Inflammation, Hyperglycemia, and Adverse Outcomes in Individuals With Diabetes Mellitus Hospitalized for COVID-19

As of October 2021, there have been >44 million confirmed cases and 700,000 deaths attributed to coronavirus disease 2019 (COVID-19) in the United States (1). The ongoing COVID-19 pandemic disproportionately affects individuals with diabetes mellitus (DM) (2). More than 40% of hospitalized individuals with COVID-19 have DM, which is a major risk factor for adverse outcomes in this patient population (2–7). The reasons underlying the susceptibility of individuals with DM to severe COVID-19 remain unclear.

DM is characterized by chronic low-grade inflammation (8), which promotes insulin resistance and hyperglycemia, processes important in the development of chronic complications (9). This chronic inflammatory state is thought to stimulate stronger immune and inflammatory responses in individuals with DM exposed to COVID-19 compared with those without DM, promoting cytokine release and hyperglycemic surges (2). Hyperglycemia further upregulates inflammatory and oxidative stress markers in a vicious cycle (10,11). The interplay between inflammatory cytokines and hyperglycemia may be a major factor in the development of multiorgan damage and mortality in individuals admitted for COVID-19 (12). Understanding the relationship among DM, inflammation, and hyperglycemia in individuals hospitalized for COVID-19 is instrumental in devising targeted strategies for improving outcomes in this high-risk patient population.

To that end, we leveraged the International Study of Inflammation in COVID-19 (ISIC), a large, multicenter, observational study of individuals admitted specifically for COVID-19 in whom inflammatory biomarkers were measured on admission. Our study objectives were to characterize the impact of DM on COVID-19–related outcomes in relation to inflammation, identify the determinants of risks in individuals with DM, and examine the interplay among inflammatory biomarkers, hyperglycemia, insulin use, and in-hospital outcomes.

ISIC is a multicenter observational study with the primary objective of characterizing the role of inflammatory biomarkers in COVID-19–related adverse outcomes (13). Participating centers and site investigators are listed in the Supplementary Material. Institutional review board approvals and consent procedures were obtained separately at each site according to local institutional policies.

Individuals were eligible if they met the following inclusion criteria: 1) adult (≥18 years old) hospitalized specifically for COVID-19, 2) confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection by RT-PCR testing of nasopharyngeal or oropharyngeal samples, and 3) at least one blood sample collected during hospitalization. Individuals with a positive SARS-CoV-2 test who were hospitalized for non–COVID-19 reasons were excluded. All patients were monitored until hospital discharge or death. Extensive clinical data were collected through electronic health records using established data mining tools and reviewed for accuracy by at least two reviewers per site. All data were entered into the secure web-based repository REDCap.

Data collected were medical history, including DM type (type 1 or type 2); demographics; laboratory tests; medications; clinical characteristics; inpatient medical therapy; hospitalization course; and outcomes. DM was defined as a documented diagnosis in the medical record, treatment with hypoglycemic agents, or a hemoglobin A_1c (HbA_1c) ≥6.5% within 1 year before admission. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation.

The Michigan Medicine COVID-19 Cohort (M²C²) is the largest ISIC subcohort. The M²C² comprises consecutive, systematically enrolled adults (≥18 years) with confirmed SARS-CoV-2 infection hospitalized specifically for COVID-19 at the University of Michigan from 1 February 2020 to 1 June 2021. In addition to the variables collected for ISIC, serial laboratory measurements, frequently monitored blood glucose levels, and daily insulin dose administered throughout hospitalization were collected for M²C² as part of a standardized inpatient management protocol for hyperglycemia (14).

Blood samples were collected and analyzed for several inflammatory biomarkers, including soluble urokinase plasminogen activator receptor (suPAR), interleukin-6 (IL-6), C-reactive protein (CRP), D-dimer, ferritin, lactate dehydrogenase, and procalcitonin levels within 48 h of admission. CRP, ferritin, D-dimer, lactate dehydrogenase, and procalcitonin levels were measured by the central laboratory at the respective institution of enrollment at the request of the clinical team. Residual samples were collected for suPAR and IL-6 measurement, which were measured in batches using a commercially available ELISA (suPARnostic ELISA [ViroGates, Birkerød, Denmark], Human IL-6 Quantikine QuicKit [R&D Systems, Minneapolis, MN]). Serum samples used for suPAR and IL-6 measurements were collected and kept frozen at −80°C until the time of measurement, which was no longer than 3 months. Samples underwent up to two thaw cycles. Both suPAR and IL-6 are highly stable in frozen samples stored for >5 years and are not affected by repeated freeze/thaw cycles. Technicians performing assays were blinded to clinical data.

