Impacts of COVID-19 on Glycemia and Risk of Diabetic Ketoacidosis

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was recently demonstrated in two separate studies to infect pancreatic β-cells in vitro (1,2), raising concerns as to whether SARS-CoV-2 infection (coronavirus disease 2019 [COVID-19]) may impact glycemic control (3). Type 2 diabetes (T2D) is an established risk factor for COVID-19 severity, and preinfection hemoglobin A_1c (HbA_1c) was shown to be an important predictor of COVID-19 severity in multiple studies (4–6). Although HbA_1c has been reported as an important risk factor for COVID-19, the impact of COVID-19 on HbA_1c is unknown (6–8). Previous studies have indicated that COVID-19 may increase HbA_1c postinfection, but the cohorts were small and lacked COVID-19(−) control subjects for comparison (9,10). Furthermore, case studies of patients developing insulin-dependent diabetes and diabetic ketoacidosis (DKA) after COVID-19 have been reported (11,12). Misra et al. (13) demonstrated an increase in hospitalization for DKA in patients with T2D and a reduction in DKA hospitalization in patients with type 1 diabetes (T1D) during the course of the COVID-19 pandemic. However, this study also lacked a negative control population to determine whether the risk was increased in patients who were infected with SARS-CoV-2 or was due to other factors. Here, we present the first evidence of COVID-19 impacts on HbA_1c in a large, real-world clinical cohort of >8,500 patients by testing the following hypotheses: 1) COVID-19 infection increases HbA_1c, 2) COVID-19 infection increases risk of developing T2D, and 3) COVID-19 infection increases risk of DKA.

We used the institutional review board–approved COVID-19 data registry established by Cleveland Clinic, which contains data from 81,093 patients tested at the Cleveland Clinic who were COVID-19(+) between March 2020 and May 2021 (Supplementary Table 1). In addition, the Cleveland Clinic registry included 153,034 matched patients who tested COVID-19(−) during the same time-period. A 1:2 matched case:control approach was implemented to define a matched cohort of COVID-19(−) patients and was performed based on location of the Cleveland Clinic facility, sex, race, and age within 5 years and in time slices of 14 days, but expanded to 21 days during the peak period of 8 November 2020 and 30 January 2021. Matching was performed with the R optmatch package (14), which allocates matching based on minimizing the aggregate sum of square differences. Sex and race are self-reported in the Cleveland Clinic’s electronic health record data. Eligible control subjects were defined as patients who tested COVID-19(−) and had no record of previously testing COVID-19(+). This is analogous to control subject selection for a grouped-time, nested case-control study, although the status of patients who do not report for testing during any given time slice is unknown.

HbA_1c from both pre– and post–COVID-19 test determined change in HbA_1c in COVID-19(+) and COVID-19(−) groups. To investigate the impact of using up to 12 months between HbA_1c testing and COVID-19 testing, we also analyzed two subcohorts for the same outcomes with a maximum time of 6 months and 3 months between HbA_1c record date and the date of COVID-19 test.

To evaluate the risk of DKA onset post–COVID-19 infection, we identified two subcohorts of patients with preexisting T1D. Patients with a prior history of DKA at any time point prior to their COVID-19 test were excluded from this cohort. Patients were determined to have T2D based on a modified form of the Electronic Medical Records and Genomics (eMERGE) algorithm, as described by Kho et al. (15) The eMERGE algorithm is designed to determine, with a high degree of specificity, which patients in an electronic health record have T2D based on a combination of ICD codes, medications, and laboratory values (e.g., blood glucose) and excludes patients with T1D. However, the original eMERGE algorithm used only ICD-9 codes, so the algorithm used here was modified to also include ICD-10 codes. T2D control patients were those who did not meet the eMERGE T2D criteria and had no history of T1D. Patients were considered to have T1D as per the Phenotype Algorithm for Type 1 Diabetes – eMERGE Phase-IV Program (16). The T2D(+) cohort was further stratified into groups based on baseline insulin usage (yes or no), race, and sex for DKA outcome assessment. Additionally, outcome of hyperosmolar hyperglycemic syndrome (HHS) was also assessed separately in T2D(+) cohort excluding patients with a prior history of HHS at any time prior to their COVID-19 test.

