Comparison of the Characteristics and Outcomes of SARS-COV 2 Critically Ill Patients After Emergence of the Variant of Concern 20H/501Y.V2: A Comparative Cohort Study

There are currently no data regarding characteristics of critically ill patients with SARS-CoV-2 Variant of Concern (VOC) 20H/501Y.V2. We therefore aimed to describe changes of characteristics in critically ill patients with Covid-19 between the rst and the second wave when viral genome sequencing indicated that VOC was largely dominant in Mayotte Island (Indian Ocean). Methods Consecutive patients with Covid-19 and over 18 years admitted in the unique Intensive Care Unit (ICU) of Mayotte during wave 2 were compared with an historical cohort of patients admitted during wave 1. We performed a Logistic Regression (LR) comparing wave 1 and wave 2 as outcomes. To complete analysis of changes in characteristics and outcomes, we built a Random Forest model (RF) -i.e., a machine learning classication tool- using the same variable set as that of the LR. Results We included 156 patients, 41 (26.3%) and 115 (73.7%) belonging to the rst and second waves respectively. Univariate analysis did not nd difference in demographic data or in mortality. Our multivariate LR found that patients in wave 2 had less fever (absence of fever aOR 5.23, 95% CI 1.89 – 14.48, p=0.001) and a lower SAPS II (aOR 0.95, 95% CI 0.91 – 0.99, p=0.007) at admission; at 24 hours, the need of invasive mechanical ventilation was higher (aOR 3.49, 95% CI 0.98 – 12.51, p=0.055) and pO2/FiO2 ratio was lower (aOR 0.99, 95 % CI 0.98-0.99, p=0.03). Patients in wave 2 had also an increased risk of VAP (aOR 4.64, 95% CI 1.54-13.93, p=0.006). Occurrence of VAP was also a key variable to classify patients between wave 1 and wave 2 in the variable importance plot of the RF model.


Abstract Background
There are currently no data regarding characteristics of critically ill patients with SARS-CoV-2 Variant of Concern (VOC) 20H/501Y.V2. We therefore aimed to describe changes of characteristics in critically ill patients with Covid-19 between the rst and the second wave when viral genome sequencing indicated that VOC was largely dominant in Mayotte Island (Indian Ocean).

Methods
Consecutive patients with Covid-19 and over 18 years admitted in the unique Intensive Care Unit (ICU) of Mayotte during wave 2 were compared with an historical cohort of patients admitted during wave 1. We performed a Logistic Regression (LR) comparing wave 1 and wave 2 as outcomes. To complete analysis of changes in characteristics and outcomes, we built a Random Forest model (RF) -i.e., a machine learning classi cation tool-using the same variable set as that of the LR.

Results
We included 156 patients, 41 (26.3%) and 115 (73.7%) belonging to the rst and second waves respectively. Univariate analysis did not nd difference in demographic data or in mortality. Our multivariate LR found that patients in wave 2 had less fever (absence of fever aOR 5.23, 95% CI 1.89 -14.48, p=0.001) and a lower SAPS II (aOR 0.95, 95% CI 0.91 -0.99, p=0.007) at admission; at 24 hours, the need of invasive mechanical ventilation was higher (aOR 3.49, 95% CI 0.98 -12.51, p=0.055) and pO2/FiO2 ratio was lower (aOR 0.99, 95 % CI 0.98-0.99, p=0.03). Patients in wave 2 had also an increased risk of VAP (aOR 4.64, 95% CI 1.54-13.93, p=0.006). Occurrence of VAP was also a key variable to classify patients between wave 1 and wave 2 in the variable importance plot of the RF model.

Conclusion
Our data suggested that VOC 20H/501Y.V2 could be associated with a higher severity of respiratory failure at admission and a higher risk for developing VAP. We hypothesized that the expected gain in survival brought by recent improvements in critical care management could have been mitigated by increased transmissibility of the new lineage leading to admission of more severe patients. The immunological role of VOC 20H/501Y.V2 in the propensity for VAP requires further investigations.

