INTRODUCTION
Of the non-communicable diseases, diabetes mellitus (DM) is one of the four leading causes of mortality worldwide.(1) One in every 11 adults has DM with an estimated prevalence of DM in South Africa (SA) of 12.8%, of which approximately 60% of people are undiagnosed.(2) In patients diagnosed with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus, DM has been identified as an important risk factor for severe disease and the merging of these two epidemics has brought to the fore the relationship between DM and the severity of COVID-19.
DM was associated with severe disease and an increased mortality in both severe acute respiratory syndrome (SARS) and the Middle East respiratory syndrome (MERS), where admission hyperglycaemia (HG) was shown to be an independent predictor of mortality.(3–5)
A potential mechanistic role is direct COVID-19-mediated pancreatic β-cell dysfunction resulting in worsening glycaemic control in pre-existing DM or causing new onset DM.(6,7) Another possible explanation for the poor prognosis of COVID-19 in patients with DM is the chronic inflammatory state associated with DM together with multiple concurring risk factors, including obesity, hypertension and other cardiovascular (CV) comorbidities.(8)
The aim of this study was to assess whether admission or in-hospital HG with or without DM was associated with poor outcomes as compared to euglycaemia (EG) in patients admitted with COVID-19 infection. A secondary aim was to characterise those with new onset DM and identify clinical and laboratory predictors of severity and mortality.
METHODS
Study design
We performed a retrospective single-centre observational study evaluating patients hospitalised at the Charlotte Maxeke Johannesburg Academic Hospital (CMJAH) with reverse-transcriptase polymerase chain reaction (RT-PCR)-confirmed COVID-19 infection between 6 March and 31 August 2020. The study was approved by the University of the Witwatersrand Human Research Ethics Committee (Medical).
All relevant patient information was captured onto an electronic database. Additional registry data was collected by manual chart review for an assessment of mean blood glucose (MBG) and mean insulin dose for the duration of hospital admission. Data was supplemented with laboratory information from the National Health Laboratory Service lab system (TrakCare, InterSystems, Cambridge, MA, USA).
Sample population
Patients were eligible for inclusion in the study if they presented with symptoms of COVID-19, with a confirmed RT-PCR. Pregnant patients and those ≤18 years were excluded from this analysis.
Definitions
DM was defined as a glycated haemoglobin (HbA1c) of ≥6.5% during the index presentation or as having pre-existing DM. Uncontrolled HG was defined as two or more blood glucose (BG) recordings >10 mmol/L within 24 h with an HbA1c ² 6.5% or if an HbA1c was not done. A patient was classified as a euglycaemic non-diabetic (EG) if BG readings were <10 mmol/L over 24 h with an HbA1c ² 6.5%. Diabetic ketoacidosis (DKA) was diagnosed by the presence of a BG ≥ 14 mmol/L with a high anion gap metabolic acidosis, indicated by a pH < 7.3 or a bicarbonate (HCO3) <18 mmol/L and ketonaemia or ketonuria. DKA was further graded as mild (pH 7.25–7.30; HCO3 15–18 mmol/L), moderate (pH 7–7.24; HCO3 10–14 mmol/L) and severe (pH < 7; HCO3 < 10 mmol/L).
Body mass index (BMI) was classified as underweight (BMI < 18.5 kg/m2), normal weight (BMI 18.5–24.9 kg/m2), overweight (BMI 25–29.9 kg/m2) and obese (≥30 kg/m2). In those patients in whom a BMI could not be calculated either due to the severity of illness or logistical limitations, we reported an endomorphic phenotype, verified clinically or radiologically.
Subset analysis
Based on glycaemic status during the hospital admission, we divided the cohort into the following subgroups (Figure 1):
Outcome measures
The primary endpoint of the study was in-hospital mortality, the need for invasive mechanical ventilation and admission to an intensive care unit (ICU). Secondary endpoints included the need for non-mechanical ventilatory support (non-rebreather mask, high-flow nasal cannula, non-invasive ventilation) and length of stay (LOS).
Statistical analysis
Non-parametric statistical tests were used because of non-normal distributions. Continuous variables are presented as medians with an interquartile range (IQR) and categorical variables as numbers and percentages. The Mann–Whitney test was used to compare continuous variables between two groups and the Kruskal–Wallis test used for comparison between three groups. Pearson's chi-square test was used to analyse categorical data between the groups and if the frequency was ≤5, a Fisher's exact (two-tailed) test was used. A P-value <0.05 was considered statistically significant. Box-and-whisker plots displayed comparative data between groups. Univariate and multivariate logistic regression models were used to explore the relationship between demographic factors, glycaemic groups, comorbidities and laboratory results with primary outcomes. Comorbidities were selected for the logistic models based on the results of the comparative analysis. Variables were included if associated with the primary outcome in the univariate analysis and were selected in the final model after a stepwise backward regression using a threshold P-value <0.2. HbA1c was not associated with the primary outcome on univariate analysis and was excluded from the multivariate model. BMI was excluded from the multivariate model due to insufficient observations. C-reactive protein (CRP) and lactate dehydrogenase (LDH) were further analysed categorically as a predictor of primary outcome. Results are reported as an odds ratio (OR) with 95% confidence intervals (CI).