The following outcomes were investigated for association with HbA_1c in COVID-19(+) patients: T2D onset (based on eMERGE algorithm described above), hospitalization, admission to intensive care units (ICU), assisted breathing requirements, ventilation, and mortality. Additionally, DKA diagnosis was determined based on the ICD-9/ICD-10 codes (ICD- 9, 250.1X; ICD-10, E08.1X, E09.1X, E10.1X, E11.1X, and E13.1X) and had “above reference values” for both anion gap and β-hydroxybutyrate laboratory tests in addition to ICD-9/ICD-10 codes as an indicator for DKA. HHS was determined based on the ICD-9/ICD-10 codes (ICD-9, 250.2X; ICD-10, E11.0X).

Statistical analyses were performed with R, version 4.1.1 (17). Pre– and post–COVID-19 test HbA_1c values were compared between patients 1) with and without T2D and 2) with and without COVID-19(+) test results with paired Student t tests. Logistic regression was used to test for differences in new T2D diagnoses among COVID-19(+) and COVID-19(−) patients. We tested for association of HbA_1c with time to hospitalization, ventilation, assisted breathing, ICU admission, mortality, and T2D onset using Cox proportional hazards (CoxPH) models. Mortality was used as a competing risk in all CoxPH models, except for when mortality was the outcome. Restricted cubic spline CoxPH models were used to investigate nonlinear relationships between HbA_1c and outcomes of interest (18–20) (Supplementary Figs. 3–5 and Fig. 2). Age, sex, race, BMI, pre–COVID-19 HbA_1c, Charlson comorbidity index score (21), medications (including antidiabetes therapies prior to infection and during the study window), prior comorbidities, and the time intervals between the COVID-19 test and the pretest HbA_1c and the post-test HbA_1c were used as model covariates for all analyses. For DKA and HHS onset, when we did not include stratification by insulin use, we adjusted for insulin at baseline, in addition to the other covariates described above. We also performed stratified analyses of baseline HbA_1c associations with hospitalization, assisted breathing, ventilation, ICU admission, and COVID-19 test status regarding risk of DKA in the T2D(+) cohort based on six common drug hypoglycemia drug classes (i.e., biguanide, sulfonylureas, dipeptidyl peptidase 4 inhibitors [DPP4i], glucagon-like peptide 1 receptor agonists [GLP1RA], sodium–glucose cotransporter 2 inhibitors [SGLT2i], and thiazolidinediones [TZD]). HbA_1c changes >0.5% were considered clinically significant. All reported P have been adjusted for multiple comparisons with use of a false discovery rate approach, and adjusted P < 0.05 was considered statistically significant (22).

The data that support the findings of this study are available from the Cleveland Clinic COVID-19 registry, but restrictions apply to the availability of these data, which were used under institutional review board license for the current study and therefore are not publicly available. Data are, however, available from the authors on reasonable request.

Of the total COVID-19(+) cohort (N = 81,093) and the control COVID-19(−) cohort (N = 153,034), 8,755 (10.7%) and 11,998 (7.8%) patients, respectively, had at least one prior HbA_1c measurement available within the 12 months preceding their COVID-19 test and served as the primary cohort for analysis. We performed secondary analyses to determine the impact of this 12-month window by investigating associations based on the selection of 3- and 6-month windows. A total of 1,357 (1.7%) COVID-19(+) and 1,730 (1.1%) COVID-19(−) and 3,395 (4.2%) COVID-19(+) and 4,985 (3.3%) COVID-19(−) patients met inclusion criteria with use of the 3- and 6-month pre–COVID-19 test HbA_1c windows.

We identified 6,322 patients with T2D among the 81,093 who were COVID-19(+) who also had corresponding HbA_1c measurements for analysis. The patients in the COVID-19(+) and the matched COVID-19(−) cohorts (N = 11,998) were similar in age, income, BMI, race, and pre–COVID-19 HbA_1c (Table 1). Similar observations were observed in the DKA and HHS cohort (Supplementary Table 2).

