Admission hyperglycaemia as a predictor of mortality in patients hospitalized with COVID-19 regardless of diabetes status: data from the Spanish SEMI-COVID-19 Registry

Abstract Background Hyperglycaemia has emerged as an important risk factor for death in coronavirus disease 2019 (COVID-19). The aim of this study was to evaluate the association between blood glucose (BG) levels and in-hospital mortality in non-critically patients hospitalized with COVID-19. Methods This is a retrospective multi-centre study involving patients hospitalized in Spain. Patients were categorized into three groups according to admission BG levels: <140 mg/dL, 140–180 mg/dL and >180 mg/dL. The primary endpoint was all-cause in-hospital mortality. Results Of the 11,312 patients, only 2128 (18.9%) had diabetes and 2289 (20.4%) died during hospitalization. The in-hospital mortality rates were 15.7% (<140 mg/dL), 33.7% (140–180 mg) and 41.1% (>180 mg/dL), p<.001. The cumulative probability of mortality was significantly higher in patients with hyperglycaemia compared to patients with normoglycaemia (log rank, p<.001), independently of pre-existing diabetes. Hyperglycaemia (after adjusting for age, diabetes, hypertension and other confounding factors) was an independent risk factor of mortality (BG >180 mg/dL: HR 1.50; 95% confidence interval (CI): 1.31–1.73) (BG 140–180 mg/dL; HR 1.48; 95%CI: 1.29–1.70). Hyperglycaemia was also associated with requirement for mechanical ventilation, intensive care unit (ICU) admission and mortality. Conclusions Admission hyperglycaemia is a strong predictor of all-cause mortality in non-critically hospitalized COVID-19 patients regardless of prior history of diabetes. KEY MESSAGE Admission hyperglycaemia is a stronger and independent risk factor for mortality in COVID-19. Screening for hyperglycaemia, in patients without diabetes, and early treatment of hyperglycaemia should be mandatory in the management of patients hospitalized with COVID-19. Admission hyperglycaemia should not be overlooked in all patients regardless prior history of diabetes.


Introduction
Since the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic emerged in China, more than 300,000 confirmed cases and approximately 28,500 coronavirus disease 2019 (COVID-19) related deaths have been reported in Spain as of the date of writing of this article [1].
Acute hyperglycaemia has been associated with inhospital complications in non-critically ill patients with and without diabetes mellitus (DM) [2]. In regard to COVID-19, elevated blood glucose (BG) levels are also associated with in-hospital complications, including mechanical ventilation requirement, intensive care unit (ICU) admission and death [3][4][5][6][7]. Thus, hyperglycaemia, particularly upon admission, could be a marker of poor prognosis regardless of diabetes status. This suggests that elevated BG may play a decisive role in the severity of the disease at an early stage [8][9][10].
There are at least two plausible reasons why hyperglycaemia, particularly acute hyperglycaemia, could be harmful in patients with COVID-19 [11]. First, SARS-CoV-2 could infect endocrine pancreas cells through their expression of angiotensin-converting enzyme 2 (ACE2) receptors, resulting in an impairment in b-cell insulin secretion [12]. Second, inflammation during COVID-19 could also generate insulin resistance. Both mechanisms combined could induce hyperglycaemia in early stages of the disease. Once a patient presents with hyperglycaemia, it could play a direct role in worsening the infection. Hyperglycaemia upregulates ACE2 expression and induces glycosylation ofACE2, facilitating the invasion of cells by SARS-CoV-2 [13].
Although several studies and meta-analyses have shown that patients with diabetes have a significantly higher risk of severe COVID-19 and increased mortality rates [14][15][16], the impact of hyperglycaemia itself, rather than the presence of DM, has not been sufficiently described in non-critically patients hospitalized with COVID-19.
Therefore, our study sought to evaluate whether acute hyperglycaemia at admission, independently of diabetes status, was associated with all-cause in-hospital mortality in a large cohort of patients with SARS-CoV-2 infection in Spain. Additionally, we explored the relationship between high BG and length of stay (LOS), ICU admission and/or mechanical ventilation.

Study design and population
The SEMI-COVID- 19 Registry is an ongoing nationwide, multicentre, observational, retrospective cohort registry involving 109 hospitals in Spain. It includes consecutive patients ! 18 years of age hospitalized from 1 March 2020 to 31 May 2020, who were admitted with COVID-19 confirmed microbiologically by reverse transcription polymerase chain reaction (RT-PCR) testing of a nasopharyngeal sample, sputum or bronchoalveolar lavage samples. Exclusion criteria were subsequent admissions of the same patient and denial or withdrawal of informed consent. Patients were treated at their attending physician's discretion, according to local protocols and their clinical judgement.

