Association of metformin, dipeptidyl dipeptidase-4 inhibitors, and insulin with COVID-19-related hospital outcomes in patients with type 2 diabetes

Abstract

Objective

The effects of diabetes medications on COVID-19 hospitalization outcomes have not been consistent. We sought to determine the effect of metformin, DPP-4 inhibitors (DPP-4i), and insulin on admission to the intensive care unit (ICU), need for assisted ventilation, development of renal insufficiency, and mortality, in patients admitted with COVID-19 infection after controlling for clinical variables and other relevant diabetes-related medications in patients with type 2 DM.

Methods

This was a retrospective study of patients hospitalized with COVID-19 from a single hospital system. Univariate and multivariate analyses were performed that included demographic data, HbA1c, kidney function, smoking status, insurance, Charlson comorbidity index, number of diabetes medications, use of angiotensin-converting enzyme inhibitors (ACEi) and statin prior to admission, and glucocorticoids during admission.

Results

A total of 529 patients with type 2 DM were included in our final analysis. Neither metformin nor DPP4i prescription was associated with ICU admission, need for assisted ventilation, nor mortality. . Insulin prescription was associated with increased ICU admission. Insulin prescription was not associated with need for assisted ventilation nor mortality. There was no association with any of these medications with development of renal insufficiency.

Conclusions

In this population limited to type 2 diabetes, and controlling for multiple variables that have not been consistently studied (such as a measure of general health, HbA1c, insurance status), insulin prescription was associated with increased ICU admission. Metformin and DPP4i prescription did not have an association on the outcomes.

Introduction:

Diabetes mellitus (DM) is associated with negative outcomes in COVID-19 infection (1,2). A review of 72,314 COVID-positive patients in China showed a case fatality rate of 7.3% in patients with DM compared to a 2.3% overall mortality rate (3). In the United States, the Centers for Disease Control and Prevention (CDC) COVID-19 Response Team previously reported that DM was the most common comorbidity in COVID-positive patients with a prevalence of 10.9% (4).

Early speculation pointed to possible mitigating effects of various diabetes medications in patients with diabetes. Dipeptidyl dipeptidase-4 inhibitors (DPP-4i) were thought to reduce risk for COVID-19 infection by preventing viral cell entry, and it was posited that metformin and insulin were protective due to their anti-inflammatory benefits (with the caveat that patients on insulin might represent populations with more advanced disease or less glycemic control and hence could be associated in this manner with worse outcomes) (5, 6, 7, 8). Subsequently, several studies had shown a protective association with metformin, a neutral effect with DPP-4i, and a negative impact with insulin on COVID-19 health-related outcomes in hospitalized patients with diabetes, yet not all studies are in keeping with these outcomes (6, 7, 8). However these studies either do not focus on, or include, the type of diabetes (important as patients with type 2 diabetes may have different physiology than patients with type 1), previous glycemic control (HbA1c), a measure of general health (such as the Charlson Comorbidity Index which we used here), or type of insurance (relevant in the U.S. setting). We sought to determine the effect of pre-admission metformin, DPP-4i, and insulin, on admission to the intensive care unit (ICU), need for assisted ventilation, development of renal insufficiency, and 30-day post-discharge mortality, in patients with type 2 diabetes admitted with COVID-19 infection after controlling for patient characteristics and other relevant diabetes-related medications.

Research Design and Methods:

