Outcomes of COVID-19 patients intubated after failure of non-invasive ventilation: a multicenter observational study

The efficacy of non-invasive ventilation (NIV) in acute respiratory failure secondary to SARS-CoV-2 infection remains controversial. Current literature mainly examined efficacy, safety and potential predictors of NIV failure provided out of the intensive care unit (ICU). On the contrary, the outcomes of ICU patients, intubated after NIV failure, remain to be explored. The aims of the present study are: (1) investigating in-hospital mortality in coronavirus disease 2019 (COVID-19) ICU patients receiving endotracheal intubation after NIV failure and (2) assessing whether the length of NIV application affects patient survival. This observational multicenter study included all consecutive COVID-19 adult patients, admitted into the twenty-five ICUs of the COVID-19 VENETO ICU network (February–April 2020), who underwent endotracheal intubation after NIV failure. Among the 704 patients admitted to ICU during the study period, 280 (40%) presented the inclusion criteria and were enrolled. The median age was 69 [60–76] years; 219 patients (78%) were male. In-hospital mortality was 43%. Only the length of NIV application before ICU admission (OR 2.03 (95% CI 1.06–4.98), p = 0.03) and age (OR 1.18 (95% CI 1.04–1.33), p < 0.01) were identified as independent risk factors of in-hospital mortality; whilst the length of NIV after ICU admission did not affect patient outcome. In-hospital mortality of ICU patients intubated after NIV failure was 43%. Days on NIV before ICU admission and age were assessed to be potential risk factors of greater in-hospital mortality.

Fraction of inspired oxygen PaO 2 Arterial partial pressure of oxygen PaCO 2 Arterial partial pressure of carbon dioxide CVVH Continuous venous-venous hemofiltration ED Emergency department OR Odds ratio CI Confidence interval ETI Endotracheal intubation The efficacy of non-invasive ventilation (NIV), including both Biphasic Positive Airway Pressure (BiPAP) and non-invasive Continuous Positive Airway Pressure (CPAP), in patients with acute respiratory failure (ARF) secondary to coronavirus disease 2019 (COVID-19) is still debated 1,2 .
On the one hand, some authors believe that NIV represents a questionable option and controlled mechanical ventilation should be established as soon as possible because of the risks of patient self-inflicted lung injury and delayed intubation 3 . On the other hand, solid evidence in favor of early intubation in COVID-19 ARF is still lacking, as several investigations failed to reveal a significant difference in mortality according to the time of intubation 4,5 .
Therefore, we designed this study aiming to investigate the incidence of in-hospital mortality in ICU patients receiving endotracheal intubation after NIV failure and to ascertain whether the length of NIV application before intubation may affect patient survival.

