Study population
We retrospectively included 202 patients admitted for COVID-19 (positive SARS-CoV-2 PCR) during the first epidemiological peak between February and June 2020 in a large tertiary care academic center. These patients were distributed in two groups: T2D or non-diabetic as controls. Control patients were matched for age and sex with diabetic patients. Blood analysis at admission included glycemia, leukocyte, polymorphonuclear neutrophils (PMNs), lymphocyte and platelet counts, C-reactive protein, troponin-T, fibrinogen, creatinine, glomerular filtration rate estimation, AST and ALT.
Multidetector chest computed tomography (CT) was performed at admission to evaluate COVID-19 pneumonia severity. Subsequently, admitted patients were either hospitalized in a conventional medical unit (CMU) or in an intensive care unit (ICU) as required according to clinical severity criteria.
This ancillary monocentric observational study was based on a COVID-19 cohort approved by the local ethics committee CER-SU 2020-14 and registered as NCT04320017 (ClinicalTrials.gov). According to local legislation all study participants could withdraw their participation in the study.
Outcome data
All clinical and biological data was collected from digital hospital admission and follow-up records. Outcome was collected at day 21 from the centralized hospital data recording ICU admission and mortality. The outcome endpoint was a composite criteria including death or ICU admission at 21 days of hospital admission.
CT acquisition protocol
All patients underwent non-ECG gated helicoidal thoracic acquisitions at 120 kV with 0.6 mm collimation on either a dual source SOMATOM Definition Flash or EDGE scanner (Siemens Healthineers, Erlangen, Germany). Acquisitions were performed with or without contrast media injection.
CT image analysis
COVID-19 related lung involvement was measured semi-quantitatively and reported according to the parenchymal extension of lung lesions using a standardized visual scale including ground glass and/or condensation as: minimal (< 10%), moderate (10–25%), extensive (25–50%) and severe (> 50%).
Since acquisitions were non-ECG-gated, we used the validated CAC-DRS score to quantify coronary calcium burden [12] graded from 0 to 3 as follows: 0: very low calcium; 1: mildly increased calcium; 2: moderately increased calcium; 3: moderately to severely increased calcium.
We assessed adipose tissue imaging biomarkers related to different fat components using a semi-automated AI based segmentation method (Siemens Healthineers Frontier) including the following parameters (typical segmentation results are illustrated in Fig. 1):
-
The total cardiac adipose tissue (CAT) in mL was measured after automated AI-based segmentation of the cardiac area with thresholding centered on the density range of adipose tissue values (-190 to -30 Hounsfield Units).
-
The visceral abdominal fat (VAT) and the subcutaneous fat (SAT) were measured semi-automatically between L1 and L2 vertebral bodies. In this stack of images, draw external and internal contours of the abdominal wall were drawn, and an automated segmentation of visceral and subcutaneous fat was performed centered on the density range of adipose tissue values (-190 to -30 Hounsfield Units).
These adipose tissue volumes were normalized to body surface area providing indexed parameters CATi, VATi and SATi respectively, in mL/m2.
The average time to perform thoracic and abdominal adipose fat tissue post processing was < 10 min by patient including image data management and in the order of 10 s for computation. for CAT alone. In Fig. 1, we illustrate the CT imaging biomarkers in two patients with favorable and unfavorable courses.
Statistical analysis
Numeric variables were reported as median and interquartile range (IQR) and qualitative variables as frequencies and percentages. To compare differences in summary statistics between groups, we used Mann–Whitney rank-sum and chi2 tests, respectively.
We performed subgroup analyses in patients with T2D and in the control group. For each group, we performed univariate logistic regression with adjustment for age, male sex, BMI, CRP, dyslipidemia, adipose tissues (CATi, VATi, SATi), severe lung lesions (> 50%), leukocytes, eGFR, C-reactive-protein, Troponin-T and glycemia to analyze independent relationships of adipose tissue parameters and the primary endpoint (composite outcome of ICU or Death). Finally, we build different multivariate logistic regression models according to the clinical relevance, univariate statistical associations, and the absence of collinearity.
To assess the prognostic performance of the adipose tissue biomarkers (CATi/VATi/SATi) for the prediction of outcome (ICU admission or death), ROC curves were generated for each subgroup. We further determined the optimal threshold values for maximization of the sensitivity and specificity according to Youden’s index. P-values < 0.05 were considered statistically significant. All statistical analysis were performed using JMP Pro version 16 (SAS Institute Inc., Cary, NC, USA, 1989–2022) and graphs were created with GraphPad Graphics software (San Diego, California USA).