Sodium Glucose Co-Transporter 2 Inhibitor Use in Hospitalised People with Type 2 Diabetes: Cohort Analysis of Those Who Developed Diabetic Ketoacidosis Compared to Those Who Did Not

1. Background

Sodium-glucose cotransporter-2 inhibitors (SGLT2i), among the newer anti-hyperglycaemic agents, inhibit renal tubular glucose reabsorption thus promoting urinary glucose excretion and improved glucose control (1). They also powerfully exhibit cardioprotective properties including reductions in cardiovascular mortality and heart failure admissions (2); and renal protection through slowing chronic kidney disease progression, proteinuria and delaying time to renal replacement therapy (3). This drug class has become the preferred second-line therapy for T2DM, especially with concurrent cardiac and/or renal disease (4). The latter effects are so powerful, the class is now also recommended for people with heart failure and /or chronic kidney disease but without diabetes (5). Given these extensive benefits, prescriptions for SGLT2i have rapidly increased.

However. in people with diabetes us of SGLT2i is also associated with a potentially life-threatening complication of diabetic ketoacidosis (DKA) (6, 7, 8). SGLT2i-induced DKA is known to be precipitated by factors including fasting, surgery, significant intercurrent illnesses, infection, reduced caloric intake, dehydration, and non-adherence with insulin dosages, and typically is seen in an acute inpatient setting among hospitalised people with diabetes (9, 10, 11).

Mechanistically, precipitating factors cause an excessive release of counter-regulatory hormones including glucagon, catecholamines and cortisol, which then increase the ratio of glucagon to insulin, resulting in lipolysis and ketogenesis. This often occurs on the background of enhanced glycosuria induced by SGLT2i and typically coupled with glycogen depletion from prolonged fasting. Collectively these pathways ultimately lead to the development of DKA (12). While much is known about the pathophysiology of SGLT2i-induced DKA, its diagnosis remains challenging due to its relatively low incidence, its relative euglycemia that reduces treating clinicians’ suspicion for ketoacidosis, and its atypical presentation compared to other usual causes of DKA (13, 14).

Furthermore, there remains a paucity of knowledge about the clinical characteristics and demographics of the people who experienced SGLT2i-induced DKA while hospitalised. To date no studies have compared the clinical and hospital profiles of people with diabetes, on SGLT2i and who when hospitalised developed DKA to those with diabetes, admitted to the same institution and on the same class of medication, but who did not experience DKA. Characteristics such as their demographics, diabetes management, reasons for hospital admission including risk factors that may precipitate DKA, the severity of DKA and actions taken by the medical teams to resolve the DKA can help guide risk stratification for DKA in people on SGLT2i admitted to hospitals and allow for more timely and effective diagnosis and management of hospital associated DKA events.

This study aims to investigate these clinical parameters for those who developed SGLT2i induced DKA: patient demographics, clinical characteristics, severity of DKA, and the actions taken by the medical team to resolve the event in inpatients, and to compare these parameters between those on SGLT2i who experienced DKA to those who did not.

2. Methodology

2.1. Study Design

A clinically pragmatic convenience audit was conducted between February 2022 to August 2023 with all cases of SGLT2i-induced DKA, referred to the inpatient Diabetes team at Liverpool Hospital, New South Wales, Australia (among the state’s largest hospitals and a quaternary referral service), placed into an audit database. Note: from July to September 2022 inclusive, due to clinician error, none of the referred cases were recorded into the audit database.

Data of all inpatients prescribed SGLT2i, excluding those who developed SGLT2i DKA over the corresponding period was correspondingly audited (noting pharmacy prescribing records for periods of March, April and October were missing). While not a perfect overlap with the SGLT2i DKA cohort, with national rules and institutional trends of SGLT2i prescribing stable throughout 2022, the SGLT2i DKA was then compared to the all SGLT2i prescription cohort.

Ethics approval was obtained from the SWS Local Health District Human Research Ethics Committee (2022/ETH02606: Prescription and appropriate use of SGLT2 inhibitors in hospital).

2.2. Data Collection

Clinical measures of patient age and gender; clinical diabetes measures encompassing type and duration of diabetes, glycated haemoglobin (HbA1c) levels within three months of admission, and regular diabetes medications was collected from the electronic medical records of all studied patients. Furthermore, data related to their admission characteristics including admission ingress. admitting home team, duration of hospital stay, need for fasting and whether planned or unplanned, involvement of high dependency/ intensive care services, surgery during the admission, and any post admission acute clinical issues / risk factors for SGLT2i-DKA were recorded.