The primary outcome was the composite end point of in-hospital death, need for mechanical ventilation, and need for renal replacement therapy. Secondary outcomes included each individual component of the primary outcome.

We included individuals hospitalized for COVID-19 during the period of 1 February 2020 to 1 June 2021, the date the database was locked (N = 2,044). We report clinical characteristics for the overall cohort stratified by DM history using categorical variables expressed as a number and percentage and continuous variables expressed as mean (SD) and median (25th–75th interquartile range) for normally and nonnormally distributed continuous data, respectively. The characteristics between individuals with and without DM were compared using χ² tests for categorical variables and unpaired t tests or Mann-Whitney U tests for normal and nonnormal continuous variables, respectively. The incidences of the individual outcomes (in-hospital death, need for mechanical ventilation, and need for renal replacement therapy) were compared between individuals with and without DM at admission using χ² tests.

To determine whether DM history was independently associated with higher thromboinflammation marker levels, separate linear regression models were created, with each biomarker as the dependent variable and DM, age, sex, BMI, race, hypertension, coronary artery disease, and heart failure as independent variables. Each model variable was standardized by subtracting the mean of each variable and dividing by its SD, resulting in a distribution with mean = 0 and SD = 1. Standardization was performed to compare the effect sizes for DM across each biomarker.

We examined the association between DM and the composite outcome using stepwise logistic regression models. Model 0 was unadjusted. Model 1 included age, sex, and race. Model 2 included the variables in model 1 as well as clinical characteristics, including BMI, history of hypertension, coronary artery disease, congestive heart failure, and admission eGFR. Model 3 included the variables in model 2 as well as suPAR, which was log₂ transformed (interpreted as per 100% increase) given the nonnormal distribution. We repeated this analysis to explore the association between DM and each outcome individually. We then performed mediation analysis to assess whether the effect of DM on the composite outcome is mediated by suPAR, after adjusting for the clinical variables in model 3 (age, sex, race, BMI, and history of hypertension, coronary artery disease, congestive heart failure, and admission eGFR) (15).

We used logistic regression to identify risk factors for the composite outcome among individuals with DM (n = 686). We first examined the association of each clinical characteristic and the composite outcome in univariable analysis. In addition to variables associated with the outcome in the univariable analysis, we included the following variables in the multivariable analysis on the basis of biologic plausibility and clinical knowledge: age, sex, race, BMI, smoking status, hypertension, coronary artery disease, heart failure, chronic kidney disease, admission eGFR, and glucose range at admission (<54–69, 70–180, 181–250, and >250 mg/dL). Biomarkers of inflammation (log₂ transformed) were each examined in the multivariable risk model separately. We also explored risk factors for each individual outcome using the same multivariable risk model. Finally, we calculated the relative importance of clinical characteristics, biomarkers of inflammation, and glucose levels for predicting the composite outcome on the basis of the Gini index using a random forest approach (16).

We used the M²C² subset of ISIC (n = 1,608), in which serial data were collected, to determine whether glucose levels and insulin administered during hospitalization were associated with inflammatory biomarkers and outcomes. We assessed the following exposures: glucose level on admission, coefficient of variation for glucose during hospitalization, percentage of glucose measurements in target glucose range (70–180 mg/dL) during hospitalization per patient, percentage of glucose measurements >180 mg/dL during hospitalization per patient, and average daily amount of insulin administered during hospitalization per patient adjusted for body weight. The coefficient of variation is expressed as the SD divided by the mean of all glucose measurements during hospitalization. The average amount of insulin administered was calculated as the total insulin dose (units) divided by patient weight (kilograms) multiplied by the total number of in-hospital days. We used Spearman rank correlation to examine the correlation between each biomarker of inflammation with glucose coefficient of variation and the average insulin dose received during hospitalization. To assess the association between each exposure and the composite outcome, we used multivariable regression models. Each variable was modeled continuously and as a categorical variable using the following categories as references for each variable: 0 for glucose coefficient of variation, 100% for glucose values in the normal range, 0% for glucose values in the high range, and 0 units/kg/day for average insulin dose. For glucose variables, the coefficients are expressed as a 10-unit difference, whereas insulin dose is expressed as a difference in 0.1 unit/kg/day. Models were adjusted for age, sex, race, BMI, and history of hypertension, coronary artery disease, and congestive heart failure. Separate models were additionally adjusted for suPAR within 48 h of admission and corticosteroid use.

