One third of cases of new‐onset diabetic ketosis in adults are associated with ketosis‐prone type 2 diabetes—A systematic review and meta‐analysis

Ketosis‐prone type 2 diabetes was defined by the World Health Organization in 2019. According to the literature, the diagnosis is based on the presence of ketosis, islet autoantibody negativity and preserved insulin secretion. Our meta‐analysis assessed the prevalence and clinical characteristics of ketosis‐prone type 2 diabetes among patients hospitalised with diabetic ketoacidosis (DKA) or ketosis.


| INTRODUCTION
Diabetes mellitus is now acknowledged as a spectrum disease: overlapping syndromes of classical type 1 and type 2 diabetes exist with considerable heterogeneity. 1 Diabetes classification in clinical practice is strongly based on clinical features, such as age at onset, body mass index (BMI) and diabetic ketoacidosis (DKA), which may be insufficient for adequate long-term treatment. 1abetic ketoacidosis has been traditionally thought to indicate type 1 diabetes and to only occur occasionally without provoking conditions in type 2 diabetes. 2However, the recent classification by the World Health Organization (WHO) introduced ketosis-prone type 2 diabetes as one of the hybrid forms of diabetes. 3Based on the literature, it is defined by the onset of ketoacidosis or ketosis with the absence of diabetes-related autoantibodies and with preserved insulin secretion. 4Insulin administration is mandatory in the management of DKA.However, as regards long-term treatment, unlike in type 1 diabetes, patients with ketosis-prone type 2 diabetes may not require insulin therapy. 4Therefore, the first step in individualising the glycaemic control strategy in diabetes is to establish its taxonomic frame accurately.
Nevertheless, the current classification provided by the American Diabetes Association does not include ketosis-prone type 2 diabetes, but a similar entity-idiopathic type 1 diabetes-affecting individuals of African or Asian ancestry, who lack autoimmune markers and show varying degrees of beta cell insufficiency, resulting in ketosis proneness. 2Moreover, a substantial European cohort study from Sweden, which involved 14,755 newly diagnosed adult patients and aimed to refine diabetes classification to allow treatment individualisation, also reported no ketosis-prone type 2 diabetes. 5tosis-prone type 2 diabetes is mainly diagnosed in patients of non-Caucasian ancestry 6 ; it accounts for up to one half of the newonset DKA cases among Afro-Americans and Hispanics. 7,8[11] A recent publication showed that one fifth of adult patients admitted for DKA were newly diagnosed with diabetes.However, patients with type 2 diabetes were excluded from this study. 12The major diagnostic challenge in adult patients with new-onset diabetes presenting with ketoacidosis or ketosis is to elucidate whether it is caused by type 1 or ketosis-prone type 2 diabetes.It is essential to increase awareness and facilitate an understanding of ketosis-prone type 2 diabetes among clinicians, as it is still a new entity with an incompletely characterised prevalence and clinical course.Therefore, we conducted the first meta-analysis to assess the prevalence and describe the characteristics of ketosis-prone type 2 diabetes among patients presenting with DKA or ketosis.

| METHODS
We report our systematic review and meta-analysis based on the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guideline. 13We followed the Cochrane Handbook guidance for study conceptualisation. 14The study protocol was registered on the International Prospective Register of Systematic Reviews before initiation (registration No: CRD42021283369).The protocol was implemented without alterations.

| Eligibility criteria
We used the CoCoPop framework for prevalence outcomes and the PECO framework for clinical characteristic outcomes to formulate our clinical questions and establish eligibility criteria.In the former,

| Sources of information
A systematic search was performed on 15 October 2021 in five databases-MEDLINE, Embase, Web of Science, Scopus, and CENTRAL-without filters (Supplementary Material/Methods/Search strategy).We also manually screened the reference lists of eligible articles to check for further eligible reports.

| Selection process
The pool of articles provided by the advanced search was downloaded to the Endnote citation manager (Clarivate Analytics, Philadelphia, PA, USA).Duplicates were removed first automatically and then manually by one researcher (AK).The selection was performed by two independent review authors (AK and DP).Cohen's Kappa coefficient (κ) was calculated to evaluate inter-rater agreement during each selection step (Supplementary Material/Methods//Selection process).Disagreements were resolved by a third independent investigator (SB) at each selection step.

