Association Between SGLT2 Inhibitor Treatment and Diabetic Ketoacidosis and Mortality in People With Type 2 Diabetes Admitted to Hospital With COVID-19

Increasing age, socioeconomic deprivation, male sex, chronic diseases including diabetes, and non-White ethnicity have been associated with worse outcomes from COVID-19 (1). The risk of hospital-related COVID-19 mortality is increased threefold with type 1 diabetes and twofold with type 2 diabetes (T2D) (2). Diabetic ketoacidosis (DKA) has been reported with COVID-19 and is associated with a 50% increased risk of mortality (3). DKA is a recognized life-threatening consequence of COVID-19 in those with type 1 diabetes. It is most frequently encountered in people with type 1 diabetes but can also occur in people with T2D, typically during an acute illness, such as sepsis, myocardial infarction, or stroke; surgery and other stressors; use of glucocorticoids; alcohol consumption; or reductions in calorie intake (4).

Sodium glucose cotransporter 2 inhibitors (SGLT2is) reduced risk of cardiovascular and kidney events in large randomized trials of people with T2D (5). However, the U.S. Food and Drug Administration published a safety update in 2015 about the association between SGLT2i use and risk of DKA (6). During the pandemic, a greater proportion of people presenting with DKA have had T2D than before the pandemic (7). Recent consensus statements have recommended that SGLT2is are safe for use in routine clinical care during the COVID-19 pandemic but that SGLT2is should be stopped in symptomatic individuals infected with SARS-CoV-2 who might be at risk for euglycemic DKA, particularly if admitted to hospital (8).

A number of observational studies have reported that the use of SGLT2-inhibitor therapies are not associated with worse outcomes in people with COVID-19 (9,10). A recent population-based study in England of nearly 3 million people reported an 18% lower risk of COVID-19–related mortality in people with T2D prescribed SGLT2is, compared with people prescribed other glucose-lowering therapies, although the authors concluded that their findings were likely to be confounded by indication for SGLT2 prescribing (11). In addition, the recent results from the Dapagliflozin in Respiratory Failure in Patients With COVID-19 (DARE-19) randomized controlled trial of dapagliflozin, SGLT2 inhibitor, among people with at least one cardiometabolic risk factor (i.e., hypertension, T2D [present in 51% of the population], atherosclerotic cardiovascular disease, heart failure, and/or chronic kidney disease) demonstrated that the incidence of DKA in severely unwell patients admitted with COVID-19 was very low, with only two cases among a total of 312 participants with history of T2D randomized to treatment with dapagliflozin (12). These cases were identified through protocol-mandated laboratory monitoring and both events resolved rapidly and completely. There were also no other safety signals and no statistically significant association between dapaglifozin and mortality (for the whole population, hazard ratio 0.77; 95% CI 0.53–1.16). However, the incidence of DKA and mortality in people with T2D who were prescribed SGLT2i prior to hospital admission with COVID-19 has not been well documented. The aim of this study was secondary analysis of data from a large, robust national audit to determine the association between prior prescription of SGLT2is and DKA incidence or mortality in people with T2D hospitalized with COVID-19.

Data were collected through a nationwide audit from a network of 40 centers in the U.K. conducted by the Association of British Clinical Diabetologists (ABCD) (13). The ABCD has a long-established infrastructure for nationwide audits to allow diabetes specialist teams across the U.K. to collect real-world data on specific therapies and technologies (13). The COVID-19 and diabetes nationwide audit commenced in September 2020 and data were collected up to 8 December 2020. Centers were asked to collate data from patient records from the start of the pandemic in March 2020 and to transfer the anonymized data to the National Institute for Health Research (NIHR) Health Informatics Collaborative Coordinating Centre within the Oxford University Hospitals National Health Service (NHS) Foundation Trust. Diabetes specialist team members at each participating hospital identified patients with diabetes admitted with COVID-19 and positive SARS-CoV-2 test. An audit data collection form was provided to each participating center, where pseudonymized data were held on secure servers.

Data were transferred securely using the National Health Service (NHS) network. Submissions were checked by the NIHR Health Informatics Collaborative team and additional information was sought from contributing centers where necessary to ensure completeness and accuracy. The cleaned data were locked on an MS SQL server and made available for analysis by a Labkey portal. The data were processed and analyzed on a secure server at Oxford University Hospital (OUH). Cases of DKA were identified by diabetes specialists at each center. The primary analysis included every case of contributor-identified DKA, whether occurring on admission to hospital, or subsequently during inpatient treatment. Adjudication of DKA cases by two experienced clinicians (Fig. 1) and a sensitivity analysis were also performed, including only those cases in which biochemistry results on admission to hospital met the diagnostic criteria of the Joint British Diabetes Societies Inpatient Group guideline on diagnosis and management of DKA (14).