2.3. Definition of SGLT2i DKA

Beyond the Long, et al. (15) and Mathew, et al. (16) diagnostic criteria of SGLT2i-induced DKA, which encompasses a triad of relative euglycemia (BGLs < 14 mmol/L or < 250 mg/dl), increased anion gap metabolic acidosis (arterial blood pH < 7.30 and sodium bicarbonate < 18 mEq/L) and ketosis (measurable ketonemia or ketonuria), the study cohort includes all referrals of hospitalised people with diabetes who developed DKA while taking SGLT2i (that is those whose initial BGLs were > 14mmol/L). For this cohort both euglycemic and hyperglycaemic DKA are collectively referred to as SGLT2i-induced DKA. Further details around the event of initial and worse blood pH, ketone level, BGL, urea, creatinine, bicarbonate, lactate were recorded, as were time from fasting to the detection then and resolution of the event, and actions taken by both the home team and consulting specialist diabetes team to resolve the event were captured. Information regarding discharge disposition including any restart of SGLT2i, documentation about the event in discharge papers and discharge location were recorded.

2.4. Statistical Analyses

Statistical analyses were conducted using Microsoft Excel and Statistical Package for Social Sciences Version 27 (SPSS Inc., Chicago, Illinois, USA). Descriptive statistics were generated and reported for both cohorts using frequencies (percent), mean ± standard deviation and ranges, as appropriate. For direct comparative analyses between the two cohorts, unpaired T-tests or Chi-square tests were conducted for matched normative continuous or categorical variables, respectively. Results with a P < 0.05 were considered statistically significant. Patients with missing data of a particular variable were excluded from that specific statistical analysis.

3. Results

3.1. Characteristics of Patients With and Without SGLT2i-DKA

The characteristics of patients with and without SGLT2i-DKA are shown on Table 1. There were 35 unique cases (54.3% / n=19 male) of SGLT2i-DKA (47% DKA vs 53% eDKA) with mean age 65 ± 13 years. The rate of SGLT2i-DKA over this period was 4.3% (35 out of 819 all SGLT2i prescriptions) None of those experiencing SGLT2i-DKA had recurrent episodes during the audit timeframe. All had T2DM (100%) with mean duration of 12 ± 8 years and mean HbA1c of 9.4 ± 2.3%. Empagliflozin was implicated in 33 (94%) cases and dapagliflozin in the remaining 2 (6%). No case was prescribed more than one SGLT2i agent. Aside from SGLT2i use, all (100%) cases were concurrently on metformin; of other diabetes medications, 15 (43%) were on dipeptidyl peptidase-4 inhibitors, 7 (20%) on sulfonylureas, 13 (37%) on insulin and 3 (9%) were on glucagon-like peptide-1 receptor agonist agents.

Almost 90% of the cohort were acute clinical presentations admitted through the accident & emergency department; of those admitted under surgical or medical teams a majority of SGLT2i—DKA cohort were admitted under surgery teams vs the whole cohort (40% vs 19%, P < 0.001). Sixteen cases (46%) were fasting (for impending surgery or procedures); of these 14 were unplanned and 2 were planned fasting; the latter had with their SGLT2is withheld >=48hrs prior to their scheduled procedure.

Sixteen cases (46%) had intercurrent infection while three had (9%) sepsis during their admissions. Other acute hospital issues that preceded DKA included acute kidney injury, acute cardiovascular or cerebrovascular diseases, new glucocorticoid use and recent surgery (identified in 5, 2, 4, 3 and 2 cases, respectively).

For other demographics of those with and without SGLT2i-DKA, both groups were similar for mean age and their duration of diabetes and entry to hospital (88% via Accident & Emergency both groups).

There was a non-significant trend to higher proportion of females in the SGLT2i-DKA group (45.7% vs 31.2%, P = 0.072). Significant differences were found for prevalence of T2DM between SGLT2i-DKA and SGLT2i all use, (100% vs 80.4%, P = 0.015); SGLT2i- DKA group also had worse glucose control (HbA1c 9.4 ± 2.3% vs 8.4 ± 2.7%, P = 0.013); and higher percentage on empagliflozin compared to all SGLT2i use (94% vs 50%, P < 0.001). Differences were also found in intercurrent sepsis (8.6% vs 2%, P = 0.012), acute cardiovascular disease (5.7% vs 55.4%, P < 0.001), and surgery during the admission (37.1% vs 18.1%, P < 0.001).

