Impact of diabetes on COVID‐19 mortality and hospital outcomes from a global perspective: An umbrella systematic review and meta‐analysis

Abstract Introduction To date, COVID‐19 has claimed 4.9 million lives. Diabetes has been identified as an independent risk factor of serious outcomes in people with COVID‐19 infection. Whether that holds true across world regions uniformly has not been previously assessed. Methods This study offers the first umbrella systematic review and meta‐analysis to analyse the collective and geographically stratified mortality, ICU admission, ventilation requirement, illness severity and discharge rate amongst patients with diabetes. Five databases (EMBASE, MEDLINE, CAB Abstracts, PsychInfo and Web of Science) and 3 additional sources (SSRN's eLibrary, Research Square and MedRxiv) were searched from inception to 30 August 2021. Prospective and retrospective cohort studies, reporting the association between diabetes and one or more COVID‐19 hospitalization outcomes, were included. This meta‐analysis was registered on PROSPERO, CRD42021278579. Abbreviated MeSH terms used for search were as follows: (Diabetes) AND (2019 Novel Coronavirus Disease), adapted per database requirements. Exclusion criteria exclusion criteria were as follows: (1) none of the primary or secondary outcomes of meta‐analysis reported, (2) no confirmed COVID‐19 infection (laboratory or clinical) and (3) no unexposed population (solely patients with diabetes included). Quality of the included studies were assessed using the Newcastle‐Ottawa Scale (NOS) whilst quality of evidence by the GRADE framework. Studies that were clinically homogeneous were pooled. Summative data and heterogeneity were generated by the Cochrane platform RevMan (V. 5.4). Results Overall, 158 observational studies were included, with a total of 270,212 of participants, median age 59 [53–65 IQR] of who 56.5% were male. A total of 22 studies originated from EU, 90 from Far East, 16 from Middle East and 30 from America. Data were synthesized with mixed heterogeneity across outcomes. Pooled results highlighted those patients with diabetes were at a higher risk of COVID‐19‐related mortality, OR 1.87 [95%CI 1.61, 2.17]. ICU admissions increased across all studies for patients with diabetes, OR 1.59 [95%CI 1.15, 2.18], a result that was mainly skewed by Far East‐originating studies, OR 1.94 [95%CI 1.51, 2.49]. Ventilation requirements were also increased amongst patients with diabetes worldwide, OR 1.44 [95%CI 1.20, 1.73] as well as their presentation with severe or critical condition, OR 2.88 [95%CI 2.29, 3.63]. HbA1C levels under <70 mmol and metformin use constituted protective factors in view of COVID‐19 mortality, whilst the inverse was true for concurrent insulin use. Conclusions Whilst diabetes constitutes a poor prognosticator for various COVID‐19 infection outcomes, variability across world regions is significant and may skew overall trends.


| INTRODUC TI ON
COVID-19, a novel coronavirus identified in late 2019, has rapidly spread worldwide resulting in the first pandemic experienced in the modern world since 1918. 1 Currently, more than 220 million have been infected, with 4.9 million deaths as of 18 October 2021.
Metabolic conditions, and primarily diabetes, have emerged since the beginning of the pandemic as significant risk factors for poor COVID-19 outcomes. 2 A wealth of observational studies and consequently meta-analyses have attempted to quantify the association of diabetes as an independent risk factor of poor COVID-19 outcomes and consistently found that diabetes is associated with poorer outcomes across this patient group.
Until present and to the best of our knowledge, an umbrella systematic review and meta-analysis has not been conducted to collectively assess available meta-analyses. Furthermore, whilst patient ethnicity as well as global discrepancies of healthcare facilities and antidiabetic medication access are well-established variables, [3][4][5][6][7] no previous work has factored in, study geographical origin to assess the potential impact of these parameters on COVID-19 outcomes in patients with diabetes.
We primarily aim to quantify the overall impact of diabetes in COVID-19 across three main outcomes: mortality, ICU admission and ventilation (invasive and non-invasive). Secondary outcomes include illness severity, discharge rate, identification of putative geographical variability across outcomes and associated factors of poorer or improved prognosis, amongst patients with diabetes.  Infection OR SARS-CoV-2 Infection).mp. limit to (English language and humans). The same search strategy was adapted for the remaining databases.