| Data collection process
Data were extracted by one review author (AK) and checked by a second reviewer (DP) to ensure precision.In the case of overlapping populations across studies, we selected articles which had enrolled a relatively high number of patients and reported the outcomes noted above.

| Data items
The following data were extracted from each eligible article and collected in a standardized Excel document (Microsoft Corporation, Redmond, WA, USA): publication details-authors, year of publication, country of origin and digital object identifier; study characteristics-study type and sample size; prevalence assessment -number of patients in each group; patients' characteristics-age, sex and BMI; laboratory findings-fasting and/or stimulated Cpeptide and haemoglobin A 1c levels on admission and at 1-year follow-up, insulin dependence at 1-year follow-up and serum lipid levels on admission; and family history of any kind of diabetes.If necessary, we converted the measurement units as follows: Cpeptide-level results were extracted in ng/mL and lipid parameters in mmol/L.

| Study risk-of-bias assessment
Two authors (AK and DP) independently performed the risk-of-bias assessment of the eligible studies.Disagreements were resolved by consensus.We used the Joanna Briggs Institute Critical Appraisal Tools (JBI Critical Appraisal Checklist for Studies Reporting Prevalence Data) for the prevalence outcomes and the Quality In Prognosis Studies tool for the clinical characteristics. 16,17

| Effect measures
The overall effect size was expressed by prevalence and odds ratio (OR) with 95% confidence intervals (CI) for categorical outcomes.
The OR was calculated as odds in the ketosis-prone type 2 group divided by odds in the type 1 diabetes group.The summary measure was reported by mean difference (MD) with CIs for continuous outcomes.Mean difference was calculated by subtracting the mean for the type 1 diabetes group from the mean for the ketosis-prone type 2 diabetes group.

| Synthesis methods
As we anticipated considerable between-study heterogeneity, a random effects model was used to pool effect sizes.Results were statistically significant if the 95% CI did not contain the null value (which represented a 5% significance level).Between-study heterogeneity was described by Higgins and Thompson's I 2 statistics.The results are graphically summarised in forest plots.Influential analyses were carried out by performing leave-one-out analyses.Publication bias was assessed by Egger's test (at a significance level of 10%) and by visual inspection of the funnel plots.All statistical analyses were conducted with R (R Core Team 2021, v4.1.1)9][20] We report 95% CI abbreviated as CI (see further details in Supplementary Material/Methods/ Synthesis methods).

| Certainty assessment
The certainty of the evidence was assessed by the GRADEpro Guideline Development Tool as recommended by the Cochrane Handbook. 21

| Study selection and characteristics
Out of the 16,962 records found, 9278 hits were excluded based on title and abstract contents after duplicates were removed, and 106 full texts were assessed for eligibility.Finally, 11 articles fulfilled our eligibility criteria and were selected for systematic review and metaanalysis (Figure S13).The inter-rater agreement was characterised as κ = 0.69 and κ = 0.60 for title and abstract and for full-text selection, respectively.Altogether, data from 2010 patients were analysed in our study.Most of the selected studies were observational and conducted in Asian [22][23][24][25][26][27][28] and African 29,30 countries.One study reported on patients of Caucasian ethnicity, 10 and one originated from the USA, 31 where mainly African-Americans and Hispanic patients KOVACS ET AL.
were enrolled.In Barkai et al, 10 Nalini et al 31 and Xu et al, 28 only the prevalence outcomes were assessed, since the first two did not report on the clinical characteristics of the comparison group (type 1 diabetes), while the latter reported on the features of an overall atypical diabetes group, from which only the new-onset cases were eligible for the ketosis-prone type 2 diabetes definition (Table S1).
For better visualisation, we summarised the results in Table 1 for outcomes which could not be assessed quantitatively due to limited data.26,30 At 1 year after onset of disease, both fasting and stimulated C-peptide levels were in the normal range and significantly higher in patients with ketosis-prone type 2 diabetes. 23,25In accordance, better glycaemic control was reported in the ketosis-prone type 2 diabetes group at 1 year. 23,25Long-term insulin treatment was shown in one study: only 23% of patients with ketosis-prone type 2 diabetes were still on insulin therapy at 1-year follow-up. 25e results of the sensitivity analyses are summarised in the Supplementary Material/Results/Influential analysis and Publication bias assessment.Although some potential and influential studies were identified based on statistical parameters, they did not change the pooled effect sizes in a relevant manner.