RESULTS
Baseline characteristics of the total cohort by diabetic status: DM compared to non-DM
Of the total cohort of 690 patients with COVID-19, 350 (51%) were female and 340 (49%) were male. The median age was 50 years (IQR 38–60) and the majority of patients were of black ethnicity (n = 561, 81%). DM was confirmed in 258 (37%) of patients and 432 (63%) were classified as non-DM. However, within the non-DM cohort, there were 245 patients with incomplete glycaemic records, and these were thus excluded from further outcome analysis (Figure 1). The baseline characteristics of the study population are summarised according to diabetic status in Table 1. Symptoms are shown in Supplementary Table 1.
Baseline characteristics of 690 hospitalised COVID-19 patients with and without DM
| Diabetic | Non-diabetic | P | |
|---|---|---|---|
| N | 258 | 432 | |
| Age (years), median (IQR) | 56 (47–63) | 44(34–57) | <0.0001 |
| Age categories (years), n (%) | |||
| <50 | 76 (29.5) | 271 (62.7) | |
| 50-59 | 87 (33.7) | 69 (16) | |
| 60-69 | 67 (26) | 51 (11.8) | |
| >70 | 28 (10.9) | 41 (9.5) | |
| Gender, n (%) | 0.06 | ||
| Male | 139 (53.9) | 201(46.5) | |
| Female | 119 (46.1) | 231 (53.5) | |
| Ethnicity, n (%) | |||
| Black | 208 (80.6) | 353 (81.7) | 0.14 |
| White | 14 (5.4) | 40 (9.3) | |
| Coloured | 6 (2.3) | 10 (2.3) | |
| Indian | 29 (11.2) | 29 (6.7) | |
| Asian | 1 (0.4) | 0 | |
| BMI (kg/m2), median (lQR) | 32.9 (29.4–35.9) | 30.5 (24.2–34) | 0.002 |
| BMI category, n (%) | 0.83 | ||
| Underweight | 1 (1) | 2 (3.9) | |
| Normal weight | 8 (8.2) | 13 (25.5) | |
| Overweight | 17 (17.3) | 10 (19.6) | |
| Obese | 72 (73) | 26 (51) | |
| Endomorphic phenotype | 67 (26) | 46 (10.7) | <0.0001 |
| HbA1c (%), median (IQR) | 8.3 (6.9–11.6) | 6.0 (5.7–6.4) | <0.0001 |
| Comorbidities, n (% ) | |||
| Hypertension | 144 (55.8) | 114 (26.4) | <0.0001 |
| Human immunodeficiency virus (HIV) | 30 (11.6) | 80 (18.5) | 0.017 |
| Ischaemic heart disease | 10 (3.9) | 6 (1.4) | 0.06 |
| Chronic kidney disease | 26 (10.1) | 24 (5.6) | 0.08 |
| Renal replacement therapy | 21 (0.08) | 11 (0.02) | 0.001 |
| Cancer | 7 (2.1) | 24 (5.6) | 0.08 |
| Chronic medication, n (%) | |||
| Thiazide diuretics | 31 (12) | 18 (4.2) | 0.0001 |
| Beta blocker | 12 (4.7) | 20 (4.6) | 0.99 |
| Calcium channel blocker | 41 (15.9) | 39 (9.0) | 0.006 |
| Alpha blocker | 2 (0.8) | 5 (1.2) | 0.63 |
| Loop diuretic | 17 (6.6) | 15 (3.5) | 0.06 |
| Mineralocorticoid-R antagonist | 6 (2.3) | 12 (2.8) | 0.72 |
| ACE-inhibitor | 29 (11.2) | 27 (6.3) | 0.20 |
| Admission vitals, median (IQR) | |||
| Respiratory rate (bpm) | 22 (18.5–27) | 20 (18–24) | 0.0002 |
| Heart rate (bpm) | 102 (92–114) | 98 (86–115) | 0.0375 |
| Systolic blood pressure (mmHg) | 129 (116–144) | 125 (114–137) | 0.0366 |
| Diastolic blood pressure (mmHg) | 77 (68–89) | 77 (68–87) | 0.72 |
| Temperature (°C) | 36.5 (36.3–37) | 36.7 (36.4–37.2) | 0.72 |
| SpO2 (%) | 87 (77–93) | 91 (85–96) | <0.0001 |
| Glasgow Coma Scale | 15 | 15 | 0.20 |
All bold values are values of statistical significance P <0.05.
