A systematic literature review on the burden of diabetic ketoacidosis in type 2 diabetes mellitus

Abstract

Aim

To understand the existing literature on the epidemiology and clinical, humanistic, and economic burden of diabetic ketoacidosis (DKA) in people living with type 2 diabetes mellitus (T2DM).

Materials and Methods

MEDLINE, Embase and the Cochrane library were systematically searched for studies published between 1 January 2014 and 14 December 2023. Clinical trials and observational studies, conducted in people living with T2DM, were included if they provided data on DKA epidemiology, morbidity, mortality, hospitalizations or patient‐reported outcomes. Studies of DKA‐associated costs in T2DM were also included. Data were summarized descriptively.

Results

Overall, 197 publications were included. We found wide variations in DKA prevalence (0.0%–50.0%; 5th–95th percentile: 0.02%–26%; 126 publications) and incidence (0.0–24.5 events per 1000 patient years; 5th–95th percentile: 0.004–7.6 events per 1000 patient years; 37 publications). Populations at increased risk of DKA included patients using sodium–glucose cotransporter‐2 inhibitors, those using insulin and those with poor glycaemic control. The most common precipitating factors were infection and non‐adherence to treatment. There was limited evidence on the humanistic burden of DKA, but the results highlighted a high burden of complications including acute kidney injury or failure. The length of hospital stay ranged from days to several weeks.

Conclusions

DKA is associated with a high clinical burden in people living with T2DM. Resources to screen for and potentially prevent DKA may reduce the burden of DKA for patients with T2DM and the healthcare system.

1. INTRODUCTION

Currently, an estimated 537 million adults worldwide are living with diabetes, more than 95% of whom have type 2 diabetes mellitus (T2DM). ¹ , ² People living with diabetes are at risk of developing diabetic ketoacidosis (DKA), a life‐threatening complication that is characterized by hyperglycaemia (in most cases), hyperketonaemia and acidosis, and requires timely care. ³ , ⁴ DKA risk varies according to diabetes treatment, comorbidities and age. ⁵ Subgroups at risk of DKA would benefit from ketone monitoring, but few people with diabetes regularly monitor their ketone levels because of a lack of awareness of the importance of ketone monitoring and how to interpret and take action on ketone levels. ⁶ The current methods for ketone monitoring can be costly and burdensome and are often not reimbursed. ⁶ In addition, these methods do not indicate the onset of ketosis or ketoacidosis, but rather confirm if it is already in progress. ⁷

Although people living with type 1 diabetes mellitus (T1DM) and T2DM are both susceptible to DKA, ³ the burden of DKA in T2DM is not well defined. To understand the health impact of DKA on patients and the healthcare system, a systematic literature review (SLR) was designed to identify and summarize existing evidence on the prevalence, incidence and burden of DKA in T2DM and to identify those with T2DM at risk of DKA.

2. METHODS

2.1. Objectives

The objective of this SLR is to understand the existing literature on the burden of DKA in people living with T2DM, including epidemiology (incidence, prevalence and risk factors) and clinical, humanistic, and economic burden. Specifically, patients receiving sodium–glucose cotransporter‐2 inhibitors (SGLT2is) and those using insulin were investigated, because DKA is known to be associated with SGLT2i use and missed or inadequate doses of insulin in patients with T2DM, as noted in the American Diabetes Association (ADA) “Standards of Care in Diabetes”. ⁸ Several additional factors (age, race/ethnicity, body weight, prior DKA, glycaemic control, ketosis, ketosis‐prone T2DM, pregnancy status, renal status, alcohol consumption, starvation, high blood lactate, newly diagnosed T2DM or duration of T2DM) were specified post hoc (after data extraction).

2.2. Searches and screening

MEDLINE, Embase and the Cochrane Library were searched via Ovid for the period of 1 January 2014 to 14 December 2023. The searches were designed to identify clinical trials and observational studies, conducted in people living with T2DM, that provided data on DKA epidemiology, monitoring, morbidity and mortality, hospitalizations with DKA and patient‐reported outcomes/quality of life in those with DKA. The searches were also designed to identify studies of the costs associated with DKA in T2DM. Case reports and non‐English language publications were excluded. Search strings are presented in Tables S1–S3.

We also identified the following as congresses of interest: ADA 2021–2023; Advanced Technologies & Treatments for Diabetes 2021–2023; and European Association for the Study of Diabetes (EASD) 2021–2023. The proceedings of all congresses of interest were indexed in Embase and captured in the electronic searches.

The abstracts of the identified publications were screened to determine whether they met predefined eligibility criteria (Table S4). If they did, the full‐text versions of those papers were then reviewed against the same criteria to confirm their eligibility. Both screening stages were performed in a single‐blind manner by one reviewer; 8% of the title/abstract screening decisions were checked by a second, senior reviewer. Discrepancies were noted in only 1% of the title/abstract screening sample. Any uncertainties on screening decisions were resolved by the senior reviewer.

2.3. Data extraction

Data regarding study design, patient characteristics and measures of the clinical, economic, humanistic and epidemiologic burden of DKA were extracted from the included studies. Epidemiologic data included DKA incidence, prevalence, risk factors (factors associated with an increased likelihood of developing DKA) and precipitating factors (events or conditions [e.g., infection] that trigger a DKA episode).

Data were extracted into a Microsoft Excel document by one reviewer and quality‐checked by a second reviewer. Outcomes were reported descriptively as ranges with 5th and 95th percentiles (obtained by numerically sorting estimates from individual studies and discarding the top and bottom 5% without weighting) to limit the impact of outliers. No meta‐analysis was performed.

3. RESULTS

Initial searching identified 1397 hits. After removal of duplicates, 1245 studies underwent title/abstract screening, 392 studies underwent full‐text review and 197 publications were identified as eligible for data extraction (Figure 1). Most of the studies were published from 2021 up to the cut‐off date of 14 December 2023 (Figure 2). Studies were distributed approximately equally among global regions with the exception of Latin America (Figure 2). Similarly, there was variation in study designs. Most studies were retrospective (78%); other designs included prospective studies, randomized controlled trials and single‐arm studies (Figure 2).

FIGURE 1.

PRISMA flow diagram. ^†Population with T1DM only or mixed T1DM and T2DM with no subgroup data. ^‡Not diabetes or not DKA, ketosis, ketoacidosis or acidosis. ^§Case studies/reports. ^¶A definition of DKA was reported, but there were no other outcomes of interest in the publication (i.e., no epidemiology, clinical, humanistic or economic burden outcomes). DKA, diabetic ketoacidosis; NMA, network meta‐analysis; PRISMA, Preferred Reporting Items for Systematic Review and Meta‐analyses; SLR, systematic literature review.

FIGURE 2.

Overview of included publications by year of publication, region and study design. ^†Others include pooled RCTs (3.6%), publications with both retrospective and prospective elements (2.0%), and unclear observational designs (1.5%). ^‡Published in 2014 (3%), 2015 (2%), 2016 (4%), 2017 (7%), 2018 (3%), 2019 (8%) and 2020 (5%). ^§The cut‐off date for the literature searches was 14 December 2023. APAC, Asia–Pacific; EMEA, Europe, the Middle East and Africa; LATAM, Latin America; NR, not reported; RCT, randomized controlled trial.