Registry information and data collection
Information on the registry is available in a previous published study, which includes all necessary details about procedures and describes the baseline characteristics of all patients included. Data are collected retrospectively and include approximately 300 variables grouped under various headings, such as epidemiological data, medical history and use of medications, symptoms and physical examination findings at admission, laboratory and diagnostic imaging tests, pharmacological treatment and so on [17]. Patients were classified as diabetic when diabetes was present in the history profile or they took antidiabetic drugs before admission. The management of hyperglycaemia during hospitalization was using insulin in basal bolus regimen according to local protocols.
The Spanish Agency of Medicines and Medical Devices (AEMPS, for its initials in Spanish) has ruled that due to the nature of the registry, the study only required the approval of Ethics Committees. The SEMI-COVID-19 Registry was approved by the Provincial Research Ethics Committee of M alaga (Spain). Informed consent was obtained from all patients. When it was not possible to obtain informed consent in writing due to biosafety concerns or if the patient had already been discharged, informed consent was requested verbally and noted on the medical record.
The STROBE statement guidelines were followed in the conduct and reporting of the study.

Study outcomes
The primary endpoint was all-cause mortality during hospitalization according to BG levels at admission. The secondary outcomes were LOS and the composite of the invasive or non-invasive mechanical ventilation, ICU admission or death. The follow-up period was from admission to discharge or death.

Statistical analysis
BG levels obtained from a sample taken at admission were categorized into three groups according to standard glycaemic targets in hospitalized patients: <140 mg/dL, 140-180 mg/dL and >180 mg/dL. Hypoglycaemia was defined as a BG level below 70 mg/ dL. Baseline characteristics were described according to these groups. Continuous variables were tested for normal distribution using Kolmogorov-Smirnov's test.
Results are shown as means (standard deviation, SD) or medians (25th to 75th percentile) for continuous variables and numbers (%) for categorical variables.
To compare baseline demographic data and clinical characteristics among the BG level groups, we used analysis of variance (ANOVA) or the Kruskal-Wallis test for continuous variables. Differences in proportions were analysed using the chi-square test. Correlations between admission serum glucose levels and the main quantitative variables were calculated using Spearman's rank correlation coefficient test. In addition, the association between BG levels and death was analysed using Kaplan-Meier's survival curves; the log-rank test was used to compare survival curves according to the BG level groups. A multivariate Cox proportional hazard regression model with BG levels as the predictor variable for all-cause mortality was used to estimate hazard ratios (HRs) with a 95% confidence interval (CI). Multivariate model was developed by using a forward stepwise method, including all variables with p<.1 on the univariate analysis. We also used a logistic regression to evaluate the relationship between BG levels and the composed secondary endpoint. All statistical analyses were performed using SPSS software (version 26.0, Chicago, IL). A two-sided p value <.05 was considered statistically significant.

Baseline characteristics and correlation with blood glucose
Clinical characteristics were collected from a total of 11,312 participants out of the 12,826 confirmed cases of COVID-19 included in the registry as of 29 May 2020. A flowchart illustrating patient inclusion is shown in Figure 1. Baseline clinical features and the most relevant laboratory variables grouped according to admission BG levels are listed in Table 1. Overall, the mean age was 67.06 years (SD 16.4) and 57.1% of patients were male. The prevalence of diabetes was 18.9%.
Patients with higher admission BG levels were older, predominantly male, and more frequently had a prior history of diabetes, hypertension and other comorbidities. In addition, lymphocytes <800/mm 3 , LDH >400 U/L, D-dimer >1000 ng/mL, and elevated serum creatinine and C-reactive protein (CRP) levels were more frequent in patients with elevated admission BG. No differences were found in serum ferritin or interleukin-6 levels according to admission BG levels ( Table 1).

Association between blood glucose and outcomes
In total, 2289 (20.2%) patients died during hospitalization. Main outcomes according to BG levels are showed in Table 2. All-cause mortality was higher in patients with admission BG levels >180 mg/dL (41.1%) compared to patients with levels of 140-180 mg/dL (33.0%) or <140 mg/dL (15.7%) (Figure 2(a)). Indeed, there was a gradual increase in all-cause mortality as admission BG levels increased, and there were no differences in mortality rates within each category of BG levels between patients with or without a previous history of diabetes (Figure 2(a)).
Kaplan-Meier's survival curves according to admission BG levels are shown in Figure 2(b) (log-rank p<.001). Similarly, survival curves after classifying the cohort according to diabetes status did not show any changes (log-rank p<.001) (Figure 2(c,d)).