Clinical features of 690 hospitalised COVID-19 patients with and without DM
| Diabetic | Non-diabetic | P | |
|---|---|---|---|
| N | 258 | 432 | |
| Clinical features, n (%) | |||
| Headache | 25 (9.7) | 49 (11.3) | 0.50 |
| Fever | 86 (33.3) | 139 (32.2) | 0.73 |
| Sore throat | 33 (12.8) | 67 (15.5) | 0.33 |
| Cough | 180 (69.8) | 230 (53.2) | <0.0001 |
| Asymptomatic | 1 (0.4) | 59 (13.7) | <0.0001 |
| Diarrhoea | 21 (8.1) | 26 (6.0) | 0.28 |
| Dyspnoea | 174 (67.4) | 211 (48.8) | <0.0001 |
| Fatigue | 63 (24.4) | 66 (15.3) | 0.0029 |
| Malaise | 54 (20.9) | 62 (14.4) | 0.0202 |
| Myalgia | 46 (17.8) | 71 (16.4) | 0.64 |
| Rhinitis | 6 (2.3) | 7 (1.6) | 0.22 |
| Vomiting | 30 (11.6) | 34 (7.9) | 0.10 |
All bold values are values of statistical significance P <0.05
Baseline characteristics and outcomes by classification of known DM compared to new onset DM
Of the 258 diabetic patients, 152 (59%) had pre-existing DM and 106 (41%) had new onset DM. Baseline characteristics of the two groups, including laboratory investigations, are outlined in Supplementary Table 2. Majority of the diabetic cohort had Type 2 DM (T2DM) (n = 247, 96%). Only 11 (4%) patients had Type 1 DM.
Baseline characteristics of patients with known diabetes compared to new onset diabetes
| Known-diabetes | New-onset diabetes | P | |
|---|---|---|---|
| N | 152 | 106 | |
| Age (years), median (IQR) | 56 (48–63.8) | 55 (44.8–62) | 0.12 |
| Diabetes classification | |||
| Type 2 DM | 145 (95.4) | 102 (96.2) | 0.75 |
| Type 1 DM | 7 (4.6) | 4 (3.8) | 0.76 |
| Gender, n (%) | 0.78 | ||
| Male | 83 (54.6) | 56 (52.8) | |
| Female | 69 (45.4) | 50 (47.2) | |
| BMI (kg/m2), median (IQR) | 32.9 (29.3–34.8) | 34.4 (29.7–34.4) | 0.23 |
| Endomorphic phenotype | 42 (27.6) | 24 (22.6) | 0.37 |
| Admission laboratory investigations, median (IQR) | |||
| LDH (U/L) (Reference range [RR] 100–190) | 406 (274–584) | 467 (327–467) | 0.17 |
| C-reactive protein (CRP) (mg/L) (RR < 10) | 71 (25–174) | 78 (25–178) | 0.74 |
| D-dimer (mg/L) (RR 0–0.25) | 0.7 (0.38–1.7) | 0.83 (0.38–1.7) | 0.85 |
| Interleukin-6 (IL6) (pg/mL) (RR <7) | 98.5 (17.2–447.8) | 78.8 (15.8–143.3) | 0.28 |
| Glycaemic parameters, median (IQR) | |||
| Fasting plasma glucose (mmol/L) | 15.7 (10.7–20) | 10.2 (7.6–14.7) | <0.0001 |
| Mean blood glucose (mmol/L) | 12.5 (10.1–14.6) | 9.8 (7.6–13.6) | <0.0001 |
| HbA1c (%), median | 9.1 (6.9–11.7) | 7.2 (6.7–9.1) | 0.012 |
| Pre-admission HbA1c (%) median | 8.4 (6.9–10.8) | – | |
| Metabolic complications, n (%) | 67 (44.1) | 44 (41.5) | 0.68 |
| Diabetic ketoacidosis (DKA), n (%) | 18 (11.8) | 11 (10.4) | 0.71 |
| Mild DKA | 7 (4.6) | 4(3.8) | 0.74 |
| Moderate DKA | 6 (3.9) | 2 (1.9) | 0.35 |
| Severe DKA | 5 (3.3) | 5 (4.7) | 0.56 |
| Duration to DKA resolution (h) | 36 (24–48) | 36 (36–72) | 0.38 |
| Hyperosmolar hyperglycemic state, n (%) | 2 (1.3) | 2 (1.9) | 0.72 |
| Hypoglycemia on admission, n (%) | 3 (1.9) | 0 | 0.15 |
| Acute kidney injury, n (%) | 52 (34.2) | 36 (34) | 0.97 |
| Acute haemodialysis, n (%) | 6 (3.9) | 2 (1.9) | 0.37 |
| Pre-admission anti-glycaemic treatment, n (%) | |||
| Metformin | 97 (64) | – | |
| Metformin dose, median | 1700 | – | |
| Sulfonylurea (glimepiride) | 9 (5.9) | – | |
| Glimepiride dose, median (IQR) | 2 (1-4) | – | |
| Sodium–glucose co-transporter-2 inhibitor | 2 (1.3) | – | |
| Insulin | 51 (34) | – | |
| Total insulin dose (IU), median, IQR | 44 (30–68) | – | |
| In-hospital treatment, n (%) | |||
| In hospital metformin | 58 (38.2) | 19 (17.9) | 0.0005 |
| Total daily insulin, median (IQR) | 30 (16.5-31) | 23 (10.5-49.5) | 0.49 |
| Discharge treatment, n (%) | |||
| Metformin | 62 (40.8) | 20 (18.9) | 0.0004 |
| Metformin dose, median | 1700 | 1700 | 0.34 |
| Total insulin dose (IU), median, IQR | 30 (21–44) | 29(16–42) | 0.32 |
| Total daily insulin (IU/kg), median, IQR | 0.34 (0.27–0.56) | 0.29 (0.19–0.52) | 0.23 |
| Primary outcome, n (%) | |||
| Mortality | 44 (28.9) | 25 (23.6) | 0.45 |
| ICU admission | 35 (23.0) | 20 (18.9 | 0.58 |
| Invasive mechanical ventilation | 22 (14.5) | 11 (10.4) | 0.52 |
| Secondary outcome | |||
| Non-mechanical ventilation, n (%) | 84 (55.3) | 63 (59.4) | 0.98 |
| Length of stay (days), median (IQR) | 7 (4-11) | 8 (5–13) | 0.07 |
All bold values are values of statistical significance P <0.05.