Procedures

This study was approved by our institution’s IRB. A retrospective review was conducted on patients included in our institution’s Health System COVID19 Registry from January 1 to June 30, 2020 who met the following criteria: age >18 years old, admission during that time period with a positive COVID19 viral polymerase chain reaction (PCR) test, established diagnosis of diabetes type 2 (using ICD-10 codes E11.xx), documentation of diabetes medication upon admission, and medication exposure to metformin, DPP4i, and insulin for at least 90 days before admission. The Cleveland Clinic developed a COVID-19 registry in a prospective manner by using standardized clinical templates across the healthcare system to gather data on patients seeking care for COVID-19-related concerns. A detailed description of the registry was previously published by Jehi et al and the flow diagram used in our study is included in the supplementary files (9). In brief, there were 11,672 unique individuals in the database, 9,961 of whom did not meet eligibility criteria, and 791 who did not have pre-admission medications. This left us with 529 unique patients with type 2 diabetes. We selected the following variables from the registry and the electronic health record based on their potential as risk factors for poor COVID-19 outcomes in patients with diabetes: demographic information, HbA1c, kidney function, smoking status, insurance, Charlson comorbidity index, number of diabetes medications, use of angiotensin-converting enzyme inhibitors (ACEi) and statins prior to admission, and glucocorticoids during admission. The hospitalization outcomes studied in the registry included admission to the ICU, need for assisted ventilation, development of renal insufficiency (defined as renal replacement therapy - use of peritoneal dialysis or hemodialysis), and 30-day post-discharge mortality. These variables were used in our analysis to investigate their association with COVID-19 outcomes in this patient population.

Statistical Analysis

Categorical variables were described using frequencies and percentages, and comparisons were assessed using Pearson’s Chi-square and Fisher’s exact tests, or Kruskal-Wallis tests for ordered variables. Continuous variables were described using means and standard deviations, and comparisons were assessed using t-tests. Comparisons of multiply imputed datasets on baseline characteristics used linear and generalized linear models. To compare outcomes of admission to ICU, assisted ventilation, and mortality on metformin, insulin and DDP4i groups, logistic regression models were used.

Given the number of variables included in the model, Firth’s penalty was employed to avoid overfitting. Linear regression models were used to compare these groups on mean glucose levels. Models adjusted for all other variables, except medication count, which was too collinear with medication group. In addition, insulin use was highly correlated with DDP-4i use (of the 40 patients with DPP4i use, 39 were also on insulin), and was not included in those multivariable models. Analyses were calculated using SAS® Software (version 9.4; Cary, NC). Missing data on adjustment factors were imputed using fully conditional specification. Ten imputation datasets were created, and analyses was performed for each dataset and pooled using Rubin’s rules. We included this in the Methods section (statistical analysis).The final model was a penalized multivariable model including all factors in Table 1 (except medication count). No variable selection was performed.

Table 1.

Overall Summaries of Adjustment and Outcome variables (total n=529)

Factor	N	Statistics
Age (yrs), mean ± SD	529	67.0 ± 14.5
Gender, n (%)	529
Male		279 (52.7)
Female		250 (47.3)
Race, n (%)	522
White		261 (50.0)
Black		233 (44.6)
Other		28 (5.4)
Most Recent BMI Result, mean ± SD	527	32.2 ± 8.5
HbA1c, mean ± SD	471	7.2 ± 1.9
Smoking, n (%)	508
Current		47 (9.3)
Former		208 (40.9)
Never		253 (49.8)
Insurance Group, n (%)	529
Government		412 (77.9)
Private		90 (17.0)
Other/Unknown		27 (5.1)
GFR Category, n (%)	526
<30 mL/min		90 (17.1)
30-59 mL/min		154 (29.3)
>60 mL/min		282 (53.6)
CCI Grouping, n (%)	526
0		1 (0.19)
1		111 (21.1)
3		138 (26.2)
5		276 (52.5)
Insulin, n (%) Metformin, n (%) DPP-4 inhibitor, n (%) ACE inhibitor, n (%)	529 529 529 529	377(71.3) 148 (28.0) 32 (6.0) 134 (25.3)
Statin, n (%)	529	333 (62.9)
Glucocorticoid, n (%)	508	74 (14.6)

Results:

Five hundred and twenty-nine patients were included in the final analysis. Mean age was 67.0 ± 14.5 years, with 52.7% male. Half were white and 46% were Black. Mean BMI was 32.2 ± 8.5 kg/m², and mean HbA1c was 7.2 ± 1.9 %. Of these patients, 38.1% had a mean blood glucose during the hospital stay between 101-140 mg/dL, 37.4% between 141-180 mg/dL and 15.2% above 180 mg/dL. The majority of our patients (62.9% ) were on statins, and 25.3% were on an ACEi, at least 90 days prior to admission. Of the admitted patients, 40.3% required ICU admission, 40.6% needed assisted ventilation, 16.4% died, and 6% developed renal insufficiency. Table 1 shows the characteristics of the patients. There were only 12 patients on GLP1 receptor agonists 90 days prior to admission, and was not a sufficient number to run an analysis on.

