Methods
This is a cross-sectional observational study conducted at an ICU in the Central-West region of Brazil. The study was approved by the local Research Ethics Committee on February 26, 2021 (protocol number: 4.563.056/2021; Certificate of Presentation for Ethical Appraisal (CAAE) number: 43454621.2.0000.5077). Data collection was performed from March 12, 2021 to May 26, 2022.
Patients admitted to the ICU, diagnosed with SARS-CoV‑2 infection, requiring IMV, older than 18 years, and of both sexes, were included in the study. Patients who had three or more incompletely analyzed variables in their medical records, who had no record of death outcome due to transfer to another institution, or who demonstrated unclear association between deaths and SARS-CoV‑2 infection were excluded from the study.
Patients who had clinical dyspnea and required supplemental oxygen, and the ratio of partial arterial pressure of oxygen (PaO2) and fraction of inspired oxygen (FiO2) ≤ 300 mmHg were considered as having ARF7.
Data regarding sociodemographic characteristics, chronic diseases, lifestyle, treatment, and hospital death were collected from medical records. In-hospital death was considered an outcome or a dependent variable. The considered sociodemographic variables were sex (male or female) and age. The considered chronic disease variables were systemic arterial hypertension; diabetes mellitus (DM); hypothyroidism; hyperthyroidism; chronic obstructive pulmonary disease; cardiovascular disease (CVD); cerebrovascular disease; mental health-related disease; history of thrombosis; hepatitis A, B, and C; asthma; and liver failure. The considered lifestyle variables were smoking, alcohol consumption, and illicit drug use. The considered treatment variables were the use of the prone position during hospitalization and the use of drugs, such as dexamethasone, azithromycin, chloroquine, orhydroxychloroquine, and low-molecular-weight heparin (enoxaparin).
For the sample size calculation, expected prevalence (or rate) of death of 45% was used,8 with an error margin of 5.0% and a 95% confidence interval (95% CI) for a population of 300,000 inhabitants residing in the city of Rio Verde, State of Goiás, Brazil, and its neighboring districts. Thus, the sample size required for this analysis (prevalence calculation) was found to be 380 patients.
All statistical analyses were performed using the Stata statistical package, version 16.0 (StataCorp LLC, College Station, TX, USA). The variables were presented as absolute numbers (n) and relative frequencies (%) with mean and standard deviation. Poisson’s regression was used to calculate the prevalence ratio and 95% CI, andp -values was obtained using the Wald test. Variables with ap -value of <0.20 in bivariate analysis were included in multiple hierarchical Poisson regression analyses, with robust variance based on a hierarchical model9. The independent variables in this hierarchical analysis were classified as follows: (I) demographic data (gender and age), (II) chronic diseases (DM and mental illness), (III) clinical features (pronation), and (IV) medication use (azithromycin and heparin). Statistical significance was established using a cutoff value of p  < 0.05. Notably, thep -value of 0.000 indicates that the relationship is statistically significant at the α = 0.05 level.
Variables without statistical power; that is, variables that after the bivariate analysis presented n < 10 in any stratum, were excluded from the multiple regression analysis10. Additionally, the variables of hyperthyroidism, mental illness, asthma, liver failure, smoking, and use of chloroquine/hydroxychloroquine were excluded.