Background
The coronavirus disease 2019  pandemic continues in the world, with recent estimates of more than 268 million cases diagnosed and more than 5 million deaths [1]. Multiple variants of the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), the causative agent of Covid-19, have been observed worldwide. A variant has one or more mutations that differentiate it from the ancestral Ribonucleic Acid (RNA) virus and from other variants. Among them, variant 20H/501Y.V2 (also known as variant Beta or B.1.351) has been characterized as a Variant Of Concern (VOC) i.e. a variant for which there is evidence of an increase in transmissibility, an increase in symptom severity (e.g., increased hospitalizations or deaths), signi cant reduction in neutralization by antibodies generated during previous infection or vaccination, reduced effectiveness of treatments or vaccines, or diagnostic detection failures [2].
Variant 20H/501Y.V2 is associated with multiple mutations in spike protein [3] and emerged in September 2020 in South Africa [4] before spreading worldwide. Variant 20H/501Y.V2 was the most common strain in some French regions between April and May 2021 [5].
There are currently limited data regarding clinical differences between Covid-19 disease caused by different variants especially for VOC 20H/501Y.V2. This VOC has been associated with an increased severity of the disease and increased in-hospital mortality [6,7] especially in patients older than 60 years [8]. However, clinical characteristics of critically ill patients remain undescribed.
Mayotte Island is an overseas department of France located in the Indian Ocean. The rst case of Covid-19 was diagnosed in week 11, 2020 and the epidemic peak of the rst wave was reached in May 2020. A few weeks after the emergence of the VOC 20H/501Y.V2 in South Africa, Mayotte Island experienced a new and more dramatic second wave, which peaked in week 5, 2021.
During week 5 and week 6, all SARS-Cov-2 positive samples were analyzed and more than 80% of them (150/172) pointed to 20H/501Y.V2 variant. Moreover, in the same period viral whole genome sequencing was performed for 23 critically ill patients: 87% (N=20) were infected by the 20H/501Y.V2 variant [9,10]. Therefore VOC 20H/501Y.V2 was considered as dominant during the second wave in Mayotte Island. The Intensive Care Unit (ICU) of the Mayotte's Hospital, a 16 beds unit, is the unique structure for critically ill patients on the island. This speci city allowed us to make a rigorous comparison between both waves and avoided interhospital variability previously reported in epidemiological studies for Covid-19 patients [11] We therefore aimed to describe changes in the characteristics of critically ill patients with Covid-19 after the emergence of the VOC 20H/501Y.V2 by comparing the clinical pro le of patients hospitalized during the rst and second waves of the SARS-Cov 2 pandemic.

Study design, waves and subjects de nitions
We performed a comparative cohort study in the Mayotte Hospital. All consecutive patients over 18 years of age admitted in ICU with laboratory-con rmed Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV 2) infection were retrospectively included in wave 1 and prospectively included in wave 2. The waves periods were de ned using ICU admission data. The rst wave was the period from the start of the pandemic i.e., March 15, 2020 (W12-2020) to September 27, 2020 (W39-2020); and the second wave was de ned from February 01, 2021 (W5-2021 to March 01, 2021 (W9-2021). Wave 2 was considered starting when viral whole genome sequencing results indicated that VOC 20H/501Y.V2 was predominant in Mayotte Island [9,10] as it has been in South Africa [3]. Patients admitted to ICU between these two periods were excluded to avoid risk of overlap between infection by Beta variant and the wild type virus. Laboratory con rmation for SARS-CoV-2 was de ned as a positive result of real-time reverse transcriptase-polymerase chain reaction (RTPCR) assay from either nasal or pharyngeal swabs, or lower respiratory tract samples.
Our study received approval from the ethical committee of the French Society of Anesthesia and Intensive Care Medicine (IRB 00010254 -2021 -080) in accordance with our local regulations. All patients or close relatives were informed that their data were included in the study when possible.
Reporting of the study followed the STROBE guidelines. [12] Data collection and outcomes We collected in health care records demographic data, medical history, comorbidities, the clinical frailty scale (a patient with a score ≥ 5 was considered frail [13]) and symptoms before admission. We also collected signs at admission, ROX index for patients under oxygen therapy [14], laboratory results and computed Tomographic (CT) scans (visual quanti cation was used to classify patients as per the percentage of lung parenchyma affected [15,16]) at admission.
Disease severity at ICU admission was evaluated using SOFA score, WHO Clinical Progression Scale [17], and SAPS II score. SOFA score and WHO Clinical Progression Scale were also evaluated at 24 hours of ICU admission.
Clinical evolution at 24-hours, use of adjunct measures (i.e., corticosteroids, neuromuscular blockers, nitric oxide, invasive therapies, prone positioning) and in-hospital complications (i.e., myocarditis, bradycardia under 50 bpm, neurocovid de ned as SARS-Cov-2 positive samples in cerebrospinal uid, thrombosis, coinfections and bacterial cultures results) were also recorded. A diagnosis of Ventilator-Associated Pneumonia (VAP) was made in case of clinical and radiological signs associated to bacterial documentation.
Finally, we evaluated the following outcomes: duration of mechanical ventilation, Ventilator-free days at 28 days (VFD28) [17] (de ned as the number of days with successful cessation (alive and free) from invasive mechanical ventilation for at least 48 h without reintubation in patients who survived 28 days after ICU admission, whereas for patients ventilated 28 days or more, or who died within 28 days ventilator-free days were zero), the need for medical evacuation, the dates of ICU and hospital discharge, and vital status at 28 days after ICU admission.
In case of Aero-medical evacuation for patient, the three receiving hospitals (University Teaching Hospital of Saint Pierre, La Réunion, University Teaching Hospital of Saint Denis, La Réunion and Hospital of Western Réunion) collected data using the same de nitions.

