Open access Original research Risk factors for severe outcomes in people with diabetes hospitalised for COVID-19: a cross-­sectional database study Emilio Ortega,1,2,3 Rosa Corcoy,4,5,6 Mònica Gratacòs,1 Francesc Xavier Cos Claramunt,1,7 Manel Mata-­Cases ‍ ‍,1,8,9 Ramon Puig-­Treserra,1 Jordi Real,1 Bogdan Vlacho,1 Esmeralda Castelblanco ‍ ‍,1,9 Pere Domingo,10 Kamlesh Khunti,11 Josep Franch-­Nadal,1,9,12 Didac Mauricio ‍ ‍1,4,9,13 To cite: Ortega E, Corcoy R, ABSTRACT Gratacòs M, et al. Risk factors Aim This study’s objective was to assess the risk of Strengths and limitations of this study for severe outcomes in people severe in-­hospital complications of patients admitted for with diabetes hospitalised for ►► A major strength of our study is the thorough COVID-19 and diabetes mellitus (DM). COVID-19: a cross-­sectional methodological approach to analyse the risk of in-­ Design This was a cross-­sectional study. database study. BMJ Open hospital COVID-19-­ related complications based 2021;11:e051237. doi:10.1136/ Settings We used pseudonymised medical record data on the presence of diabetes mellitus (DM) or overt bmjopen-2021-051237 provided by six general hospitals from the HM Hospitales hyperglycaemia. group in Spain. ►► Prepublication history and ►► We were limited by not having access to the pa- Outcome measures Multiple logistic regression analyses additional supplemental material tients’ medical history prior to admission, and the were used to identify variables associated with mortality and for this paper are available low number of registers for some important DM vari- the composite of mortality or invasive mechanical ventilation online. To view these files, ables (such as glycosylated haemoglobin (Hb1Ac)), please visit the journal online (IMV) in the overall population, and stratified for the and the lack of data on weight or body mass index (http://​dx.​doi.​org/​10.​1136/​ presence or absence of DM. Spline analysis was conducted (only the presence of obesity). bmjopen-​2021-​051237). on the entire population to investigate the relationship ►► The selection of subjects with DM was made based between glucose levels at admission and outcomes. Received 15 March 2021 on a proxy algorithm including DM diagnosis during Results Overall, 1621 individuals without DM and 448 Accepted 01 July 2021 the hospital stay, antidiabetic treatment, and HbA1c with DM were identified in the database. Patients with DM and blood glucose levels at admission. were on average 5.1 years older than those without. The ►► We used random blood glucose on admission for overall in-­hospital mortality was 18.6% (N=301), and was spline analyses, thus preventing the distinction be- higher among patients with DM than those without (26.3% tween stress-­related hyperglycaemia and uncon- vs 11.3%; p<0.001). DM was independently associated trolled pre-­existing DM. with death, and death or IMV (OR=2.33, 95% CI: 1.7 to 3.1 and OR=2.11, 95% CI: 1.6 to 2.8, respectively; p<0.001). In subjects with DM, the only variables independently associated with both outcomes were age >65 years, male pandemic, a public health emergency of sex and pre-­existing chronic kidney disease. We observed international importance. A few days later, a non-­linear relationship between blood glucose levels the respiratory disease caused by SARS-­CoV-2 at admission and risk of in-­hospital mortality and death was officially named COVID-19.1 2 The first or IMV. The highest probability for each outcome (around 50%) was at random glucose of around 550 mg/dL (30.6 person diagnosed as positive in Spain was © Author(s) (or their mmol/L), and the risks flattened above this value. confirmed on 31 January 2020, on the island employer(s)) 2021. Re-­use Conclusion The results confirm the high burden of La Gomera.3 The median age of hospi- permitted under CC BY-­NC. No associated with DM in patients hospitalised with COVID-19 talised patients infected with SARS-­ CoV-2 commercial re-­use. See rights and permissions. Published by infection, particularly among men, the elderly and those is 46.2 years, men comprise about 60% of BMJ. with impaired kidney function. Moreover, hyperglycaemia patients, and the average incubation period For numbered affiliations see on admission was strongly associated with poor outcomes, is 5.7 days.4 As of 8 February 2021, approxi- end of article. suggesting that personalised optimisation could help to mately 3 million people have been infected improve outcome during the hospital stay. with SARS-­CoV-2 in Spain since the start of Correspondence to the COVID-19 pandemic, and 62 295 persons Dr Didac Mauricio; INTRODUCTION ​didacmauricio@​gmail.​com and have died. On 30 January 2020, the WHO declared the Dr Josep Franch-­Nadal; Several meta-­analyses have reported that outbreak of the novel SARS-­CoV-2 coronavirus ​josep.​franch@​gmail.​com the most severe and fatal cases of COVID-19 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 1 Open access occur among the elderly and in patients with under- tocilizumab in patients with COVID-19-­related systemic lying comorbidities.5–7 Indeed, those with two or more inflammation.26 concomitant diseases have a significantly higher risk In Spain, DM is a highly prevalent disease in people of admission to an intensive care unit (ICU), invasive over 18 years of age (13.8% of the population).27 Given ventilation, or death compared with those with a single the high prevalence of DM and the additional challenging concomitant disease, or without comorbidities.8 The scenario that COVID-19 poses to healthcare professionals most prevalent comorbidities associated with increased in this particular population, it is crucial to accumulate COVID-19-­related morbidity and mortality are the pres- and share information and data from different countries ence of diabetes mellitus (DM), cardiovascular diseases and regions.28 Therefore, the main objective of this study (CVDs), chronic lung disease, chronic kidney disease was to assess the risk