The definition of DKA was generally consistent across publications (91 publications) ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ , ⁷² , ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ , ⁷⁷ , ⁷⁸ , ⁷⁹ , ⁸⁰ , ⁸¹ , ⁸² , ⁸³ , ⁸⁴ , ⁸⁵ , ⁸⁶ , ⁸⁷ , ⁸⁸ , ⁸⁹ , ⁹⁰ , ⁹¹ , ⁹² , ⁹³ , ⁹⁴ , ⁹⁵ , ⁹⁶ , ⁹⁷ , ⁹⁸ , ⁹⁹ : blood glucose >250 mg/dL, pH <7.3, anion gap >10 mmol/L, bicarbonate level ≤18 mmol/L and ketonemia/ketonuria. The most frequently used International Classification of Diseases (ICD) diagnostic codes were ICD, Ninth Revision 250.1 (6 publications) ²⁶ , ²⁷ , ²⁹ , ³¹ , ³³ , ⁹² and ICD, Tenth Revision E11.1 (6 publications). ²⁹ , ⁴⁴ , ⁴⁷ , ⁸⁴ , ¹⁰⁰ , ¹⁰¹

3.1. Prevalence and incidence

The prevalence and incidence of DKA in patients with T2DM varied widely. For prevalence (126 publications), ¹⁰ , ¹¹ , ¹² , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ , ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ , ⁶⁹ , ⁷⁰ , ⁷¹ , ⁷² , ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ , ⁷⁷ , ⁷⁸ , ⁷⁹ , ⁸⁰ , ⁸¹ , ⁸² , ⁸³ , ⁸⁴ , ⁸⁵ , ⁸⁶ , ⁸⁷ , ⁸⁸ , ⁸⁹ , ⁹⁰ , ⁹¹ , ⁹² , ⁹³ , ⁹⁴ , ⁹⁵ , ⁹⁶ , ⁹⁷ , ⁹⁸ , ⁹⁹ , ¹⁰⁰ , ¹⁰¹ , ¹⁰² , ¹⁰³ , ¹⁰⁴ , ¹⁰⁵ , ¹⁰⁶ , ¹⁰⁷ , ¹⁰⁸ , ¹⁰⁹ , ¹¹⁰ , ¹¹¹ , ¹¹² , ¹¹³ , ¹¹⁴ , ¹¹⁵ , ¹¹⁶ , ¹¹⁷ , ¹¹⁸ , ¹¹⁹ , ¹²⁰ , ¹²¹ , ¹²² , ¹²³ , ¹²⁴ , ¹²⁵ , ¹²⁶ , ¹²⁷ , ¹²⁸ , ¹²⁹ , ¹³⁰ , ¹³¹ , ¹³² , ¹³³ , ¹³⁴ , ¹³⁵ , ¹³⁶ , ¹³⁷ , ¹³⁸ , ¹³⁹ the range was 0.0%–50.0% (5th–95th percentile: 0.02%–26%). ¹⁰ , ¹³ , ¹⁴ , ¹⁷ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ⁴⁰ , ⁴¹ , ⁴⁵ , ⁴⁶ , ⁵¹ , ⁵² , ⁵³ , ⁵⁶ , ⁵⁷ , ⁵⁹ , ⁶² , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ , ⁷⁰ , ⁷⁶ , ⁷⁸ , ⁷⁹ , ⁸² , ⁸⁴ , ⁸⁷ , ⁸⁹ , ⁹⁰ , ⁹¹ , ⁹⁴ , ⁹⁵ , ⁹⁶ , ⁹⁷ , ⁹⁸ , ⁹⁹ , ¹⁰¹ , ¹⁰² , ¹⁰³ , ¹⁰⁴ , ¹⁰⁵ , ¹⁰⁶ , ¹⁰⁷ , ¹⁰⁸ , ¹⁰⁹ , ¹¹⁰ , ¹¹¹ , ¹¹² , ¹¹³ , ¹¹⁴ , ¹¹⁵ , ¹¹⁶ , ¹¹⁷ , ¹¹⁸ , ¹¹⁹ , ¹²⁰ , ¹²¹ , ¹²² , ¹²³ , ¹²⁴ , ¹²⁵ , ¹²⁶ , ¹²⁷ , ¹²⁸ , ¹²⁹ , ¹³⁰ , ¹³¹ , ¹³² , ¹³³ , ¹³⁴ , ¹³⁵ , ¹³⁶ , ¹³⁷ , ¹³⁸ , ¹³⁹ , ¹⁴⁰ , ¹⁴¹ , ¹⁴² , ¹⁴³ , ¹⁴⁴ , ¹⁴⁵ , ¹⁴⁶ , ¹⁴⁷ , ¹⁴⁸ , ¹⁴⁹ , ¹⁵⁰ , ¹⁵¹ , ¹⁵² , ¹⁵³ , ¹⁵⁴ , ¹⁵⁵ , ¹⁵⁶ , ¹⁵⁷ , ¹⁵⁸ , ¹⁵⁹ , ¹⁶⁰ , ¹⁶¹ , ¹⁶² , ¹⁶³ , ¹⁶⁴ , ¹⁶⁵ , ¹⁶⁶ , ¹⁶⁷ , ¹⁶⁸ , ¹⁶⁹ , ¹⁷⁰ , ¹⁷¹ , ¹⁷² , ¹⁷³ , ¹⁷⁴ , ¹⁷⁵ , ¹⁷⁶ For incidence (37 publications), ¹⁰ , ¹¹ , ¹⁷ , ¹⁸ , ³⁸ , ⁴² , ⁴⁶ , ⁴⁷ , ⁵³ , ⁵⁶ , ⁷¹ , ⁷⁴ , ⁷⁵ , ⁸⁰ , ⁸³ , ⁸⁴ , ¹⁰² , ¹⁰³ , ¹⁰⁴ , ¹⁰⁵ , ¹⁰⁶ , ¹¹¹ , ¹²² , ¹²⁴ , ¹²⁵ , ¹³⁰ , ¹³⁴ , ¹³⁸ , ¹⁴⁰ , ¹⁴¹ , ¹⁴² , ¹⁴³ , ¹⁴⁴ , ¹⁴⁵ , ¹⁴⁶ , ¹⁴⁷ , ¹⁴⁸ the range was 0.0–24.5 events per 1000 patient years (PY) (5th–95th percentile: 0.004–7.6 events per 1000 PY), ⁹ , ²⁹ , ³⁰ , ³¹ , ³⁵ , ⁴³ , ⁵² , ⁵⁶ , ⁵⁹ , ⁶⁰ , ⁸⁴ , ⁸⁹ , ⁹⁰ , ⁹¹ , ¹⁰¹ , ¹¹⁴ , ¹¹⁵ , ¹²⁰ , ¹²³ , ¹³⁴ , ¹³⁷ , ¹⁴² , ¹⁵³ , ¹⁵⁴ , ¹⁵⁵ , ¹⁵⁶ , ¹⁵⁷ , ¹⁶¹ , ¹⁷¹ , ¹⁷⁴ , ¹⁷⁵ , ¹⁷⁷ , ¹⁷⁸ , ¹⁷⁹ , ¹⁸⁰ , ¹⁸¹ , ¹⁸² with an outlier of 63 events per 1000 PY in Ethiopia. ⁹ The authors of the Ethiopian study—which was a retrospective analysis of data from 2016 to 2020—commented that the high incidence could be due to their hospital‐based study design and disparities in economic status, level of education and access to healthcare facilities compared with other study populations. ⁹