There were no significant differences in inflammatory biomarkers between the two groups. However, significantly higher HbA1c (9.1% vs. 7.2%, P = 0.012), admission fasting plasma glucose (FPG) (15.7 vs. 10.2 mmol/L, P < 0.0001) and MBG (12.5 vs. 9.8 mmol/L, P < 0.0001) were observed in those with known DM as compared to new onset DM.
The frequency of diabetic associated metabolic complications was similar between the two groups; DKA occurred in 18 (11.8%) of those with known DM and 11 (10.4%) with new onset DM. The median duration to resolution of DKA was 36 h. There was no difference in the mortality rate in those with DKA compared to those with DM without DKA (32% vs. 26%, P = 0.45). There were only four documented cases of hyperglycaemic hyperosmolar syndrome. Three patients with known DM presented with hypoglycaemia. The median pre-admission HbA1c in those with hypoglycaemia was 6.3% with a median admission HbA1c of 7.2%. One patient was on metformin and a sulfonylurea and the other two on biphasic human insulin. All three had chronic kidney disease with a median serum creatinine of 559 µmol/L (IQR 278–796). Two of the three patients demised and one required ICU admission, ventilation, acute dialysis and inotropic support.
During hospitalisation, metformin was stopped in 112 (74%) of those with known DM. Pre-admission, inpatient as well as discharge diabetic treatment is outlined in Supplementary Table 2. There was no difference in MBG based on the different types of corticosteroids prescribed.
In those with known DM that had a pre-admission HbA1c recorded (n = 48), there was a significant difference in pre-admission HbA1c and admission HbA1c (8.4%, [IQR 6.9–10.8] vs. 9.7%, [IQR 8–12.3], P = 0.013). The loss of glycaemic control during the lockdown from COVID-19 probably resulted from either a sedentary lifestyle and lack of exercise or lack of access to health care. However, there was no significant difference in pre-admission HbA1c, and primary outcome based on the following pre-HbA1c categories <7, 7–9, >10%.
Subset analysis
Primary and secondary outcome analysis was performed on three groups: (1) DM, (2) uncontrolled HG, (3) non-DM with EG (Table 2).
Primary and secondary outcomes of patients with DM, HG and EG
| Diabetes | Uncontrolled hyperglycaemia | Non-diabetic euglycaemia | P | |
|---|---|---|---|---|
| N (%) | 258 | 39 | 148 | |
| Primary outcome, n (%) | ||||
| Mortality | 67 (26) | 21 (53.8) | 18 (12.2) | <0.0001* |
| ICU admission | 55 (21.3) | 17 (43.6) | 27 (18.2) | 0.0003* |
| Invasive mechanical ventilation | 35 (13.6) | 13 (33.3) | 20 (13.5) | 0.0135† |
| Secondary outcome | ||||
| Non-mechanical ventilation | 146 (56.6)** | 21 (53.8) | 5 (3.4) | <0.0001* |
| Length of stay, median (IQR) | 10.5 (7–14) | 8 (6–14) | 6 (4–8.5) | <0.0001 |
*All outcomes are significant for DM vs. EG and HG vs. EG.
†Outcome significance for HG vs. EG.
**Significance between DM vs. HG.
All bold values are values of statistical significance P <0.05
Outcome analysis of glycaemic status in ICU
A total of 150 patients were admitted to the COVID ICU. However, only 117 of the patients were captured on the database for analysis. The mortality rate was 50% (n = 58). A reason for the higher than expected mortality could be due to the poor patient selection for critical resources prior to ICU admission. Of the ICU cohort, 55 (47%) were diabetic, 17 (15%) had HG, and 28 (24%) were non-diabetic with EG. We excluded 17 from the outcome analysis as no glycaemic observations were available. There were no significant differences between the groups for the primary outcome of death (52% vs. 65% vs. 43%, P = 0.633) and ventilation (62% vs. 71% vs. 21%, P = 0.96). There was also no difference in the secondary outcome of non-mechanical ventilation (33% vs. 24% vs. 22%, P = 0.33). No significant differences were observed in clinical or laboratory parameters except for HbA1c, FPG and MBG (Supplementary Table 3).