of in-­hospital COVID-19-­ related (CKD), hypertension, cancer and obesity.5–7 In addition, complications based on the presence of DM or overt the AB0 blood type may play a role in the susceptibility hyperglycaemia at admission in Spain. and severity of COVID-19 infection, which could be of importance in patients with underlying high-­risk condi- tions.8 For instance, it has been reported that non-0 blood group patients with hypertension have significantly higher METHODS values of pro-­thrombotic indexes and increased rates of Study design and settings cardiac injury and deaths compared with 0 patients.9 This was a cross-­sectional study in hospitalised individ- SARS-­CoV-2 uses ACE-2 as a cellular entry receptor, uals infected with SARS-­CoV-2, stratified by presence or and the spike protein of the virus needs to be cleaved by absence of DM. Data were obtained from pseudonymised cellular proteases (specifically TMPRSS2) to fuse with the electronic health records provided by six general hospi- cellular membrane.10 Although it was initially assumed tals from the HM Hospitales group (Spain). The data- that ACE inhibitors and angiotensin receptor blockers base included information related during the hospital to treat hypertension or cardiovascular conditions stay (diagnosis and procedure codes, prescribed medica- might exacerbate COVID-19 infection and lead to worse tions, vital signs and laboratory parameters), from 2310 outcomes, the most recent available meta-­analysis did not subjects during the first COVID-19 wave with hospital confirm this higher risk.11 Finally, it has been suggested admission between 27 January and 24 April 2020 (study that modulating TMPRSS2 expression through specific start and end date, respectively). Subjects were followed antibodies or non-­coding RNAs could prevent virus entry from admission to hospital discharge or death. Detailed into host cells,11 12 but these potential therapeutic options information related to the database is presented in the are still under investigation. online supplemental material (database description). Previous studies have reported that people with DM The REporting of studies Conducted using Observational are prone to new infections and recurrence, particularly Routinely-­collected Data Checklist is presented as online influenza and pneumonia, due to impaired defences and supplemental material. disease complications.13–16 Although the estimated preva- The study data were collected by medical professionals lence of DM in COVID-19-­infected patients varies greatly of the HM Hospitales group (Spain) during the first by geographical region, it is considered similar to the DM wave of the COVID-19 pandemic. In order to promote prevalence in the general population, thus not repre- COVID-19-­ related research, the HM Hospitales group senting a risk factor for infection.17 However, the preva- pseudonymised the medical history of SARS-­ CoV-2-­ lence of diabetes among COVID-19 hospitalised subjects infected patients and created a project titled: ‘Covid Data is higher than the overall diabetes prevalence.17 18 A study Save Lives’. Before access was granted, a formal petition, conducted in England found that one-­third of in-­hospital specific study protocol and ethics committee approval deaths occurred in people with type 2 DM and that these were obtained. patients had greater odds of COVID-19-­related in-­hos- pital death than those without DM.19 This observation Inclusion and exclusion criteria has been confirmed in a meta-­analysis showing that DM The study enrolled people over 18 years of age with is associated with a twofold higher risk of dying from SARS-­ CoV-2 infection (COVID-19 positive) microbio- COVID-19,20 and a second study reporting that patients logically proven by reverse transcription PCR (RT-­PCR). with pre-­ existing DM have a threefold greater risk of Those with DM were identified in the database if they: (1) in-­hospital mortality.21 had any ICD-10 (International Statistical Classification of Early reports showed that about half of patients with Diseases) diagnostic code for type 1 or type 2 DM (ie, severe COVID-19 presented acute hyperglycaemia, with E.10 and E.11), (2) were on treatment with antidiabetic no more than 10% of them having a prior diagnosis of drugs, (3) had a register of insulin use within the first 24 DM.22 23 Following these observations, two meta-­analyses hours after admission, or (4) had a glycosylated haemo- concluded that hyperglycaemia at hospital admission is globin (HbA1c) value ≥6.5% (48 mmol/mol; first avail- associated with severe complications and mortality, regard- able record after admission) or baseline blood glucose less of diabetes status.24 25 Moreover, hyperglycaemia also (BG) values ≥200 mg/dL (11.1 mmol/L; recorded has a negative impact on the therapeutic response to within the first 24 hours of admission). Subjects with no 2 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 Open access confirmation of SARS-­CoV-2 infection and those younger analyses were performed using the R statistical software than 18 years were excluded from the study. V.3.6.1 (https://www.​r-​project.​org/). Study variables Patient and public involvement The following baseline variables were collected: age and Patients or the public were not involved in the design, sex; SARS-­CoV-2 diagnosis (positive RT-­PCR); comorbid- or conduct, or reporting, or dissemination plans of our ities (ie, hypertension, hyperlipidaemia, obesity (body research. mass index (BMI) ≥30 kg/m2), CVD, heart failure, cerebrovascular diseases, ischaemic heart disease, CKD, chronic obstructive pulmonary disease (COPD), asthma, RESULTS mental disorders and cancer); blood laboratory param- Baseline characteristics eters (ie, HbA1c, BG, electrolytes, renal function, liver Of the 2306 subjects admitted to hospital within the function, haematology and coagulation, inflammation period of study, 2069 were over 18 years of age and had a markers and gas tests); clinical parameters (ie, systolic positive diagnostic test for SARS-­CoV-2 (figure 1). Among and diastolic blood pressure, heart rate and tempera- them, 448 (21.7%) were identified as having DM and ture); and concomitant medications (ie, baseline insulins, 1621 (78.3%) without DM (non-­DM group). The char- systemic corticosteroids, antimicrobials, anticoagulants acteristics of the two populations at hospital admission and antiplatelet agents, and antihypertensive and lipid-­ are shown in table 1. Subjects with DM were on average lowering drugs). 