The prevalence and incidence of DKA were reported for 28 and 11 countries/regions, respectively. Both varied between countries and across publications within the same country (Table 1). The highest prevalence of DKA was reported in China (50.0%); however, the publication exclusively included patients with ketosis‐prone T2DM (diabetes without precipitating illness and with ketosis or DKA in the absence of autoantibodies at diagnosis)—these patients had a high likelihood of DKA by definition. ⁹⁷ The next highest prevalence values were reported in the USA, in patients newly diagnosed with T2DM during the COVID‐19 pandemic (2020; 47.1%) ¹³⁰ and in patients aged <21 years with T2DM who were hospitalized with COVID‐19 (46%) ¹⁰² ; delays in accessing medical care during the COVID‐19 pandemic and (in the latter study) the hospital‐based setting and presence of infection may have contributed to the high prevalence of DKA. The lowest prevalence of DKA (0.0%) was reported in subpopulations of two studies in China ¹¹² , ¹²⁸ and one study in the USA. ¹⁴⁷ One of the studies in China included 2654 hospitalized patients with T2DM without ketosis (by definition, this subgroup would not have DKA) ¹¹² ; the other Chinese study included 508 patients with T2DM aged ≥65 years who received at least one dose of dapagliflozin and were followed for 24 weeks (the sample size and duration of follow‐up may have contributed to the low prevalence in this study). ¹²⁸ The US‐based study included 32 White paediatric patients diagnosed with T2DM during the COVID‐19 pandemic; the lack of DKA in these patients may be due to the small sample size, but the same study also found a DKA prevalence of 26% in 25 Black patients, suggesting disparity between racial groups. ¹⁴⁷ The prevalence of DKA was 0.01%–37.5% in other study populations in the USA (42 publications) ¹⁴ , ¹⁹ , ²¹ , ²³ , ²⁴ , ²⁹ , ³¹ , ³² , ⁶² , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁷⁸ , ⁷⁹ , ⁸² , ⁹⁰ , ⁹¹ , ¹⁰⁸ , ¹⁰⁹ , ¹¹⁰ , ¹¹⁷ , ¹¹⁸ , ¹²³ , ¹³⁰ , ¹³⁶ , ¹³⁸ , ¹⁴⁰ , ¹⁴³ , ¹⁴⁵ , ¹⁴⁶ , ¹⁵⁰ , ¹⁵¹ , ¹⁵³ , ¹⁵⁴ , ¹⁵⁵ , ¹⁵⁶ , ¹⁵⁷ , ¹⁶³ , ¹⁶⁵ , ¹⁶⁶ , ¹⁶⁸ , ¹⁷² and 0.01%–5.6% in European countries (17 publications). ³³ , ⁴⁰ , ⁵¹ , ⁵⁶ , ⁸⁹ , ¹⁰¹ , ¹¹⁶ , ¹²⁶ , ¹²⁷ , ¹³⁵ , ¹³⁹ , ¹⁴⁴ , ¹⁴⁹ , ¹⁵² , ¹⁵⁹ , ¹⁶² , ¹⁷⁵ The incidence of DKA was 0.55–9.1 per 1000 PY in the USA (13 publications) ¹⁸ , ¹⁹ , ²⁹ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁷⁷ , ¹⁴³ , ¹⁴⁹ , ¹⁵⁰ , ¹⁵² , ¹⁵⁸ , ¹⁵⁹ and 0.001–19.8 per 1000 PY in European countries (5 publications). ⁵⁶ , ⁸⁹ , ¹⁰¹ , ¹⁷⁵ , ¹⁷⁹ There was no clear relationship between the number of patients included in each study and the incidence estimate, suggesting that major publication bias towards high incidence estimates in small studies is unlikely (Figure S1).

TABLE 1.

Summary of DKA prevalence and incidence by country/region.

Country/region	Prevalence or incidence, range	Citation
Prevalence (%)
Australia	0.3–15.4	30, 46, 64
China	0.0–50.0	34, 36, 41, 57, 94, 95, 97, 98, 99, 106, 112, 113, 128, 129, 169, 176
Croatia	4.8	139
Denmark	0.1–0.5	56, 159
Ethiopia	1.8–28.4	53, 107, 124
France	0.2–1.7	101, 126, 127, 162
Germany	0.1–0.3	89
Ghana	2–43.8	13, 105
Hong Kong	0.04–20.7	59, 115, 141
Hungary	0.7	135
India	1–13.3	68, 70, 187
Indonesia	14.6	158
Italy	1	33
Japan	0.2–31.4	45, 84
Malaysia	13.4	173
Philippines	3.9	164
Qatar	0.3–21.4	17
Saudi Arabia	0.2–22.1	22, 103, 104
Singapore	0.6–0.7	125
South Korea	0.04–7	132, 133, 134
Spain	0.01–5.6	116, 149
Sri Lanka	0.13	111
Taiwan	10.8	96
Thailand	9.7–13.7	160, 167
The Gambia	2.6	10
Turkey	2.4	152
UK	0.2–3.0	40, 51, 144, 175
USA	0.0–47.1	14, 19, 21, 23, 24, 29, 31, 32, 62, 65, 66, 67, 78, 79, 82, 90, 91, 102, 108, 109, 110, 117, 118, 123, 130, 136, 138, 140, 143, 145, 146, 147, 150, 151, 153, 154, 155, 156, 157, 163, 165, 166, 168, 172
Incidence (events per 1000 PY)
Australia	0.83	30
Denmark	0.53–3.47	56
Ethiopia	63	9
France	11.2–19.8	101
Germany	0.5–2.8	89
Hong Kong	0.2–24.5	59, 60, 115
Japan	0.38–0.48	84, 182
South Korea	0.6–2.7	43, 134
Switzerland	0.001–0.07	179
UK	0.85	175
USA	0.55–9.1	18, 19, 29, 47, 48, 49, 77, 143, 149, 150, 152, 158, 159

Note: Ranges include data from patient subgroups if the prevalence/incidence for the overall population was not reported. Treatment subgroups are included. For DKA prevalence, the UK category includes one publication that includes data from Scotland only. Incidence estimates reported in units other than events per 100 PY, per 1000 PY, per 10 000 PY or per 100 000 PY are excluded.

Abbreviations: DKA, diabetic ketoacidosis; PY, patient years.