| Search strategy and selection criteria
Prospective and retrospective cohort studies were extracted from eligible systematic reviews and meta-analyses to enable umbrella systematic review of available data as described in Aromataris et al., 8 examining COVID-19 mortality, ICU admission, ventilation requirement, disease severity and discharge in the context of diabetes (Table S1).
Restrictions included English language and human. After removing duplicates (EndNote V.20), citations were screened by title and abstract; then, full texts were appraised to determine their eligibility by two authors (SK and MP) ( Figure 2). Two authors (SK and MP) independently conducted the abstract and full-text screening. Disagreements were resolved by a consensus meeting. Peer-reviewed full-text papers that reported one or more of the primary outcomes were selected. Full-text exclusion criteria were as follows: (1) none of the primary or secondary outcomes of meta-analysis reported, (2)  Quality of the included studies were assessed by two independent reviewers (SK and MP) using the Newcastle-Ottawa Scale (NOS) for observational studies. 9 Studies were of high quality if a NOS score ≥6 was achieved. 10 Adequate follow-up was ≥30 days (Table S1). Overall grading the quality of evidence was assessed by the GRADE framework. 11 Heterogeneity was assessed using I 2 .

| Study outcomes
Study primary outcomes included mortality, ICU admission and ventilation requirement events. These were defined as the proportion of people with an event, of each respective outcome, in comparison to people without the event, in the same population. Secondary outcomes were disease severity [mild, moderate and severe/critical] (events) and discharge events amongst patients with diabetes vs. adjusted HRs were presented with associated 95% confidence intervals (CI) ( Figure S3). For crude HRs, antidiabetic medication brand, dose and duration of action were not possible to factor in, due to lack of data reporting in individual studies. Adjusted HR (95% CI) of mortality amongst patients with diabetes was adjusted for age, gender, cardiovascular comorbidities, biochemical findings, smoking/alcohol use, immunocompromised status and medications ( Figure S4).  (Table S1). with shock or respiratory failure, mechanical ventilation requirement, or combined with other organ failure, requiring admission to intensive care unit (ICU). Individual severity definition per study is presented in Table S1.

| Data analysis
Clinical context and design were compared and where appraised as homogeneous, studies were considered as suitable for pooling. 14 The meta-analysis was conducted by computing the pooled odds ratio (OR) as per Haensel-Mantel model or Hazard ratio (HR) as per inverse variance analysis, random effects (RE) with Review Manager (RevMan) V 5.4 software. Statistical heterogeneity was quantified using I 2 statistics and Cochrane Q tests.

| Assessment of heterogeneity and subgroups to explain differences
Only studies that are clinically homogeneous were pooled.
Heterogeneity was assessed using I 2 , and I 2 greater than 70% was explored using subgroups. 14 The following subgroups were used to explain the heterogeneity: risk of bias; age, geography, study design (prospective). Asymmetry was assessed by funnel plot, and asymmetry was assessed formally by rank correlation test (Begg's test; RevMan V. 5.4). Sensitivity analyses were conducted to assess the impact of individual potential confounding variables. Publication bias was assessed visually by funnel plot, and asymmetry was formally assessed, by rank correlation test (Begg's test). 15

| RE SULTS
Following the PRISMA guidelines on systematic review search, we identified 53 eligible meta-analyses studies for study extraction. Post-individual study extraction and duplicate study removal, we identified 185 studies eligible for full-text screening ( Figure 2).

| Risk of bias
We (SLK and MP) employed the NOS for quality assessment. 9 Ninety-nine (99) studies were graded as good, forty-two (42) as fair and eighteen (18) studies as poor according to independent grading as per NOS selection, comparability and outcome parameters (Table S1). Overall quality of evidence was assessed with the GRADE framework and was found to be high (Table S1). 11

| Mortality
A total of 136 studies were included in the analysis of mortality as an outcome ( Figure 3A; Figure S2A, Table S1). Overall, studies supported the previously reported increased mortality in patients with diabetes,  Table S1).
Mortality was explored amongst patients with type 1 vs. type 2 diabetes. Only two studies 82,117 reported crude numbers of patients with type 1 or type 2 diabetes deaths, suggesting that patients with type 2 diabetes had worse outcomes in respect to mortality, OR 0.68 [95% CI 0.24, 1.87], I 2 = 0%. albeit the lack of statistical significance, possibly due to the limited sample size (p = .45, N: 308) ( Figure S3D).

| ICUadmission
A total of 59 studies were included in the analysis of ICU admission as an outcome ( Figure 3B; Table S1). Overall, studies supported and American studies did not reach statistical significance for this outcome ( Figure 3B).

| Ventilation requirement
A total of 83 studies were included in the analysis of ventilation requirement as an outcome amongst patients with diabetes vs. without ( Figure 4A, Table S1). Overall, studies supported the previ-  Table S1). Of note, American studies indicated a decrease of ventilation requirement in patients with diabetes, albeit the lack of statistical significance.

