insulin deficiency may assist in clinical decision‐making prior to prescribing an SGLT2i in these patients. 25
The strengths of our study include its large sample size and real‐world population which enhance the generalizability of our findings, and the inclusion of a phenotypically insulin deficient population, providing actionable insights for clinicians. However, some limitations of our study should be acknowledged. As an observational study, it cannot establish causality, and residual confounding may influence the results. Although our design and analytic approach aimed to minimize time‐related bias, some residual bias may remain. Additionally, the retrospective design precluded access to detailed data on insulin dosing or dose adjustments, which may have affected the outcomes. Moreover, due to the lack of routine measurements of insulin or c‐peptide in our population, as well as lack of anti‐GAD or other T1DM‐related antibodies testing in our cohort, these data were not shown. Patients in our study were selected due to their clinical phenotype indicating a high probability of insulin deficiency—yet there is no biochemical data to support this observation.
In addition, there were no standardized protocols for DKA monitoring or prevention during the study period, potentially introducing variability in event detection. We lacked access to glucose, ketone or pH values at the time of hospital admission and DKA events were identified solely based on diagnostic codes. Although these codes are reviewed for accuracy by trained personnel after hospital discharge, this limitation may have led to under‐ascertainment or misclassification, particularly of euglycaemic DKA cases, which are more common among SGLT2i users and may not present with marked hyperglycaemia. Furthermore, the possibility of detection bias must be considered, as heightened clinical awareness of DKA risk among SGLT2i users could have increased the likelihood of DKA being investigated or diagnosed in this group.
Finally, the absence of an active comparator group is an important limitation. While such a design could have further reduced confounding by indication, it was not feasible in our cohort, as most patients with an insulin‐deficient phenotype who initiated SGLT2i also received DPP‐4 inhibitors. Creating mutually exclusive treatment groups would have substantially reduced the sample size and limited generalizability.
In conclusion, caution should be exercised when prescribing SGLT2i to patients with an insulin‐deficient phenotype, particularly those with higher HbA1c and lower BMI. While these agents offer significant benefits in glycaemic management, their associated DKA risk necessitates a thoughtful, evidence‐based approach to their use. Future research should focus on validating risk mitigation strategies and optimising patient education to ensure the safe integration of SGLT2i into clinical practice.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
PEER REVIEW
The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/dom.16545.
Supporting information
Data S1. Supporting information.
ACKNOWLEDGEMENTS
The authors have nothing to report.
Tsur A, Leibowitz G, Cohen MJ, Cahn A, Pollack R. Predictors of diabetic ketoacidosis in patients with insulin‐deficient diabetes phenotype initiating SGLT2 inhibitors. Diabetes Obes Metab. 2025;27(9):4989‐4997. doi: 10.1111/dom.16545
Avivit Cahn and Rena Pollack contributed equally to this study.
Funding information This research received no specific grant from any funding agency in the public, commercial or not‐for‐profit sectors.
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
REFERENCES
- 1. Zelniker TA, Wiviott SD, Raz I, 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. [DOI] [PubMed] [Google Scholar]
- 2. Fitchett D, Zinman B, Wanner C, et al. Heart failure outcomes with empagliflozin in patients with type 2 diabetes at high cardiovascular risk: results of the EMPA‐REG OUTCOME(R) trial. Eur Heart J. 2016;37(19):1526‐1534. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Heerspink HJL, Stefansson BV, Correa‐Rotter R, et al. Dapagliflozin in patients with chronic kidney disease. N Engl J Med. 2020;383(15):1436‐1446. [DOI] [PubMed] [Google Scholar]
- 4. Jhund PS, Solomon SD, Docherty KF, et al. Efficacy of dapagliflozin on renal function and outcomes in patients with heart failure with reduced ejection fraction: results of DAPA‐HF. Circulation. 2021;143(4):298‐309. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Perkovic V, Jardine MJ, Neal B, et al. Canagliflozin and renal outcomes in type 2 diabetes and nephropathy. N Engl J Med. 2019;380(24):2295‐2306. [DOI] [PubMed] [Google Scholar]
- 6. Neal B, Perkovic V, Mahaffey KW, et al. Canagliflozin and cardiovascular and renal events in type 2 diabetes. N Engl J Med. 2017;377(7):644‐657. [DOI] [PubMed] [Google Scholar]