Discussion
After analysing the data from 11,312 consecutive noncritically patients with confirmed COVID-19 admitted to Spanish hospitals, we found that the admission BG level was an independent predictor of all-cause mortality during hospitalization.
A small number of observational studies have analysed the relationship between glycaemic control and clinical outcomes in patients hospitalized due to COVID-19 with and without diabetes. All of them provide clinical evidence of a correlation between uncontrolled hyperglycaemia with poor prognosis and a particularly high mortality rate [3][4][5][6][7]. To date, our study has included by far the greatest number of valid patients. Furthermore, our study was conducted  Similarly, our study also showed 1.63-fold increase in mortality in patients with diabetes on the univariate analysis. However, Zhu et al. did not include patients without diabetes in their study, so the effect of hyperglycaemia in the non-diabetic population with COVID-19 cannot be analysed.
A retrospective observational study (n ¼ 1122) by Bode et al. [5] from 88 hospitals in the USA studied the clinical outcomes of 451 patients with diabetes (HbA1c ! 6.5%) and/or uncontrolled hyperglycaemia (defined as !2 BG values > 180 mg/dL in a 24-hour period with HbA1c 6.5% or no A1c available) using data abstracted from Glytec's data warehouse. Among the 570 patients who died or were discharged, the mortality rate was 28.8% in the group with diabetes and/or uncontrolled hyperglycaemia (n ¼ 184) compared to 6.2% in the control group (n ¼ 386) (p<.001). The mortality rate was particularly high among patients with uncontrolled hyperglycaemia without diabetes (41.7%) compared to patients with diabetes (14.8%) (p<.001). This finding is consistent with our results (a mortality rate of 43.3% in non-diabetic patients with admission BG >180 mg/dL). Overall, these findings suggest that stress hyperglycaemia could play a crucial role in the prognosis of patients hospitalized with COVID-19.
Two observational studies conducted in Wuhan, China compared the relationship between hyperglycaemia and outcomes in patients hospitalized with confirmed COVID-19. Zhang et al. [6] evaluated the relationship between hyperglycaemia and outcomes in patients (n ¼ 166) hospitalized with confirmed COVID-19 with diabetes and secondary hyperglycaemia classified into three groups: control (group 1: no history of diabetes and basal BG <126 mg/dL), secondary hyperglycaemia (group 2: no history of diabetes and basal BG !126 mg/dL and HbA1c <6.5%) and diabetes (group 3: basal BG !126 mg/dL and history of diabetes or HbA1c !6.5%). The mortality rates in groups 2 and 3 were significantly higher compared to group 1 (21.3%, 14.3% and 9.5%, respectively; p<.05 for both). The composite outcome of ICU admission, use of either invasive or non-invasive mechanical ventilation, or death occurred in 38.1% of patients in group 2 and in 27.9% of patients of group 3. After adjusted for confounding variables, the OR for composite outcomes was 5.47 (95%CI 1.56-19.82) and 2.61 (95%CI 0.86-7.88) in groups 2 and 3 compared to group 1. These results suggest that stress hyperglycaemia is also related to poor prognosis and mortality in patients infected by SARS-CoV-2.
Wu et al. [7] collected data on 2041 patients admitted to two medical centres. They compared non-critical patients at admission (n ¼ 1690) to critical patients at the time of admission or at the time of transition from the non-critical group (n ¼ 697). Elevated admission BG levels (defined as admission BG !110 mg/dL) were an independent risk factor for progression to critical status (ICU admission, mechanical ventilation, compromised hemodynamic) or death among noncritical patients (OR 1.30; 95%CI 1.03-1.63, p¼.026). Higher median BG levels after a non-critical patient becomes critical were also independently associated with higher in-hospital mortality (OR 2.39; 95%CI 1.41-4.07, p¼.001). Similar to our results, hyperglycaemia but not diabetes was associated to worse outcomes and death on the multivariate analysis.
Sardu et al. [4] studied 59 patients with COVID-19 hospitalized in two Italian hospitals. Patients were divided into a hyperglycaemic group (n ¼ 25), defined as patients with admission BG >140 mg/dL, and a normoglycaemic group (n ¼ 34). In the hyperglycaemic group, 18 (72%) patients had prior history of diabetes and 15 (60%) were treated with an insulin infusion until they reached a BG level <140 mg/dL. At baseline, interlukin-6 and D-dimer levels were also significantly higher in the hyperglycaemic group than in the normoglycaemic group (p<.001). After adjusting the model for confounding variables, patients with hyperglycaemia treated with an insulin infusion had a lower risk of complications than patients who did not receive an insulin infusion. Additionally, interleukin-6 and D-dimer levels were reduced after treating the hyperglycaemia. The authors concluded that optimal glycaemic control during hospitalization could be associated with reduced risk of severe disease and death.
Another recent study showed that COVID-19 patients with recently diagnosed diabetes had the highest risk of all-cause mortality compared to those who had diabetes for a longer time (HbA1c !6.5%), hyperglycaemia or normoglycaemia [18]. Wang et al. [10] showed, in a relatively small sample (n ¼ 695), that fasting BG >126 mg/dL at admission was an independent predictor for 28-day mortality in patients without a previous diagnosis of diabetes.