Statistical analysis
No statistical sample size calculation was made and sample size was equal to the number of patients admitted to the ICU during both de ned periods. Characteristics of patients were described as frequencies and percentages for categorical variables, as means and standard deviations (SD) or medians and interquartile ranges (IQR) for continuous variables. Categorical variables were compared by Chi-square or Fisher's exact test. After testing for normality (kolmogorov-Smirnov test) and for equality of variances (Fisher-Snedecor test), continuous variables were compared either by Student's t test, Welch's t-test or Wilcoxon's ranksum test. Tests were two-sided and values of p less than 0.05 were considered statistically signi cant. Missing data was not imputed.
In order to compare the characteristics of Covid-19 patients admitted to ICU during wave 1 and wave 2, we performed a Logistic Regression (LR). Variables signi cantly associated in bivariate analysis with a value of p<0.10 and with less than 10% missing data were included in the LR model. Results were presented as adjusted odds ratios (aOR) with their 95% con dence intervals (OR, 95% CI).
To complete analysis of changes in characteristics and outcomes, we built a Random Forest model (RF) -i.e., a machine learning classi cation tool-using the same variable set as that of the LR.
Random Forest built a multitude of decision trees based on bootstrap sampling of the training dataset. The variables in each tree were randomly selected. Algorithm aggregated the results of multiple decision trees via majority voting. [19] We set the number of multiple decision trees as 5000 and the number of variables used in a decision tree was 2. The RF model can provide importance ranking of the predictor variables using the Mean Decrease Accuracy (MDA) plot and the Mean Decrease in Gini (MDG) coe cient. The MDA plot expresses for all variable how much accuracy the model losses by excluding the variable. The more the accuracy suffers, the more important the variable is for classi cation. The MDG coe cient is a measure of how a variable contributes to the homogeneity of the nodes and leaves in the Random Forest model. The higher the value of MDA of MDG for a variable, the higher the importance of the variable in the model.
The prediction performance of the models was assessed by the area under the receiver operating characteristic curve (AUROC).
The statistical analysis was implemented using SAS (Enterprise Guide 7.11) and R (1.1.463).

Results
We included 156 patients among 190 assessed for eligibility, 41 during the rst wave and 115 during the second wave. Thirty-four patients were excluded for risk of overlap between variant and wild-type virus. Six patients enrolled were lost to follow-up at Day-28. (Figure 1

Univariate analysis of characteristics and outcomes
We did not observe signi cant differences in basic demographic data and comorbidities between both waves. (Table  1) However, during the second waves, patients had less fever before admission and at ICU admission (47 vs 73.2% before admission, p=0.007 and 31.9 vs 58.5% at admission, p=0.005). The rate of patients with invasive mechanical ventilation and pO2/FiO2 ratio < 150 mmHg at admission (WHO Clinical Progression Scale) was higher during wave 2 (14.6 vs 21.7% p =0.03). We did not nd difference concerning laboratory results and comparison of percentage of lung parenchyma affected was not possible because of the proportion of missing data during the rst wave (44.7%).
Analysis of adjunct measures showed differences between both waves with much more days of continuous neuromuscular blockade (7.51 vs 6.89 days, p=0.029), more use of corticosteroids (100 vs 20%, p<0.001), and nitric oxide use (20 vs 2.4%, p=0.015) during wave 2. The second wave was also associated with an increased rate of Ventilator Associated Pneumonia (VAP) (57.4 vs 26.8%, p=0.001).
We did not nd signi cant difference for outcomes except for the rate of Aero-medical evacuation that was signi cantly higher during wave 2 (58.3 vs7.7%, p<0.001). (Table 2)   Table 1 Demographic, clinical, and ventilatory support characteristics at admission of patients according to the wave.   (Table 3) The AUROC for LR model was 0.797, and accuracy was 82%. (Table 4) Table 3 Multivariate Analysis of Factors Associated with Wave 2

Contribution of machine learning model in determination of waves characteristics
Variable importance plot based on Random Forest model showed that absence of fever at admission and VAP were the most important variables to classify patients between wave 1 and wave 2. (Figure 3) The AUROC for RF model was 0.678. (Table 4)   Table 4 Models, prediction performance and parameters.