| Risk-of-bias assessment
Most of the studies involved in the prevalence analyses are at high risk of bias due to the tertiary centre study setting and the F I G U R E 1 One third of patients admitted with diabetic ketoacidosis or ketosis at the onset of diabetes had ketosis-prone type 2 diabetes.Overall effect size expressed by prevalence with 95% confidence interval.Heterogeneity estimated with I 2 statistics with 95% confidence interval.DKA, diabetic ketoacidosis; KPT2D, ketosis-prone type 2 diabetes mellitus.small number of participants (Figure S20).With regard to clinical characteristics, we found a low overall risk of bias except for the family history of diabetes (Figures S21-S27).In this case, the risk of bias was moderate because of the uncertainty of the method used to assess the family history of diabetes in the different articles.

| Certainty of evidence
Because of the observational study design, high between-study heterogeneity and low number of enrolled patients, the certainty of the evidence is very low for every outcome (Table S2).

| DISCUSSION
We conducted the first meta-analysis on the prevalence of ketosisprone type 2 diabetes among patients with DKA or ketosis.Our results revealed that 35% of the DKA or ketosis cases at the onset of disease are not caused by type 1, but rather by ketosis-prone type 2 diabetes.

New-onset diabetes presenting with ketosis in adult patients
requires differentiation between type 1 and ketosis-prone type 2 diabetes.Our meta-analyses showed that patients with ketosis-prone type 2 diabetes were older by 11.5 years and had a higher BMI by 5.48 kg/m 2 compared with those with type 1 diabetes.In accordance with our findings, the importance of phenotypic data in diabetes classification was also highlighted in a recent review by Herder and Roden. 32sed on these two phenotypic parameters in a new diagnosis of diabetes with the onset of keto(acido)sis, clinicians should be aware of the possibility of ketosis-prone type 2 diabetes and should assess diabetes-related autoimmune markers and C-peptides.
The WHO classification of diabetes mellitus in 2019 defined a new entity: ketosis-prone type 2 diabetes. 3According to the literature, it is characterised by unprovoked ketosis at the onset of diabetes with autoimmune negativity and preserved insulin secretion. 4eviously, ketosis-prone type 2 diabetes was classified as idiopathic type 1 diabetes, since both forms of diabetes share similar clinical characteristics, such as occurrence in primarily African and Asian populations and showing intermittent ketosis in some cases. 33though idiopathic type 1 diabetes has been a taxonomic unit in the diabetes spectrum for decades and repeatedly included in the ADA F I G U R E 2 Patients with ketosis-prone type 2 diabetes were older than those with type 1 diabetes.Overall effect size expressed as mean difference with 95% confidence interval.Heterogeneity estimated with I 2 statistics with 95% confidence interval.MD, mean difference; KPT2D, ketosis-prone type 2 diabetes mellitus; T1D, type 1 diabetes mellitus.