We performed a complete case analysis for multivariable models. There were no missing data for any demographic or clinical characteristic. A two-sided P < 0.05 was used to determine statistical significance. All analyses were performed using R 4.1.0 statistical software (R Foundation for Statistical Computing, Vienna, Austria).

The overall cohort had a mean (SD) age of 60 (16) years, 42% were female, and 20.5% were Black. One-third of the cohort (n = 686, 33.6%) met the criteria for DM, of whom 98.5% had type 2 DM and 1.5% had type 1 DM. Compared with individuals without DM, those with DM were older (mean age 64 vs. 58 years) and were more likely to be Black (27.6% vs. 16.9%), to be obese (mean BMI 33 vs. 31 kg/m²), and to have a greater comorbidity burden, including hypertension (81.2% vs. 45.7%), coronary artery disease (24.6% vs. 8.9%), heart failure (17.2% vs. 7.7%), and chronic kidney disease (26.8% vs. 10.8%; P < 0.001 for all) (Table 1). On hospital admission, individuals with DM were less likely to present with fever (59.6% vs. 65.8%) but more likely to present with altered mental status (13.4% vs. 6.5%) compared with those without DM (P < 0.001).

In unadjusted analyses, the levels of several inflammatory biomarkers, including suPAR, CRP, procalcitonin, and D-dimer, were higher on admission in individuals with DM than in those without DM (Table 1). In multivariable analyses, only suPAR levels were independently associated with DM (standardized β = 0.10 [95% CI 0.05, 0.15]) (Supplementary Table 1). On average, participants with DM had 20.7% higher suPAR levels than those without DM.

Overall, the primary composite outcome was observed in 647 (31.7%) individuals. There was a total of 288 (14.1%) in-hospital deaths, 550 (26.9%) individuals who required mechanical ventilation, and 182 (8.9%) individuals who required renal replacement therapy. In unadjusted analyses, individuals with DM had a significantly higher cumulative incidence of the primary composite outcome (37.8% vs. 28.8%; P < 0.001) as well as the individual components of in-hospital death (16.9% vs. 12.7%; P = 0.01), need for mechanical ventilation (31.6% vs. 24.5%; P = 0.001), and need for renal replacement therapy (12.4% vs. 7.1%; P < 0.001) compared with those without DM (Table 1). In multivariable analyses, adjusting for demographics (model 1) and clinical characteristics (model 2) heavily attenuated the association between DM and the primary outcome (adjusted odds ratio [aOR] 1.23 [95% CI 1.00, 1.52]), which became nonsignificant after including suPAR in the model (aOR 1.03 [95% CI 0.78, 1.37]) (Fig. 1). When these outcomes were examined individually, a similar pattern was seen (Fig. 1).

In mediation analysis, the average causal mediation effect (also known as indirect effect) of DM on the primary outcome through suPAR was significant (P = 0.008), while the average direct effect of DM on the primary outcome was not significant (P = 0.73). The proportion of the effect of DM on the primary outcome mediated by suPAR was 84.2%.

When examining predictors of the composite outcome in the subgroup of individuals with DM (n = 686), we found that higher BMI (aOR 1.18 [95% CI 1.06, 1.31]), lower eGFR (aOR 1.07 [95% CI 1.03, 1.10]), and admission glucose levels >180 mg/dL (aOR 1.85 [95% CI 1.20, 2.83]) were associated with the primary composite outcome (Supplementary Table 2). We found similar associations when examining outcomes individually, with a few notable exceptions (Supplementary Table 3). Older age was strongly associated with in-hospital death (aOR 1.44 [95% CI 1.17, 1.77]), and male sex was associated with the need for renal replacement therapy (aOR 2.33 [95% CI 1.31, 4.12]). Type 1 DM, prior insulin use, and medications for hyperglycemia were not associated with an increased odds in the primary outcome. Levels of all inflammatory biomarkers were associated with an increased odds of the primary outcome when examined separately in a multivariable model adjusted for demographic and clinical risk factors (Supplementary Table 2).

We identified suPAR level as the most important variable associated with the primary outcome in individuals with DM and COVID-19, followed by BMI, admission glucose, and age in descending order of importance (Fig. 2). Individuals with DM with a suPAR level <5.94 ng/mL (first quartile) had a 23.9% incidence of the primary outcome compared with 53.8% in individuals with suPAR ≥14.8 ng/mL (fourth quartile).