3.2. Biochemical and Hospital Stay Characteristics of the SGLT2i- DKA Event

The biochemical and hospitals stay characteristics of SGLT2i-induced DKA cases were illustrated in Table 2. Of the SGLT2i- DKA cohort, mean length of hospital stay until DKA onset was 0.9 days. The initial and worst pH was 7.29 ± 0.11 (range: 7 – 7.44) and 7.08 ± 1.20 (range: 6.92 – 7.42), respectively, confirming at least moderate range acidosis. Ketosis was confirmed with initial ketone level 4.09 ± 1.58 mmol/L (range: 1.5 – 7.2 mmol/L), which increased to 4.43 ± 1.64 mmol/L (range: 1.9 – 7.3 mmol/L) at its peak. The average initial BGL indicated hyperglycaemia, with initial level 16.8 ± 12.0 mmol/L (range: 4.8 – 50 mmol/L) and peak level of 19.5 ± 11.9 mmol/L (range: 4.8 – 50 mmol/L). The mean values of other biochemical parameters are for urea, 8.81 ± 6.55 mmol/L; creatinine, 81.7 ± 43.2 μmol/L; bicarbonate, 16.2 ± 6.6 mmol/L; and lactate, 1.66 ± 0.69 mmol/L.

3.3. Resolution of SGLT2i-DKA and Discharge Outcomes

The actions to resolve the SGLT2i-DKA event and their discharge disposition are shown in Table 3. The mean time from detection to resolution of SGLT2i-DKA was 12.4 hours. Insulin was administered in all but three (9%) cases either as a stat dose, regular injections, or via intravenous route. Other actions such as administrating intravenous fluid, breaking fasting or consulting the ICU team were undertaken.

Regarding post SGLT2i-DKA outcomes, 20% required transfer to high dependency / ICU level care; 20% had their procedures delayed or cancelled; 11% worsened, to more severe range DKA or development of hyperosmolar hyperglycaemic state#, and 5.7% developed new renal injury (reflected in progress biochemistry) and 5.7% had new infections diagnosed. While beyond the scope of this pragmatic convenience audit to review these complications in detail, importantly no patient deaths were recorded among the cohort.

In terms of the discharge actions, 57% of the patients were prescribed new regular insulin on discharge while only 31% had a SGLT2i agent restarted. Nearly 80% of the cohort were scheduled for follow-up appointments with outpatient diabetes and endocrine services. On their discharge summaries while nearly 80% had the SGLT2i-DKA event documented, only 50% of discharge summaries contained specific written guidance to primary care practitioners about any formal decision about ceasing or restarting SGLT2i agents.

4. Discussion

To the authors’ knowledge, this is the first study to investigate differences in the clinical profile between hospitalised patients on SGLT2is who developed SGLT2i-DKA and those who on therapy did not. Within the context of a pragmatic convenience study, conceived and conducted by active clinicians, this was as designed to be as robust as possible, with a well adjudicated and characterised study cohort. Acknowledging limitations of missing and non-contiguous data, the data sets were reviewed by two senior researchers in order to minimise avoidable input error.

4.1. Key Findings

This cohort study analysed 35 cases of SGLT2i-DKA over 12 months. While some biochemical measurements were missing in one case, the key data variables around the SGLT2i-DKA event were recorded for all index cases.

• Firstly, the baseline event rate of SGLT2i-DKA for this institution was 4.3% (35 / 819).
• Secondly despite the wide range of lowest pH (6.92 – 7.42) amongst the SGLT2i-induced DKA group, the lowest pH of some patients reached the range of moderate-to-severe acidosis.
• Thirdly, in our cohort, 47% of the patients had DKA with hyperglycaemia (i.e., initial BGL >14 mmol/L) with many patients had BGLs in the mild to moderately hyperglycaemic range as indicated by a wide range of average initial and peak BGL (range: 4.8 – 50 mmol/L). Only half of the patients had euglycemic DKA
• Fourthly, comparing the event cohort and those on SGLT2i who did not develop DKA, the event cohort had poorer baseline diabetes control (HbA1c 9.4 ± 2.3% vs 8.4 ± 2.7%, P = 0.013).
• Fifthly, the majority of SGLT2i-induced DKA cases occurred in non-elective or emergency admissions with sepsis and surgery identified as precipitating factors.
• Lastly, in terms of treatment and discharge actions, while this serious drug mediated event was described in nearly 80% of discharge summaries, less than one-third (31%) of patients were restarted on an SGLT2i agent prior to discharge, and under 50% of discharge summaries had clear advice to primary care practitioners around this important issue.