Methods
The protocol was approved by the Institutional Ethical Committee of each participating centre (Ref: 4853AO20). The study was conducted in accordance with the Helsinki declaration and national regulation on study involving humans. Informed consent was obtained for each patient in compliance with national regulation and the recommendations of the Institutional Ethical Committee of Padova University Hospital.
We screened the records of all adult patients with confirmed SARS-CoV-2 infection, admitted into the twentyfive ICUs belonging to the COVID-19 VENETO ICU network 12 , between February 28 and April 28, 2020. We deemed eligible for analysis only patients who received endotracheal intubation after experiencing NIV (either CPAP or BiPAP) failure 12 . Patients exclusively receiving conventional and/or high-flow oxygen therapy or NIV, intubated after high-flow oxygen therapy, experiencing invasive mechanical ventilation without previous noninvasive treatments, with incomplete records or defined 'do not intubate' were excluded. Details on NIV setting, hospital organization and criteria for intubation are described in the supplementary material (Additional file, Methods).
The diagnosis of COVID-19 was made according to the WHO interim guidance (http:// www. who. int/ docs/ defau lt-source/ coron aviru se/ clini cal-manag ement-of-novel-cov. pdf). Laboratory confirmation of SARS-CoV-2 was defined as a positive result of real-time reverse transcriptase-polymerase chain reaction assay of nasopharyngeal swabs.
The following variables were collected: i) demographic data (age, gender, body mass index (BMI), onset of symptoms); ii) medical history (chronic diseases and long-term therapies, Charlson comorbidity index unadjusted for age 24 ); iii) laboratory findings at ICU admission (blood count with formula, coagulation tests, C-reactive protein (CRP), procalcitonin, coagulation tests) and in-hospital treatments (i.e., ongoing therapies, including antiviral drugs and corticosteroids); iv) sequential organ failure assessment (SOFA) score at ICU admission; v) respiratory parameters before endotracheal intubation, i.e., positive end-expiratory pressure (PEEP), inspiratory pressure support above PEEP, fraction of inspired oxygen (FiO 2 ), pH, arterial partial pressure of oxygen (PaO 2 ), PaO 2 /FiO 2 , arterial partial pressure of carbon dioxide (PaCO 2 ) and respiratory rate; vi) length of NIV application, either overall, before and after ICU admission; vii) the hospital location where NIV was applied, i.e., when NIV was applied exclusively in medical wards, respiratory high dependency units or emergency departments (ED), patients were included in the 'out-of-ICU' group. When NIV was applied exclusively after ICU admission patients were included in the 'in-ICU' group. When NIV was applied before and after ICU admission, patients were included in the 'out-and in-ICU' group; viii) complications occurred during the ICU stay (see full description listed in the additional file, Table 1); ix) ICU and hospital lengths of stay; x) hospital location before ICU admission (medical wards, respiratory high dependency units or ED); xi) hospital mortality.
For patients being readmitted or moved to a different hospital, only data from the first admission were considered. This study followed the 'Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement guidelines for observational cohort studies' 25 (Additional files, Table 2). Each investigator had a personal username and password and entered data into a pre-designed online data acquisition system (www. covid 19ven eto. it). Patients' privacy was protected by assigning a de-identified patient code. Prior to data analysis, two independent investigators and a statistician screened the database for errors against standardized ranges and contacted local investigators with any queries. Validated or corrected data were then entered into the database for final analysis. Categorical data were presented as absolute numbers and percentages; for continuous data, normality was tested by Skewness and Kurtosis tests. Means and standard deviations were used when the variables were normally distributed, while medians and interquartile ranges were used in case of non-normally distributed variables. No imputation for missing data was planned.
Univariate analysis was used to investigate any difference between in-hospital survivors vs. non-survivors, concerning clinical characteristics, respiratory parameters before endotracheal intubation and the length of NIV application, both overall, before and after ICU admission.
Then, the independent predictors of in-hospital mortality have been identified through a stepwise multivariable regression model. This approach combines forward and backward selection methods in an iterative procedure (with a significance level of 0.05 both for entry and retention) to select predictors in the final multivariable model 26 . Independent variables used in the stepwise approach, and selected considering their clinical relevance, were age, Charlson comorbidity index, SOFA score at ICU admission, PaO 2 /FiO 2 , length of NIV application before, after ICU admission and the overall length of NIV.
Curves of cumulative incidence of in-hospital mortality were drawn to describe in-hospital mortality stratified by: i) patients' characteristics (age); ii) length of NIV application prior to intubation; iii) and hospital location initially providing NIV. The median age and median length of NIV application, prior to intubation, of nonsurvivors were used as cut-off values for stratifying patients in two groups, as previously done 16 . Since discharge must be considered an 'informative' censoring 27 , cumulative incidence was calculated using methods accounting for competing risks and conventionally reported at 60-days. The Gray's test was used to assess the difference between cumulative incidence functions. The observation period started at the day of endotracheal intubation. All statistical tests were 2-tailed, and statistical significance was defined as p < 0.05. Ethical approval. This was a multicenter, observational study performed in twenty-five hospitals of Veneto Region, Northern Italy, listed on the Acknowledgements. All the participating centers obtained Ethics Committee approval for the present research project, initially approved by the Institutional Ethical Committee of Padova University hospital on the 21st April, 2020 (Ref: 4853AO20). Local investigators were responsible for ensuring data integrity and validity. The study was conducted in accordance with the Helsinki declaration and national regulation on study involving humans. Informed consent was obtained for each patient in compliance with national regulation and the recommendations of the Institutional Ethical Committee of Padova University Hospital.