Discussion
To our knowledge, this is the rst study to report changes in the characteristics of critically ill patients with acute respiratory failure due to Covid-19 after the emergence of the VOC 20H/501Y.V2. Demographic characteristics were similar between both waves, and our multivariate analysis associated with a machine learning method found that patients of wave 2 had less fever and had a lower illness severity score (SAPS II) at admission, but they were also more hypoxemic and more mechanically ventilated at 24 hours after admission and nally developed more VAP during their ICU stay.
Experience gained during the rst wave and speci c therapies have been associated to a better outcome among critically ill Covid-19 patients admitted during second waves across the world. [20][21][22] Among these therapies, the introduction of corticosteroids have been associated to a lower proportion of patients requiring invasive mechanical ventilation and to reduced mortality [25].
Interestingly, despite implementation of a protocol for corticosteroids in our ICU, we did not found difference of mortality between both waves.
It has been suggested that VOC 20H/501Y.V2 could be associated to a higher hospital morbidity and mortality rate [26,27], therefore in our study the expected gain in survival brought by improvements in critical care management could have been mitigated by increased severity of the new lineage. However, changes in morbidity and mortality rates of patients with SARS-CoV-2 infection should be interpreted with caution. Independently of virus virulence, Covid-19 outcomes can largely be affected by the epidemiological context i.e. population structure, climate, and social practice for example.
[28] The observational nature of previous studies implies possible unmeasured confounders factors. Moreover, hospital mortality and ICU mortality can differ largely because of very different fatality rates. In our study, higher rate of VAP, increased duration of invasive mechanical ventilation and of ICU length of stay could have directly contributed to mitigate outcome improvement.
We observed that despite similar SOFA score and slightly decreased SAPS II at admission, patients in wave 2 needed more invasive mechanical ventilation (aOR 3.49, 95% CI 0.98 -12.51, p=0.055) and had a lower pO2/FiO2 ratio at 24 hours (aOR 0.99, 95 % CI 0.98-0.99, p=0.0309). These results are in line with a recent study of Carbonell et al, comparing mortality in ICU between waves in a multicenter retrospective cohort. Authors found a signi cant lower illness severity at admission but also a trend for an increased severity of respiratory failure at admission (lower pO2/FiO2 ratio) during wave 2 and 3 in Europe when higher incidence of cases led to ICU overload. [29] During wave 2, when VOC 20H/501Y.V2 was the dominant lineage, Mayotte Island experienced a rapid and intense deterioration of the epidemiological situation. At the epidemic peak, reached in the rst weeks of February, Incidence Rate (IR) was 851/100,000, (i.e., 2,378 new con rmed cases in the week) and Positive Rate (PR) was 28%.
[9] The unique hospital of Mayotte was rapidly overloaded with a peak of new admissions reached in W6-2021 (225 patients hospitalized including 30 to ICU). For comparison during the peak of wave 1, 54 patients were hospitalized including 8 in ICU in a week [10]. This epidemiological situation could be explained by an increased transmissibility of VOC 20H/501Y.V2 compared to wild lineage [4].
We observed that delay between rst signs and hospital admission was increased during wave 2, 7 days (4-9) vs 6 days (3-8). As previously suggested, in an overwhelmed health system, individuals might avoid seeking care until later stage of disease [27]. Likewise, recent studies have reported association between hospital and ICU load and increased mortality. [30,31] Therefore, we hypothesize that increased transmissibility of VOC 20H/501Y.V2 during wave 2 led to hospital and ICU capacity overload and thus to admission of patients with more severe respiratory failure.
An important nding of the study was that, using the same de nition of VAP between both cohorts, we found a much higher incidence of VAP during the second wave compared to the rst wave (57.4% vs 26.8%, p=0.001). VAP was independently associated with wave 2 (aOR 4.64, 95 CI 1.54-13.93, p =0.0063) after adjusting for invasive ventilation and pO2/FiO2 ratio at 24 hours. In a recent multicentric cohort comparing mortality between three waves in critically ill patients, Carbonell et al, also found an increased rate of VAP during second and third waves in Europe. [29]. The large use of immunosuppressive agents (corticosteroids) during wave 2 could explain this nding. However, Ritter et al, showed in an observational study, after adjusting for competing risks, that corticosteroids seemed to have no impact on the likelihood of developing VAP [32].