Glycaemic characteristics of ICU patients with DM, HG and EG
| Diabetes | Uncontrolled hyperglycaemia | Non-diabetic euglycaemia | P | |
|---|---|---|---|---|
| N (%) | 55 | 17 | 28 | |
| Admission FPG (mmol/L) (IQR) | 15.8 (10.6–23.9) | 13.3 (10.8–15.9) | 7.6 (6.1–11.1) | 0.028 |
| Median MBG (mmol/L) (IQR) | 12.2 (10–14.3) | 10.8 (9.3–11.4) | 7.7 (6.4–10.8) | <0.0001 |
| HbA1c (%) IQR | 8.3 (6.9–12.9) | 5.9 (5.7–6.2) | 6.3 (5.6–6.4) | <0.0001 |
All bold values are values of statistical significance P <0.05
Differences in laboratory tests and inflammatory markers between three groups: DM, HG, non-diabetic EG
All inflammatory markers, CRP, LDH, ferritin, interleukin-6 (IL-6), as well as neutrophil: lymphocyte ratio, d-dimer, procalcitonin and creatinine were significantly higher in those with HG (DM and HG) as compared to EG. Oxygen saturation (SpO2), S:F ratio and P:F ratio were significantly lower in those with HG as compared to EG. Specific inflammatory markers and glycaemic parameters are represented in Supplementary Figure 1.
Outcome analysis in DM and HG subgroups based on MBG <10 and >10 mmol/L
Within the group of DM and HG, a further subset outcome analysis was conducted based on an MBG above and below 10 mmol/L as per the American Diabetes Association upper limit for in-hospital glycaemic target (Table 3A). For the total stay MBG < 10 mmol/L, those with HG had a significantly higher mortality (42% vs. 16%, P = 0.03), a higher rate of invasive mechanical ventilation (67% vs. 8%, P < 0.0001) and ICU admission (67% vs 18%, P = 0.0002). For the subset analysis of MBG > 10 mmol/L (Table 3B), the mortality was also significantly higher in those with HG (59% vs. 34%, P = 0.01), but there were no differences in the need for invasive mechanical ventilation or ICU admission between the two groups. A significant difference in mortality was observed in those with DM with an MBG < 10 as compared to >10 mmol/L (15.7% vs. 34.3%, P = 0.0011). No significant difference in mortality was observed in those with HG and an MBG above or below 10 mmol/L (P = 0.31).
Outcome analysis in DM vs. HG based on MBG < 10 mmol/L
| Diabetes | Uncontrolled hyperglycaemia | P | |
|---|---|---|---|
| N | 258 | 39 | |
| Total stay MBG < 10 mmol/L | |||
| N (%) | 115 | 12 | |
| Primary outcome | |||
| Mortality | 18 (15.7) | 5 (41.7) | 0.03 |
| ICU admission | 21 (18.3) | 8 (66.7) | 0.0002 |
| Invasive mechanical ventilation | 9 (7.8) | 8 (66.7) | <0.0001 |
| Secondary outcome | |||
| Non mechanical ventilation | 63 (54.8) | 4 (33.3) | 0.16 |
| Length of stay (days), median (IQR) | 6 (4-11) | 11 (7-22.8) | 0.0048 |
| Glycaemic parameters | |||
| Admission FPG (mmol/L) | 8.5 (7.1-10.3) | 12 (9.4-16.3) | 0.028 |
| Median MBG (mmol/L) (IQR) | 8.3 (7.1-9.1) | 9.3 (8-10.3) | 0.033 |
| HbA1c (%) | 7 (6.6-16.6) | 5.9 (5.8-6.4) | <0.0001 |
All bold values are values of statistical significance P <0.05.
Outcome analysis in DM vs. HG based on MBG > 10 mmol/L
| Diabetes | Uncontrolled hyperglycaemia | P | |
|---|---|---|---|
| N | 258 | 39 | |
| Total stay MBG > 10 mmol/L | |||
| N (%) | 143 | 27 | |
| Primary outcome | |||
| Mortality | 49 (34.3) | 16 (59.3) | 0.01 |
| ICU admission | 34 (23.8) | 9 (33.3) | 0.34 |
| Invasive mechanical ventilation | 26 (18.2) | 5 (18.5) | 0.97 |
| Secondary outcome | |||
| Non mechanical ventilation | 83 (58.0) | 17 (63) | 0.63 |
| Length of stay | 8 (5-13) | 8 (6-12) | 0.89 |
| Glycaemic parameters | |||
| Admission FPG (mmol/L) | 16.6 (12.6-22.5) | 12.6 (10.2-14) | <0.0001 |
| Median MBG (mmol/L) | 13.6 (12-15.7) | 11.3 (10.8-15.5) | <0.0001 |
| HbA1c (%) | 10.65 (8-13.6) | 5.85 (5.7-6.2) | <0.0001 |
All bold values are values of statistical significance P <0.05.