Table 2 shows univariable and multivariable outcome analysis by medication prior to multivariable imputation. Per univariable analysis, neither metformin nor DPP4i prescription was associated with ICU admission, need for assisted ventilation, nor mortality. Multivariable outcome analysis revealed the same lack of association with these outcomes.

Table 2.

Univariable and multivariable outcome analysis by medication, n=529

Outcomes	n (%)	p-value on univariate analysis	p-value on multivariate analysis	Odds ratio (95% CI) on multivariate analysis
Metformin Admission to ICU	54 (36.5)	0.26	0.72	0.92 (0.60,1.43)
Need for assisted ventilation	58 (39.2)	0.67	0.87	1.04 (0.67,1.59)
Mortality	18 (12.2)	0.098	0.98	1.01 (0.54,1.87)
Renal insufficiency	5 (3.4)	0.090	n/a	n/a
	n (%)	p-value on univariate analysis	p-value on multivariate analysis	Odds ratio (95% CI) on multivariate analysis
DPP4i Admission to ICU	10 (31.3)	0.28	0.44	0.73 (0.33,1.63)
Need for assisted ventilation	13 (40.6)	0.99	0.97	1.01 (0.47,2.18)
Mortality	8 (25.0)	0.18	0.46	1.42 (0.56,3.59)
Renal insufficiency	1 (3.1)	0.71	n/a	n/a
	n (%)	p-value on univariate analysis	p-value on multivariate analysis	Odds ratio (95% CI) on multivariate analysis
Insulin Admission to ICU	169 (44.9)	<0.001	0.006	1.86 (1.20,2.91)
Need for assisted ventilation	162 (43.0)	0.086	0.14	1.39 (0.90,2.14)
Mortality	63 (16.7)	0.800	0.57	1.19 (0.66,2.13)
Renal insufficiency	26 (6.9)	0.32	n/a	n/a

On univariable analysis, insulin prescription was associated with increased ICU admission, which was maintained upon multivariable analysis. Insulin prescription was not associated with need for assisted ventilation nor mortality, both on univariable and multivariable analyses.

None of the medications was related to development of renal insufficiency on univariable analysis, and there were too few events to allow for multivariable analysis. While we did not include renal insufficiency in our multivariable analyses, we ensured that all other outcomes did adjust for GFR, so that the impact of impaired renal function on the association between medication prescription and other outcomes was still accounted for.

Discussion:

Patients with diabetes are disproportionately affected by COVID-19 compared to people without diabetes. Proposed mechanisms of protective or harmful effects of diabetes medications on COVID-19-related outcomes are important to tease out in populations as they may assist in predictive models for care. However, current studies either include all types of diabetes – making it difficult to make conclusions for either type 1 or type 2 diabetes, do not include measures of general health, glucose control, or insurance type (relevant in the U.S. setting). In this study, we investigated the effects of the various anti-diabetes medications on COVID -19 patient outcomes in patients with type 2 diabetes, including the aforementioned variables.

Only insulin prescription was associated with one outcome – increased ICU admission. Neither metformin nor DPP4i was associated with any of the hospital outcomes.

Metformin acts via activation of AMP-activated protein kinase (AMPK), and has been shown to regulate angiotensin-converting enzyme 2 (ACE2) protein stability in several lung injury rodent models. Since the discovery that the ACE2 receptor is implicated in the pathogenesis of SARS CoV-2 through its effect on viral cell entry, numerous studies have investigated the possibility that metformin may positively affect outcomes in COVID-19 pneumonia. Similarly, the SARS CoV-2 spike glycoprotein predicts interactions with DPP4, the expression of which is associated with diabetes (10). Insulin is known to have inflammatory effects, but is also often used as surrogate for diabetes duration or severity. Limiting the study population to type 2 diabetes reduces the concern that the different physiology of type 1 diabetes affects the results.