The higher rate of VAP during wave 2 could also be explained by an increased rate of invasive mechanical ventilation during the rst 24 hours and with an increased duration of mechanical ventilation.
Finally, it is possible that VOC 20H/501Y.V2 itself played a role immunologically in the propensity for VAP. Indeed, it has been reported that critically-ill patients with Covid-19 patients suffered from a considerable burden of immunoparesis, due to impaired immune cell function [33,34]. Since the beginning of the surge, the incidence of VAP is found higher in patients with SARS-CoV-2 infection, as compared to patients with in uenza pneumonia, or no viral infection. [35,36] However, to date, speci c effect on immune response of variants in critically-ill patients remains non-investigated.
The rate of comorbidities in our population was in line with previous studies except for diabetes, more frequent than previously reported [21,37]. Mayotte is the poorest and most densely populated territory in France with poor socio-economic and health conditions and prevalence of comorbidities such as diabetes is high [38]. Interestingly, although second waves across the world have been frequently associated to younger patients with fewer comorbidities [20] we did not observe difference in demographic data between both waves. It has been proposed that the most vulnerable patients as older people and those with comorbidities were likely to die during rst waves [39]. However, wave 1 in Mayotte was much less intense than in other regions of the world.
Therefore, speci cities of the epidemic in Mayotte could explain the absence of change in pro le of Covid-19 critically-ill patients between both waves.
Our study has some important strengths. First, to our knowledge, we describe for the rst-time clinical characteristics and outcomes of critically ill patients with VOC 20H/501Y.V2. Second, we performed a detailed report of physiological, clinical features, and ventilatory management using the recent recommended WHO Clinical Progression Scale that has been developed to facilitate data pooling across cohort studies and clinical trials. [17] Third, we found similar typology of the patients at admission during both waves making relevant the comparison of outcomes. Fourth, the proportion of missing data was very low. Fifth, we performed the comparison between both waves in the only ICU of the island avoiding hospital variability previously described for Covid-19 patients [11]. Lastly, addition of a machine learning method to the logistic regression allowed to improve characterization of variables associated to wave 2. Contrary to other machine learning classi cation tools, the logic of RF algorithm is understandable for clinicians, moreover RF model can provide importance ranking of the variables.
We acknowledge several limitations to our study. First, we performed a retrospective analysis with risk of mis-classi cation bias. Second, our study took place Mayotte Island, potentially limiting generalizability to other hospitals. Indeed, this population is known to have higher prevalence of cardiometabolic comorbidities and socioeconomic vulnerabilities. Studies among more ethnically and geographically diverse cohorts are needed to con rm our hypothesis concerning VOC 20H/501Y.V2. Third, we performed a description of changes in the characteristics of critically ill patients with Covid-19 in the rst and second wave as a proxy for dominant lineage and we have individual-level data on lineage only for 23 critically-ill patients. However, during two days in the rst half of February, all usable SARS-Cov-2 positive samples were screened and more than 80% of them (150/172) pointed to 20H/501Y.V2 variant. Lastly, we could not assess statistically the association of the ICU overload and outcomes.

Conclusion
Our data described emergence of SARS-Cov-2 VOC 20H/501Y.V2 in critically ill patients. It was associated with an increased severity of respiratory failure on admission and more risk for VAP without an increase of mortality. We hypothesized that the expected gain in survival brought by recent improvements in critical care management could have been mitigated by increased transmissibility of the new lineage leading to more severe patients on admission in ICU. The immunological role of VOC 20H/501Y.V2 in the propensity for VAP requires further investigations. Studies with geographically diverse cohorts and complete individual lineage data are needed to con rm our ndings about this variant.

Declarations
Ethical Approval and Consent to participate All patients or close relatives were informed that their data were included in the study when possible.

Consent for publication
Not applicable.
Availability of supporting data Epidemic curve of laboratory con rmed severe acute respiratory syndrome coronavirus 2 during wave 1 and 2.

Figure 3
Variable importance plot to predict waves based on Random Forest model.