The prevalence and incidence of DKA also ranged widely in SGLT2i users. The prevalence in SGLT2i users was 0.0%–3.6% (5th–95th percentile: 0.02%–1.4%; 45 publications), ¹⁷ , ²⁹ , ³¹ , ³⁵ , ⁴⁶ , ⁵¹ , ⁵² , ⁵⁶ , ⁵⁹ , ⁶⁴ , ⁹⁰ , ⁹¹ , ⁹⁸ , ⁹⁹ , ¹⁰³ , ¹¹¹ , ¹¹⁴ , ¹¹⁷ , ¹¹⁸ , ¹¹⁹ , ¹²⁰ , ¹²¹ , ¹²³ , ¹²⁵ , ¹²⁸ , ¹²⁹ , ¹³¹ , ¹³² , ¹³⁴ , ¹³⁷ , ¹³⁸ , ¹⁴² , ¹⁴⁴ , ¹⁴⁸ , ¹⁵² , ¹⁵³ , ¹⁵⁴ , ¹⁵⁵ , ¹⁵⁶ , ¹⁵⁷ , ¹⁶¹ , ¹⁶⁵ , ¹⁷⁰ , ¹⁷¹ , ¹⁷⁴ with two outliers of 15.4% ⁶⁴ and 21.4% ¹⁷ (in 39 patients undergoing cardiac surgery ⁶⁴ and 14 patients receiving canagliflozin, ¹⁷ respectively). The incidence of DKA in SGLT2i users was 0.0–9.1 events per 1000 PY (5th–95th percentile: 0.2–7.6 events per 1000 PY; 28 publications), ²⁹ , ³¹ , ³⁵ , ⁴³ , ⁵² , ⁵⁶ , ⁵⁹ , ⁶⁰ , ⁹⁰ , ⁹¹ , ¹¹⁴ , ¹²⁰ , ¹²³ , ¹³⁴ , ¹³⁷ , ¹⁴² , ¹⁵³ , ¹⁵⁴ , ¹⁵⁵ , ¹⁵⁶ , ¹⁵⁷ , ¹⁶¹ , ¹⁷¹ , ¹⁷⁴ , ¹⁷⁷ , ¹⁷⁸ , ¹⁸⁰ , ¹⁸¹ with an outlier of 24.5 events per 1000 PY (in 1087 patients undergoing emergency surgery). ⁶⁰

In insulin users, the prevalence of DKA was 0.1%–5.2% (11 publications), ²¹ , ³² , ⁵⁶ , ⁸⁹ , ¹¹⁰ , ¹²³ , ¹²⁶ , ¹²⁷ , ¹³⁴ , ¹⁴¹ , ¹⁴⁶ with an outlier of 20.7% (in 58 Chinese patients) ¹⁴¹ and the incidence of DKA was 0.5–4.9 events per 1000 PY (3 publications). ⁵⁶ , ⁸⁹ , ¹²³ The types of insulin used in these study populations were mostly mixed or unspecified (7 publications) ²¹ , ³² , ⁵⁶ , ¹¹⁰ , ¹²³ , ¹³⁴ , ¹⁴¹ ; single publications reported on patients using multiple daily injections or an insulin pump, ¹²⁶ basal insulin, ¹²⁷ insulin glargine, ¹⁴⁶ and human insulin/neutral protamine Hagedorn or basal insulin analogue. ⁸⁹

Excluding newly diagnosed patients, the mean duration of T2DM was 0.6–18.4 years (30 publications), ¹² , ²¹ , ²⁵ , ²⁶ , ²⁷ , ³⁵ , ⁵⁴ , ⁵⁶ , ⁵⁸ , ⁵⁹ , ⁶⁸ , ⁶⁹ , ⁷⁷ , ⁷⁸ , ⁸⁶ , ⁸⁹ , ⁹³ , ⁹⁷ , ¹¹¹ , ¹¹² , ¹¹³ , ¹²⁸ , ¹²⁹ , ¹³¹ , ¹³⁵ , ¹⁴¹ , ¹⁴⁴ , ¹⁵³ , ¹⁶¹ , ¹⁸³ whereas the median duration was 0.1–15 years (17 publications). ¹¹ , ¹⁶ , ³⁷ , ⁴² , ⁴⁵ , ⁶³ , ⁷¹ , ⁸⁷ , ⁹⁸ , ¹⁰³ , ¹¹⁵ , ¹²⁰ , ¹⁶⁰ , ¹⁶⁴ , ¹⁷⁰ , ¹⁷⁶ , ¹⁸⁴ There was no clear relationship between the incidence of DKA and the mean or median duration of T2DM (Figure S2).

3.2. Risk factors

3.2.1. SGLT2i use

Many publications reported an increased risk (Figure 3, Table S5) or incidence (Figure S3) of DKA in SGLT2i users compared with non‐users—this was not seen in all comparisons, likely due to small sample sizes in some studies. ²⁹ , ³¹ , ³⁵ , ⁴² , ⁴³ , ⁵¹ , ⁵⁶ , ⁵⁹ , ⁶⁰ , ⁶⁴ , ⁹⁰ , ⁹¹ , ¹²³ , ¹²⁵ , ¹³⁴ , ¹⁴³ , ¹⁵³ , ¹⁵⁴ , ¹⁷⁸ , ¹⁸⁵ , ¹⁸⁶

FIGURE 3.

Risk of DKA in SGLT2i (mixed or unspecified type) users. Data shown are from studies that report risk associated with the SGLT2i class overall without specifying the type of SGLT2i (i.e., excluding publications that only reported use of specific types of SGLT2i: Canagliflozin, dapagliflozin, empagliflozin or sotagliflozin). Excludes an outlier: McCann, 2022, ⁶⁴ who report OR = 38 (95% CI 9.1–158.7). Excludes Caparrotta, 2023 ¹⁸⁵ and McCoy, 2021, ¹⁴³ who do not report the comparator. Excludes 2 publications that reported results for subgroups only: SGLT2i vs. non‐SGLT2i in subgroups of patients undergoing emergency vs. elective operations, ⁶⁰ and SGLT2i vs. DPP‐4i in subgroups by data source (CCAE vs. MDCD vs. MDCR vs. Optum database). ^{90 †}Propensity score‐matched cohorts. ^‡Adjusted for age (continuous), sex, diabetes duration (continuous) and deciles of time‐conditional propensity score. ^§Adjusted for sex, age, diabetes duration, daily dose of insulin and year of becoming at risk. ^¶Adjusted for age, sex, ethnicity, admission blood glucose level, insulin administration, and micro‐ and macrovascular disease. Data from D'Andrea, 2023, ²⁹ Dawwas, 2022, ³¹ Dawwas, 2021, ¹⁷⁸ Douros, 2020, ³⁵ Fu, 2022, ¹⁸⁶ Fu, 2023, ¹²³ Goh, 2022, ¹²⁵ Hamblin, 2019, ⁴² Han, 2021, ⁴³ Khunti, 2022, ⁵¹ Kim, 2018, ¹³⁴ Laursen, 2021, ⁵⁶ Lui, 2023, ⁵⁹ Patil, 2023, ¹⁵³ Patorno, 2021, ¹⁵⁴ and Wang, 2017. ⁹¹ CCAE, IBM MarketScan Commercial Claims and Encounters; CI, confidence interval; DKA, diabetic ketoacidosis; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; GLP‐1RA, glucagon‐like peptide‐1 receptor agonist; HR, hazard ratio; IRR, incidence rate ratio; MDCD, IBM MarketScan Multi‐State Medicaid Database; MDCR, IBM MarketScan Medicare Supplemental and Coordination of Benefits; OR, odds ratio; RR, risk ratio; SGLT2i, sodium–glucose cotransporter‐2 inhibitor.

