This multicenter retrospective cohort study included 2563 COVID-19 patients with type 2 diabetes admitted to 16 hospitals in Hubei, China between December 30, 2019 and April 23, 2020. The study design was approved by the central ethics boards and was accepted or approved by each collaborating hospital. The requirement for informed consent was waived by the ethics boards of the hospitals.

Patients with COVID-19 were diagnosed based on chest computed tomography (CT) manifestations or reverse transcription-polymerase chain reaction (RT-PCR) according to the New Coronavirus Pneumonia Prevention and Control Program (5th edition) published by the National Health Commission of China and WHO interim guidance[11,12]. The status of coexisting type 2 diabetes was determined based on the admitting diagnosis and disease history. The exclusion criteria included incomplete medical records (e.g., transfer to any other hospital), age < 18 or ≥ 85 years, pregnancy, acute lethal organ injury (e.g., acute myocardial infarction, severe acute pancreatitis, or acute stroke), death from surgery and a surgical complication, decompensated or end-stage chronic organ dysfunction (e.g., decompensated liver cirrhosis, estimated glomerular filtration rate (eGFR) < 30 mL/min per 1.73 m², or chronic renal insufficiency above stage III), and prehospital confusion or mental disorders. Patients not taking oral hypoglycemic drugs during the observation period were also excluded.

The medical data of the included patients were collected, including demographic information, physical examination results, laboratory findings, radiographic data from CT, history of comorbidities, medication records, and clinical outcomes during hospitalization. We extracted information on the use of glucose-lowering drugs from the patients’ medication records. The laboratory findings included fasting blood glucose (FBG) and glucose (GLU) reflecting glycemic control, leukocyte count, neutrophil count, neutrophil percentage, red blood cell count, C-reactive protein (CRP), procalcitonin, D-dimer, and serum biochemical markers representing liver injury, kidney injury, and cardiac dysfunction. All the information was collected by anonymizing the personal identifying information (e.g., name and ID) by giving each participant a new study ID through a coding system. A team of professional physicians carefully interpreted and double-checked all data to guarantee the accuracy of the data.

The onset of COVID-19 was defined as the day when the first symptom was noticed. Patients with type 2 diabetes who received DPP4i during hospitalization were classified as a DPP4i group (142 participants). Patients with type 2 diabetes who never received DPP4i but orally took other anti-diabetic drugs during hospitalization were classified as a non-DPP4i group (1115 participants).

The primary endpoint was 28-d all-cause death. The secondary endpoints were defined as acute respiratory distress syndrome (ARDS), septic shock, acute cardiac injury, acute kidney injury, and acute liver injury. ARDS and septic shock were defined according to the WHO interim guideline “clinical management of severe acute respiratory infection when novel coronavirus (2019-nCoV) infection is suspected”[13]. Acute cardiac injury was defined when the serum level of cardiac troponin I (cTNI), cardiac troponin T (cTNT), or high sensitivity cardiac troponin I (hs-cTNI) was above the upper limit of normal (ULN). Acute kidney injury was defined as an elevation in serum creatinine level equal to or above 26.5 μmol/L within 48 h[14]. Acute liver injury was defined when acutely increased levels of serum alanine aminotransferase (ALT) or alkaline phosphatase (ALP) above three times the upper limit of normal (ULN) were observed[15]. The increase or decrease of biochemical indicators was defined as exceeding or below their normal range, respectively, according to the laboratory standard of each hospital.

The adverse effects during hospitalization included hyperglycemia, hypoglycemia, metabolic acidosis, and lactic acidosis. Hyperglycemia was defined as FBG ≥ 6.1 mmol/L or 2-h postprandial blood glucose (2hPBG) ≥ 7.8 mmol/L or GLU ≥ 11.1 mmol/L[16]. Hypoglycemia was defined when FBG < 3 mmol/L[17]. Metabolic acidosis was defined as a blood pH < 7.35 and a decrease in plasma bicarbonate concentration (HCO3⁻) < 22 mmol/L, and lactic acidosis was defined when blood pH < 7.35 and the arterial lactate level > 5 mmol/L[18].

Variables that were expected to be potential confounders for DPP4i use and for the association of in-hospital DPP4i use with COVID-19 outcomes were addressed using the propensity score matching (PSM) model. These variables included age, sex, heart rate, blood pressure, pulse oxygen saturation (SpO₂) < 95%, CT-diagnosed bilateral lung lesions, incidence of increased leukocyte count, neutrophil count, CRP, procalcitonin, ALT, D-dimer, and low density lipoprotein cholesterol (LDL-c), decreased lymphocyte count and eGFR, comorbidities (chronic obstructive pulmonary disease, cerebrovascular diseases, chronic liver disease, and chronic renal diseases), number of oral hypoglycemic agents, and proportion of insulin usage. A nonparametric missing value imputation based on the missForest procedure in R was adopted to explain the missing laboratory data. A random forest model based on the other variables in the data set was applied to predict the missing values and to estimate the internally cross-validated errors. According to the propensity scores, the DPP4i and non-DPP4i users were paired at a ratio of 1:3 through exact matching with a caliper size of 0.05. The balance between covariates was evaluated by estimating standardized differences before and after matching, and those with absolute values less than 0.1 were regarded as successfully balanced. A sensitivity analysis was performed to assess the robustness of our findings.