- 7. The E‐KCG , Herrington WG, Staplin N, et al. Empagliflozin in patients with chronic kidney disease. N Engl J Med. 2023;388(2):117‐127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Dandona P, Mathieu C, Phillip M, et al. Efficacy and safety of dapagliflozin in patients with inadequately controlled type 1 diabetes: the DEPICT‐1 52‐week study. Diabetes Care. 2018;41(12):2552‐2559. [DOI] [PubMed] [Google Scholar]
- 9. Mathieu C, Dandona P, Gillard P, et al. Efficacy and safety of Dapagliflozin in patients with inadequately controlled type 1 diabetes (the DEPICT‐2 study): 24‐week results from a randomized controlled trial. Diabetes Care. 2018;41(9):1938‐1946. [DOI] [PubMed] [Google Scholar]
- 10. Rosenstock J, Marquard J, Laffel LM, et al. Empagliflozin as adjunctive to insulin therapy in type 1 diabetes: the EASE trials. Diabetes Care. 2018;41(12):2560‐2569. [DOI] [PubMed] [Google Scholar]
- 11. Garg SK, Henry RR, Banks P, et al. Effects of sotagliflozin added to insulin in patients with type 1 diabetes. N Engl J Med. 2017;377(24):2337‐2348. [DOI] [PubMed] [Google Scholar]
- 12. Dandona P, Mathieu C, Phillip M, et al. Efficacy and safety of dapagliflozin in patients with inadequately controlled type 1 diabetes (DEPICT‐1): 24 week results from a multicentre, double‐blind, phase 3, randomised controlled trial. Lancet Diabetes Endocrinol. 2017;5(11):864‐876. [DOI] [PubMed] [Google Scholar]
- 13. Henry RR, Thakkar P, Tong C, Polidori D, Alba M. Efficacy and safety of canagliflozin, a sodium‐glucose cotransporter 2 inhibitor, as add‐on to insulin in patients with type 1 diabetes. Diabetes Care. 2015;38(12):2258‐2265. [DOI] [PubMed] [Google Scholar]
- 14. Buse JB, Garg SK, Rosenstock J, et al. Sotagliflozin in combination with optimized insulin therapy in adults with type 1 diabetes: the North American inTandem1 study. Diabetes Care. 2018;41(9):1970‐1980. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Danne T, Cariou B, Banks P, et al. HbA(1c) and hypoglycemia reductions at 24 and 52 weeks with sotagliflozin in combination with insulin in adults with type 1 diabetes: the European inTandem2 study. Diabetes Care. 2018;41(9):1981‐1990. [DOI] [PubMed] [Google Scholar]
- 16. Taylor SI, Blau JE, Rother KI, Beitelshees AL. SGLT2 inhibitors as adjunctive therapy for type 1 diabetes: balancing benefits and risks. Lancet Diabetes Endocrinol. 2019;7(12):949‐958. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Fralick M, Redelmeier DA, Patorno E, et al. Identifying risk factors for diabetic ketoacidosis associated with SGLT2 inhibitors: a nationwide cohort study in the USA. J Gen Intern Med. 2021;36(9):2601‐2607. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Dennis JM, Shields BM, Henley WE, Jones AG, Hattersley AT. Disease progression and treatment response in data‐driven subgroups of type 2 diabetes compared with models based on simple clinical features: an analysis using clinical trial data. Lancet Diabetes Endocrinol. 2019;7(6):442‐451. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Ahlqvist E, Storm P, Karajamaki A, et al. Novel subgroups of adult‐onset diabetes and their association with outcomes: a data‐driven cluster analysis of six variables. Lancet Diabetes Endocrinol. 2018;6(5):361‐369. [DOI] [PubMed] [Google Scholar]
- 20. Ferrannini E, Muscelli E, Frascerra S, et al. Metabolic response to sodium‐glucose cotransporter 2 inhibition in type 2 diabetic patients. J Clin Invest. 2014;124(2):499‐508. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Rosenstock J, Ferrannini E. Euglycemic diabetic ketoacidosis: a predictable, detectable, and preventable safety concern with SGLT2 inhibitors. Diabetes Care. 2015;38(9):1638‐1642. [DOI] [PubMed] [Google Scholar]
- 22. Goldenberg RM, Gilbert JD, Hramiak IM, Woo VC, Zinman B. Sodium‐glucose co‐transporter inhibitors, their role in type 1 diabetes treatment and a risk mitigation strategy for preventing diabetic ketoacidosis: the STOP DKA protocol. Diabetes Obes Metab. 2019;21(10):2192‐2202. [DOI] [PubMed] [Google Scholar]
- 23. Danne T, Garg S, Peters AL, et al. International consensus on risk management of diabetic ketoacidosis in patients with type 1 diabetes treated with sodium‐glucose cotransporter (SGLT) inhibitors. Diabetes Care. 2019;42(6):1147‐1154. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Ehrmann D, Kulzer B, Roos T, Haak T, Al‐Khatib M, Hermanns N. Risk factors and prevention strategies for diabetic ketoacidosis in people with established type 1 diabetes. Lancet Diabetes Endocrinol. 2020;8(5):436‐446. [DOI] [PubMed] [Google Scholar]
- 25. Kang YM, Melloni GEM, Cahn A, et al. Clinical features and C‐peptide levels associated with diabetic ketoacidosis in patients with type 2 diabetes on dapagliflozin vs. placebo—insights from DECLARE‐TIMI 58. Diabetes. 2024;73(Supplement_1):1896‐LB. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. Supporting information.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.