A study by Li et al. included 131 COVID-19 patients with DM, showed in a multivariate analysis that 94.6% of patients who were on metformin survived compared to only 22.3% patients who were not on metformin (p=0.023) (1). However, use of possible confounding medications such as statins and glucocorticoids was not analyzed. In a study on 6,256 outpatients with type 2 diabetes, metformin prescription was associated with decreased mortality in women hospitalized with COVID-19, but not in men. The study included both type 1 and type 2 diabetes, which might affect the results, and HbA1c and type of insurance were not included (11). In a study of 283 Chinese patients with DM, in-hospital mortality rate was significantly lower in patients taking metformin (2.9% compared to 12.3% of patients who were not on metformin) (12). However, HbA1c was not included – an important parameter associated with COVID-19-related mortality (13).

DPP4i prescription in our study did not show protective nor deleterious outcomes, consistent with multiple studies (8, 14, 15).

Patient prescription of insulin retained its association with increased likelihood of ICU admission even after multivariable analysis, suggesting that these patients at baseline have poorer outcomes. Our finding is consistent with a large study by Wander et al. that showed a higher ICU admission risk in COVID-19 patients who were prescribed insulin than those not prescribed this drug (8). This effect may be a reflection of later stage disease state in many insulin-requiring patients compared to patients who are on oral agents alone. However, similar to many studies, we do not have duration of diabetes readily available. Unlike the meta-analysis by Chen et al where mortality was high in the patients prescribed insulin before admission, our study did not support this finding (16). Type of diabetes and a measure of glycemic control was not reported as parameters that they extracted from the included studies.

Our study has several limitations. There is a lower-than-expected proportion of metformin prescription in our population (28%) compared to the general US population. Since the method of medication extraction for all 3 medications was similar, this might be a reflection of the COVID-19 population admitted. In addition, the low proportion of metformin is not too dissimilar from other studies – Crouse et al 34%, Chen et al 35%, and Bramante et al 36% (2, 6, 11). It is possible that our cohort includes patients, who, once placed on insulin, have had their oral medications, including metformin, discontinued. We did not perform chart reviews, therefore could not ascertain the diagnosis of type 2 (as opposed to type 1 or other diabetes -- for example, that due to drug or other underlying condition). However, the number of patients we excluded because of ICD codes for type 1 diabetes (n= 109), and for the rest/other (n=271) fairly reflects the usual predominant proportion of type 2 diabetes. Approximately 30% of patients were on multiple medications, and while multivariable analysis took this into consideration where possible, correlation among use especially among those with concomitant DDP4 and insulin use, limits the ability to make strong conclusions about use of DDP4 alone. We could not ascertain duration of diabetes, and insulin prescription is often used as a surrogate for patients with type 2 diabetes. However, we have applied multiple sub-analyses to potentially address the characteristics of the patients on insulin.

The strengths of our study include a large population size, and multiple variables that may have an impact on outcomes – smoking, glucocorticoid use, statin use, HbA1c, Charlson comorbidity index, and insurance type. Insurance status/type is an important factor in the US setting, and understandably not reported in non-US studies where healthcare is socialized. In a study addressing this, in-hospital mortality was highest in the Medicare group (17). Prescription of the medications (diabetes medication, statin, ACEi) for at least 3 months prior to hospitalization was ascertained. Our study population was derived from the earlier part of the pandemic, and did not include antiviral medications tried as these were still not standard. However, the population used in our study was contemporaneous with the other publications cited, with the standard of care for COVID19 at that time likely being similar.

After controlling for multiple variables, metformin and DPP4i upon hospital admission did not modify the risk of adverse COVID-19 hospital outcomes while insulin was associated with increased risk of ICU admission.