Proportion of newly diagnosed diabetes in COVID-19 patients: A systematic review and meta-analysis.

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1111/dom.14269.

The world is currently grappling with a dual pandemic of diabetes and coronavirus disease 2019 (COVID‐19). Several articles published in the recent issues of Diabetes, Obesity and Metabolism and elsewhere have raised concerns about a bi‐directional relationship between these two health conditions. , , , , , , , It is now undoubtedly proven that diabetes is associated with a poor prognosis of COVID‐19. , , , , , On the other hand, COVID‐19 patients with diabetes frequently experience uncontrolled hyperglycaemia and episodes of acute hyperglycaemic crisis, requiring exceptionally high doses of insulin. , , , , , More intriguingly, recent reports show that newly diagnosed diabetes is commonly observed in COVID‐19 patients. , , , However, this has not been systematically studied before. Therefore, we performed a systematic review and meta‐analysis to examine the proportion of newly diagnosed diabetes in COVID‐19 patients.

This study was conducted and reported in accordance with the Preferred Reporting Items for Systematic reviews and Meta‐Analyses (PRISMA) and Meta‐analyses Of Observational Studies in Epidemiology (MOOSE) guidelines (see Figure S1 and Table S1 for checklists), and is registered with PROSPERO (registration no. CRD42020200432). Two authors (TS and YC) independently searched PubMed, MEDLINE, Embase and Scopus databases and preprint servers (medRxiv and Research Square) until 2 November 2020. We considered observational studies providing data on the number or proportion of COVID‐19 patients (laboratory confirmed or clinically diagnosed) with newly diagnosed diabetes. We excluded observational studies that were conducted only among patients with diabetes, case reports, case series, letters, editorials, commentaries and review articles. Newly diagnosed diabetes was defined as new‐onset diabetes (no prior history of diabetes with fasting plasma glucose [FPG] ≥ 7.0 mmol/L or random blood glucose [RBG] ≥ 11.1 mmol/L and HbA1c < 6.5%) or previously undiagnosed diabetes (FPG ≥ 7.0 mmol/L or RBG ≥ 11.1 mmol/L and HbA1c ≥ 6.5% or HbA1c ≥ 6.5% only). We used the search terms ‘new‐onset diabetes’, ‘newly diagnosed diabetes’, ‘incident diabetes’, ‘transient hyperglycaemia’ and ‘secondary hyperglycaemia’ in conjunction with ‘COVID‐19’ (see the supporting information for search strategies). No language restrictions were applied. We also checked the reference list of relevant articles to identify additional eligible studies. If the study cohorts overlapped (i.e. patients from the same hospital with similar time periods of data collection), then the study with the largest sample size was selected. Data on first author name, country, study design, hospital name, study period, age, sex, total number of patients, number of patients with newly diagnosed diabetes, definition of newly diagnosed diabetes, time of diagnosis and type of diabetes were extracted independently by the same two authors (TS and YC) using a data extraction form that was adapted from the Cochrane Collaboration. We did not contact the authors of the included studies to obtain missing data because of time constraints. The National Institutes of Health Quality Assessment Tool for observational studies was used to assess the quality of the included studies. Disagreements in study selection, data extraction and quality assessment were resolved by consensus between the two authors (TS and YC) or by discussion with a third author (RJT).

We pooled the proportion of newly diagnosed diabetes across studies with the DerSimonian and Laird random effects model. The variances of the proportions were stabilized with the Freeman–Turkey Double Arcsine Transformation method. The 95% confidence interval (CI) for the proportion in individual studies was calculated using the exact method. Heterogeneity between the studies was assessed using the Cochranʼs Q Test (P < .01 for heterogeneity) and Higgins I2 statistic (low: <25%, moderate: 25%‐50% and high: >50%). To investigate the sources of between‐study heterogeneity, we performed a subgroup analysis by the country of origin of studies, and univariate random effects meta‐regression models were fitted for mean age (median age was used if mean was not reported), sex (proportion of males) and sample size of studies. We did not assess for publication bias, as there were fewer than 10 studies in this meta‐analysis. , Analyses were performed using Stata software version 16.1 (StataCorp, College Station, TX, USA).

A total of 148 studies were retrieved during the search, of which 100 studies were duplicates, and 30 were case reports, case series, letters, commentaries or review articles. After full‐text review, a further 10 studies were excluded, including three overlapping cohorts, three studies that were conducted only among patients with diabetes, and four that did not satisfy the criteria of newly diagnosed diabetes. A total of eight studies were included in the final analysis , , , , , , , (see Figure S2 for the PRISMA flow diagram).