The audit was registered with the OUH, and a Data Protection Impact Assessment was carried out and approved by the OUH Caldicott Guardian and the Public Benefit and Privacy Panel in Scotland (reference no. 2021–0111); there was no requirement for approval by a research ethics committee. ABCD audits have a long track record of supporting clinical practice in the NHS, which itself supports such work with clear guidance for contributing centers on the secure use of routine health care data. The ABCD audit remains open and the original data collection sheet can be obtained from the ABCD secretariat.

We collected demographic and clinical data. Demographic data included age (in years) at hospital admission, sex, ethnicity, and Index of Multiple Deprivation decile, an area-based measure of socioeconomic status derived using the postcode (U.S. zip code equivalent) for place of residence. Clinical characteristics included weight and height, or BMI; smoking status; type and duration of diabetes; diabetes complications, including diabetic foot ulcer, diabetic nephropathy, diabetic peripheral neuropathy, diabetic retinopathy, peripheral vascular disease, ischemic heart disease (namely, myocardial infarction and/or heart failure), cerebrovascular disease (i.e., stroke or transient ischemic attack), and other significant comorbidities, including hypertension, dementia, asthma, chronic obstructive pulmonary disease, and malignant neoplasm. Medication history included antidiabetic medications and other prespecified therapy classes (namely, angiotensin-converting enzyme inhibitor or angiotensin-receptor antagonist, oral corticosteroid, statin, antiplatelet, anticoagulant, and regular nonsteroidal antiinflammatory drug). Laboratory data included latest preadmission HbA_1c and serum creatinine level, and admission blood glucose, pH, bicarbonate, lactate, serum creatinine, and capillary blood ketone levels. Dates of the start and finish (if applicable) of each hospital admission were collected, along with the date of positive SARS-CoV-2 test, which was a prerequisite for inclusion in the study. Recorded outcomes included vital status and admission to an intensive care unit (ICU). For this analysis, we only included people in the audit with T2D.

Baseline clinical characteristics are reported as frequency and percentages for categorical variables, and as mean and SD for continuous variables by exposure status, defined as record or absence of record of SGLT2i use prior to hospital admission. The primary outcome of the study was DKA. The secondary outcomes were mortality in the whole cohort, in those with or without insulin administration, and in those with DKA. Crude, age- and sex-adjusted, and multivariable logistic regression models were used to generate odds ratios (ORs) and 95% CIs. For the logistic regression analysis, continuous variables, including age and admission blood glucose value, were used to generate ORs. Binary categorical variables were created to define sex, ethnicity (White or not), micro- and macrovascular complications, DKA status, mortality, SGLT2i prescription, and insulin prescription. Sample-size calculations were not performed, because this was an exploratory analysis of data from a large nationwide audit that was originally designed to describe risk factors associated with adverse outcomes among people with diabetes who were admitted to hospital with COVID-19.

All statistical analysis was performed using R, version 3.3. P < 0.05 was considered statistically significant.

The audit included 3,067 people with T2D who were admitted to hospital with COVID-19 and who had complete data. Of these, 230 (7.5%) were prescribed SGLT2is (Table 1). The mean age of the cohort was 72 years, 62.3% were men, and 34.9% (n = 994 of 2,845) were prescribed insulin. SGLT2is were more likely to be prescribed to male patients (73%) than to female patients (61%), to patients who were using insulin (49% vs. 34% not using insulin), people with diabetic foot ulcers (20% vs. 7% without), those with diabetic nephropathy (40% vs. 23% without), with peripheral vascular disease (28% vs. 11% without), with retinopathy (40% vs. 23% without), and to those with cardiovascular disease (47% vs. 36% without). Overall, 86 of the 3,067 people in the study with T2D (2.8%) had DKA and 1,082 of 3,039 people (35.6%) died. Table 2 shows the ORs (95% CIs) for DKA and for death before and after adjusting for age, sex, ethnicity, admission blood glucose level, insulin administration, and micro- or macrovascular disease in people prescribed SGLT2is compared with those not prescribed SGLT2is. The odds of DKA did not differ significantly different between those prescribed SGLT2is and those not (adjusted OR 0.56; 95% CI 0.16–1.97). The odds of mortality associated with SGLT2is were similar in the total study population (OR 1.13; 95% CI 0.78–1.63), in the subgroup prescribed insulin (OR 1.02; 95% CI 0.59–1.77), and in the subgroup in which DKA developed (OR 0.21; 95% CI 0.01–8.76). In a sensitivity analysis including only those cases in which admission biochemistry met the Joint British Diabetes Societies Inpatient Group diagnostic criteria for DKA (14), the adjusted OR for adjudicated DKA associated with a prescription of SGLT2is and adjudicated DKA not so associated was 1.58 (95% CI 0.35–7.08).