The statistical methods of this study were reviewed by Dr. Jing Chen from Department of Mathematics and Physics, Wuhan Institute of Technology. Statistical analyses were conducted with R-3.6.3 (R Foundation for Statistical Computing, Vienna, Austria) and SPSS Statistics (version 23.0, IBM, Armonk, NY, United Stated). Continuous variables are presented as medians and interquartile ranges (IQRs), and categorical variables are presented as frequencies and percentages (%). Statistical differences between two groups were analyzed using t tests (normally distributed data) or the Kruskal-Wallis rank sum test (non-normally distributed data) for continuous variables, and comparisons of categorical variables were performed by Fisher’s exact test or the χ² test.

The association between DPP4i use and the risk of COVID-19 outcomes was analyzed using a logistic regression model and a mixed-effect Cox model. The corresponding adjusted odds ratio (OR) and hazard ratio (HR) were calculated after adjusting for the incidence of increased CRP between the DPP4i and non-DPP4i groups. For the mixed-effect Cox model, we modeled site as a random effect, and the proportional hazard assumptions were verified using correlation testing based on Schoenfeld residuals. Considering the possibility of type 1 error due to multiple comparisons, the findings from the analyses of the secondary endpoints should be interpreted as exploratory.

This study included a total of 2563 participants with type 2 diabetes who were admitted to hospitals due to COVID-19. After excluding 148 patients without complete electronic medical records, 78 aged ≥ 85 or < 18 years, 148 pregnant patients or patients with ineligible diseases or contraindications, 62 with confusion or mental disorders, 547 without hypoglycemic therapy, and 323 not taking oral hypoglycemic agents during hospitalization, 1257 patients [aged 64 years (IQR, 56-69); 52.03% male] were included for the retrospective longitudinal analysis (Figure 1). The 142 participants [aged 63 years (IQR, 55-67); 46.48% male] who received DPP4i drugs were classified into the DPP4i group, while the remaining 1115 [aged 64 years (IQR, 56.5-69); 52.74% male] were included in the non-DPP4i group (Table 1). All of the included participants were followed for a total of 28 d after admission. Compared to that in the non-DPP4i group, a higher percentage of participants in the DPP4i group took three or more types of anti-diabetic agents (36.62% in the DPP4i group vs 12.11% in the non-DPP4i group for 3 types; 13.38% vs 1.08% for ≥ 4 types). Compared to that in the non-DPP4i group, the incidence of elevated ALT was lower but the incidence of elevated LDL-c was higher in the DPP4i group at the time of admission (Table 1).

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Flow chart of study inclusion. T2D: Type 2 diabetes; DPP4i: Dipeptidyl peptidase-4 inhibitors.

To demonstrate the associations between in-hospital DPP4i use and COVID-19-related mortality, we performed PSM analysis and successfully matched 111 participants in the DPP4i group with 333 non-DPP4i users at a ratio of 1:3. The clinical characteristics of the DPP4i group vs the non-DPP4i group on admission after PSM are presented in Table 1.

       In the logistic regression model, we did not observe any significant difference between patients taking DPP4i drugs and those taking other oral hypoglycemic drugs regarding the incidence or risk of 28-d all-cause mortality (1.8% vs 3.3%; OR = 0.58 95%CI: 0.12–2.68, P = 0.48). The incidences and risks of the secondary outcomes and the occurrences of septic shock, ARDS, or acute organ (kidney, liver, and cardiac) injury were comparable between the DPP4i and non-DPP4i groups (Table 2).

       After further adjusting for imbalanced variables (the incidence of elevated CRP) in the mixed-effect Cox model using site as a random effect, the results remained nonsignificant for the association of DPP4i use with 28-d all-cause mortality in patients with COVID-19 (adjusted HR = 0.44, 95%CI: 0.09-2.11, P = 0.31) or with the abovementioned secondary outcomes (Table 2). The insignificant association between the in-hospital use of DPP4i and 28-d mortality due to COVID-19 was further validated in the sensitivity analyses (adjusted HR = 0.45, 95%CI: 0.10-2.08, P = 0.31 in the first sensitivity analysis with caliper size at 0.04; adjusted HR = 0.35, 95%CI: 0.07-1.71, P = 0.19 in the second sensitivity analysis without matching procalcitonin).

Our recent study demonstrated that patients with well-controlled glucose levels had a significantly reduced risk of COVID-19-related mortality compared to those with poorly controlled glucose levels among patients with coexisting diabetes[2]. We thus further analyzed whether DPP4i treatment had a better performance in maintaining glucose control than non-DPP4i treatment. However, the median, maximum, and minimum FBG and GLU were all approximately equivalent between the patients who received DPP4i drugs and those who received non-DPP4i drugs during hospitalization (Table 3). Although the relative duration of FBG above 7.0 mmol/L was slightly longer in the DPP4i group than in the non-DPP4i group, there was no statistically significant difference (Table 3). Likewise, the glucose control-related adverse outcomes, including hyperglycemia requiring treatment, hypoglycemia, acidosis, lactic acid elevation, and the probability of discontinuing the current regimen, were all comparable among patients with diabetes in the DPP4i and non-DPP4i groups (Table 4).