3.2.2. Insulin use

Patients receiving insulin had numerically higher prevalence ⁵⁶ , ¹²³ , ¹³⁴ and incidence ⁵⁶ , ¹²³ of DKA than those who did not receive insulin (Figures S4 and S5). The risk of DKA in patients using insulin was assessed in three large studies, each of which found an increased risk in users compared with non‐users (odds ratio [OR] for hospitalization with a diagnosis of DKA = 1.8 ¹¹⁸ ; post‐operative DKA incidence rate ratio [IRR] = 2.878 ⁵⁹ ; DKA IRR [non‐insulin vs. insulin] = 0.31 ¹⁴³ ).

3.2.3. Post hoc subgroup analyses

As the post hoc analyses were non‐systematic, they are only briefly summarized here. Two publications showed that patients with a history of DKA/hyperglycaemic crises are at increased risk of DKA. ¹¹⁸ , ¹⁴³ Patients with elevated glycated haemoglobin (five publications), ⁵⁹ , ⁶⁰ , ⁶⁸ , ¹⁴⁰ , ¹⁴³ history of hypoglycaemia (two publications), ¹¹⁸ , ¹⁴³ and uncontrolled T2DM (1 publication) ³³ were also shown to be at increased risk of DKA. One of these publications reported a linear increase in risk of DKA with increasing blood glucose level (indicating increasingly poor glycaemic control). ¹⁴³

In two studies, Black race was associated with a higher prevalence of DKA than White race at the time of diagnosis of T2DM. ¹⁹ , ¹⁴⁷ The incidence of DKA in patients with T2DM increased steadily with increasing age from 37 years onwards (1 study) ¹⁷⁹ ; analysis of the prevalence and incidence of DKA by age group across studies (based on study inclusion criteria) showed no consistent pattern (Table S6), perhaps owing to confounding. Two publications reported no association between body mass index/obesity and risk of DKA ³³ , ⁵⁹ ; in a third publication (describing a paediatric study), body mass index z‐score was significantly associated with DKA (p = 0.004). ¹⁴⁰

The prevalence of DKA in patients with CKD was 0.2%–0.7% ¹²³ , ¹³² , ¹³⁸ and the incidence was 2.5–4.9 events per 1000 PY. ¹²³ , ¹⁸⁰ Nephropathy was associated with DKA (IRR = 1.29) in a US analysis. ¹⁴³ In another publication, renal failure was not a risk factor for DKA, whereas diabetes‐related renal complications were associated with a reduced risk of DKA (OR = 0.31). ³³ In short, the association of DKA with renal disease was inconclusive based on post hoc analysis of the published literature.

Among patients with newly diagnosed or new onset T2DM, 0.3%–50.0% presented with DKA (29 publications) ¹³ , ¹⁴ , ¹⁹ , ²² , ²³ , ²⁴ , ³² , ³⁶ , ⁶² , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁷⁹ , ⁸⁷ , ⁹⁴ , ⁹⁷ , ¹⁰² , ¹¹³ , ¹³⁰ , ¹³³ , ¹⁴⁰ , ¹⁴⁵ , ¹⁴⁹ , ¹⁵⁰ , ¹⁶⁰ , ¹⁶³ , ¹⁶⁷ , ¹⁶⁸ , ¹⁸⁷ ; the upper end of the range (50.0%) was from a study of patients with ketosis‐prone T2DM. ⁹⁷ In two other studies of patients with ketosis‐prone T2DM (new onset or previously diagnosed), the prevalence of DKA was 6.6%–30.2%. ³⁶ , ¹¹³

In a study of hospitalized patients with T2DM and ketosis, the prevalence of DKA was 12.1%. ¹¹²

We found no data on the association between DKA and pregnancy, alcohol consumption, starvation or lactate level.

3.3. Precipitating factors

Precipitating factors for DKA were reported in 31 publications. ¹¹ , ¹⁶ , ¹⁷ , ²⁵ , ³⁷ , ⁴² , ⁴⁵ , ⁴⁸ , ⁵² , ⁵⁴ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷³ , ⁷⁵ , ⁷⁷ , ⁸⁰ , ⁸³ , ⁸⁶ , ⁸⁸ , ⁹³ , ⁹⁵ , ¹¹³ , ¹³¹ , ¹⁸⁸ , ¹⁸⁹ , ¹⁹⁰ , ¹⁹¹ , ¹⁹² , ¹⁹³ , ¹⁹⁴

Infection (of any type) was the most common precipitating factor in 20 publications (65%). ¹⁶ , ¹⁷ , ²⁵ , ³⁷ , ⁴² , ⁵⁴ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷⁵ , ⁷⁷ , ⁸⁶ , ⁸⁸ , ⁹³ , ⁹⁵ , ¹¹³ , ¹⁸⁹ , ¹⁹⁰ , ¹⁹² , ¹⁹⁴ Of the 20 publications, five had a study period that substantially overlapped with the COVID‐19 pandemic ²⁵ , ⁶⁹ , ¹⁸⁹ , ¹⁹⁰ or was unknown. ⁸⁸ One of these studies assessed the role of COVID‐19 and found that infections were the precipitating factor for new‐onset DKA in 30.5% and 36.4% of patients before and during the COVID‐19 pandemic, respectively, with COVID‐19 becoming the most common type of precipitating infection during the pandemic. ¹⁹⁰

Non‐adherence to insulin was the precipitating factor for DKA in 10.5% of patients on average (range: 0.0%–23.4%; 7 publications), ¹¹ , ¹⁷ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷⁵ , ⁸⁶ and was less common as a precipitating factor than infection in all but one of the identified publications. ¹⁷ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷⁵ , ⁸⁶ In the remaining publication, non‐adherence to insulin and infection were equally common as precipitating factors. ¹¹

Non‐compliance to any treatment (insulin or other medications) was the precipitating factor for DKA in 14.1% of patients on average (range: 0%–38.0%; 18 publications). ¹¹ , ¹⁶ , ¹⁷ , ²⁵ , ³⁷ , ⁴² , ⁵⁴ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷³ , ⁷⁵ , ⁷⁷ , ⁸⁶ , ⁸⁸ , ⁹⁵ , ¹⁸⁹ , ¹⁹²

Cardiovascular events were the precipitating factor for DKA in 5.8% of patients on average (range: 0.0%–13.5%; 4 publications) ⁶⁹ , ⁷¹ , ⁸⁰ , ⁹⁵ ; two publications specified acute coronary syndrome as a precipitating factor (in 5.4% and 12% of patients). ⁶⁹ , ⁸⁰

More than half of the publications that reported precipitating factors for DKA (16/31) were unable to conclusively identify precipitating factors for some patients. The cause of DKA was unknown in 0.0%–51.1% of patients, suggesting an evidence gap in what triggers DKA. ¹¹ , ¹⁷ , ²⁵ , ³⁷ , ⁴² , ⁴⁸ , ⁵⁴ , ⁵⁸ , ⁶⁹ , ⁷¹ , ⁷³ , ⁷⁵ , ⁸⁸ , ⁹⁵ , ¹⁸⁸ , ¹⁹² On average, about 10% of patients had no obvious cause of DKA, ²⁵ , ⁴² , ⁴⁸ , ⁶⁹ , ¹⁹² and in nearly a quarter of patients, the exact triggering factor was unknown. ¹¹ , ¹⁷ , ³⁷ , ⁴⁸ , ⁵⁴ , ⁵⁸ , ⁷¹ , ⁷³ , ⁷⁵ , ⁸⁸ , ⁹⁵ , ¹⁸⁸