5.1 years older than subjects without DM, and more We considered the following variables as events or frequently men (67.9% vs 58.6%). Moreover, individuals complications during the hospital stay: death, acute respi- in the DM group had a poor comorbidity profile, with a ratory distress syndrome (ARDS), pulmonary thrombosis, higher frequency of all assessed prior conditions except neurological complications (including encephalopathy, for cerebrovascular diseases and asthma. encephalitis, myelitis and encephalomyelitis), throm- Regarding laboratory parameters on admission (online botic complications identified by ICD-10 diagnostic supplemental table 1), the DM group had slightly codes (phlebitis and thrombophlebitis), admission to lower estimated glomerular filtration rates (eGFRs) ICU and invasive mechanical ventilation (IMV) identi- (73.5±26.5 mL/min/1.73 m2 vs 81.2±23.9 mL/min/1.73 fied by ICD-10 procedure codes. The composite primary m2; p<0.001), and higher levels of serum creatinine outcome was defined as death or IMV. (1.09±0.72 mg/dL vs 0.94±0.51 mg/dL; p<0.001) than the non-­DM group. Regarding markers of inflammation and infection, we observed higher levels of C reactive protein Statistical methods and procalcitonin in the DM group (97.1±107 mg/L The demographic and clinical characteristics of the vs 75.9±82.5 mg/L and 0.66±1.30 mg/L vs 0.39±1.30 two groups of hospitalised patients (ie, with or without mg/L, respectively; p<0.001). We also observed higher DM) were compared and summarised at the quantitative levels of D-­dimer, a marker of endothelial and coagu- (minimum, maximum, median, first and third quartile, lation dysfunction, in the DM group (3990±10 800 ng/ mean and SD) or categorical level (frequency, number mL vs 2340±6720 ng/mL, respectively). Regarding the and %). pharmacological therapy used during the hospital stay, The association between the study outcomes (ie, we observed differences and increased use of almost all mortality and mortality or mechanical ventilation) and drugs of interest among subjects with DM, compared with DM was performed using logistic regression analyses those without DM, especially for diuretics, systemic corti- adjusted for sex, age and associated risk factors. In addi- costeroids and tocilizumab. tion, several models of interest were tested (a model with basic clinical variables such as age and sex, a model Events and complications during in-hospital stay adding obesity, hypertension and hyperlipidaemia, and A total of 301 (14.5%) subjects positive for SARS-­CoV-2 a model adding organ lesion variables, such as CVD, died in hospital, 118 (26.3%) out of 448 in the DM group heart failure, CKD, COPD), namely with the sequential and 183 (11.3%) out of 1621 in the non-­ DM group inclusion of different covariates and the estimated differ- (p<0.001; figure 2). All studied events, except pulmonary ences expressed as OR and the respective 95% CIs. We embolism and thrombotic or neurological complications, evaluated goodness of fit of the logistic regression models were significantly more frequent among patients with DM with Hosmer-­Lemeshow test. To analyse the non-­linear than those without (figure 2). The most frequent outcome relationship of random BG levels on admission with the was the composite of death or IMV (31% in the DM group two study outcomes, we used an adjusted semiparametric vs 14% in the non-­DM group; figure 2) followed by death model (generalised additive model) calculating the spline (26.3% vs 11.3%), admission to ICU (21% vs 6.9%), IMV curves with two df (knots) using the mgcv package in R, (10.7% vs 4.2%) and ARDS (3.8% vs 1.5%). V.1.8–3129 with adjustment for potential confounders. We The frequency of events by group and age showed analysed the entire database available and no statistical that, in both subjects with and without DM, death and power was calculated. Data management and statistical the composite of death or IMV were significantly more Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 3 Open access Figure 1 Flow chart diagram. frequent among those >65 years (online supplemental The multiple logistic regression models were repeated figure 1). In contrast, the proportion of subjects requiring to rule out the potential interaction of DM with different IMV and ICU admission was significantly higher among clinical conditions (ie, obesity, hyperlipidaemia, obesity those ≤65 years and with DM, while age was not significant and hyperlipidaemia, heart failure, CKD and COPD) in those without DM. When stratifying the results by sex, for the in-­hospital death outcome. The results showed we did not observe differences except for admission to that none of these conditions affected the relationship ICU, which was significantly more frequent among male between the risk of death and DM (online supplemental subjects with DM (online supplemental figure 1). Within table 4). the diabetes group, when we stratified by pre-­existing DM (DM codes and/or HBA1c ≥6.5% and/or antidia- Factors associated with hospital death and death or IMV by betic treatment) and ‘stress’ hyperglycaemia/unknown comorbid diabetes diabetes (glucose ≥200 mg/dL or insulin use within the A subanalysis was performed separately for subjects with first 24 hours period after admission), we observed higher or without DM. In the DM group, the only variables percentages for death, death or IMV, ARDS, admission independently associated with the risk of both mortality to