Study design
This was a retrospective observational study of prospectively collected data relating to a cohort of hospitalised COVID-19 patients.
Setting
The study was conducted at the Department of Infectious Diseases and the intensive care unit of Luigi Sacco Hospital in collaboration with the Department of Internal Medicine of Fatebenefratelli Hospital. Luigi Sacco Hospital, which is located on the outskirts of Milan, is one of the city’s major infectious diseases centres, and has been at the forefront of the hospitalisation of COVID-19 patients since the first days of the pandemic in Italy [13,14,15,16]. The Department of Internal Medicine of Fatebenefratelli Hospital, which is located in the inner city, was rapidly converted to a COVID-19 centre when the pandemic struck.
As laid down in Italian healthcare regulations, urgent and essential healthcare is provided free of charge to Italians and immigrants regardless of their legal status.
Participants
The study enrolled all of the adult patients with a diagnosis of COVID-19 confirmed by reverse-transcriptase polymerase chain reaction on a nasopharyngeal swab who were admitted to our hospitals between 21 February and 31 November 2020; observation of the cohort was censored on 28 February 2021.
Data source and management
The characteristics of the data management have been fully described elsewhere [13,14,15,16]. In brief, the data were extracted from the patients' clinical charts on a daily basis, and were stored in an ad hoc database. The collected data were the patients’ date and place of birth, and biological sex; the time between symptom onset and hospital admission; co-morbidities (including diabetes, lung diseases, heart diseases, renal diseases, immune system diseases, liver diseases, and obesity defined as a body mass index of ≥ 30) [17]; the burden of co-morbidities (0, 1, 2, and 3 +); whether there was a need for supportive oxygen therapy upon hospital admission; disease severity upon hospital admission (defined as mild, moderate, severe or critical in accordance with the World Health Organisation (WHO) guidelines for the management of COVID-19) [13, 18]; and hospitalisation outcome (date and cause of death, discharge, or transfer to another facility). The vital status of the patients discharged or transferred before the censoring date was ascertained by means of telephone calls.
Outcomes and variables
The main outcome of interest was COVID-19-related mortality, and the principal variable of interest was place of birth. The patients were classified as natives (if they were born in Italy) or immigrants (sub-divided into four regions of origin: central/eastern Europe, Africa, Latin America, and Asia).
The baseline variables known to be clinically relevant to the outcome of interest [3, 13, 14] and included in the analysis as potential confounders were age, biological sex, the number of days between symptom onset and hospital admission, co-morbidities (including obesity), and disease severity upon hospital admission.
Statistical analysis
The descriptive statistics of the categorical variables are given as proportions, and those of the continuous variables as median values and interquartile range (IQR). The baseline demographic and clinico-epidemiological characteristics of the Italians and immigrants were compared using the χ2 or Fisher's exact test for the categorical variables, and Wilcoxon’s rank-sum test for the continuous variables; the characteristics of the immigrants from different regions of origin were also compared in the same way.
The Kaplan–Meier method was used to plot the survival curves of the Italians and the immigrants as a whole or stratified on the basis of their region of origin (central/eastern Europe, Africa, Latin America, and Asia). Survival curves adjusted for the potential confounders of age, biological sex, time from symptom onset, obesity, and disease severity upon hospital admission were generated using Cox’s model.
The association between the patients’ origins and the risk of COVID-19-related death was investigated using uni- and multivariable Cox proportional hazard ratios (HRs) and their 95% confidence intervals (CIs). All of the variables were entered in the multivariable model by adjusting the effect of immigrant status and region of origin for all of the other co-variates as potential confounders. The discriminative ability of the model was assessed using the integrated area under the receiver-operating characteristic curve (AUC-ROC), which averages all of the available AUC statistics over time. The presence of multi-collinearity among the explanatory variables was verified using the generalised variance inflation factor (GVIF) tool.
All of the statistical analyses were made using SAS software, version 9.4, and differences with a P value of < 0.05 were considered statistically significant.
The study was approved by our Ethics Committee (Comitato Etico Interaziendale Area 1, Milan, Italy). All of the patients signed a written informed consent form except for those undergoing mechanical ventilation upon admission for whom it was allowed to be waived.