Table 1 presents the characteristics of the included studies. All eight studies were retrospective cohort studies, consisting of four from China, , , , two from Italy , and two from the United States. , All studies were conducted during the first 5 months of the pandemic (i.e. January‐May 2020). The mean or median age of patients in these studies varied from 47 to 65.5 years. All the studies (except for two with no data on sex) , had more males than females, with the proportion of males ranging from 52.1% to 67.1%. Data on new‐onset diabetes were available in two studies , and three studies (or cohorts) had previously undiagnosed diabetes cases. , , In six studies (or cohorts), , , , , , HbA1c was not performed for all participants, so it was not possible to differentiate between new‐onset and previously undiagnosed diabetes. In the majority of studies (n = 6), , , , , , the exact time of detection of newly diagnosed diabetes was not reported, whereas, in two studies, , the diagnosis was made within 24 hours to 3 days after hospital admission. Only one study reported on the type of diabetes (i.e. type 2 diabetes). The quality of studies was either fair or good, with most (n = 6; 75%) studies being of good quality. With a total of 3711 COVID‐19 patients with 492 cases of newly diagnosed diabetes from eight studies, the random effects meta‐analysis estimated a pooled proportion of 14.4% (95% CI: 5.9%‐25.8%) with a high degree of heterogeneity (I2: 98.6%, P < .001) (Figure 1). The pooled proportion was non‐significantly lower in China than in other countries (13.4% vs. 15.4%, P = .87; Figure S3). Meta‐regression models found no significant association between the pooled proportion and mean study age (P = .84), the proportion of males (P = .89) and sample size (P = .81) (Table S2).

TABLE 1

Study characteristics

Author and country	Study design	Study period	Study setting	Age (y), mean (SD or range) or median (IQR)	Male, %	N a	n b (%)	Definition of newly diagnosed diabetes	Time of diagnosis	Type of diabetes	Study quality c
New‐onset diabetes
Li et al., 4 China	Retrospective cohort	22 Jan‐17 Mar 2020	Wuhan Union Hospital	61 (49‐68)	52.1	453	25 (5.5)	No prior diabetes history, FPG ≥7.0 mmol/L and HbA1c <6.5%	Within 3 d after hospital admission	NR	Good
Zhou et al., 31 China	Retrospective cohort	Jan‐Mar 2020	Anhui Provincial Hospital	47 (35‐56)	60	80	22 (27.5)	No prior diabetes history, RBG ≥11.1 mmol/L and HbA1c<6.5%	Exact time of diagnosis not reported	NR	Good
Previously undiagnosed diabetes
Li et al., 4 China	Retrospective cohort	22 Jan‐17 Mar 2020	Wuhan Union Hospital	61 (49‐68)	52.1	453	38 (8.4)	No prior diabetes history, FPG ≥7.0 mmol/L and HbA1c ≥6.5%	Within 3 d after hospital admission	NR	Good
Yi et al., 30 China	Retrospective cohort	Jan‐Feb 2020	Jinyintan Hospital in Wuhan, Ruijin Hospital in Shanghai, Tongren Hospital in Shanghai, and Tongling Peopleʼs Hospital in Anhui	NR	NR	521	3 (0.6)	No prior diabetes history and HbA1c ≥6.5%	Exact time of diagnosis not reported	NR	Good
Smith et al., 28 United States	Retrospective cohort	16 Mar‐2 May 2020	Saint Barnabas Medical Center	64.4 (range: 21‐100)	53.3	184	85 (46.2)	No prior diabetes history and HbA1c ≥6.5%	Exact time of diagnosis not reported	NR	Fair
New‐onset or previously undiagnosed diabetes d
Li et al., 4 China	Retrospective cohort	22 Jan‐17 Mar 2020	Wuhan Union Hospital	61 (49‐68)	52.1	453	31 (6.8)	No prior diabetes history and FPG ≥7.0 mmol/L	Within 3 d after hospital admission	NR	Good
Wang et al., 29 China	Retrospective cohort	24 Jan‐10 Feb 2020	Wuhan Union West Hospital and Wuhan Red Cross Hospital	NR	NR	1101 e	176 (16.0)	No prior diabetes history and FPG ≥7.0 mmol/L	Within 24 h after hospital admission	NR	Fair
Fadini et al., 25 Italy	Retrospective cohort	Feb‐Apr 2020	Hospital in North‐East Italy	64.9 (15.4)	59.3	413	21 (5.1)	No prior diabetes history, HbA1c ≥6.5% or RBG ≥11.1 mmol/L with signs and symptoms of hyperglycaemia	Exact time of diagnosis not reported	NR	Good
Smith et al., 28 United States	Retrospective cohort	16 Mar‐2 May 2020	Saint Barnabas Medical Center	64.4 (range: 21‐100)	53.3	184	29 (15.8)	No prior diabetes history, persistently elevated FPG ≥7.0 mmol/L and requiring insulin therapy	Exact time of diagnosis not reported	NR	Fair
Sieglie et al., 27 United States	Retrospective cohort	11 Mar‐30 Apr 2020	Massachusetts General Hospital	63.9 (16.5)	57.6	450	13 (2.9)	No prior diabetes history and HbA1c ≥6.5% or RBG ≥11.1 mmol/L	Exact time of diagnosis not reported	Type 2 diabetes	Good
Lampasona et al., 26 Italy	Retrospective cohort	25 Feb‐19 Apr 2020	IRCCS San Raffaele Hospital	65.5 (55‐74.5)	67.1	509	49 (9.6)	No prior diabetes history and mean FPG ≥7.0 mmol/L during hospitalization	Exact time of diagnosis not reported	NR	Good

Abbreviations: FPG, fasting plasma glucose; IQR, interquartile range; NR, not reported; RBG, random blood glucose; SD, standard deviation.