4.2. Consequences of SGLT2-Induced DKA

In keeping with other studies, patients with SGLT2i-induced DKA experienced moderate-to-severe range acidosis (9, 13, 17, 18, 29) and 20% of cases required transfer to HDU/ICU level care. This highlights that the acidosis of SGLT2i-DKA is not necessarily mild, with cases having moderate to severe range acidosis, and accompanied by life-threatening consequences of reduced cardiac output, arrhythmias; hypotension; altered mental status and shock (19).

While no SGLT2i-DKA deaths were reported in this study, a mortality rate of 1.24% and 1.54% is recorded in the Australian Therapeutic Goods Administration Database of Adverse Event Notifications (20) and the United State Food and Drug Administration Adverse Event Reporting System (21) respectively.

The difference in mortality rates seen in this study is in part attributable to Liverpool Hospital’s status as a quaternary level hospital with multiple levels of medical services and subspecialities, high levels of medical staffing and ready access to a large critical care facility. It is possible that a higher risk of harm from SGLT2i- DKA may occur in lower acuity facilities, where this level of support is not as readily available and access to diabetes specialist care reduced. This also highlights the role of increased understanding of hospital specific risk factors for SGLT2i-DKA as part of SGLT2i prescribing risk stratification and SGLT2i-DKA prevention policy development.

4.3. Characteristics of Patients With and Without SGLT2i-Induced DKA

When comparing the characteristics of inpatients on SGLT2i with and without SGLT2i-DKA, both cohorts exhibited similarities for age and duration of diabetes. Whereas, SGLT2i-DKA cohort had poorer baseline control of their diabetes than those who did not. This noted difference may be due to differences in adherences with their DM medications, especially insulin, as this is a commonly reported precipitating factor for SGLT2i-DKA (12, 22, 23). One other local issue contributing to reduced glucose control observed in both cohorts are the social economic health determinants of the geographic catchment area that the hospital services with lower health literacy consistently associated with poorer diabetes knowledge and health outcomes (24).

Additionally, 47% of the cohort had an initial BGL of >14 mmol/L with the majority of patients experiencing SGLT2i-DKA having BGLs at least in the mild hyperglycaemic range. This result is consistent with a systematic review of 34 SGLT2i-associated DKA case reports carried by Burke, Schumacher (11) which also described a mean BGL of 14.7 mmol/L. Contrawise to the name SGLT2i euglycemic DKA, this finding of hyperglycaemic range BGLs, which remained consistent over the 12 months of audit, reflect the impact of hospitalisation on glucose control through changed diet, new medications, and glycaemic stress from intercurrent illnesses, recent surgery or acute infection (25, 26), which were all factors seen in this cohort and in part why we have termed the condition SGLT2i-DKA.

Statistically significant differences were illustrated between both groups in rates of sepsis and surgery during the admission, akin in keep with a meta-analysis of 169 unique cases of SGLT2i-induced DKA by Juneja, Nasa (27) with acute severe infection (including sepsis) the most common triggering factor for SGLT2i-DKA; a systemic review by Thiruvenkatarajan, Meyer (9) outlined the pathophysiological association between surgical stress and SGLT2i-induced DKA.

A relative strength of this study is the focus on acute hospital care and the identification of SGLT2i-DKA risk factors among common hospital related presentations and morbidities, while the comparison between SGLTA2i-DKA cohort and all SGLT2i use while hospitalised again demonstrated the association between elevated A1c and SGLT2i-DKA risk reported in the large meta-analysis on SGLTG2i-DKA in T2D by Wang, Qin et al. (28)

4.4. Discharge Actions

While guidelines are still evolving, fewer than one-third (31%) of patients were restarted on an SGLT2i agent prior to discharge. For remaining patients, any resumption of SGLT2i would need to be initiated by their regular primary care practitioners, or their other specialists, during their post admission outpatient care. These medical practitioners may be waiting on diabetes specialist advice prior to any restart and while 77% of SGLT2i- DKA patients had scheduled follow-up outpatient clinic appointments with the hospital’s Diabetes and Endocrine service, as per internal departmental attendance metrics, at least 33% of those referred to post-admission clinic fail to attend appointments. Additionally, less than half (48.6%) of the cohort were provided with clear advice in their discharge paperwork to guide their own medical practitioners on whether it was safe to restart on SGLT2i agents. Without specific instructions on discharge, or reference to diabetes specialist advice, it is very possible that the patient’s regular health practitioners would not feel confident or significantly delay any restart of these medications, or for those whose therapy cessation was deliberate, restart therapy. Impacts of this include failing to restart SGLT2is negatively impacting glycaemic control, and by not utilising their cardio- and reno-protective properties, worsening long term patient risk for adverse cardiac and renal outcomes; for those who should stay off therapy, restart exposes such patients to increased risk of genital urinary infection and even sepsis.

