Overall, 2,878 patients hospitalised for COVID-19 in 24 French centres were included between February 26 and April 20, 2020 (list of participating centres in Supplemental table S1). Among them, 27 were excluded from analysis for missing data for outcomes or diabetes status (Figure 1). Clinical and biological characteristics of the 2,851 patients included for analysis overall and according to the diabetes status are presented in Table 1. The mean (±SD) age was 67 ± 17 years and 1203 (42.2%) were female. The prevalence of comorbidities at admission was: hypertension (50.7%), diabetes (23.7%), cardiovascular diseases (12.6%), chronic heart failure (10.9%) and chronic obstructive pulmonary disease (5.6%). Regarding laboratory findings, levels of C - reactive protein were elevated whereas most of other laboratory findings at admission were in the normal range. Abnormalities on chest computed tomography were classified as severe in 427 (19%) patients. The median time from symptom onset to admission was 7.0 [3.0;10.0] days and duration of hospital stay was 8.0 [5.0;12.0] days (Table 2). Antibiotic therapy was the most prescribed drug during hospitalization (74.4%). Overall, anticoagulation therapy was prescribed during hospitalization in 88.2% of patients including prophylaxis-dose (63.2%) and therapeutic-dose (18.5%).
Risk factors associated with outcomes in the overall population are reported in Supplemental Table S2. Briefly patients who experienced outcomes, compared with those who did not, had more often diabetes and diabetes related comorbidities such as hypertension, diabetes, dyslipidemia, chronic kidney disease, history of cardiovascular disease and heart failure.
Comparison according to diabetes status in the unmatched cohort
Compared to patients without diabetes, those with diabetes were older (mean age 71 ± 13 years vs. 65 ± 18 years), were less often female (38% vs. 44%) and more likely to have medical history of hypertension (79% vs 42%), coronary heart disease (23% vs 9%), stroke (13% vs 8%), chronic heart failure (17% vs 9%), chronic kidney disease (26% vs 10%), and chronic obstructive pulmonary disease (7.26% vs 5%) (Table 1). Accordingly, the use of cardioprotective drugs at baseline such as blood pressure lowering, lipid lowering drugs, antiplatelet and anticoagulant agents were present more often among patients with diabetes. Laboratory findings showed a higher level of white cell count and C - reactive protein in patients with diabetes than those without diabetes. Glomerular filtration rate was lower in diabetes group. Prevalence of severe lung injury was slightly but not significantly higher in patients with diabetes compared to those without diabetes (21.8% vs 18.4% respectively) (Table 1). The use of pharmacological agents during follow-up was similar in patients with and without diabetes except for therapeutic-dose anticoagulation which was more often prescribed in patients with diabetes (21.9% vs 17.4%) (Table 2).
Outcomes in the unmatched and the matched cohorts
During a median follow-up of 19.0 [6.0;27.0] days, the primary outcome occurred in 830 (29.1%) patients, including 283 (10.0%) patients who died without transfer in ICU and 547 (19.2%) transferred in ICU. The overall rate of in-hospital death was 12.6% (358 patients) including 75 (2.6%) deaths in ICU. At the end of the follow-up, 1979 (69.6%) patients were discharged and 507 (17.8%) were still hospitalized.
Compared to patients without diabetes, those with diabetes had shorter durations from admission to occurrence of primary outcome (15.0 days [4.0;25.0] vs. 20.0 days [7.0;27.0] respectively; p<0.001) but not from admission to death (7.0 days [4.0;11.0]) vs 7.0 days [4.0;11.0]; p=0.63). The primary outcome occurred in 584 (36.4%) patients with diabetes compared to 246 (26.8%) in those without diabetes (p<0.001). We also observed a significant higher rate of secondary outcomes (overall mortality, transfer in ICU, invasive mechanical ventilation) in patients with diabetes versus those without (p<0.05 for all) (Table 2). The Kaplan-Meier survival curves for primary outcome and overall mortality according to diabetic status are presented in Supplemental Figure S1. The crude cox proportional hazard model showed a significant higher risk of primary outcome and overall mortality in patients with diabetes compared with those without (HR 1.44 [95%CI 1.24 – 1.67] for the primary outcome) (Supplemental Table S3).
PSM analysis was further performed to avoid confounding variables that could have interfered with the association between outcomes and diabetic status. After PSM, characteristics between groups were better balanced with no significant difference between patients with and without diabetes (Table 1).
As shown in Supplemental Table S2, in the PSM cohort, patients who experienced primary outcome, compared with those who did not, had more often diabetes and diabetes related comorbidities such as hypertension, diabetes, dyslipidemia, chronic kidney disease, history of cardiovascular disease and heart failure.
The incidence of the primary outcome was not statistically different between the two groups. Indeed, primary outcomes occurred in 214 (35.5%) patients with diabetes versus 192 (31.8%) (OR 1.18, 95%CI 0.93 – 1.50; p=0.20) (Table 2). No differences regarding the secondary outcomes were found between patients with and without diabetes. In-hospital death during follow-up encountered in 111 (18.4%) patients with diabetes compared to 91 (15.1%) patients without diabetes (OR 1.27, 95%CI 0.94 – 1.72; p=0.14). Figure 2 showed the survival curves for primary outcome and overall mortality in patients with and without diabetes. Cox proportional hazards survival regression analyses did not show any significant association between diabetes and the incidence of either the primary outcome (HR 1.16, 95%CI 0.95-1.41, p=0.14) or each of the secondary outcome (Table 3).
As sensitivity analysis, we compared the incidence of outcomes in diabetes and non-diabetes groups in two other PSM cohorts (one including personal characteristics and comorbidities and one including admission vitals and laboratory findings) to better assess the impact of each type of covariates on the association between diabetes and events. In the first PSM cohort in which patients with and without diabetes were matched based on their personal characteristics and comorbidities, no difference was found for vital signs, laboratory and radiological findings between both groups (Supplemental Table S4). In the second PSM, including for matching vital signs and biological findings only, patients with diabetes were older, more frequently male, had a higher BMI and a higher prevalence of comorbidities as observed in the unmatched cohort. Cox regression analysis showed that in both PSM cohorts, associations between diabetes status and risk of severe outcomes were non-significant (Supplemental Table S5).