Number of COVID‐19 patients.

Number of newly diagnosed diabetes cases.

Study quality was assessed using the National Institutes of Health Quality Assessment Tool.

HbA1c was not performed for all participants, so it was not possible to differentiate between new‐onset and previously undiagnosed diabetes.

157 patients who were transferred to another hospital were excluded.

FIGURE 1

Forest plot for pooled proportion of newly diagnosed diabetes in COVID‐19 patients

Li et al.,

Zhou et al.,

Li et al.,

Yi et al.,

Li et al.,

While newly diagnosed diabetes in COVID‐19 patients could be attributed to the stress response associated with severe illness or treatment with glucocorticoids, the diabetogenic effect of COVID‐19 should also be considered. This is supported by reports showing exceptionally high insulin requirements in severely or critically ill COVID‐19 patients with diabetes. These appear disproportionate when compared with critical illness caused by other conditions. , Further, it has been noted that diabetic ketoacidosis and hyperosmolar hyperglycaemic state are unusually common in COVID‐19 patients with diabetes. , , , , Severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2), the virus that causes COVID‐19, by attaching to angiotensin‐converting enzyme‐2 (ACE2) receptors in beta cells of the pancreas, could cause acute impairment in insulin secretion. Indeed, an organoid study has shown that SARS‐CoV‐2 can enter and damage the pancreatic beta cells. SARS‐CoV‐2 may also injure the beta cells by triggering a plethora of pro‐inflammatory cytokines (e.g. interleukin‐6) or by enhancing autoimmunity in genetically predisposed people. In addition to defective insulin secretion, COVID‐19 patients also present with a high degree of insulin resistance, particularly those with severe illness. It is not known whether this is because of insulin receptor defects in the key metabolic organs associated with glucose metabolism or interference with the insulin receptor signalling by the virus. ACE2 receptors are expressed in the liver, adipose tissue and skeletal muscle, and binding of SARS‐CoV‐2 to these receptors may impair responses to insulin. The insulin receptor signalling could also be impaired by the pro‐inflammatory cytokines induced by SARS‐CoV‐2, or by enhanced actions of angiotensin II resulting from the downregulation of ACE2 after the virus enters the cells. , , Similar mechanisms with other viral infections, such as hepatitis C, leading to type 2 diabetes, have been described previously.

This is the first systematic review and meta‐analysis to study the extent of newly diagnosed diabetes in COVID‐19 patients. We used robust and standard methods, and the search was comprehensive (including grey literature); the literature search, study screening, selection, data extraction and quality assessment were performed independently by two researchers, and the quality of most studies was good. Finally, we removed overlapping cohorts from our analysis, a step not commonly performed in many other systematic reviews and meta‐analyses conducted in COVID‐19 patients. However, our study has some limitations. The true proportion is unknown as all studies were hospital‐based, and the patients were mostly severely or critically ill. Further, of the eight studies, 50% were from China, and the rest were from only two other countries, which limits the generalizability of the findings. Finally, the subgroup and meta‐regression analyses lack sufficient power to detect associations between variables, as they are limited to the use of study‐level data, and the number of studies was small. , These limitations clearly emphasize the need for more studies with larger samples, including those conducted in community settings where mild cases are treated, from several regions of the world.

In conclusion, this meta‐analysis of eight studies with more than 3700 patients shows a pooled proportion of 14.4% for newly diagnosed diabetes in hospitalized COVID‐19 patients. Recent reports have shown that newly diagnosed diabetes may confer a greater risk for poor prognosis of COVID‐19 than no diabetes or pre‐existing diabetes. , Therefore, COVID‐19 patients with newly diagnosed diabetes should be managed early and appropriately and closely monitored for the emergence of full‐blown diabetes and other cardiometabolic disorders in the long term.3, 34 In this regard, the establishment of the CoviDiab Registry (covidiab.e-dendrite.com) is timely and should provide valuable insights into issues regarding COVID‐19‐related diabetes. We are now seeing a classic example of a lethal intersection between a communicable and a non‐communicable disease.

CONFLICT OF INTEREST

The authors have no conflicts of interest to declare.

AUTHOR CONTRIBUTIONS

TS conceived the idea, conducted the literature search, screened, selected and assessed the quality of the articles, and wrote the first draft of the manuscript. NK reviewed and edited the manuscript. YC conducted the literature search and study screening, selection, data extraction and quality assessment; she also reviewed and edited the manuscript. RJT resolved any disagreements between TS and YC regarding the selection and quality assessment of the studies. RJT also helped TS in addressing the reviewersʼ comments and revising the manuscript. PZ conceived the idea along with TS and reviewed and edited the manuscript.

Appendix S1. Supplementary file.

Click here for additional data file.