Consent for publication.
Informed consent was obtained for each patient in compliance with national regulation and the recommendations of the Institutional Ethical Committee of Padova University Hospital.
Baseline demographic and clinical characteristics of the study population are presented in Table 1 or listed in the Additional files, Table 1.
At univariate analysis, Charlson comorbidity index, SOFA score at ICU admission, FiO 2 , PaO 2 /FiO 2 , PaCO 2 and the length of NIV before ICU admission were significantly related to in-hospital mortality (Table 1). On the contrary, at the multivariable logistic regression model, only age and the length of NIV before ICU admission were confirmed as independent predictors of in-hospital mortality ( Table 2).
Finally, in-hospital mortality was higher in patients exclusively treated with 'out-of-ICU' NIV, as opposed to those exclusively treated with 'in-ICU' NIV (cumulative incidence 51% vs 24%, p < 0.01) or treated with NIV both outside and inside the ICU (cumulative incidence 51% vs 41%, p = 0.04) (Fig. 4).

Discussion
This study, conducted during the first wave of COVID-19 pandemia, shows 43% in-hospital mortality among patients who underwent endotracheal intubation after NIV failure for SARS-CoV-2. Moreover, length of NIV application outside the ICU exceeding 48 h and age above 73 years were associated with greater mortality.
To the best of our knowledge, this is the first study focusing on the outcome of COVID-19 ICU patients intubated after NIV failure. Noteworthy, patients intubated after NIV failure showed a mortality rate no different from 292 patients receiving intubation without a previous NIV trial (42% vs 43%, p = 0.66) (Fig. 1), which suggests that attempting NIV did not worsen outcome even in case of intubation after failure.
While previous investigations were focused on the outcome of NIV delivered out of ICU 15,16,19,[21][22][23]28 , our study provides detailed information on the outcome of intubation after NIV failure. Worth remarking, our data do not allow drawing any conclusion on the benefits of the application of NIV outside the ICU, as we do not consider the multitude of patients successfully treated with NIV in settings other than ICU in Veneto region during the study period 12 .
Our study presents some limitations. First of all, like many of the investigations on COVID-19, it is an observational study, thus it bears the limits of this study design. Second, in keeping with previous guidelines, we did not distinguish between patients treated with CPAP or BiPAP 1,29 , nor between patients supported with helmet or facial mask, nor between continuous or intermittent treatments. Therefore, our data do not allow to separately evaluate the benefits of BiPAP vs. CPAP or helmet vs. facial mask. Irrespective of the mode and interface, however, NIV guarantees maintenance of airway defence mechanisms and allows flexibility in applying and removing ventilatory assistance 30 . Third, NIV was mainly delivered through helmets, which made impossible measuring tidal volume 31 and predicting the risk of patient self-inflicted lung injury 32 . Finally, it is worth remarking that the observed outcomes do not necessarily reflect those of patients treated outside a pandemic condition. www.nature.com/scientificreports/ In conclusion, 43% of ICU patients receiving intubation after NIV failure died. Length of NIV before ICU admission and age were independent predictors of in-hospital mortality. Our findings suggest that prompt intubation is advisable in the case of lack of improvement after 2 days of NIV delivered outside ICU.   www.nature.com/scientificreports/ Figure 2. In-hospital mortality stratified by age (≤ or > 73 years). p Value Gray's test was used for calculating equality of cumulative incidence function. The median age of non-survivors (= 73 years) was considered as the cut-off value for stratifying patients in two groups. NIV: non-invasive ventilation; ETI: endotracheal intubation.

Figure 3.
In-hospital mortality stratified by length of NIV application before ICU admission (≤ or > 2 days). p Value Gray's test was used for calculating equality of cumulative incidence function. The median length of NIV application before ICU admission of non-survivors (= 2 days) was considered as the cut-off value for stratifying patients in two groups. NIV: non-invasive ventilation; ICU: intensive care unit; ETI: endotracheal intubation.

Data availability
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Figure 4.
In-hospital mortality stratified by hospital location. p value Gray's test was used for calculating equality of cumulative incidence function. When NIV was applied exclusively in medical wards, respiratory high dependency units or Emergency Department, patients were included in the 'out-of-ICU' group. When NIV was applied before and after ICU admission, patients were included in the 'out-and in-ICU' group. When NIV was applied exclusively after ICU admission patients were included in the 'in-ICU' group. NIV: non-invasive ventilation; ICU: intensive care unit; ETI: endotracheal intubation.