F I G U R E 3
Higher body mass index (BMI) in ketosis-prone type 2 diabetes compared with type 1 diabetes.Overall effect size expressed as mean difference with 95% confidence interval.Heterogeneity estimated with I 2 statistics with 95% confidence interval.MD, mean difference; KPT2D, ketosis-prone type 2 diabetes mellitus; T1D, type 1 diabetes mellitus.classification, there is growing evidence of a potential overlap with ketosis-prone type 2 diabetes. 6Before the recent WHO classification, ketosis-prone type 2 diabetes had been referred to as Flatbush diabetes, ketosis-prone diabetes mellitus or, lately, "A-ßþ" ("A-": negative islet cell antibodies; "ßþ": preserved beta cell function) diabetes. 4 autoantibody-negative, insulin-dependent diabetes subtype has previously been described in Tanzania and recently in Uganda. 34,35The latter showed that approximately one third of newly diagnosed adult sub-Saharan African patients with diabetes with negative islet autoimmunity pattern are lean and have reduced insulin secretory capacity. 35These cases are categorised as idiopathic type 1 diabetes and differ from ketosis-prone type 2 diabetes.
Although the pathophysiology of ketosis in ketosis-prone type 2 diabetes is not understood comprehensively, the most relevant theories reported amino and fatty acid metabolic deteriorations combined with the impairment of ketone oxidation as a cause of ketosis proneness. 36In 2016, Schwartz et al proposed a beta cell-centric classification of diabetes. 37 addition to differentiating ketosis-prone type 2 diabetes from autoimmune type 1 diabetes, it is essential to distinguish it from the "real" idiopathic type 1 diabetes characterised by a negative islet antibody pattern and insulin deficiency as well as from another subtype of type 1 diabetes-fulminant type 1 diabetes-in which ketosis is caused by a rapid, non-autoimmune-mediated beta-cell destruction. 38A B L E 1 A systematic review of lipid parameters, beta cell function, long-term glycaemic control and insulin dependence.Undoubtedly, at onset, ketosis-prone type 2 diabetes mainly resembles type 1 diabetes.However, during the natural course of the disease, it seems to be closer to "typical" type 2 diabetes. 8These similarities are most probably due to the clinical phenotype of obesity and hepatic steatosis, resulting in insulin resistance. 39Glucoselowering medications with high efficacy for weight reduction, such as GLP-1 RAs and GIP/GLP-1 RA, are therefore also promising therapeutic options for long-term, insulin-free management of ketosis-prone type 2 diabetes. 4The use of SGLT2 inhibitors, on the other hand, requires precautious control, for example, urine and/or blood ketone self-monitoring, to avoid euglycemic ketoacidosis in this ab ovo ketosis-prone population. 4,40Insulin secretory capacity and insulin sensitivity seem to deteriorate comparably over the course of the disease in both ketosis-prone and "typical" type 2 diabetes. 8unger age at onset along with poor HbA 1c and reduced beta cell recovery within the first year after diagnosis may indicate a relatively early need for long-term insulin therapy in ketosis-prone type 2 diabetes. 41other recent suggestion for diabetes classification came from Sweden, where Ahlqvist et al performed a data-driven cluster analysis of clinical characteristics on a large cohort of newly diagnosed adult patients. 5Based on six variables (BMI, age at onset of diabetes, HbA 1c , homoeostatic model assessment two estimates of beta cell function and insulin resistance (HOMA2-B and HOMA2-IR, respectively) and the presence of glutamate decarboxylase antibodies), they identified five diabetes types-severe autoimmune diabetes, severe insulin-deficient diabetes (SIDD), severe insulin-resistant diabetes (SIRD), mild obesity-related diabetes (MOD) and mild age-related diabetes (MARD). 5However, ketosis proneness was not evaluated in this survey.