3.4. Clinical burden

Overall mortality (due to DKA or other causes such as comorbidities) for patients with T2DM and DKA was reported in 46 publications and showed a wide range (Table 2). ¹¹ , ¹² , ¹³ , ¹⁶ , ¹⁷ , ²⁵ , ³² , ³⁷ , ³⁸ , ³⁹ , ⁴⁰ , ⁴¹ , ⁴² , ⁴⁷ , ⁴⁹ , ⁵¹ , ⁵⁴ , ⁵⁵ , ⁵⁹ , ⁶¹ , ⁶³ , ⁶⁸ , ⁶⁹ , ⁷¹ , ⁷² , ⁷⁴ , ⁷⁷ , ⁸⁶ , ⁸⁸ , ⁹⁰ , ⁹² , ⁹³ , ⁹⁵ , ⁹⁶ , ¹⁰⁰ , ¹⁰⁴ , ¹⁰⁹ , ¹¹⁵ , ¹²⁴ , ¹⁷⁹ , ¹⁸² , ¹⁸⁹ , ¹⁹² , ¹⁹⁴ , ¹⁹⁵ , ¹⁹⁶ Focusing on the large studies (N > 2000), without COVID‐19 and not linked to the COVID‐19 pandemic, gives a mortality of 1.0%–4.4%. ⁵⁵ , ¹⁰⁰ , ¹⁸² , ¹⁹⁶

TABLE 2.

Summary of mortality in patients with T2DM and DKA.

Author, year	Subgroup	Sample size (N)	Mortality (%)
Large studies (N > 2000), without COVID‐19 or related to the pandemic
Bertasi, 2022 100	N/A	153 475	4.4
Kumar, 2023 55	N/A	2175	2.5
Sato, 2021 182	N/A	13 835	4.3
Shaka, 2021 196	N/A	19 675	1.0
Studies comparing subgroups a
Almazrouei, 2023 11	N/A	55	5.5
	SGLT2i	17	0.0
	Non‐SGLT2i	38	7.9
Ata, 2021 17	N/A	43	2.3
	SGLT2i (dapagliflozin)	31	3.2
	SGLT2i (empagliflozin)	9	0.0
	SGLT2i (canagliflozin)	3	0.0
Ata, 2023 16	N/A	442	0.9
	New onset	176	1.1
	Pre‐existing T2DM	266	0.8
Ebrahimi, 2022 179	Male, 0–9 years	6	0.0
	Male, 10–19 years	9	0.0
	Male, 20–29 years	27	0.0
	Male, 30–59 years	416	2.6
	Male, 60–90 years	528	10.8
	Female, 0–9 years	5	0.0
	Female, 10–19 years	10	10.0
	Female, 20–29 years	17	0.0
	Female, 30–59 years	197	0.5
	Female, 60–90 years	482	9.8
Fan, 2023 115	<20 years at diagnosis	NR	0.0
	20–39 years at diagnosis	NR	0.4
	≥40 years at diagnosis	NR	0.03
Fu, 2022 41	DKA, acute pancreatitis	27	3.7
	No DKA, acute pancreatitis	109	0.0
Hamblin, 2019 42	SGLT2i	37	5.4
	Non‐SGLT2i	125	8.8
Kumar, 2023 55	Hyperlactataemia	726	4.4
	Normolactataemia	1449	1.5
Lui, 2023 59	SGLT2i	43	2.3
	Non‐SGLT2i	339	11.5
Nakhleh, 2023 b	SGLT2i	16	12.5
	Non‐SGLT2i	55	20.0
Rashid, 2017 77	N/A	128	12.5
	Known T2DM	118	13.6
	New onset	10	0.0
Wu, 2020 93	N/A	65	0.0
	Hyperosmolar hyperglycaemia	19	5.3
Wang, 2019 90	SGLT2i, CCAE database	218	0.0
	DPP‐4i, CCAE database	171	0.0
	SGLT2i, MDCD database	32	3.1
	DPP‐4i, MDCD database	76	1.3
	SGLT2i, MDCR database	37	2.7
	DPP‐4i, MDCR database	57	3.5
Studies with COVID‐19 or related to the pandemic
Almistehi, 2021 104	COVID‐19	20	25.0
Dell'Aquila, 2023 32	New onset, COVID‐19	118	49.2
	Non‐insulin dependent, COVID‐19	29	27.6
Farzadfar, 2022 39	COVID‐19	50	62.0
	No COVID‐19	50	4.0
Field, 2023 40	Insulin, COVID‐19	28	21.4
	No insulin, COVID‐19	23	52.2
	COVID‐19	57	36.8
	SGLT2i, COVID‐19	6	33.3
Kempegowda, 2021 49	COVID‐19	15	26.7
	No COVID‐19	2	0.0
	Pre‐pandemic	8	0.0
Khunti, 2022 51	COVID‐19	86	33.7
	SGLT2i, COVID‐19	7	28.6
	No SGLT2i, COVID‐19	79	34.2
Misra, 2021 195	Pandemic (Nov 2020–Feb 2021)	3912	19.2
	Pandemic (Mar–Jun 2020)	2831	17.1
	Pandemic (Jul–Oct 2020)	2613	11.6
	Pre‐pandemic (Nov 2017–Feb 2020)	2608	10.7
	Pre‐pandemic (Mar–Jun 2017–2019)	2010	10.2
	Pre‐pandemic (Jul–Oct 2017–2019)	2011	8.9
Mondal, 2021 68	COVID‐19	26	11.5
Other studies
Almazrouei, 2022 12	N/A	48	6.3
Ameyaw, 2017 13	N/A	16	0.0
Charoenpiriya, 2022 25	N/A	53	0.0
Cymbaluk, 2021 109	N/A	14	0.0
Eledrisi, 2022 37	N/A	1011	7.4
Elkituni, 2021 38	N/A	141	0.4
Gizaw, 2015 124	N/A	16	6.3
Gnaneshwari, 2023 189	N/A	30	10.0
Injinari, 2023 47	N/A	11	18.2
Kruljac, 2018 54	N/A	137	44.5
Ma, 2022 61	N/A	136	0.7
Masuda, 2022 63	N/A	73	1.4
Mondal, 2023 69	N/A	37	8.1
Ndebele, 2018 192	N/A	41	29.3
Nunes, 2021 c	N/A	9	22.2
Patel, 2021 74	N/A	26	61.5
Thewjitcharoen, 2019 86	N/A	47	8.5
Tiwari, 2017 194	N/A	50	4.0
Tomar, 2022 88	N/A	23	4.3
Wong, 2016 92	N/A	16	0.0
Xu, 2016 95	N/A	294	3.4
Yo, 2014 96	N/A	27	29.6

Abbreviations: CCAE, IBM MarketScan Commercial Claims and Encounters; COVID‐19, coronavirus disease 2019; DKA, diabetic ketoacidosis; DPP‐4i, dipeptidyl peptidase‐4 inhibitor; MDCD, IBM MarketScan Multi‐State Medicaid Database; MDCR, IBM MarketScan Medicare Supplemental and Coordination of Benefits; N/A, not applicable; NR, not reported; SGLT2i, sodium–glucose cotransporter‐2 inhibitor; T2DM, type 2 diabetes mellitus.