4.5. Clinical Implications

It is important to note that SGLT2i are increasingly prescribed by a wider range of medical professionals beyond endocrinologists and diabetes specialists; by primary care practitioners, cardiologists and nephrologists and for people without known diabetes (29) as seen by Lin, Pearson (29), which showed a 216-fold increase in number of SGLT2i users from January 2014 to July 2022. Given the rising ubiquity of this drug class, it is imperative for all medical prescribers be aware of and know how to recognise SGLT2i-DKA, to advise their patients of risk factors and mitigation strategies, and for hospital based acute care staff to be aware of risk factors, presenting signs and symptoms and treatment pathways for SGLT2i-DKA among their patients on these increasingly common agents.

With reference to the most recent guideline on the use of SGLT2i for T2DM in acute inpatient setting published by NSW Health’s Agency for Clinical Innovation (30), frequent monitoring of the blood ketone levels plays an key role in the recognition and diagnosis of DKA and glucometers that can simultaneously measure both capillary blood glucose and ketones should be made available on acute care wards (31). There have also been international proposals to provide funding and subsidies for simultaneous glucose-ketone testing kits for patients with T2DM who are prescribed SGLT2i (31). More recently, ketone monitoring as part of continuous glucose monitor devices has emerged, with the potential to significantly strengthen comprehensive self-monitoring, thereby, more effectively treat and manage DKA associated with SGLT2i use (32).

Education of junior and senior medical staff and staff working in theatres, emergency departments and critical care services around this class of agent and their commonly used combinations of their generic and trade names, and formalised instructions as to withhold them when patients are unwell, have reduced oral intake or fasting for procedures / surgeries should be mandatory. Implementing System wide prescription safety checks via changes to electronic prescribing is another area of potential risk reduction.

People prescribed the agents themselves should be educated on ‘sick day rules’ and be familiar with the safe use and risks associated with SGLT2i therapy, and encouraged to follow formalised rules around ketone testing and food consumption such as proposed by Garg, Peters (33) to mitigate and manage DKA.

Due to the predominant occurrence of emergency admissions in patients with SGLT2i- DKA, the implementation of routine point of care combined glucose and blood ketone testing when admitted and baseline HbA1c measurements in the Emergency Department could potentially reduce the risk of this event among SGLT2i users. Routine blood glucose and HbA1c assessment improving detection of previously undiagnosed diabetes (34); point of care blood ketone testing may aid detection of insulin insufficiency beyond early detecting of DKA in patients on SGLT2i therapy (35).

4.6. Limitations

As noted earlier this study has methodological limitations of missing data across the study period resulting in 3 months of non-collection over in 2022; the non-contiguous nature of the SGLT2i -DKA vs all use SGLT2i cohorts and finally it is possible that very mild SGLT2i-induced eDKA cases were not recognised by staff, or that staff did not follow up testing of lower range blood or urinary elevations in such cases which resolved with usual standard care.

There are also inherent limitations in the pragmatic convenience design; this relies on medical and nursing clinical staff entering clinical parameters of ketones and BGLs and it is possible errors arose. Lastly, this study only captures data from a single high level centre that is based in a heavily urbanized region with a noticeably higher prevalence of diabetes (7.7%) than that of wider Sydney (5.7%) or Australia wide (5.2%) (36). While perhaps less applicable to other parts of Australia, the study is similar in characteristics to other major metropolitan centres.

5. Conclusions

SGLT2i-induced DKA was observed in hospitalised patients with poorer glycaemic control and was more prevalent among those admitted through the Emergency Department who had experienced sepsis and undergone surgery. When identified, all SGLT2i- DKA episodes were treated promptly and typically resolved within 24 hours. Although no deaths had been reported in this study, it is essential to emphasize that SGLT2i- DKA should not be underestimated, as it can lead to severe consequences.

Therefore, strategies to risk stratify among those on SGLT2is, to aid in early detection and effective management are crucial, especially given the increasing use of SGLT2i in both people with diabetes but also those without known diabetes but concurrent cardiovascular and renal diseases, conditions themselves nevertheless associated with diabetes mellitus.