It has been reported previously that DPP4i might exert a comprehensive anti-inflammatory effect by blocking inflammatory cell activation and cytokine production[4]. However, the box plot demonstrates that there were no significant differences in neutrophil count, leukocyte count, or cytokine concentrations (e.g., tumor necrosis factor-α [TNF-α] and interleukin-6 [IL-6]) between the DPP4i and non-DPP4i groups among patients with COVID-19 with coexisting type 2 diabetes (Figure 2).

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Box plots showing the median levels of inflammatory response indicators in dipeptidyl peptidase-4 inhibitor and non-dipeptidyl peptidase-4 inhibitor groups (n = 444). A-F: Box chart analysis of median levels of neutrophil count, neutrophil percentage, leukocyte count, tumor necrosis factor-α, interleukin (IL)-6, and IL-10 between the dipeptidyl peptidase-4 inhibitor (DPP4i) and non-DPP4i groups. The median levels of those parameters of each participant during the follow-up duration were applied and were normalized according to their upper limits of normal range of each hospital. DPP4i: Dipeptidyl peptidase-4 inhibitors; TNF-α: Tumor necrosis factor-α; IL-6: Interleukin-6; IL-10: Interleukin-10.

Diabetes is one of the most important comorbidities linked to the remarkably increased severity and mortality of coronavirus disease 2019 (COVID-19). Our most recent retrospective study based on one of the largest scale patient cohorts has demonstrated that glucose control is the key point to diminish COVID-19 mortality for patients with co-existing diabetes. Dipeptidyl peptidase 4 (DPP4) is commonly targeted for glucose lowering and has outstanding roles in inflammatory and immune regulation. As the well-established Middle East respiratory syndrome coronavirus cellular receptor, DPP4 was predicted to have a tight interaction with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike glycoprotein in the recent molecular modeling study.

The latest observations sparked intense debates about whether DPP4 inhibitors might be promising therapeutic agents against COVID-19 among patients with diabetes. However, there is no direct clinical evidence supporting the associations between DPP4i usage and COVID-19 prognosis so far.

To investigate the potential impacts of DPP4i on COVID-19 among patients with diabetes, and to clarify whether continuation of DPP4i for diabetes management among patients in the setting of COVID-19 is an appropriate option.

We enrolled 15451 confirmed COVID-19 cases, including 2563 inpatients with type 2 diabetes from 16 hospitals in Hubei Province, between December 30, 2019 and April 23, 2020. After excluding those ineligible individuals, 142 patients receiving DPP4i and 1115 cases receiving non-DPP4i oral anti-diabetic were included in the subsequent analysis. Then we performed a strict propensity score matching (PSM) analysis where age, sex, comorbidities, number of oral hypoglycemic agents, heart rate, blood pressure, pulse oxygen saturation (SpO₂) < 95%, CT diagnosed bilateral lung lesions, laboratory indicators, and proportion of insulin usage were matched. Finally, 111 participants treated with DPP4i were successfully matched with 333 non-DPP4i users. Then logistics linear model and mixed-effect Cox model were applied to analyze the associations between in-hospital use of DPP4i and adverse outcomes of COVID-19.

We found that there was no significant association between patients taking DPP4i vs other oral hypoglycemic drugs regarding the incidence or risk of 28-d all-cause death (adjusted hazard ratio = 0.44, 95%CI: 0.09-2.11, P = 0.31). Likewise, the incidences and risks of secondary outcomes including septic shock, ARDS, or acute organ (kidney, liver, and cardiac) injuries were also comparable between the DPP4i and non-DPP4i groups. And the performances of DPP4i on glucose control (e.g., the median level of fasting blood glucose and random blood glucose) and inflammatory regulation were all kept approximately equivalent in the DPP4i vs non-DPP4i group. Furthermore, we did not observe substantial side effects on uncontrolled glycaemia or acidosis due to DPP4i application compared to non-DPP4i agents in the study cohort.

These data demonstrated that there are no significant correlations between DPP4i treatment and adverse clinical outcomes among COVID-19 patients with diabetes, and DPP4i does not show any other significant adverse effects on anti-diabetic treatment. Thus, DPP4i might be a relative safe option for managing pro-existing type 2 diabetes patients infected by SARS-CoV-2, which may provide certain reference value for clinical decision-making. Our study also paves the way for perspective studies and randomized controlled clinical trials on DPP4i therapy for COVID-19.

Large-scale prospective cohort studies and randomized clinical trials with ethnically and geographically diverse cohorts are needed to better understand the effects of DPP4i agents on mortality, glycemic control, and side effects among COVID-19 individuals.