Among the M²C² subset with longitudinal serial glucose and insulin data, we found only modest correlations between biomarkers of inflammation and both glucose coefficient of variation (r = 0.05–0.02) and average insulin dose (r = 0.09–0.02) during hospitalization (Supplementary Table 4). We also examined whether glucose ranges, glucose variation, and insulin requirements were associated with the primary outcome. The glucose coefficient of variation in individuals with DM was 17.0%, with an average of 53.9% of glucose measurements falling in the target range (70–180 mg/dL) and 44.8% of glucose values >180 mg/dL. The glucose coefficient of variation, a greater percentage of glucose values outside the target range, a greater percentage of high glucose values, and a higher required insulin dose were all associated with a greater odds of the primary outcome in individuals with DM (Fig. 3 and Supplementary Fig. 1). Per each 10% higher glucose coefficient of variation, the odds of the primary outcome was 1.30 (95% CI 1.11, 1.54]) (Supplementary Fig. 1). Per every 0.1 unit/kg/day of insulin administered, the odds of the primary outcome was 1.18 (95% CI 1.11, 1.25). Including suPAR or corticosteroid use in the models did not affect estimates significantly (Supplementary Table 5).

In this in-depth examination of the interplay among DM, inflammation, hyperglycemia, and outcomes in individuals hospitalized for COVID-19, we found that the impact of DM on outcomes is tightly linked to heightened inflammation. First, individuals with DM had a greater incidence of in-hospital outcomes and higher levels of inflammatory markers (notably suPAR) compared with those without DM. The association between DM and outcomes was abrogated, however, by including suPAR in the model, with mediation analysis suggesting that the effect of DM on outcomes is largely mediated by suPAR. Among individuals with DM, suPAR, BMI, admission glucose levels, and age were the most important risk factors (in that order). The correlation between inflammatory markers and hyperglycemia was modest at best, while hyperglycemia and higher insulin requirements during hospitalization were associated with worse outcomes. This association was not attenuated after adjusting for suPAR, implying that hyperglycemia affects COVID-19–related outcomes through noninflammatory processes.

DM is a well-established risk factor for COVID-19 (2,17); however, the underlying mechanisms are unclear. In susceptible individuals, SARS-CoV-2 infection is thought to trigger a prolonged hyperinflammatory response, dubbed the cytokine storm (4,18–22). DM, as a chronic inflammatory condition, may predispose individuals to a heightened inflammatory response (23,24). Mitochondrial disruption, rather than changes to glucose metabolism, has been found to lead to altered T-cell cytokine production (notably by T-helper 17 cells) in type 2 DM (23). Consistently, we found that individuals with DM had higher levels of inflammatory biomarkers, including suPAR, CRP, procalcitonin, and D-dimer. After adjusting for comorbidities, we noted a singular association between DM and suPAR, suggesting that suPAR represents the inflammatory biomarker most reflective of the hyperinflammatory state in DM and COVID-19. Our mediation analysis supports this finding in that we found that suPAR levels accounted for 84.2% of the effect of DM on the outcomes. Conversely, another study found that CRP accounted for only 32.7% (12).

SuPAR is an immune-derived signaling glycoprotein, which is notorious for its role in kidney disease (25–27), cardiovascular disease (28–30), and most recently, COVID-19 (13,31). Blood suPAR levels are notably high in individuals with type 1 or type 2 DM, even in the nonacute setting, and are strongly predictive of DM-related outcomes, such as nephropathy and atherosclerotic events (28,32,33). Several studies have identified a correlation between T-helper 17 cells and suPAR levels (34,35), which may explain the predilection for individuals with DM to have higher suPAR levels (23,36). SuPAR differs from other biomarkers of inflammation in that it is not an acute-phase reactant: Levels remain stable in highly proinflammatory situations, such as acute myocardial infarction or cardiac surgery (27). An increased suPAR level, however, is triggered by specific stimuli, such as smoking and RNA viruses (e.g., SARS-CoV-2), and is highly expressed in lung tissue (37). Accordingly, individuals with DM and COVID-19 have four- to eightfold higher suPAR levels (median 8.82 ng/mL) than healthy individuals (median 2.40 ng/mL). We found that suPAR was the most important predictor of outcomes in individuals with DM, which mediated at least 80% of the effect of DM on outcomes. Overall, these findings suggest that suPAR levels may reflect more specifically the burden of inflammation in COVID-19 compared with other biomarkers.