^a Non‐treatment subgroups (e.g., by age, by sex) were defined after the protocol stage and were, therefore, not specified in the protocol or the DET. Data are added to the report where relevant, but this is a limitation of the SLR and some data for these subgroups may have been missed.

^b See publication for long‐term all‐cause mortality.

^c See publication for 2‐year mortality.

Fifteen publications reported DKA‐related complications. ¹¹ , ¹² , ²⁵ , ²⁶ , ²⁷ , ³⁷ , ⁴⁹ , ⁶⁹ , ⁷¹ , ⁷³ , ⁸¹ , ⁹² , ⁹⁵ , ¹⁷⁹ , ¹⁹⁶ Common complications occurring during/after DKA were acute kidney injury or failure (12.6%–64.3%; 4 publications), ⁶⁹ , ⁸¹ , ⁹⁵ , ¹⁹⁶ hypokalaemia (12.3%–56%; 4 publications), ¹¹ , ²⁵ , ³⁷ , ⁹² and arrhythmias (10.8%–28.6%; 2 publications). ⁶⁹ , ⁹⁵ In the largest of the 15 studies (N = 19 675), the most common complication was acute kidney failure (42.9%). ¹⁹⁶

3.5. Economic burden

Medical costs for DKA were reported in 8 publications (China, 3; Indonesia, 1; USA, 4; Table S7). Some publications noted that a high prevalence or severity of comorbidities in patients with T2DM and DKA may contribute to high costs. ¹⁰⁰ , ¹⁹⁶ In a study of hospitalized patients in China with T2DM and diabetic complications (DKA, coma, peripheral neuropathy, diabetic kidney disease, eye disorder, foot damage, cerebrovascular disease, cardiovascular disease and peripheral vascular disease), subgroups with and without DKA showed no difference in total hospitalization costs (median 7794.17 RMB with DKA vs. 8020.00 RMB without DKA), whereas the presence of ≥4 complications significantly increased costs (median 12177.62 RMB with ≥4 complications vs. 6349.72 RMB without complications). ¹⁰⁶ In the USA, mean hospital costs for patients hospitalized with DKA were reported to be 36 000 USD ¹⁹⁶ and 68 927 USD. ¹⁰⁰ In a population covered under a Medicaid insurance plan, annual DKA hospital admission costs per plan member (averaged across all members with and without DKA hospital admissions) were estimated to be 500 USD in those using a FreeStyle Libre glucose monitoring system and 883 USD in those using blood glucose monitoring. ¹⁹⁷

The median length of hospital stay in patients hospitalized with DKA ranged from 65.1 h (~2.7 days) to 30.0 days (19 publications), ¹¹ , ¹⁵ , ¹⁷ , ³⁷ , ³⁹ , ⁴¹ , ⁴² , ⁴⁴ , ⁴⁹ , ⁶¹ , ⁶⁴ , ⁷³ , ⁸¹ , ⁸⁶ , ⁹² , ⁹⁵ , ¹⁹¹ , ¹⁹³ , ¹⁹⁸ while the mean length of hospital stay was 2.2–24.6 days (13 publications). ¹² , ¹⁶ , ⁴⁷ , ⁵⁹ , ⁶⁹ , ⁷¹ , ⁷⁷ , ⁸⁸ , ⁹³ , ¹⁰⁰ , ¹⁷⁹ , ¹⁹⁶ , ¹⁹⁹ In the USA, the length of stay ranged from 2.7 to 30.0 days (both median values). ³⁹ , ⁴⁴ , ⁸¹ , ¹⁰⁰ , ¹⁹⁶ , ¹⁹⁹ Outside of the USA, the length of stay showed a similarly broad range, from 2.2 to 24.6 days (both mean values). ¹¹ , ¹² , ¹⁵ , ¹⁶ , ¹⁷ , ³⁷ , ⁴¹ , ⁴² , ⁴⁷ , ⁴⁹ , ⁵⁹ , ⁶¹ , ⁶⁴ , ⁶⁹ , ⁷¹ , ⁷³ , ⁷⁷ , ⁸⁶ , ⁸⁸ , ⁹² , ⁹³ , ⁹⁵ , ¹⁷⁹ , ¹⁹¹ , ¹⁹³ , ¹⁹⁸ Overall, the shortest hospital stays were reported in youths admitted with isolated DKA (without severe hyperglycaemia or hyperosmolality) ⁸¹ and in adult patients treated with SGLT2is who had been diagnosed with T2DM ≥3 years before the admission with DKA and who had not received insulin treatment within 3 years of their diabetes diagnosis. ⁷¹ The longest hospital stays were reported in adult patients with COVID‐19 being treated with a continuous insulin infusion for DKA ³⁹ and in adult patients who had post‐operative DKA after elective or emergency surgery without pre‐operative SGLT2i treatment. ⁵⁹

3.6. Humanistic burden

There is an evidence gap for information on humanistic burden from DKA. The one study identified suggested that a third of patients with DKA need ‘total assistance’ for daily living before being admitted to hospital (Figure S6); however, no details of the impact of DKA on health‐related quality of life were reported. ¹⁸²

4. DISCUSSION

Our SLR shows a high clinical burden of DKA in patients with T2DM. The included studies generally used a consistent definition of DKA, but DKA incidences ranged widely. Populations at increased risk of DKA included patients using SGLT2is, those using insulin and those with poor glycaemic control. The most common precipitating factors for DKA were infection (of any type) and non‐adherence to treatment. Despite limited evidence on the impact of DKA on patients' health‐related quality of life, the results highlight a high burden of complications. In addition, the length of hospital stay ranges from days to several weeks.

DKA has been considered an occasional complication of T2DM, ²⁰⁰ but our SLR results indicate that it occurs more frequently than has generally been perceived, especially in certain subgroups such as SGLT2i users. Consistent with our findings, two recent meta‐analyses of real‐world studies of patients with T2DM showed a significantly increased risk of DKA associated with the use of SGLT2is. ²⁰¹ , ²⁰² SGLT2is lower blood glucose by inhibiting glucose reabsorption in the kidneys. ²⁰³ The mechanisms underlying the increased risk of DKA in SGLT2i users are unclear, but dehydration and insulinopenia may contribute, with dehydration provoking increases in glucocorticoid and catecholamine concentrations, leading to adipose tissue lipolysis in the setting of insulinopenia. ²⁰⁴ Decreased insulin levels have also been suggested to divert hepatic free fatty acids towards beta‐oxidation by increasing free fatty acid flux into the mitochondria via carnitine palmitoyltransferase 1. ²⁰⁵ The risk of DKA must be weighed against the benefits associated with SGLT2i use—as well as lowering blood glucose levels, SGLT2is have been shown to reduce cardiovascular risk in patients with T2DM. ²⁰⁶ SGLT2is (and/or glucagon‐like peptide‐1 receptor agonists) with demonstrated cardiovascular benefit are recommended by the ADA for patients with T2DM and established atherosclerotic cardiovascular disease or indicators of high atherosclerotic cardiovascular disease risk, heart failure or chronic kidney disease, with consideration of patient‐specific factors. ²⁰⁷