Outcome
There was only one study in our meta-analysis that analysed a population of Caucasian ethnicity exclusively; the prevalence of ketosis-prone type 2 diabetes was 23% (95% CI: 16%-33%) among new-onset cases with ketosis. 10Although the prevalence is lower than in other ethnic groups, this form of diabetes also needs to be considered in the Caucasian population.
Notable heterogeneity was observed in the prevalence analyses.
Since finding the reasons for these differences may reveal the risk factors for ketosis-prone type 2 diabetes, we investigated between-study heterogeneity.As we applied no age restriction for inclusion, studies with children and adolescents were also involved.Based on the available data, we assumed that a younger age at diagnosis was associated with a higher prevalence of ketosis-prone type 2 diabetes, as reported by Du et al, 22 Liu et al 24 and Ekpebegh et al. 29 The phenomenon seen in these three studies can be explained with two arguments: firstly, the prevalence of typical immune-mediated type 1 diabetes is low in non-Caucasians where ketosis-prone type 2 diabetes is more frequent, and secondly, ketosis-prone type 2 diabetes is usually diagnosed at an earlier age compared to type 2 diabetes. 6,42 has repeatedly been shown that there is a trend towards male predominance in ketosis-prone type 2 diabetes. 6,87][28] In these studies, the prevalence may have been underestimated as the classification was not exclusively based on the assessment of betacell function and the presence of islet cell autoantibodies but also on treatment history.
Our study is the first to quantitatively address the prevalence of ketosis-prone type 2 diabetes among DKA or ketosis cases in a metaanalysis and to identify the clinical features of this type of diabetes compared with type 1 diabetes (Table 2).We applied a robust methodology following the recommendations of the Cochrane collaboration 14 and the PRISMA 2020 Statement. 13wever, there are some major limitations in our study.population of patients with DKA or ketosis.As our eligibility criteria contained no age restriction, we also report data on children and adolescents (which were the minority).It was not possible to perform a subgroup analysis, as data were not provided separately on different age groups in the primary studies.Lastly, we are aware of the limitations that arise from the observational study design and the overall low number of participants in the eligible articles.
Diagnosis of diabetes mellitus is based on blood glucose level and/or HbA 1c . 2,3The treatment is established according to the type of diabetes and the accompanying metabolic status.Though DKA or ketosis requires prompt insulin therapy, long-term treatment depends on disease type, therefore an accurate classification is essential. 43e Co(ndition) was diabetic keto(acido)sis, Co(ntext) referred to ketosisprone type 2 diabetes, and Pop(ulation) involved patients with diabetes mellitus.For the clinical characteristics, the population (P) consisted of patients with diabetes mellitus hospitalised for DKA or ketosis.The Exposed group (E) included patients with ketosis-prone type 2 diabetes, while the Comparison group (C) consisted of type 1 diabetes cases.The assessed outcomes (O) encompassed the difference in clinical characteristics between the two groups as reported in each study, including demographic data, family history of diabetes, BMI, the presence of endogenous insulin secretion, quality of glycaemic control, lipid profile and long-term need for insulin treatment.Ketosis-prone type 2 diabetes was defined as newly diagnosed patients with DKA (based on American Diabetes Association (ADA) criteria; Supplementary Material/Methods/Eligibility criteria) 15 or ketosis (established by ketonuria or serum ß-hydroxybutyrate) at the onset of disease without any diabetes-related autoantibodies and with preserved insulin secretion measured by fasting and/or stimulated serum C-peptide level (C-peptide in the given reference range).No age restriction was applied; therefore, children and adolescents were also included.Studies in which there were no data on C-peptide level or autoantibody status were excluded.With regard to study design, systematic reviews and case reports were excluded.

T A B L E 2
Abbreviations: DKA, diabetic ketoacidosis; KPT2D, ketosis-prone type 2 diabetes mellitus; T1D, type 1 diabetes mellitus.a Data based on present meta-analysis results.
updated WHO classification has introduced a new concept of hybrid forms of diabetes based on the overlapping clinical phenotypes of type 1 and type 2 diabetes. 3However, regarding the pathomechanism, one of the hybrid forms-slowly evolving, immunemediated diabetes in adults-is classified as type 1 diabetes along with the classical autoimmune-mediated, idiopathic and fulminant subtypes.On the other hand, the other hybrid form-ketosis-prone type 2 diabetes-is rather categorised as type 2 diabetes along with other subtypes, such as SIDD, SIRD, MOD and MARD.The new WHO hybrid category may increase clinicians' awareness of differential diagnosis.5| CONCLUSIONWe have shown that ketosis-prone type 2 diabetes accounts for one third of DKA or ketosis cases at the onset of diabetes in adults.To differentiate it from type 1 diabetes, we suggest an integrative approach in all new-onset cases with DKA or ketosis, including islet autoantibody and C-peptide assessment.Patients with ketosis-prone type 2 diabetes were found to be in their third to fourth decade of life at the time of the diagnosis and were categorised as overweight or obese based on BMI.Long-term follow-up studies on ketosisprone type 2 diabetes are needed for a better understanding and better management of the disease.