A statistically significant, but clinically insignificant, increase in HbA_1c was observed after a COVID-19(+) test (ΔHbA_1c = 0.06%, P < 0.001), whereas COVID-19(−) patients did not demonstrate an increase in HbA_1c (ΔHbA_1c = 0.02%, P = 0.05) (Supplementary Fig. 1A). For HbA_1c change between pre– and post–COVID-19 testing there was a 0.08% greater HbA_1c increase in COVID-19(+) patients compared with COVID-19(−) patients (P < 0.001) (Supplementary Fig. 1B). These HbA_1c changes were of <0.5% and therefore were not considered clinically significant. After stratification of COVID-19(+) patients by T2D status, we still observed a significant increase postinfection in patients with and without T2D (ΔHbA_1c = 0.1%, P < 0.001, and ΔHbA_1c = 0.1, P < 0.001, respectively). Importantly, no significant increase was observed for COVID-19(−) cases after stratification by T2D status (P > 0.05) (Fig. 1). Restricting the pre–COVID-19 test HbA_1c time interval from 12 months to 6 and 3 months resulted in no statistically significant differences between pre– and post–COVID-19 HbA_1c data either overall or with stratification by T2D status (P > 0.05). For the 6-month window, among the COVID-19(+) patients, there was a 0.009% increase of HbA_1c from pre- to post-test, which was also not statistically significant (P = 0.427). Among the COVID-19(−) patients, there was a 0.022% increase from pre- to post-test phase (P = 0.17). Also, with stratification by T2D status, there was no significant increase after COVID-19 test among COVID-19(+) (Δ = 0.008%, P = 0.52) or COVID-19(−) (Δ = 0.03%, P = 0.25) patients with T2D. In patients without T2D, we did not observe any statistically significant increase for COVID-19(+) (Δ = 0.042%, P = 0.21) and COVID-19(−) (Δ = 0.014%, P = 0.35) patients. Similarly, using a 3-month window, we did not observe any significant increase post–COVID-19 test, in COVID-19(+) (Δ = 0.05%, P = 0.17) and COVID-19(−) (Δ = 0.05%, P = 0.08) patients overall. After stratification by T2D status, there was no significant increase post-test among the COVID-19(+) (Δ = 0.05%, P = 0.34) and COVID-19(−) (Δ = 0.07%, P = 0.24) patients with T2D. In patients without T2D also, we did not see any significant increase among COVID-19(+) (Δ = 0.06%, P = 0.34) and COVID-19(−) (Δ = 0.03%, P = 0.40) patients.

In comparison with COVID-19(−) patients, a 0.09% increase in postinfection HbA_1c was observed in COVID-19(+) patients with T2D (P < 0.001) (Supplementary Fig. 1C). In patients without T2D, a clinically insignificant increase in HbA_1c of 0.05% in COVID-19(+) patients compared with COVID-19(−) patients (P < 0.001) was observed (Supplementary Fig. 1C). Furthermore, the 0.09% HbA_1c increase observed in patients with T2D was statistically greater compared with patients without T2D (0.05%) (P = 0.002). Comparable results were obtained after restricting the pre–COVID-19 test HbA_1c time interval from 12 to 6 and 3 months. With the 6-month window, in comparison with COVID-19(−) patients, a 0.04% increase in postinfection HbA_1c was observed in COVID-19(+) patients with T2D (P = 0.002). However, in patients without T2D, statistically insignificant increase (Δ = 0.02%, P = 0.65) was observed for COVID-19(+) patients compared with the COVID-19(−) patients. With the 3-month window, however, no significant increase was observed: compared with COVID-19(−) patients, an insignificant 0.012% (P = 0.43) increase in postinfection HbA_1c was observed in COVID-19(+) patients with T2D and in patients without T2D, and an insignificant increase of 0.08% (P = 0.35) was observed for COVID-19(+) patients compared with the COVID-19(−) patients. Although the percent change of HbA_1c increased slightly with longer time windows, these percent changes were not statistically significantly different.