To our knowledge, this is the largest multicenter study reporting data on DKA and mortality in people with T2D admitted to hospital with COVID-19 that investigates association with SGLT2is preadmission and suggests that that, despite limited power, there is no evidence for association. Only a few small studies or case studies have recently reported high rates of DKA in people hospitalized with COVID-19, particularly in those with T2D (15).

Inpatient admission has been reported as a risk for SGLT2 inhibitor–associated DKA. In a retrospective, multicenter cohort study from Australia, the risk of DKA in patients with T2D during inpatient admission was small but higher in SGLT2i versus non-SGLT2i users, with planned fasting and surgery identified as potential risk factors (16). There has also been a concern that routine SGLT2-inhibitor treatment could increase COVID-19–related risks through increased kidney expression of angiotensin-converting enzyme 2. One large, retrospective, nationwide database study of nearly 3 million people with T2D in England recently reported that the adjusted hazard ratio for COVID-19–related mortality for people using SGLT2is was 0.82 (95% CI 0.74–0.91) compared with the group not prescribed any glucose-lowering therapies (11). The study did not report DKA outcomes, and the authors concluded that overall differences in outcomes for all glucose-lowering therapies, including SGLT2is, were small and likely to be confounded by indication (11).

In the DARE-19 trial, in which approximately 51% of participants had diabetes, dapagliflozin was well tolerated and, overall, there were fewer serious adverse events in those prescribed dapagliflozin compared with placebo. The DARE-19 trial recruited 312 participants who were randomized to receive dapagliflozin, whereas the present study included 230 patients who were taking SGLT2i at hospital admission. Our cohort was older (mean age, 72 years vs. 61 years in the DARE-19 trial), with greater proportion of non-White ethnicity (31% vs. 26% in the DARE-19 trial), a similar BMI (31 kg/m²), but higher prevalence of cardiovascular disease (47% vs. 17%) and of nephropathy (33% vs. 17% CKD in the DARE-19 trial). Overall, 10.6% of patients in the dapagliflozin arm and 13.3% in the placebo arm were reported to have serious adverse events. Safety events of acute kidney injury were reported in 3.4% of patients in the dapagliflozin group and 5.5% in the placebo group. Participants in the DARE-19 trial were closely monitored for adverse events, whereas patients in our study would have had routine hospital monitoring per local protocols. Despite the cohort of patients in our study being at higher risk (e.g., age, ethnicity, and co-morbidities) compared with the participants in the DARE-19 trial, we did not observe a significant increase in incidence of DKA in our study. In view of the DARE-19 trial results suggesting dapagliflozin may reduce organ failure and death among high-risk individuals admitted with COVID-19, investigators in the U.K. recently announced an empagliflozin treatment arm in the Randomized Evaluation of COVID-19 Therapy (RECOVERY) platform trial (17), and the U.S. National Institutes of Health have added SGLT2is to the Accelerating COVID-19 Therapeutic Interventions and Vaccines 4 ACUTE (Activ4a) pragmatic trial platform, which is evaluating promising treatments in patients hospitalized with COVID-19 (18). Subsequently, findings of the Empagliflozin in Patients Hospitalized for Acute Heart Failure (EMPULSE) double-blind trial were published; in that study, 530 patients with acute de novo or decompensated chronic heart failure, of whom just under half had diabetes, were randomly assigned to empagliflozin 10 mg daily or placebo (19). Initiation of empagliflozin resulted in a statistically significant and clinically meaningful benefit at 90 days with reductions in all deaths, hospitalization for heart failure, and improvements in quality of life (stratified-win ratio 1.36; 95% CI 1.09–1.68), and no cases of DKA in the empagliflozin arm (19).

A smaller proportion of people (7.5%) included in our audit were prescribed SGLT2is, compared with 9.3% of people with T2D in the community who were prescribed them (11). Only 47% of people with cardiovascular disease and 40% with diabetic nephropathy were prescribed SGLT2is, despite international guideline recommendations for prescribing these novel therapies in high-risk populations (20). Interestingly, despite concerns about SGLT2is increasing risk of diabetic foot ulceration, 26% of people in our study with foot ulcers were prescribed SGLT2is (20).