Univariate and multivariate logistic regression models for ICU admission, invasive mechanical ventilation and mortality as shown in Table 4
Univariate and multivariate logistic regression models for primary outcomes: ICU admission, invasive mechanical ventilation and death
| Primary outcomes, OR (95% CI) | ||||||
|---|---|---|---|---|---|---|
| ICU admission | Invasive mechanical ventilation | Mortality | ||||
| Univariate model | Multivariable model | Univariate model | Multivariable model | Univariate model | Multivariable model | |
| N | 258 | 193 | 293 | |||
| Glycaemic group | ||||||
| Euglycaemia | Reference | Reference | Reference | Reference | Reference | Reference |
| Diabetes | 1.26 (0.75–2.10) | 0.72 (0.30–1.76) | 1.03 (0.57–1.86) | 0.37 (0.13–1.09) | 2.49 (1.41– 4.40)* | 1.63 (0.78–3.41) |
| Uncontrolled hyperglycaemia | 3.46 (1.62–7.39)* | 2.34 (0.70–7.81) | 3.25 (1.43–7.38)* | 0.52 (0.10–2.57) | 8.17 (3.67–18.18)* | 5.41 (1.74 –16.84)* |
| BMI category (kg/m 2 ) | ||||||
| <25 | Reference | – | Reference | – | Reference | – |
| Overweight (25 - 30) | 0.39 (0.10–1.49) | – | 0.35 (0.7–1.69) | – | 0.21 (0.02–2.15) | – |
| Obese (>30) | 0.45 (0.15–1.29) | – | 0.27 (0.08–0.94) | – | 0.72 (0.18–2.86) | – |
| Age (years) | ||||||
| <50 | Reference | Reference | Reference | – | Reference | Reference |
| 50–59 | 0.69 (0.40–1.18) | 0.52 (0.21–1.25) | 0.70 (0.37–1.32) | – | 1.74 (0.98–3.07) | 1.71 (0.78–3.75) |
| 60–69 | 0.68 (0.37–1.24) | 0.53 (0.20–1.38) | 0.92 (0.47–1.77) | – | 2.18 (1.18–3.99)* | 1.71 (0.70–4.18) |
| ≥70 | 0.10 (0.02–0.44)* | 0.06 (0.01–0.55)* | 0.08 (0.1–0.63)* | – | 3.29 (1.65–6.56)* | 3.13 (1.22–8.01)* |
| Race | ||||||
| Black | Reference | Reference | Reference | – | Reference | – |
| White | 0.94 (0.39–2.25) | 1.34 (0.26–6.97) | 1.54 (0.63–3.74) | – | 1.96 (0.92–4.16) | – |
| Coloured | 0.32 (0.04–2.50) | 0.59 (0.06–5.61) | 0.57 (0.07–4.57) | – | 0.69 (0.15–3.20) | – |
| Indian/Asian | 1.81 (0.92–3.55) | 2.52 (0.95–6.74) | 1.73 (0.80–3.73) | – | 1.39 (0.70–2.78) | – |
| Gender | ||||||
| Female | Reference | Reference | Reference | – | Reference | – |
| Male | 1.57 (0.99–2.48) | 1.23 (0.62–2.44) | 1.46 (0.86–2.47) | – | 2.09 (1.33–3.29)* | – |
| Co-morbidities** | ||||||
| Ischaemic heart disease | 0.93 (0.25–3.41) | – | 0.88 (0.19–4.04) | – | 1.26 (0.39–4.11) | – |
| Hypertension | 1.28 (0.81–2.00) | 1.30 (0.64–2.66) | 1.10 (0.66–1.85) | – | 1.09 (0.70–1.68) | – |
| COPD/Asthma | 0.91 (0.29–2.80) | – | 1.00 (0.28–3.52) | – | 2.06 (0.78–5.47) | – |
| Cancer | 0.19 (0.02–1.47) | – | 0.30 (0.04–2.31) | – | 0.96 (0.31–3.02) | – |
| Chronic kidney disease | 0.85 (0.36–2.00) | – | 1.37 (0.57–3.27) | – | 1.97 (0.96–4.06) | – |
| HIV | 0.62 (0.30–1.27) | 0.43 (0.15–1.23) | 0.42 (0.16–1.10) | 0.30 (0.07–1.26) | 0.50 (0.24–1.05) | 0.57 (0.22–1.46) |
| Glycaemic parameters | ||||||
| Euglycaemic | Reference | Reference | Reference | Reference | Reference | Reference |
| Mean blood glucose (mmol/L) | 1.05 (0.99–1.11) | – | 1.06 (1.002–1.12)* | – | 1.21 (1.14–1.28)* | – |
| Admission FPG (mmol/L) | 1.05 (1.02–1.08)* | 1.06 (1.01–1.11)* | 1.04 (1.004 –1.07)* | 1.06 (0.98–1.14) | 1.06 (1.03–1.09)* | – |
| HbA1c (%) | 1.00 (0.92–1.08) | – | 0.96 (0.87–1.05) | 0.81 (0.64–1.04) | 1.03 (0.96–1.11) | – |
| Laboratory investigations | ||||||
| LDH (U/L) | 1.002 (1.001–1.003)* | 1.002 (1.001–1.003)* | 1.002 (1.001–1.003)* | 1.002 (1.001–1.004)* | 1.002 ( 1.001–1.003)* | 1.001 (1.00–1.008)* |
| LDH ≤ 350 (U/L) | Reference | Reference | Reference | Reference | Reference | |
| LDH > 350 (U/L) | 2.98 (1.29–6.88)* | 3.68 (1.31–10.36)* | 6.99 (1.64–29.73)* | 5.77 (0.73–45.82) | 4.65 (1.79 – 12.09)* | 3.49 (1.28–9.5)* |