Our study, which analyses the largest number of patients of any study to date, is consistent with all these results and reinforces the strong association found between hyperglycaemia and in-hospital mortality in non-critically patients hospitalized with COVID-19, independently of prior history of diabetes. Acute hyperglycaemia occurs in about 22% of patients hospitalized for COVID-19, while 18.9% of patients had diabetes, but 10% of them had admission BG <140 mg/dL. Other data registry shows that acute hyperglycaemia occurs in about 50%, while the prevalence of diabetes is about 7% [19].
One question that remains to be answered is whether hyperglycaemia plays any role in the physiopathology of the disease or if it is just an inflammatory bystander. Apart from the glycosylation of ACE2 receptors that facilitates virus binding and the inflammatory process that increases insulin resistance, the hypoxia that is normally present in patients with COVID-19 is frequently accompanied by disordered cellular glucose metabolism. Under anaerobic conditions, glucose ferments into lactate, which produces a limited amount of adenosine triphosphate (ATP). Hypoxia and ATP depletion cause an elevation of blood lactate and LDH levels. In our study, elevation of the LDH level was also associated with mortality according to BG levels, a finding that is consistent with the mechanism described above. This finding suggests that an early imbalance in glucose metabolism could be involved in a crucial manner in the physiopathology of the viral respiratory infection. Only one very small study suggests that supplemental oxygen at the earliest stages of COVID-19 could be useful in correcting an anaerobic glucose metabolism imbalance [20]. Adequate oxygen delivery and BG monitoring should be carried out for patients who are asked to remain at home in the early stages of the infection in order to prevent clinical deterioration. Early correction of hyperglycaemia in the course of COVID-19 could result in a decrease in the release of inflammatory cytokines and a reduction in the virus' ACE binding capacity, consequently resulting in better outcomes [11].
Both strategiesscreening for hyperglycaemia in patients without diabetes and early treatment of hyperglycaemiashould be mandatory in the management of patients hospitalized with COVID-19 [21]. Unfortunately, insufficient evidence is available on the benefits of strict glycaemic control in patients hospitalized with COVID-19 due to the fact that glycaemic management was underestimated during outbreak and the difficulties of multiple daily insulin injections and frequent point-of-care glucose testing in areas with high burden of COVID-19 patients [22,23]. Thus, to date, early glycaemic control may be a suitable therapeutic option to reduce the poor outcomes in hospitalized hyperglycaemic COVID-19 patients with or without a previous diabetes diagnosis [24].
This study has several limitations. First, it is an observational retrospective cohort study conducted during an outbreak, so there may be residual or unmeasured confounding factors. Second, most patients did not have a HbA1c measurement and as such, some patients classified as non-diabetic could have unknown diabetes. Third, the registry is missing data on some relevant inflammatory variables such as interleukin-6, D-dimer and serum ferritin. Finally, time from hospital admission to ICU admission was not available.
On the other hand, as a strength, our registry is the largest available cohorts of non-critically ill hospitalized patients with confirmed COVID-19 in contrast to other studies focuses on critically patients and it includes data from over 11,000 patients on admission BG levels before starting any treatment.
In conclusion, our study found that admission hyperglycaemia was an independent predictor of progression to critical condition and all-cause mortality in non-critically patients hospitalized with COVID-19. Moreover, this finding is independent of a prior history of diabetes. These results provided a simple and practical way to stratify risk of death in hospitalized patients with COVID-19. Hence, admission hyperglycaemia should not be overlooked, but rather detected and appropriately treated to improve the outcomes of COVID-19 patients with and without diabetes.

Author contributions
The research initiative and study concept were taken and designed by FJCS and AHJ. FJCS, FJMM, AHJ and RGH designed the present analysis. FJCS performed statistical analysis. FJCS, FJMM and RGH interpreted the data and drafted the manuscript. All authors interpreted the results, edited and revised the manuscript and read and approved the final version of the manuscript.

Disclosure statement
The authors hereby declare that they have no conflict of interest related to this article.

Funding
The study was completely support by the Spanish Federation of Internal Medicine.

Data availability statement
The steering committee of the Spanish SEMI-COVID-19 Registry will consider reasonable requests for the sharing of data. Requests should be made to the corresponding author.