| Disease severity
A total of 43 studies were included in the analysis of disease severity (severe or critical) as an outcome amongst patients with diabetes vs. without ( Figure 4B; Figure S2D, Table S1). Overall, studies indicated increased patient numbers with diabetes pre-

Middle East world subgrouping was not feasible for this outcome
given that only one study reported this outcome. 105

| DISCUSS ION
Whilst overall patient mortality has decreased since the beginning of the pandemic, attributable to variable clinical and non-clinical factors, metabolic conditions, amongst which diabetes, have emerged as significant risk factors for poor COVID-19 outcomes. 2 The present work is the first systematic review to assess outcomes of patients with diabetes in the context of COVID-19 infection whilst accounting for geographical location of outcome reports.
Overall, our findings indicate that patients with diabetes are at a higher risk of poor hospitalization outcomes, and this is stratified by geographical region. Whilst studies originating from the Far and Middle East reported statistically significant, higher mortality across patients with diabetes, this finding was not the case for the EU, or America world regions ( Table 1). Whether healthcare and affordable F I G U R E 4 Odds associated with an increased ventilation (invasive and non-invasive) requirement in patients with diabetes (A) and patients with diabetes presenting with severe or critical condition (B). Haensel-Mantel statistical method with odds ratio (random effects) as output only for included observational studies and subgroups as per subgroup title. Summative forest plots of included observational studies of the meta-analysis (patients with Diabetes vs. without representing those with increased ventilation requirement (A) or those presenting with severe or critical illness (B) as per patient population. Illness severity definitions per included study are as presented in Table S1. Forrest and associated funnel plots ( Figure S2C,D) were generated with Review Manager V. 5.4 Cochrane Tool for meta-analysis antidiabetic medication access inequalities or whether inherent nonmodifiable (such as genetic variants) and modifiable parameters (such as obesity) across ethnic groups are responsible for this data variability, should be considered. [3][4][5][6][7] Furthermore, whilst geographical stratification did not lead to significant differences amongst world regions regarding disease severity in patients with diabetes, the need for ventilation, here defined as either invasive or non-invasive, was variable across the world. Studies from America, mostly reflecting USA trends, did not indicate higher ventilation requirements in this patient group. Whether this finding reflects overall healthcare system preparedness for catastrophic events, including pandemic emergence is not clear. 173 The present work has also highlighted those patients with overall better control of diabetes and on oral glucose-lowering medications such as metformin, had significantly improved outcomes in terms of mortality. Intriguingly, insulin use has been identified as a risk factor in COVID-19-positive, patients with diabetes. As almost the entirety of the patients with diabetes included in the present study, were patients with type 2 diabetes and given that insulin use is the final step in the control of type 2 diabetes, this finding may signify an overall decreased patient physiological reserve or poorer all-mortality outcomes, as shown in previous studies. 174 Whilst adjusted hazard ratios for medications amongst patients with diabetes still highlighted an increased risk of death in this patient group, biochemical variables including HbA1C where not consistently reported across studies to enable its inclusion in our adjusted model. Previous work has highlighted that hyperglycaemia in COVID-19 patients is notable (reviewed in Accili, 2021). 175 Thus, the literature consensus, in agreement with our findings, supports that good glycaemic control is the best way prevent COVID-19-related admissions. 175 The lack of consistent evidence across studies did not allow for robust comparison of mortality outcomes amongst the patients with type 1 vs. F I G U R E 5 Odds associated with patient discharge at the end-of study follow-up. Haensel-Mantel statistical method with odds ratio (random effects) as output only for included observational studies and subgroups as per subgroup title. Summative forest plot of included observational studies of the meta-analysis (patients with Diabetes vs. without) representing respective discharge odds between the two populations. Forrest and associated funnel plots ( Figure S2E)

| Strengths and implications for future research
To the best of our knowledge, the present work is the first umbrella meta-analysis and systematic review, to assess patients with diabetes outcomes regarding COVID-19 infection whilst accounting for geographical location of outcome reports. We have identified and addressed sources of heterogeneity by geographical and study design subgrouping sensitivity and IVR analysis. This study is the first to highlight major worldwide discrepancies and data variability worldwide in major clinical outcomes. Through this work, we highlight the overall healthcare system preparedness, medication availability and patient ethnicity-related modifiable and non-modifiable variables as putative risk factors of worldwide mortality, ICU and ventilation requirements, amongst the patients with diabetes.

| CON CLUS ION
Whilst diabetes is undoubtably a poor prognosticator of COVID-19 infection outcomes, geographical variations across world regions are notable. Whether this finding comes as a result of the variability of healthcare provisions for control and management or patient ethnicity remains to be fully elucidated.

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used and analysed during the current study are available as online Supplementary Material.