| IL6 (pg/mL) | 1.00 (0.999–1.00) | – | 1.00 (0.99–1.00) | – | 1.00 (0.99–1.00) | – |
| CRP (mg/L) | 1.004 (1.001–1.006)* | – | 1.004 (1.002–1.006)* | 1.005 (1.001–1.01)* | 1.006 (1.004–1.008)* | 1.005 (1.002–1.008)* |
| CRP ≤ 100 (mg/L) | Reference | – | Reference | Reference | Reference | Reference |
| CRP > 100 (mg/L) | 2.20 (1.38–3.52)* | – | 2.11 (1.22–3.63)* | 3.21 (1.28–8.06)* | 2.85 (1.78 – 4.55)* | 2.27 (1.20–4.29)* |
| CRP ≤ 200 (mg/ L) | Reference | – | Reference | Reference | Reference | Reference |
| CRP > 200 (mg/ L) | 2.04 (1.25–3.33)* | – | 2.11 (1.21–3.68)* | 2.74 (1.17–6.41)* | 2.68 (1.66 – 4.31)* | 2.38 (1.26–4.51)* |
| Ferritin (μg/L) | 1.00 (0.99–1.00) | – | 1.00 (0.99–1.00) | – | 1.00 (0.99–1.00) | – |
| D-dimers (mg/L) | 1.02 (1.002–1.04)* | – | 1.02 (1.002–1.04)* | – | 1.03 (1.01–1.05)* | – |
*P <0.05. **For all co-morbidities, not having the specific co-morbidity was the reference group.
ICU admission
Factors associated with ICU admission in univariate analysis were HG and higher levels of admission FPG, LDH, CRP and d-dimers. After adjusting for glycaemic group, age, race, gender, comorbidities, admission FPG and LDH, factors that were independently associated with ICU admission were admission FPG and LDH. Age of ≥70 years was associated with a reduced likelihood of ICU admission in both univariable and multivariable analyses.
DISCUSSION
The prevalence of DM within our cohort of COVID-19 positive patients was 37%. The prevalence of DM in COVID-19 varies based on age, geographical location and severity of disease. A meta-analysis of 12 Chinese studies observed a DM prevalence of 10.3% in COVID-19, similar to the overall background prevalence of the population (10.9%), whereas in Italy, a prevalence of 8.9% was noted and this was lower than that in the background population (11%).(9,10) However, the prevalence of DM is much higher in hospitalised patients with severe COVID-19 who required ICU care, ranging between 32% in the UK to 35% in the USA and 36% in Italy.(10–12) The prevalence in our cohort is much higher than the prevalence in the background population of SA. One possible explanation is that approximately 60% of our population remains undiagnosed with DM and thus being hospitalised with COVID-19 may unmask pre-existing DM, or it may mirror the international data whereby a higher prevalence of DM is seen in patients with more severe COVID-19. Our data suggests that DM does not increase the risk of infection with SARS-CoV-2, but it rather increases the severity of disease in those who contract COVID-19.
The postulated mechanism of HG and new onset DM associated with COVID-19 is that SARS-CoV-2 can lead to direct islet injury and insulin deficiency as the virus utilises the ACE2 receptors that are expressed on pancreatic β cells.(13) Insulin resistance secondary to inflammation may also play a role.(6,14,15)
Our study is in keeping with the international epidemiological data, whereby age of ≥70 years was associated with a 3.3-fold greater odds of death and male sex with a 2-fold greater odds on univariate analysis.(6) However, paradoxically our study also showed that an age of ≥70 years was associated with a significantly lower chance of ICU admission and ventilation, based on the pre-defined medical triage protocol in a resource limited setting.
Of the 258 diabetics, 41% were newly diagnosed. There were no significant differences in primary outcome between these groups; however, those with new onset DM had significantly lower admission FPG, MBG and HbA1c compared to those with known DM. This supports the possibility of the recent onset of HG. These patients could represent those that are misclassified in international studies as HG from COVID-19-mediated β-cell dysfunction.