Although well reported in the literature, we also investigated whether pre–COVID-19 HbA_1c was associated with COVID-19 severity in our cohort (i.e., hospitalization, ventilation, assisted breathing, ICU admission). In the COVID-19(+) cohort of patients with T2D, we identified a significant positive association between pre–COVID-19 HbA_1c and time to hospitalization (hazard ratio [HR] 1.07 [95% CI 1.05–1.10], P < 0.001), assisted breathing (HR 1.06 [95% CI 1.02–1.09], P = 0.001), and admission to ICU (HR 1.05 [95% CI 1.00–1.11], P = 0.007) (Supplementary Fig. 2). We observed similar associations using restricted cubic spline nonlinear CoxPH models (Fig. 2). Restricting the pre–COVID-19 test HbA_1c time interval to 6 months [COVID-19(+) = 3,395 and COVID-19(−) = 4,985] and 3 months [COVID-19(+) = 1,357 and COVID-19(−) = 1,730] still resulted in significant associations with both hospitalization (HR_{6 months} 1.08 [95% CI 1.04–1.12], P_{6 months} <0.001, N_events = 1,031, and HR_{3 months} 1.06 [95% CI 1.003–1.12], P_{3 months} = 0.02, N_events = 451) and assisted breathing (HR_{6 months} = 1.08 [95% CI 1.036–1.138], P_{6 months} < 0.001, N_events = 650, and HR_{3 months} = 1.07 [95% CI 1.004–1.147], P_{3 months} = 0.02, N_events = 295). However, no significant associations were observed for ventilation and admission to ICU.

We also investigated whether preinfection HbA_1c associations with hospitalization, assisted breathing, ventilation, or ICU admission were modified by insulin use or any of six common diabetes medication classes (i.e., biguanide, sulfonylureas, DPP4i, GLP1RA, SGLT2i, TZD) (Supplementary Fig. 2). Associations with preinfection HbA_1c and assisted breathing and ventilation were not modified by drug class (P > 0.05) (Supplementary Fig. 2). However, the association between preinfection HbA_1c and increased risk of hospitalization was significantly higher in patients taking SGLT2i (P < 0.05). Furthermore, preinfection HbA_1c was significant higher and increased risk of ICU admission was significantly greater for patients on biguanide, GLP1RA, and SGLT2i (P < 0.05) (Supplementary Fig. 2).

Pre–COVID-19 HbA_1c was a significant risk factor for receiving a T2D diagnosis for COVID-19(+) patients (HR 4.08 [95% CI 2.81–5.34], P < 0.001) and COVID-19(−) patients (HR 1.34 [95% CI 1.04–1.65], P < 0.001), indicating that higher pre–COVID-19 test HbA_1c increased risk of receiving a T2D diagnosis regardless of COVID-19 status (Fig. 3A). Similarly, risk of T2D onset was higher in individuals with greater HbA_1c increase in COVID-19(+) (HR 1.74 [95% CI 1.58–1.93], P < 0.001) and COVID-19(−) patients (HR 1.46 [95% CI 1.32–1.59], P < 0.001) (Fig. 3B). However, COVID-19(+) patients were also more likely to receive a T2D diagnosis postinfection (N = 326/2,433) compared with COVID-19(−) patients (N = 380/6,594) (odds ratio 1.4, P < 0.001). Interestingly, the likelihood of being diagnosed with T2D, per unit change in HbA_1c, was also higher for COVID-19(+) patients than COVID-19(−) patients. Using the restricted cubic spline nonlinear CoxPH models demonstrated similar association patterns (Supplementary Fig. 3). We did not observe any significant associations between mortality preinfection glycemia or postinfection HbA_1c change in either the T2D(+) or T2D(−) cohorts (Fig. 3 and Supplementary Figs. 4 and 5). The nonparametric estimates also failed to show any significant trends (Supplementary Figs. 4 and 5).

Furthermore, restricting the pre–COVID-19 test HbA_1c time interval from 12 months to 6 months, the pre–COVID-19 HbA_1c was a significant risk factor for type 2 diabetes diagnosis for both COVID-19(+) (HR 1.87 [95% CI 1.57–2.24], P = 0.0061) and COVID-19(−) (HR 1.02 [95% CI 1.01–1.03], P = 0.004) patients. Similarly, increased HbA_1c post-test was also a significant contributor to T2D diagnosis risk among the COVID-19(+) (HR 3.21 [95% CI 1.89–5.00], P = 0.005) and COVID-19(−) (HR 1.57 [95% CI 1.31–1.88], P = 0.001) patients.

With the 3-month window, pre–COVID-19 HbA_1c was a significant risk factor for T2D diagnosis for both COVID-19(+) (HR 1.87 [95% CI 1.57–2.24], P = 0.006) and COVID-19(−) (HR 1.02 [95% CI 1.01–1.03], P = 0.004) patients. Similarly, increased HbA_1c post-test was also a significant contributor to T2D diagnosis risk among the COVID-19(+) (HR 3.21 [95% CI 1.89–5.00], P = 0.005) and COVID-19(−) (HR 1.57 [95% CI 1.31–1.88], P = 0.001) patients.