We also found good evidence that insulin users have a higher incidence of DKA than non‐users. DKA in these patients may be caused by missed or inadequate doses of insulin. ⁸ Consistent with this, poor glycaemic control was identified as a risk factor for DKA in our SLR and an international consensus statement. ⁵ Insulin use may also signify more severe T2DM (with reduced endogenous insulin secretion) or longer disease duration, which may contribute to the increased risk of DKA in this group. ⁵⁶ , ⁵⁹

The highest prevalence of DKA in our SLR (50.0%) was in a cohort with ketosis‐prone T2DM. ⁹⁷ A recent SLR by Kovacs et al. showed that ketosis‐prone T2DM accounts for one third of DKA or ketosis cases at the onset of diabetes in adults. ²⁰⁸ Most of the studies in that SLR were conducted in Asian and African countries, ²⁰⁸ but one focused on a Caucasian population and found ketosis‐prone T2DM in 23% of new‐onset cases with ketosis, ²⁰⁹ indicating that ketosis‐prone T2DM also needs to be considered in Caucasian patients. ²⁰⁸

We found no clear relationship between the incidence of DKA and duration of T2DM. Limitations of these results include the reporting of T2DM duration as mean/median across studies and the presence of potential confounding factors. However, these results align with one publication that reported no significant difference in DKA incidence according to diabetes duration (mixed T1DM and T2DM) ³⁸ and another publication reporting no significant effect of diabetes duration on the risk of post‐operative DKA in SGLT2i users. ⁵⁹

Infection is a common trigger for DKA, as is non‐adherence to treatment (insulin or other medications). Infection and omission or insufficient use of insulin therapy were also identified as common triggers for DKA in an international consensus statement, ⁵ with infection being the most common precipitating factor worldwide. However, precipitating factors for DKA often remain unknown; more than half of the publications in this SLR that reported precipitating factors for DKA did not identify the trigger in some patients.

The length of stay in hospital with DKA varied widely in the studies included in our SLR. The greatest durations were reported in patient populations with other factors that likely contributed to the high length of stay (e.g., COVID‐19 or surgical treatment). The shortest durations (median 2.7 days or mean 2.2 days) are close to the mean length of stay reported in an international consensus statement (3.7 days). ⁵

While continuous glucose monitors have reduced the incidence of DKA, ¹²⁶ , ¹²⁷ there is still an unmet need for a convenient and reliable ketone monitoring tool. In 2021, a pilot study demonstrated the feasibility of continuous ketone monitoring. ⁷ Continuous glucose monitoring is already part of the ADA standards of care for patients treated with insulin and is widely available in many countries. ²¹⁰ Combining continuous glucose and ketone monitoring in a single sensor would simplify the detection of ketone levels and reduce the risk of progression to full DKA. ³ In addition to patients receiving insulin, those with other risk factors such as SGLT2i treatment, recurrent infections or ketosis‐prone T2DM may benefit. Further research may identify other subgroups that would benefit from continuous ketone monitoring.

Strengths of this SLR include its broad scope, systematic approach and large volume of data on DKA prevalence and incidence. A limitation is the variation in study designs, data sources, geographic locations (which may have differences in access to healthcare) and patient characteristics across the included studies. These differences may have contributed to the wide range of reported prevalence and incidence values for DKA. In addition, some publications reported prevalence and incidence for subgroups of patients (e.g., by treatment type), but not for the overall cohorts. These subgroups are included in the results, meaning that upper and lower bounds of ranges might be derived from specific subgroups that are at a higher and lower risk of DKA, respectively. Data for post hoc subgroups of interest were only extracted from publications in which data for the overall cohort were not reported. As such, data for post hoc subgroups were extracted and reported non‐systematically and may be incomplete. The studies included in our SLR are unlikely to capture the incidence of mild/euglycemic DKA that occurs in routine clinical practice which is frequently missed. They are also unlikely to capture the incidence of DKA versus hyperglycaemic hyperosmolar state with ketosis, which may be difficult to distinguish. Socioeconomic risk factors for DKA (e.g., low income, area‐level deprivation, housing insecurity, underinsurance or lack of insurance, and food insecurity ⁵ ) were not a focus of this SLR. SGLT2i users were compared with four distinct control groups; evidence for each comparison is limited. The included studies were not assessed for quality or risk of bias in this broad‐ranging SLR. There may be missing data in retrospective studies using medical charts, patients may be lost to follow‐up in prospective studies, and interventional studies often have stringent eligibility criteria which may mean that the study population is not representative of the general population. The screening of references by a single reviewer may have introduced bias; however, checking of a sample of the title/abstract screening decisions by a second reviewer suggests this is unlikely.

DKA in T2DM has been generally overlooked and this is reflected in the identified evidence gaps. In many cases the cause of DKA was unknown, highlighting a level of uncertainty about the aetiology of DKA. In addition, the humanistic burden and costs associated with DKA were assessed in only a few identified studies. The results of our SLR provide a rationale for further research into the causes of DKA and its humanistic and economic burden in patients with T2DM.

In conclusion, our SLR provides an overview of the epidemiology of DKA and its implications in patients with T2DM. Our findings indicate a high clinical burden of DKA in these patients. Clinicians should therefore be alert to the possibility of DKA in patients with T2DM. Further research is needed to improve our understanding of DKA and its associated complications, identify high‐risk patients and those with poor outcomes, and develop tools for convenient monitoring of ketones. This would allow for optimization of established treatments and early detection of ketone elevation, alleviating the burden of DKA in patients with T2DM and the healthcare system by preventing its onset altogether.

AUTHOR CONTRIBUTIONS

Carol Wysham, Anila Bindal and Yeesha Poon made substantial contributions to the conception and design and the analysis and interpretation of data; revised the manuscript critically for important intellectual content; gave final approval of the version to be published; and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Fleur Levrat‐Guillen revised the manuscript critically for important intellectual content; gave final approval of the version to be published; and agrees to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Desislava Kostadinova made substantial contributions to the acquisition, analysis and interpretation of data; was involved in drafting the manuscript; gave final approval of the version to be published; and agrees to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

FUNDING INFORMATION

This study was funded by Abbott.

CONFLICT OF INTEREST STATEMENT

Carol Wysham has received research funding from Abbott, AbbVie, Bayer, Eli Lilly and Company, and Novo Nordisk, and has served as a speaker/advisor for Abbott, Biomea, Eli Lilly and Company, MannKind, and Novo Nordisk. Anila Bindal, Fleur Levrat‐Guillen and Yeesha Poon are employees of Abbott. Desislava Kostadinova is an employee of Oxford PharmaGenesis, which received funding from Abbott to conduct the SLR.

Supporting information

Data S1. Supporting Information.

Wysham C, Bindal A, Levrat‐Guillen F, Kostadinova D, Poon Y. A systematic literature review on the burden of diabetic ketoacidosis in type 2 diabetes mellitus. Diabetes Obes Metab. 2025;27(5):2750‐2767. doi: 10.1111/dom.16282

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.