Our study showed significantly higher inflammatory markers in those with DM and HG as compared to those with non-diabetic EG. However, only an elevated LDH was a significant predictor of mortality, ICU admission and ventilation. An LDH > 350 U/L was associated with 3.5 times greater odds of mortality. A high CRP was also found to be an independent predictor of death and ventilation, with a CRP > 100 mg/L having 3.2 fold greater odds for ventilation and 2.3 fold greater odds for mortality. Several studies have shown that high concentrations of inflammatory markers are associated with more severe disease and mortality in COVID-19, with a CRP > 200 mg/L being a marker for critical disease.(14–16) DM and HG are pro-inflammatory states associated with a dysregulated innate and adaptive immune response, thus predisposing patients to developing a cytokine storm and multi-organ failure.(11,17–21) Furthermore, a pro-coagulant state occurs, which increases the risk of thrombosis.(18)
Hyperglycaemia also known as “stress HG” is defined as a BG > 7.8 mmol/L.(22) “Stress HG” is a well-established predictor of in-hospital mortality in patients with critical illness, acute CV events, influenza (H1N1), SARS and MERS.(3,23–25) HG is encountered in approximately 40% of hospitalised patients.(26) The study by Bode et al. showed a 41.7% mortality rate in those with HG and 14.8% in those with DM.(27) In our study, patients classified with HG in the absence of DM had a lower mean admission FPG and MBG than those with DM, yet they had a significantly higher mortality rate than those with DM. Furthermore, those with HG had a higher mortality, need for ventilation and ICU admission at an MBG < 10 mmol/L, which confirms data from other studies that HG has a more deleterious effect in critical illness than those with DM.(16,17) On univariate analysis, HG was associated with a significant 3.3-fold greater mortality and ventilation rate and a 3.5-fold greater chance of requiring ICU admission. However, on multivariate analysis, HG was significantly associated only with mortality. Patients with DM can adapt to high ranges of glucose and may be more tolerant to moderately high BG than those with HG.(16,17) Our findings suggest that acute HG is an independent risk factor for mortality in COVID-19 regardless of diabetic status, similar to data from China and the USA, whereby in-hospital HG contributed towards a higher mortality and longer LOS.(27–29)
In our study, univariate analysis identified admission FPG and total stay MBG as having significantly greater odds for all primary outcome measures. However, in the multivariate analysis, admission FPG and MBG did not remain significantly associated with the primary outcome. The association between admission HG and poorer outcomes has also been shown in other studies.(30–33) Interestingly HbA1c did not have greater odds of predicting outcome. Our study confirms the findings reported by the CORONADO study whereby there was no association between HbA1c and the primary outcome of death and ventilation.(34)
A high prevalence of DKA has been observed with SARS-CoV-2. Within our study DKA was observed in 11.2% of the patients with DM with a mortality rate of 32%. This is almost identical to a Chinese study of 658 COVID-19 positive patients, which documented a 11.6% prevalence of DKA, with a 26.7% mortality rate.(35)
There is insufficient evidence to support a clear recommendation for glycaemic targets in critically ill patients. The current recommendation is to initiate insulin therapy for those with persistent HG > 10 mmol/L and titrate to achieve a target of 7.8–10 mmol/L.(22) In our study a 2-fold greater mortality rate was observed in those with DM with an MBG > 10 mmol/L compared to those less than 10 mmol/L. In a small study of eight critically ill patients with COVID-19, glucometric analysis indicated that high doses of insulin were required with a mean of 201 IU/day (2.2 IU/kg/day) during the peak of the inflammatory response, and this decreases as the illness improves whereby four of the six patients were discharged on 18 IU/day.(36) In our study, those patients with DM in ICU were treated with an insulin infusion (mean 51 IU/day), achieving an MBG of 12.2 mmol/L (IQR 10–14.3), with a discharge insulin of 0.34 IU/kg. Further studies are required in order to assess if the significant insulin requirements that are described are due to insulin resistance secondary to inflammation or a direct effect of COVID-19 on pancreatic β cells. Furthermore, corticosteroids were standard of care in patients with hypoxaemia, and the extent that this contributed to worsening of glycaemic control requires further evaluation. Another important consideration in our study is that the primary medical team was responsible for glycaemic control and that lower insulin doses on discharge may reflect physician inertia.
In our study, metformin was stopped in 74% of those with known DM and was prescribed significantly less in those with new onset DM compared to known DM, reflecting the misconception that physicians have of not prescribing metformin for inpatient use. Metformin has been proposed as a host-directed therapy for COVID-19 because of its immuno-modulatory effects and was shown to decrease mortality in those with DM that continued metformin as compared to those where it was discontinued.(37) Thus, therapy should be individualised taking into consideration the patient's renal function and haemodynamic status. Nevertheless, HG should be recognised and treated as it has a significant impact on morbidity and mortality.
LIMITATIONS
The major limitation of this study was that not all records of COVID-19 positive patients, including the ICU admissions, were accessible and hence were not captured in the database. The study was also limited by the small sample size of the subgroups especially the group with HG. A positive correlation has been seen between BMI and severe COVID-19 disease in international studies, but the number of subjects in whom BMI was available was too small and thus BMI could not be included in the multivariate logistic regression analysis. A longitudinal study of glycaemic control, including HbA1c measurements and insulin requirements, may provide greater perspective regarding COVID-19-mediated pancreatic beta-cell injury and potential recovery.
CONCLUSION
This study found that in patients admitted with COVID-19, there was a high prevalence of DM and HG. These patients had a worse prognosis compared to euglycaemic patients suggesting a bidirectional relationship between HG and COVID-19. In addition to uncontrolled HG, patients older than 70 years, elevated LDH and CRP were independent predictors of mortality in COVID-19. In view of the high mortality associated with COVID-19-related HG with or without DM, glucose monitoring and glycaemic control should become the focus of all standard operating procedures.