We evaluated the risk of DKA onset post–COVID-19 infection, in two subcohorts of patients, with pre-existing T1D (n = 701) and T2D (n = 21,830), out of the total COVID-19(+) (n = 81,093) and COVID-19(−) population (n = 153,034). In patients with T1D, there was no statistically significant difference in risk of developing DKA between those who were COVID-19(+) and those who were COVID-19(−) (HR 0.98 [95% CI 0.46–1.56], P = 0.53) (Fig. 4). Similarly, patients with T2D who were COVID-19(+) did not have a significant difference in risk of developing DKA compared with those who were COVID-19(−) (HR 1.38 [95% CI 0.79–2.41], P = 0.26) (Fig. 4).

Notably, Black patients who were COVID-19(+) with T2D were more than twice as likely to be diagnosed with DKA compared with White patients with T2D (HR 2.46 [95% CI 1.48–6.10], P = 0.04) (Fig. 5). There was no significant difference in DKA risk between males and females with T2D who were COVID-19(+) (HR 0.93 [95% CI 0.42–2.07], P = 0.86) or COVID-19(−) (HR 0.91 [95% CI 0.52–1.34], P = 0.73) (Fig. 5). Insulin use was also not significantly associated with DKA onset in either COVID-19(+) (HR 1.27 [95% CI 0.52–3.10], P = 0.60) or COVID-19(−) (HR 0.52 [95% CI 0.21–1.28], P = 0.15) patients (Fig. 5).

We further investigated whether risk of DKA in patients with T2D was differentially associated with six common classes of diabetes medications (i.e., biguanide, sulfonylureas, DPP4i, GLP1RA, SGLT2i, TZD). However, no significant associations between DKA onset and drug classes were observed (Figs. 6 and 7). No significant associations were observed between HHS (N = 163) and COVID-19 infection status in patients with T2D (N = 22,966) (Supplementary Fig. 6).

The impact of COVID-19 on HbA_1c and glycemic impairment is still poorly understood. Yang et al. (1) demonstrated that human pluripotent stem cell–derived and adult human pancreatic β-cells were permissive to SARS-CoV-2 infection in vitro. This was further supported by findings that SARS-CoV-2 was capable of infecting human pancreatic islets ex vivo and impaired insulin secretion in an in vitro model system (2). Further raising concerns that COVID-19 might impair glycemic control, Montefusco et al. (10) found that 46% of 551 patients hospitalized for COVID-19(+) were hyperglycemic, but this analysis lacked a COVID-19(−) population for comparison. A meta-analysis of 179 patients showed an increase in blood glucose in COVID-19(+) patients that failed to reach statistical significance (9). Other forms of glycemic impairment have also been reported: in a case study it was recently reported that a patient developed autoantibody-negative insulin-dependent diabetes (type 1B subtype) that presented as DKA several weeks post–COVID-19 infection (12). A second case study of an otherwise healthy patient developing DKA while infected with SARS-CoV-2 was also reported, although the autoantibody status of this patient was not reported (11). Research to identify treatment options that can stimulate an anti–SARS-CoV-2 immune response, while reducing the impact of post–COVID-19 immune abnormalities, is ongoing (23).

Although these studies suggest the potential for COVID-19 to disrupt glycemic response, no large controlled clinical studies have been conducted to determine the extent of this risk in patients diagnosed with COVID-19. This study improves on limitations of previous studies in 1) using a large cohort of 8,755 COVID-19(+) patients, 2) investigating pre- and postinfection HbA_1c levels with respect to T2D status, and 3) including a large, matched cohort of 11,998 COVID-19(−) patients for comparison. Our findings indicate that there is an ∼0.1% increased postinfection HbA_1c among patients with and without T2D, which is statistically significant but clinically insignificant (Fig. 1). Although COVID-19(+) patients were 40% more likely to be diagnosed with T2D compared with COVID-19(−) patients (P < 0.001), we also found that COVID-19(+) patients were 274% more likely to get diagnosed with T2D for the same pre–COVID-19 test HbA_1c and 28% more likely to have the same increase in HbA_1c after their COVID-19 test (Fig. 3A and B) (P < 0.001). A possible explanation for this is that COVID-19(+) patients are receiving more intensive care that results in better identification of patients who have underlying T2D.

