This study investigated the incidence, patient profiles, outcomes, and risk factors for developing in-hospital DKA without prior insulin dependence in COVID-19 patients and compared to those in influenza patients. The major findings are: i) the incidence of DKA in hospitalized COVID-19 patients without previous insulin dependence is 1.4% and is significantly higher than those in hospitalized influenza (0.8%) over the same period and in the same health system, ii) COVID-19 patients with DKA have significantly worse disease severity, have more steroid treatment, and are more likely to die (mortality OR = 6.178) compared to those without DKA, iii) DKA patients without prior diabetes are much more likely to die compared to DKA patients with prior diabetes (OR = 7.56), iv) steroid use, pre-existing T2D status, and male sex are major risk factors for DKA, and v) DKA patients have higher rate of new insulin use at 3-months post COVID-19, suggesting SARS-CoV-2 infection could trigger new insulin dependence.
Incidence
The 1.4% incidence of DKA in COVID-19 patients without previous insulin dependence is higher than those in the general population [31]. In a 2020 study of patients in the United States using the National Inpatient Sample, there were 61.6 cases of DKA per 10,000 admissions (0.6%) nationwide in 2017, which included previously insulin-dependent patients who are susceptible to DKA [31]. There are a few reasons that our data are not that of national average under non-covid pandemic conditions, namely: our cohort excluded patients with insulin-dependent DM, patients who had COVID-19 were enriched for many major comorbidities and older age; and our patient cohort had a large proportion of Blacks and Hispanics, including those who were underserved.
Influenza, a similar respiratory virus, is arguably a very good control. The incidence of DKA in hospitalized COVID-19 patients without previous insulin dependence (1.4%) was significantly higher than those in hospitalized influenza (0.8%) whereas the demographics and comorbidities were similar between groups, suggesting virus specific activity or disease severity likely contribute to glucose dysregulation and insulin dependency. SARS-CoV-2 has proposed tropism for ACE2 receptors of pancreatic beta cells [32], and there is some evidence of tropism of the influenza virus for pancreatic beta cells, with an ability to replicate in human pancreatic cells, most often resulting in pancreatitis [33, 34]. Although influenza is also a known contributor to DKA, its relationship to DKA is likely nonspecific as an infectious process, and more likely to affect ketosis-prone diabetic patients [35, 36].
Risk factors for DKA
Steroid use, pre-existing T2D status, and male sex are risk factors for DKA. Acute steroid treatment is known to exacerbate hyperglycemia and exhaust the pancreas and result in ketosis [37, 38]. It is not surprising that steroid use and pre-existing T2D status are risk factors for DKA. Steroid was likely administered to more severe COVID-19 patients and thus it is necessary to adjust for covariates as performed. There is a preponderance of evidence that the male sex is more susceptible to a wide range of COVID-19 related morbidities and worse outcomes, including higher risk of mortality, hospitalization, critical illness, acute kidney injury and acute cardiac injury [26, 28, 39]. The exact mechanism of sex difference is unclear.
Surprisingly, COVID-19 patients with hypertension and COPD were less likely to develop DKA. A possible explanation is that we excluded previously insulin-dependent patients in our analysis, including those with advanced T2D, which may have resulted in a different profile of patients with lower rates of comorbidities. The lower prevalence of hypertension and COPD in COVID-19 patients with DKA is consistent with one study [40] but differed from another [15]. Further studies are warranted.
New-onset vs. pre-existing Type 2 Diabetes in DKA patients
In a sub-group analysis, patients with new-onset T2D were more likely to be intubated and more likely to die although patient profiles were similar between groups. It is possible that new-onset T2D group had more severe COVID-19. It is also possible that patients with pre-existing T2D were pre-conditioned or were on diabetic medication/management, resulting in resistant to DKA. Another possible explanation is complications due to mismanagement of DKA, such as fatal cerebral edema, a DKA complication in cases of new onset diabetes [41–43].
Note that new-onset and pre-existing T2D DKA patients had similar profiles, likely had similar vulnerability to insulin dysregulation due to the infection and steroid use. Only patients with insulin dependency were excluded, but not diabetic patients who were not on insulin. Thus, it is also possible that some new-onset T2D patients were not put on insulin due to unawareness or lack of care.
A1c values for new-onset and pre-existing T2D groups were unusually high for both groups. As A1c reflects the average blood glucose level over the previous 3 months, it is possible that severe DKA might have increased the proportion of glycated hemoglobin at a rate that obscures the baseline blood glucose levels [11]. Thus, baseline blood glucose levels for new-onset T2D patients cannot be determined with confidence. Further studies are needed.
New insulin use at 3 months post covid
Another novel finding of this study is that DKA patients had higher rate of new insulin use at 3-months after SARS-CoV-2 infection, consistent with the limited regenerative capacity of beta cells of the pancreas [44]. Stress hyperglycemia due to severe infection resulted in pancreatic exhaustion and damage that is difficult to reverse [44, 45].
In the sub-group analysis, we found pre-existing T2D cohort had significant new insulin dependence at 3 months (26.3%), suggesting SARS-CoV-2 triggered new-insulin dependence in patients who were at risk. Moreover, many adults without prior diabetes also developed insulin-dependent new-onset T2D at 3 months post covid (5.9%). This is concerning and warrants closer follow up. Taken together, our findings provide evidence that SARS-CoV-2 infection directly or indirectly triggers new insulin dependence. This is in addition to SARS-CoV-2 infection triggering new-onset insulin resistant diabetes [1–5].
Potential mechanisms of COVID-19 related DKA
There is some evidence that COVID-19 may contribute to insulin dependence through direct damage to pancreatic insulin-producing beta-cells via tropism of COVID-19 to ACE2 receptors on these cells. This proposed tropism has precedent in other viruses such as enteroviruses, mumps and others which have also resulted in fulminant ketoacidosis [32]. It is possible the insulin dependence of DKA is a manifestation of insulopenia directly from SARS-CoV-2 mediated beta cell dysfunction, however there is stronger evidence that indirect pathways to DKA in COVID-19 infection are more responsible.
Indirect effects from systemic hypoxia, respiratory distress, sepsis, inflammatory responses, and cytokine storm, among others, secondary to from SARS-CoV-2 infection could also cause metabolic decompensation or stress [20–24], resulting in insulin resistance DKA. This pathway to DKA is likely a result of pancreatic beta cell exhaustion due dramatic hyperglycemia from inflammation and steroids which cause acute overactivation of beta cells, likely in the context of chronic hyperstimulation due to insulin resistance in a diabetes or pre-diabetes status [37, 38, 45, 46].
Insulin-dependency in COVID-19 patients with DKA is likely some combination of direct viral-mediated beta cell damage and indirect beta cell dysfunction from severe hyperglycemia triggered by SARS-CoV-2 infection.
Limitations
This study has several limitations. It is possible some new-onset T2D patients had pre-existing diabetes of which they were unaware. If undiagnosed diabetes was present in this group, it is most likely to be T2D (but not Type 1 Diabetes) given the advanced age, male gender, and elevated BMI characteristic of T2D. As with any retrospective study, there could be other unintended patient selection bias and unaccounted confounders.