ICU and IMV events in subjects with ‘stress’ hypergly- and death or IMV were male sex, older age and CKD caemia. The results of this stratification are presented in (figure 4A and online supplemental tables 5 and 6). In online supplemental table 2. contrast, in subjects without DM, besides the aforemen- tioned variables, the odds of death were also increased Baseline demographic and clinical characteristics associated among subjects with CVD (OR=1.94, 95% CI=1.03 to 3.7), with in-hospital death and death or IMV and the odds of death or IMV among those with obesity For the overall hospitalised population, the demographic or COPD (OR=2.96, 95% CI=1.7 to 5.3 and OR=2.30, characteristics significantly associated with mortality were 95% CI=1.4 to 3.8, respectively) (figure 4B and online male sex and older age (OR=1.98, 95% CI=1.2 to 3.3 and supplemental tables 5 and 6). OR=1.10, 95% CI=1.08 to 1.11, respectively) (figure 3; Factors associated with hospital death and death or IMV by online supplemental table 3). The comorbidities inde- glucose levels at admission pendently associated with increased odds of death were We used non-­parametric logistic regression models to DM (OR=2.33, 95% CI=1.7 to 3.1), CKD (OR=2.14, 95% assess whether there was a relationship between random CI=1.2 to 3.7) and COPD (OR=1.72, 95% CI=1.1 to 2.8). BG on admission and the risk of mortality (and death When considering the composite outcome of death or IMV). We observed a marked non-­linearity in the or IMV, the same variables associated with death (ie, effect of BG on admission in the risk of both outcomes age, sex, diabetes, CKD and COPD) were identified as (figure 5A,B and online supplemental table 7). While increasing the risk. In addition, obesity emerged as an the risk was increased among subjects with high random independently associated variable (OR=1.98, 95% CI=1.5 BG levels on admission, the magnitudes of the associ- to 2.7) (figure 3, online supplemental table 3). ated mortality differed depending on the baseline 4 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 Open access Table 1 Baseline characteristics of the studied cohorts at hospital admission Overall study population Diabetes No diabetes Characteristic N=2069 N=448 N=1621 P value Age, mean (SD), years 67.8 (15.7) 71.7 (11.9) 66.6 (16.3) <0.001 Age, median (P25, P75), years 69.0 (57.0, 80.0) 72.0 (64.0, 80.0) 67.0 (55.0, 79.0) <0.001 Sex (male), n (%) 1205 (60.3) 304 (67.9) 950 (58.6) <0.001 Glucose, mean, (SD) <0.001  mg/dL 124 (47.7) 168 (74.4) 112 (24.8)  mmol/L 6.8 (2.6) 9.3 (4.1) 6.2 (1.4) Comorbidities, n (%)  Hypertension 651 (31.5) 224 (50.0) 427 (26.3) <0.001  Hyperlipidaemia 409 (19.8) 154 (34.4) 255 (15.7) <0.001  Obesity 117 (5.65) 45 (10.0) 72 (4.44) <0.001  Cardiovascular diseases 77 (3.72) 28 (6.25) 49 (3.02) 0.002  Heart failure 51 (2.46) 18 (4.02) 33 (2.04) 0.026  Cerebrovascular diseases 27 (1.30) 10 (2.23) 17 (1.05) 0.086  Ischaemic heart disease 47 (2.27) 18 (4.02) 29 (1.79) 0.009  Chronic kidney disease 76 (3.67) 30 (6.70) 46 (2.84) <0.001  Chronic obstructive pulmonary disease 112 (5.41) 34 (7.59) 78 (4.81) 0.029  Asthma 2 (0.10) 0 (0.00) 2 (0.12) 1.000  Mental disorders 114 (5.51) 35 (7.81) 79 (4.87) 0.022  Cancer 117 (5.65) 36 (8.04) 81 (5.00) 0.019 Pharmacological therapy, n (%)  Biguanides 66 (3.19) 66 (14.7) 0 (0.00) <0.001  Sulfonylureas 1 (0.05) 1 (0.22) 0 (0.00) 0.217  Dipeptidyl peptidase 4 inhibitors 11 (0.53) 11 (2.46) 0 (0.00) <0.001  Fast-­acting insulins 95 (4.5) 66 (14.7) 29 (1.79) <0.001  Intermediate-­acting insulins 9 (0.43) 7 (1.56) 2 (0.12) 0.001  Long-­acting insulins 23 (1.11) 20 (4.46) 3 (0.19) <0.001  Antibiotics 1882 (91.0) 421 (94.0) 1461 (90.1) 0.016  Antithrombotics 1752 (84.7) 396 (88.4) 1356 (83.7) 0.017  Renin-­angiotensin system agents 523 (25.3) 153 (34.2) 370 (22.8) <0.001  Beta blocking agents 316 (15.3) 104 (23.2) 212 (13.1) <0.001  Calcium channel blockers 384 (18.6) 118 (26.3) 266 (16.4) <0.001  Diuretics 508 (24.6) 185 (41.3) 323 (19.9) <0.001  Statins 256 (12.4) 88 (19.6) 168 (10.4) <0.001  Systemic corticosteroids 977 (47.2) 267 (59.6) 710 (43.8) <0.001  Tocilizumab 421 (20.3) 137 (30.6) 284 (17.5) <0.001 P25, P75=25th and 75th percentile, respectively. values, with a large increase in the log odds of death or CKD and COPD were associated with in-­hospital death IMV with values up to 200 mg/dL (11.1 mmol/L), and (online supplemental table 7). These variables were smaller increases above this level. The logistic regres- linked to death or IMV too, but obesity was an addi- sion models (figure 6A,B) showed that the highest tional risk factor (online supplemental table 7). probability of death (near 50%) was at around 550 mg/ dL (30.6 mmol/L) and, above this value, the mortality DISCUSSION risk flattened. Finally, the multivariate model showed Data from this cross-­sectional study showed that the that, beside glucose at admission, male sex, older age, COVID-19-­ related in-­ hospital death rate was higher Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 5 Open access Figure 2 Proportion of events (%) during hospitalisation according to the presence of diabetes. ***P<0.001; **p<0.01; *p<0.05. ARDS, acute respiratory distress syndrome; DM, diabetes mellitus; ICU, intensive care unit; IMV, invasive mechanical ventilation. among subjects with DM than those without. Moreover, Moreover, patients with DM were more frequently men DM was independently associated with the risk of in-­hos- and over 65 years, had more comorbid conditions, and pital case fatality and the composite outcome, death or higher levels of inflammatory, endothelial, and coag- IMV. In the DM group, both outcomes were associated ulation dysfunction markers on admission. Different with older age, male sex and pre-­existing CKD. Finally, meta-­analyses have reported that older age and male we observed a non-­linear relationship between BG levels sex are characteristics associated with severe COVID-19 on admission and the probability of death and death or