As well as the beneficial effects of SGLT2is for patients admitted to hospital, another key concern is the potential for these cardiovascular and kidney protective therapies to not be initiated following discharge from hospital. One U.S. study of patients with T2D hospitalized after a myocardial infarction showed that approximately half of the patients may not have had their glucose-lowering therapy commenced after discharge from hospital (21).

There are several strengths to our study, besides being the largest multicenter study to report the association between history of SGLT2i use prior to hospital admission with COVID-19 and DKA or mortality among people with T2D. Data from the centers were collected using a structured proforma with admission variables defined specifically for this national COVID-19 audit. Each case of DKA was reviewed by senior clinicians using established criteria (14). Limitations of the study include the opportunistic retrospective analysis of data collected in a rigorously conducted audit with limited power rather than an adequately powered primary study. The data also relate only to patients admitted to hospital, excluding those not reaching, or not requiring, hospital treatment. Although we adjusted our analysis for several key risk factors, there is potential for residual confounding. The admitted patients were a sicker group, as has been shown in previous studies, and we therefore adjusted our analysis using a number of baseline variables associated with a higher risk of mortality.

We also had a number of patients with missing data for some variables. Another key limitation is that we do not know whether SGLT2i treatment was stopped on admission, as recommended by national and international guidelines. Finally, although DKA incidence was ascertained by diabetes specialists in each contributing center, adjudication of biochemical diagnosis was possible when DKA was present on admission, but not when it occurred after admission to hospital.

The safety of SGLT2i use in people admitted with COVID-19 has not been well described during the COVID-19 pandemic. To our knowledge, this is the largest multicenter study to report safety outcomes of DKA and death in people with T2D admitted to hospital with COVID-19 with or without SGLT2i treatment. We describe a low DKA risk and high in-hospital mortality in people with T2D admitted to hospital with COVID-19 overall, but no evidence for an increase in risk of DKA or in-hospital mortality associated with prescription of SGLT2is, albeit on the basis of imprecise estimates. Furthermore, for those prescribed SGLT2is, there was no evidence of safety concerns for people with or without DKA in this population at very high risk of mortality. Previous recommendations to stop SGLT2i use during acute illness, particularly admission even during COVID-19, have been based on consensus. In view of the results of the DARE-19 and EMPULSE trials and of our study, we believe that careful consideration should be given to the risks and benefits of continuing SGLT2i treatment in people admitted to hospital, in view of SGLT2i cardiovascular and kidney benefits. If SGLT2is are continued, there should be careful monitoring for development of DKA. We recommend adequately powered randomized controlled trials and observational studies be conducted to determine the safety of SGLT2is in people acutely admitted to hospital.

Funding. K.K. is supported by the NIHR Applied Research Collaboration East Midlands and the NIHR Leicester Biomedical Research Centre (BRC). RR is supported by the NIHR Oxford BRC.

Duality of Interest. R.R. has acted as a consultant or speaker or received grants from Novo Nordisk, Eli Lilly, and Boehringer Ingelheim. K.K. has acted as a consultant or speaker or received grants for investigator-initiated studies for AstraZeneca, Novartis, Novo Nordisk, Sanofi, Lilly, Merck Sharp & Dohme, Boehringer Ingelheim, Bayer, Berlin-Chemie AG/Menarini Group, Janssen, and Napp; and is chair of the Ethnicity Subgroup of the UK Scientific Advisory Group for Emergencies (SAGE) and member of SAGE. S.H.W. attends meetings of the Scottish Study Group of Diabetes in the Young, which receives support from Novo Nordisk. S.H. has received educational funding support from Sanofi Aventis, and consulting fees from Eli Lilly and Oviva. B.C.T.F. has acted as a consultant or speaker or received grants from Abbott Diabetes, AstraZeneca, Boehringer Ingelheim, Eli Lilly, GSK, Janssen, Medtronic, MSD, Napp, Novo Nordisk, and Sanofi. R.E.J.R. has received speaker fees and/or consultancy fees and/or educational sponsorships from AstraZeneca, BioQuest, GI Dynamics, Janssen, Novo Nordisk, Sanofi, and Takeda. E.G.W. has received personal fees from Abbott Diabetes Care, Dexcom, Diasend, Eli Lilly, Insulet, Medtronic, Novo Nordisk, and Sanofi Aventis. No other potential conflicts of interest relevant to this article were reported.

The views expressed are those of the authors and not necessarily those of the NIHR, the National Health Service, or the Department of Health and Social Care.

Author Contributions. K.K. and Y.R. codesigned the study analysis, carried out the data analysis, and drafted the manuscript. Y.R. had access to all the data. All authors contributed to the interpretation of the results and critical review of the manuscript. K.K. is the guarantor of this work and takes responsibility for the integrity of the data and the accuracy of the data analysis.