Hyperglycemia has traditionally been thought to be a major driver of inflammation through several mechanisms, including increased oxidative stress (8). In our study, hyperglycemia and higher insulin requirements are independently associated with in-hospital outcomes in individuals with DM and COVID-19, consistent with earlier studies (2,38). Surprisingly, we found only a weak correlation between suPAR or other inflammatory biomarkers with hyperglycemia, and the association between hyperglycemia and outcomes was not mitigated by adjusting for suPAR. The association between hyperglycemia and COVID-19–related outcomes likely occurs through mechanisms not reflected by inflammatory biomarkers. This is consistent with a study showing that nonmitochondrial glycolysis did not affect the inflammatory signature in type 2 DM (39). Whether aggressive glucose control would improve COVID-19–related outcomes remains to be shown in a clinical trial setting (14).

This study has several important strengths. It is the largest study to investigate the role of inflammatory biomarkers in individuals with DM hospitalized for COVID-19. In addition, in contrast with other studies, it includes a diverse cohort of individuals specifically hospitalized for COVID-19 rather than defined by SARS-CoV-2 positivity alone. Blood samples were collected on admission, without being confounded by anti-inflammatory therapies, and thus, reflect more accurately the inflammatory state. The clinical data were collected through careful and adjudicated review of individual medical records rather than through administrative data sets. The study benefited from standardized glucose and insulin data collected continuously throughout the hospitalization through the Michigan Medicine hyperglycemia management protocol.

This study also had some limitations. Given the small number of patients with type 1 DM in this cohort, the findings cannot be extended to these individuals. The diagnosis of DM was based on medical chart review and available HbA_1c levels at the time of admission; thus, it is possible that some individuals classified as not having DM could have had undiagnosed DM. Finally, mechanistic studies are warranted to validate the inferences based on the epidemiologic observations noted in our study.

In summary, these data show that COVID-19–related in-hospital outcomes in individuals with DM are driven by a hyperinflammatory state reflected best by suPAR levels. SuPAR levels were the most important predictor of outcomes in individuals with DM, followed by obesity, hyperglycemia, and age. Hyperglycemia and higher insulin requirements correlated weakly with inflammatory biomarkers and were associated with outcomes independently of suPAR, suggesting that they likely impact outcomes through other mechanisms. Further study is needed to determine whether suPAR and hyperglycemia are therapeutic targets for the management of COVID-19 in individuals with DM.

Funding. A.V. is supported by a National Heart, Lung, and Blood Institute–funded postdoctoral fellowship (T32HL007853). S.S.H. is funded by National Heart, Lung, and Blood Institute grant 1R01HL153384-01, National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grants 1R01DK12801201A1 and U01DK119083-03S1, and the Frankel Cardiovascular Center COVID-19: Impact Research Ignitor award (U-M G024231). R.P.-B. is supported by NIDDK grants 1R01DK107956-01 and U01DK119083, JDRF Australia grant 5-COE-2019-861-S-B, and Michigan Diabetes Research Center pilot and feasibility NIDDK grant P30-DK020572. E.G.B. is supported by the Hellenic Institute for the Study of Sepsis. F.T. is supported through intramural funds from Charité Universitätsmedizin Berlin and the Berlin Institute of Health. S.P. is supported by the University of Michigan O’Brien Kidney Translational Core Center (NIDDK grant P30DK081943).

The funders had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; and preparation, review, or decision to publish the manuscript.

Duality of Interest. J.R. and S.S.H. are members of the scientific advisory board of Walden Biosciences. J.E.O. is a cofounder, shareholder, and chief scientific officer of ViroGates and a named inventor on patents related to suPAR. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.V. wrote the first draft. A.V., Y.H., L.Z., R.P.-B., and S.S.H. performed the statistical analyses. H.S., I.K., T.C., E.A., H.B., M.P., T.U.A., C.M., P.O., E.M., R.F., P.B., C.L., and S.S.H. collected the data and performed quality control. L.A., M.M., K.M.-S., S.P., M.K., S.H.L., A.C., F.T., E.J.G.-B., J.R., J.E.O., E.L.F., R.P.-B., and S.S.H. provided expert interpretation of the findings. All authors reviewed the initial draft and provided critical revisions and approved the final version of the manuscript. R.P.-B. and S.S.H. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Data Sharing. Study protocol, statistical code, and data set summary data are available upon request after publication through a collaborative process. Data sets can be accessed upon approval of a submitted research proposal. Please contact [email protected] for additional information.