infection and high fatality rates.17 30 31 Along the same IMV in the overall inpatient population. line, studies assessing the phenotypic characteristics In our study, the proportion of severe COVID-19 cases of patients with COVID-19 with pre-­existing DM have (eg, requiring IMV or ICU admission) in the popula- found that those with severe infection were older, had tion with DM was higher than in the non-­DM cohort. more comorbidities (ie, cerebrovascular disease, CVD, hypertension and COPD), and increased values of inflammation, endothelial and coagulation dysfunction markers (eg, D-­dimer, procalcitonin and thrombocyto- penia), than those without DM.30–35 In our study, patients with DM had significantly higher creatinine on admission, lower eGFR and more frequently pre-­existing CKD than subjects without DM. Besides, CKD was the only comorbid condition increasing the odds (threefold increase) of in-­hospital death (and death or IMV) among the DM cohort after adjusting for age, sex and confounding variables. Different meta-­analyses have identified CKD as a risk factor for severity and in-­hospital death in SARS-­CoV-2-­infected patients.7 36–38 Moreover, a recent study conducted in Danish hospital-­ diagnosed patients with COVID-19 reported that kidney insufficiency was independently associated with increased risk of severe disease or death, and the degree of renal impairment inversely correlated with the rate of adverse outcomes.39 Although it is difficult to distinguish whether poor outcomes are linked to acute kidney injury (AKI) developed during the course of the disease, or to pre-­existing CKD,39 a study conducted in Spain showed that patients with increased creatinine on admission, previous CKD or developing AKI had a higher risk of in-­hospital death than those with normal creatinine on admission.40 Of note, the authors found that older age and diabetes, but not other comorbidities, were associ- Figure 3 Clinical and demographic variables associated ated with in-­hospital death.40 Finally, a study conducted with increased risk of in-­hospital death and the composite outcome of death or invasive mechanical ventilation (IMV). in Mexico reported that patients with DM and CKD had CKD, chronic kidney disease; COPD, chronic obstructive a twofold higher rate of intubation, 56% higher ICU pulmonary disease; CVD, cardiovascular disease; DM, admission and 21% excess probability of case fatality diabetes mellitus; HF, heart failure. once admitted, than subjects with CKD alone.41 6 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 Open access Figure 5 Spline plot demonstrating a marked non-­linearity in the relationship between plasma glucose (mg/dL) levels on admission and the log odds of death (A) and death or invasive mechanical ventilation (IMV) rate (B). Tick marks above the horizontal axis indicate the values at which the observations were made. The dotted lines represent the 95% CI. The model was adjusted for age, sex, obesity, hypertension, hyperlipidaemia, history of CVD, HF, CKD and COPD. CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular disease; HF, heart failure. associated with increased mortality and poor outcome, regardless of pre-­existing DM.43 These results confirmed Figure 4 Clinical and demographic variables associated previous observations that FBG on admission, and the with increased risk of in-­hospital death and the composite odds of being admitted to the ICU, followed a logarithmic outcome of death and/or invasive mechanical ventilation association, with different magnitudes of risk depending (IMV) in subjects with diabetes (A) and without diabetes (B). on the baseline level.42 We add to the literature that, CKD, chronic kidney disease; COPD, chronic obstructive besides the previously reported effect of hyperglycaemia pulmonary disease; CVD, cardiovascular disease; HF, heart on the risk of COVID-19 severity, ICU admission and failure. mortality,24 25 BG has a non-­linear relationship with case fatality and the risk of death or IMV. It is possible that In our study, we used splines as a scientific and pref- this relationship was also accompanied by, or reflected erable alternative to the categorisation of BG levels.42 glycaemic variability and less time spent in range. Indeed, We used this approach because a recent dose–response glycaemic fluctuation has been reported to be inde- meta-­ analysis demonstrated a non-­ linear relationship pendently associated with poor prognosis and mortality between admission fasting blood glucose (FBG) level and in COVID-19 hospitalised patients.44 In the same vein, a COVID-19 severity, with high levels being significantly study on ICU patients showed that the less time spent in Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 7 Open access data on sociodemographic characteristics (ethnicity, race, economic or educational status) and toxic habits (smoking, alcohol or drug use) were not available. Second, we had very few registers for some important variables for diabetes, such as Hb1Ac (data from only 36 patients) and no data on weight or BMI. Indeed, no more than 10% of the patients had documented obesity, which is clearly lower than the expected prevalence in the general population. This was most probably related to the clinician’s under-­recording of this particular condition and to the fact that, during the first wave, obesity had not yet been identified as a significant risk factor and thus not specifically registered. Thirdly, the selection of DM subjects was made based on a proxy algorithm (including DM diagnosis during the hospital stay, antidiabetic treat- ment, and HbA1c and blood glucose levels at admission), which could have introduced selection or referral bias, potentially leading to an inaccurate estimation of DM prevalence. Besides, we had no access to the patients’ treatments prior to hospital admission. Since the propor- tion of patients identified as having diabetes and receiving glucose-­lowering agents was surprisingly low (approxi- mately 40%), this can be also attributed to antidiabetic treatment under-­ reporting at admission. Fourth, and inherent to data coming from hospital medical records, missing values could have reduced the statistical power of the study, or produced biased estimates. Fifth, we used random BG on admission for the spline analyses, thus preventing the distinction between stress-­related hyper- glycaemia and uncontrolled pre-­existing DM. This also Figure 6 Predicted probability of in-­hospital death (A) and prevented the analysis of time in range or BG variability, death or IMV (B) based on generalised smoothing splines. both being linked to increased severity, case fatality and The shaded area represents the 95% CI. The model was poor COVID-19 outcomes.42–46 Lastly, the study period adjusted for age, sex, obesity, hypertension, hyperlipidaemia, coincides with the height of the first pandemic wave in history of CVD, HF, CKD and COPD. CKD, chronic kidney Spain, when there was a shortage of ventilators and inten- disease; COPD, chronic obstructive pulmonary disease; sive care beds. At this point, age was the deciding factor CVD, cardiovascular disease; HF, heart failure; IMV, invasive for whether or not someone received potentially life-­ mechanical ventilation. saving ICU care. This might be reflected in our results, where in-­hospital death was more frequent among those range was associated with increased utilisation of a venti- over 65 years, but ICU admission was more frequent lator, prolonged mechanical ventilation and increased among those ≤65 years. mortality.45 Most importantly, a spline analysis of glucose levels in patients with DM with continuous glucose moni- toring showed a non-­ linear relationship between time CONCLUSIONS spent above range and glycaemic variability with the The results of our study confirm the high burden associ- increased likelihood of composite adverse COVID-19 ated with DM in patients hospitalised due to SARS-­CoV-2 outcomes (need for ICU admission, mechanical ventila- infection. Comorbid DM poses a challenge to health tion or critical illness).46 professionals and the system because it is associated with severe disease, higher ICU admission rates, IMV and ulti- Limitations of this study mately death, particularly among the elderly. The non-­ The findings of this study must be interpreted with linear relationship of hyperglycaemia at admission with caution and a number of limitations should be borne increased odds of death and IMV suggests that optimising in mind. First, we had limited data for SARS-­ CoV-2-­ glycaemic control during the hospital stay could help to infected persons. For instance, we did not have access to reduce in-­hospital death and the composite death/IMV. the patient’s medical history prior to admission; so the Besides, out-­of-­hospital care should be a priority to reduce possibility exists that some important medical conditions or prevent uncontrolled glycaemia among those with DM, were not included in the emergency room medical report as it could potentially help reduce poor outcomes when and, therefore, not included in the analysis. Moreover, hospitalisation is required. 8 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 Open access Author affiliations Ethics approval The study was approved by the Ethics Committee of the Primary 1 DAP-­Cat group, Unitat de Suport a la Recerca Barcelona, Fundació Institut Health Care University Research Institute (IDIAP) Jordi Gol, Barcelona (approval Universitari per a la recerca a l'Atenció Primària de Salut Jordi Gol i Gurina number: 20/089-­PCV). This study does not involve animal subjects. (IDIAPJGol), Barcelona, Spain Provenance and peer review Not commissioned; externally peer reviewed. 2 CIBER of Pathophysiology of Obesity and Nutrition (CIBEROBN), Instituto de Salud Carlos III (ISCIII), Madrid, Spain Data availability statement Data may be obtained from a third party and are not 3 Department of Endocrinology and Nutrition, Institut d’Investigacions Biomèdiques publicly available. August Pi i Sunyer, Hospital Clinic, Barcelona, Spain Supplemental material This content has been supplied by the author(s). It has 4 Department of Endocrinology and Nutrition, Hospital Universitari de la Santa Creu i not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been Sant Pau, Barcelona, Spain peer-­reviewed. Any opinions or recommendations discussed are solely those 5 Department of Medicine, Universitat Autònoma de Barcelona, Barcelona, Spain of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and 6 CIBER of Bioengineering, Biomaterials and Nanomedicine (CIBER-­BBN), Instituto de responsibility arising from any reliance placed on the content. Where the content Salud Carlos III (ISCIII), Madrid, Spain includes any translated material, BMJ does not warrant the accuracy and reliability 7 Innovation office, Institut Català de la Salut, Barcelona, Spain of the translations (including but not limited to local regulations, clinical guidelines, 8 Primary Health Care Center La Mina, Gerència d’Àmbit d’Atenció Primària terminology, drug names and drug dosages), and is not responsible for any error Barcelona Ciutat, Institut Català de la Salut, Sant Adrià de Besòs, Spain and/or omissions arising from translation and adaptation or otherwise. 9 CIBER of Diabetes and Associated Metabolic Diseases (CIBERDEM), Instituto de Open access This is an open access article distributed in accordance with the Salud Carlos III (ISCIII), Madrid, Spain Creative Commons Attribution Non Commercial (CC BY-­NC 4.0) license, which 10 Infectious Diseases, Hospital Universitari de la Santa Creu i Sant Pau, Institut de permits others to distribute, remix, adapt, build upon this work non-­commercially, Recerca Hospital Universitari de la Santa Creu i Sant Pau, Barcelona, Spain and license their derivative works on different terms, provided the original work is 11 Diabetes Research Centre, University of Leicester, Leicester, UK properly cited, appropriate credit is given, any changes made indicated, and the use 12 Primary Health Care Center Raval Sud, Gerència d’Atenció Primaria, Institut Català is non-­commercial. See: http://​creativecommons.