Interestingly, COVID-19 infection did not increase DKA risk in patients with preexisting T1D (HR 0.98 [95% CI 0.46–1.56], P = 0.53) or T2D (HR 1.38 [95% CI 0.79–2.41], P = 0.26). In a previous study investigators found that hospital admissions for DKA increased compared with preceding years in patients with T2D during the COVID-19 pandemic but decreased for patients with T1D over the same time period; however, this study did not include investigation of the COVID-19 status of the patients (13). Findings of a recent seven-center study in the U.S. suggested that DKA frequency increased among a pediatric cohort of T1D patients during COVID-19 surges; however, DKA was less common in patients using continuous glucose monitors or insulin pumps (24). We also observed a higher risk of DKA onset in the Black COVID-19(+) patients with T2D compared with the White COVID-19(+) patients with T2D (HR 2.46 [95% CI 1.48–6.1], P = 0.04) (Fig. 5). These findings, in particular the findings of increased DKA in patients with T2D who are Black, indicate that increased awareness is needed for DKA in these patient populations if infected with COVID-19.

As with any study, there are limitations that should be considered during the interpretation of the findings. To evaluate a large cohort of patients, we included patients with HbA_1c measurement for up to 12 months prior to their COVID-19 test. Although we adjusted for the time between HbA_1c and COVID-19 testing, comorbid conditions, and medications during this period, it is possible that medication changes or other factors during the 12 months impacted HbA_1c. To address this, we also evaluated the results using 3- and 6-month intervals, which yielded consistent results, indicating that these factors are unlikely to have a material impact on the analysis. Although consistent criteria were applied to the selection of COVID-19(+) and COVID-19(−) patients, patients suspected to be at increased risk of developing T2D may be more likely to have their HbA_1c tested. HbA_1c values were not available for many patients in the COVID-19 registry, and care was disrupted during the pandemic, which may bias the population available for inclusion in this study. Furthermore, although the Cleveland Clinic is a large health care system, patients may have received care at a different institution, resulting in medical histories being unavailable for this study. Additional studies at other sites and with longer follow-up data on HbA_1c levels are needed to verify our results.

In conclusion, we observed a statistically significant, but clinically insignificant, increase in HbA_1c post–COVID-19 infection, and the increased likelihood of receiving a T2D diagnosis post–COVID-19 infection may be due to the increased care provided to these patients. Notably, in this large cohort of patients who were COVID-19(+), the risk of DKA was significantly higher in patients with preexisting T2D and who are Black. Additional research is required to understand the pathogenesis of COVID-19 infection and DKA in patients with T2D, and additional clinical awareness is needed for DKA in these patient populations.

Acknowledgments. The authors acknowledge the many people who worked to establish the Cleveland Clinic COVID-19 data registry and their efforts to make this valuable resource available to investigators.

Duality of Interest. D.M.R., M.W.K., A.S., and A.Ma. have received research funding from Novo Nordisk. K.M.P. has received research support from Bayer AG, Merck & Co., Novo Nordisk, and Twin Health; consulting honoraria from AstraZeneca, Bayer AG, Corcept Therapeutics, Diasome, Eli Lilly and Company, Merck & Co., Novo Nordisk, and Sanofi; and speaker honoraria from AstraZeneca, Corcept Therapeutics, Merck & Co., and Novo Nordisk in the past 12 months. A.D.M.-H. reports grants from Novo Nordisk, grants from Merck & Co., and grants from Boehringer Ingelheim Pharmaceuticals outside the submitted work. D.M.R., K.M.P., and A.Ma. have intellectual property related to treatment decision making in the context of T2D. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.S. performed analyses and wrote the manuscript. A.Mi., and A.Ma. extracted and curated data. A.D.M.-H., M.W.K., and K.M.P. provided input into the study design and edited and reviewed the manuscript. D.M.R. designed the study and wrote the manuscript. D.M.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. A.O. and W.K. curated and confirmed the data for DKA events.

Prior Presentation. A non–peer-reviewed version of this article was submitted to the medRxiv preprint server (https://doi.org/10.1101/2022.03.08.22272041) on 10 March 2022.