​org/​licenses/​by-​nc/​4.​0/. de la Salut, Barcelona, Spain 13 ORCID iDs Departament of Medicine, University of Vic - Central University of Catalonia, Vic, Manel Mata-­Cases http://​orcid.​org/​0000-​0003-​3693-​3622 Barcelona, Spain Esmeralda Castelblanco http://​orcid.​org/​0000-​0002-​2061-​6270 Didac Mauricio http://​orcid.​org/​0000-​0002-​2868-​0250 Twitter Emilio Ortega @EmilioOrtegaMdV and Didac Mauricio @DidacMauricio Acknowledgements Covid Data Save Lives—Hospitales HM for providing database. CIBER of Diabetes and Associated Metabolic Diseases (CIBERDEM) and CIBER of physiopathology of obesity and Nutrition (CIBEROBN) are initiatives from the Instituto de Salud Carlos III, Madrid, Spain. The authors acknowledge Amanda REFERENCES Prowse (Lochside Medical communications) for providing support in paper editing. 1 World Health Organization. 15-­Novel Coronavirus(2019-­nCoV). WHO KK is supported by the National Institute for Health Research (NIHR) Applied Bull 2020:1–7. 2 Agència de Salut Pública de Catalunya (ASPCAT). Informe tècnic Research Collaboration East Midlands (ARC EM) and the NIHR Leicester Biomedical resum dels casos de covid-19 a Catalunya, 2020. Available: http://​ Research Centre (BRC). salutpublica.​gencat.​cat/​web/.​content/​minisite/​aspcat/​butlletins/​ Contributors Conceptualisation—EO, JF-­N, RC, MM-­C, BV, KK and DM. vigilanciaaspcat/​2020/​45/​INFORME-​TECNIC-​3-​COVID-​19_​020420.​ Methodology—EO, JF-­N, RC, MM-­C, BV, KK and DM. Formal analysis—RP-­T pdf 3 Linde P. Sanidad confirma en La Gomera El primer caso de and JR. Resources and data curation—RP-­T, JR and BV. Writing (original draft coronavirus en España. El País. ISSN 1134-6582, February 1, 2020. preparation)—BV and MG. Writing (review and editing)—EO, JF-­N, RC, MM-­ Available: https://​elpais.​com/​sociedad/​2020/​01/​31/​actualidad/​ C, MG, FXCC, EC, BV, KK, DM and PD. Supervision—DM, RC and JF-­N. Project 1580509404_​469734.​htm administration—BV. 4 Muniyappa R, Gubbi S. COVID-19 pandemic, coronaviruses, and diabetes mellitus. Am J Physiol Endocrinol Metab 2020;318:E736–41. Funding This study was supported by the Primary Care Diabetes Europe grant 5 Zhou Y, Yang Q, Chi J, et al. Comorbidities and the risk of severe (grant number FEr20/0020). or fatal outcomes associated with coronavirus disease 2019: a Competing interests EO has received advisory and or speaking fees from systematic review and meta-­analysis. Int J Infect Dis 2020;99:47–56. AstraZeneca, Boehringer Ingelheim, Lilly, MSD, Novo Nordisk, Sanofi and Amgen; 6 Deng G, Yin M, Chen X, et al. Clinical determinants for fatality of 44,672 patients with COVID-19. Crit Care 2020;24:179. he received research grants to the institution from MSD and Amgen. RC has 7 Mesas AE, Cavero-­Redondo I, Álvarez-­Bueno C, et al. Predictors of received advisory and/or speaking fees from Abbott, Ascensia, Lilly, MSD, Novo in-­hospital COVID-19 mortality: a comprehensive systematic review Nordisk and Sanofi. MM-­C has received advisory honorarium from AstraZeneca, and meta-­analysis exploring differences by age, sex and health Bayer, Boehringer Ingelheim, GSK, Lilly, MSD, Novartis, Novo Nordisk and Sanofi; conditions. PLoS One 2020;15:e0241742. he received speaker honoraria from AstraZeneca, Bayer, Boehringer Ingelheim, 8 Pendu JL, Breiman A, Rocher J, et al. Abo blood types and GSK, Lilly, Menarini, MSD, Novartis, Novo Nordisk and Sanofi; he received research COVID-19: spurious, anecdotal, or truly important relationships? A grants to the institution from AstraZeneca, GSK, Lilly, MSD, Novartis, Novo Nordisk Reasoned review of available data. Viruses 2021;13:160. and Sanofi. JF-­N has received advisory and or speaking fees from AstraZeneca, 9 Sardu C, Marfella R, Maggi P, et al. Implications of AB0 blood group in hypertensive patients with covid-19. BMC Cardiovasc Disord Ascensia, Boehringer Ingelheim, GSK, Lilly, MSD, Novartis, Novo Nordisk and 2020;20:373. Sanofi; he received research grants to the institution from AstraZeneca, GSK, Lilly, 10 Hoffmann M, Kleine-­Weber H, Schroeder S, et al. SARS-­CoV-2 cell MSD, Novartis, Novo Nordisk, Sanofi and Boehringer. KK has acted as a consultant, entry depends on ACE2 and TMPRSS2 and is blocked by a clinically speaker or received grants for investigator-­initiated studies for AstraZeneca, proven protease inhibitor. Cell 2020;181:271–80. Novartis, Novo Nordisk, Sanofi-­Aventis, Lilly and Merck Sharp & Dohme, Boehringer 11 Baral R, Tsampasian V, Debski M, et al. Association between renin-­ Ingelheim, Bayer, Berlin-­Chemie AG/Menarini Group, Janssen and Napp. DM has angiotensin-­aldosterone system inhibitors and clinical outcomes received advisory and/or speaking fees from AstraZeneca, Ascensia, Boehringer in patients with COVID-19: a systematic review and meta-­analysis. JAMA Netw Open 2021;4:e213594. Ingelheim, GSK, Lilly, MSD, Novartis, Novo Nordisk and Sanofi; he received research 12 Matarese A, Gambardella J, Sardu C, et al. miR-98 regulates grants to the institution from AstraZeneca, GSK, Lilly, MSD, Novartis, Novo Nordisk, TMPRSS2 expression in human endothelial cells: key implications for Sanofi and Boehringer. PD has received lecture and Advisory Board fees from COVID-19. Biomedicines 2020;8:462. Gilead Sciences, Roche, Merck, Sharp & Dohme, ViiV Healthcare, Janssen & Cilag, 13 Papazafiropoulou AK, Antonopoulos S. The COVID-19 pandemic Theratechnologies, Boehringer Ingelheim and Ferrer International. PD has received and diabetes mellitus. Archives of medical sciences. Atherosclerotic research grants from Gilead Sciences, ViiV Healthcare, GSK, Janssen & Cilag and Diseases 2020;5:e200. Boehringer Ingelheim. BV, FXCC, JR, RP-­T, MG and EC have no conflict of interest to 14 McDonald HI, Nitsch D, Millett ERC, et al. New estimates of the declare. burden of acute community-­acquired infections among older people with diabetes mellitus: a retrospective cohort study using linked Patient consent for publication Not required. electronic health records. Diabet Med 2014;31:606–14. Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237 9 Open access 15 Li S, Wang J, Zhang B, et al. Diabetes mellitus and cause-­ 32 Cariou B, Hadjadj S, Wargny M, et al. Phenotypic characteristics specific mortality: a population-­based study. Diabetes Metab J and prognosis of inpatients with COVID-19 and diabetes: the 2019;43:319–41. CORONADO study. Diabetologia 2020;63:1500–15. 16 Del Sole F, Farcomeni A, Loffredo L, et al. Menichelli D, Vicario T, et 33 Zhu L, She Z-­G, Cheng X, et al. Association of blood glucose control al. Features of severe COVID-19: A systematic review and meta-­ and outcomes in patients with COVID-19 and pre-­existing type 2 analysis. Eur J Clin Invest 2020;50:0–1. diabetes. Cell Metab 2020;31:1068–77. 17 Pugliese G, Vitale M, Resi V, et al. Is diabetes mellitus a risk factor for 34 Elamari S, Motaib I, Zbiri S, et al. Characteristics and outcomes coronavirus disease 19 (COVID-19)? Acta Diabetol 2020;19. of diabetic patients infected by the SARS-­CoV-2. Pan Afr Med J 18 Peric S, Stulnig TM. Diabetes and COVID-19. Wien Klin Wochenschr 2020;37:32. Epub ahead of print. 2020;132:356–61. 35 Yan Y, Yang Y, Wang F, et al. Clinical characteristics and outcomes of 19 Barron E, Bakhai C, Kar P, et al. Associations of type 1 and type patients with severe covid-19 with diabetes. BMJ Open Diabetes Res 2 diabetes with COVID-19-­related mortality in England: a whole-­ Care 2020;8:e001343. population study. Lancet Diabetes Endocrinol 2020;8:813-822. 36 Laguna-­Goya R, Utrero-­Rico A, Talayero P, et al. IL-6-­based mortality 20 Tian W, Jiang W, Yao J, et al. Predictors of mortality in hospitalized risk model for hospitalized patients with COVID-19. J Allergy Clin COVID‐19 patients: a systematic review and meta‐analysis. J Med Immunol 2020;146:799–807. Virol 2020. 37 Working group for the surveillance and control of COVID-19 in 21 Mantovani A, Byrne CD, Zheng M-­H, et al. Diabetes as a risk Spain, Members of the Working group for the surveillance and factor for greater COVID-19 severity and in-­hospital death: a control of COVID-19 in SpainRedondo-­Bravo L. The first wave of meta-­analysis of observational studies. Nutr Metab Cardiovasc Dis the COVID-19 pandemic in Spain: characterisation of cases and 2020;30:1236–48. risk factors for severe outcomes, as at 27 April 2020. Euro Surveill 22 Wu C, Chen X, Cai Y, et al. Risk factors associated with acute 2020;25:2001431. respiratory distress syndrome and death in patients with coronavirus 38 Tian W, Jiang W, Yao J, et al. Predictors of mortality in hospitalized disease 2019 pneumonia in Wuhan, China. JAMA Intern Med COVID-19 patients: a systematic review and meta-­analysis. J Med 2020;180:934–43. Virol 2020;92:1875–83. 23 Li X, Xu S, Yu M, et al. Risk factors for severity and mortality 39 Carlson N, Nelveg-­Kristensen K-­E, Freese Ballegaard E, et al. in adult COVID-19 inpatients in Wuhan. J Allergy Clin Immunol Increased vulnerability to COVID-19 in chronic kidney disease. J 2020;146:110–8. Intern Med 2021;290:166–78. 24 Lee MH, Wong C, Ng CH, et al. Effects of hyperglycaemia on 40 Portolés J, Marques M, López-­Sánchez P, et al. Chronic kidney complications of COVID-19: a meta-­analysis of observational disease and acute kidney injury in the COVID-19 Spanish outbreak. studies. Diabetes Obes Metab 2021;23:287–9. Nephrol Dial Transplant 2020;35:1353–61. 25 Yang Y, Cai Z, Zhang J. Hyperglycemia at admission is a strong 41 Leon-­Abarca JA, Memon RS, Rehan B, et al. The impact of predictor of mortality and severe/critical complications in COVID-19 COVID-19 in diabetic kidney disease and chronic kidney disease: a patients: a meta-­analysis. Biosci Rep 2021;41. doi:10.1042/ population-­based study. Acta Biomed 2020;91:e2020161. BSR20203584. [Epub ahead of print: 26 02 2021]. 42 Alahmad B, Al-­Shammari AA, Bennakhi A, et al. Fasting blood 26 Marfella R, Paolisso P, Sardu C, et al. Negative impact of glucose and COVID-19 severity: nonlinearity matters. Diabetes Care hyperglycaemia on tocilizumab therapy in Covid-19 patients. 2020;43:3113–6. Diabetes Metab 2020;46:403–5. 43 Lazarus G, Audrey J, Wangsaputra VK, et al. High admission 27 Soriguer F, Goday A, Bosch-­Comas A, et al. Prevalence of diabetes blood glucose independently predicts poor prognosis in COVID-19 mellitus and impaired glucose regulation in Spain: the ​Di@​bet.​es patients: a systematic review and dose-­response meta-­analysis. Study. Diabetologia 2012;55:88–93. Diabetes Res Clin Pract 2021;171:108561. 28 Caballero AE, Ceriello A, Misra A, et al. COVID-19 in people living 44 Chen L, Sun W, Liu Y, et al. Association of early-­phase in-­hospital with diabetes: an international consensus. J Diabetes Complications glycemic fluctuation with mortality in adult patients with coronavirus 2020;34:107671. disease 2019. Diabetes Care 2021;44:dc200780. 29 Wood S. mgcv: mixed GAM computation vehicle with automatic 45 Kapoor R, Timsina LR, Gupta N. Maintaining blood glucose levels smoothness estimation. R-­package version 1.8–31, 2019. Available: in range (70–150 mg/dL) is difficult in COVID-19 compared to https://​CRAN.​R-​project.​org/​package=​mgcv Non-­COVID-19 ICU Patients—A retrospective analysis. J Clin Med 30 Li G, Deng Q, Feng J, et al. Clinical characteristics of diabetic 2020;9:3635. patients with COVID-19. J Diabetes Res 2020;2020:1652403. 46 Shen Y, Fan X, Zhang L, et al. Thresholds of glycemia and the 31 Izcovich A, Ragusa MA, Tortosa F, et al. Prognostic factors for outcomes of COVID-19 complicated with diabetes: a retrospective severity and mortality in patients infected with COVID-19: a exploratory study using continuous glucose monitoring. Diabetes systematic review. PLoS One 2020;15:e0241955. Care 2021;44:dc201448. 10 Ortega E, et al. BMJ Open 2021;11:e051237. doi:10.1136/bmjopen-2021-051237