The repurposed use of anesthesia machines to ventilate critically ill patients with Coronavirus Disease 2019 (COVID-19)

Background The surge of critically ill patients due to the coronavirus disease-2019 (COVID-19) overwhelmed critical care capacity in areas of northern Italy. Anesthesia machines have been used as alternatives to traditional ICU mechanical ventilators. However, the outcomes for patients with COVID-19 respiratory failure cared for with Anesthesia Machines is currently unknow. We hypothesized that COVID-19 patients receiving care with Anesthesia Machines would have worse outcomes compared to standard practice. Methods We designed a retrospective study of patients admitted with a confirmed COVID-19 diagnosis at a large tertiary urban hospital in northern Italy. Two care units were included: a 27-bed standard ICU and a 15-bed temporary unit emergently opened in an operating room setting. Intubated patients assigned to Anesthesia Machines (AM group) were compared to a control cohort treated with standard mechanical ventilators (ICU-VENT group). Outcomes were assessed at 60-day follow-up. A multivariable Cox regression analysis of risk factors between survivors and non-survivors was conducted to determine the adjusted risk of death for patients assigned to AM group. Results Complete daily data from 89 mechanically ventilated patients consecutively admitted to the two units were analyzed. Seventeen patients were included in the AM group, whereas 72 were in the ICU-VENT group. Disease severity and intensity of treatment were comparable between the two groups. The 60-day mortality was significantly higher in the AM group compared to the ICU-vent group (12/17 vs. 27/72, 70.6% vs. 37.5%, respectively, p = 0.016). Allocation to AM group was associated with a significantly increased risk of death after adjusting for covariates (HR 4.05, 95% CI: 1.75–9.33, p = 0.001). Several incidents and complications were reported with Anesthesia Machine care, raising safety concerns. Conclusions Our results support the hypothesis that care associated with the use of Anesthesia Machines is inadequate to provide long-term critical care to patients with COVID-19. Added safety risks must be considered if no other option is available to treat severely ill patients during the ongoing pandemic. Clinical Trial Number Not applicable

A major priority amid the emergency response to surges of coronavirus disease-2019 (COVID- 19) cases has been to increase hospital capacity, particularly in terms of ICU bed availability. 1 In many hospitals, this undertaking has included converting operating rooms (OR) and post-anesthesia care units (PACU) into temporary ICUs. 1,2 Using anesthesia machines in addition to standard ICU ventilators signi cantly increases a hospital's capacity to provide critical ventilatory support. Several authors and societies have advocated using Anesthesia Machines in COVID-19 patients at institutions faced with resource limitations. 3,4,5 Critical care ventilators are designed to function as mostly unattended devices. Alarms are usually integrated with an overhead monitoring system and trigger personnel from a distance. Through sophisticated tools and software, different ventilatory modes can be applied in a wide array of critical respiratory conditions. Inspired gases are usually actively humidi ed, and exhaled breath is dispersed in the room air after ltration with a bacterial lter. 6 The ventilator apparatus attached to an anesthesia machine is designed to be closely attended by trained professionals in the OR. Anesthesia Machines usually provide ventilation only while the patient is unconscious and paralyzed for surgery, with a limited range of available ventilatory settings and monitoring features. Anesthesia Machine workstations can be used to deliver inhaled anesthetics through dedicated vaporizers. Dry compressed gases are passively humidi ed through a heat and moisture exchanger, usually with ltration properties. A unique feature of Anesthesia Machines is the ability to regulate an inlet of fresh gas ow, altering the amount of rebreathed exhaled gas via a scavenger system.
While the use of lters and a closed system might be attractive options during the pandemic to limit viral contamination of the room and to spare medical gases, long-term use of Anesthesia Machines could also pose several complications. 3,7 Overall, Anesthesia Machines can provide life-sustaining mechanical ventilation, but they were not originally designed to support critically ill patients for prolonged times. 8 At the start of the 2020 pandemic, a registry was formed to understand patterns and trends in the critical care being delivered to patients with COVID-19 requiring mechanical ventilation. In this retrospective observational study, we investigated how the care of patients who received Anesthesia Machines versus the care of patients who received ICU ventilators impacted mortality. We hypothesized that 60-day survival would be reduced in patients cared for with Anesthesia Machines compared to care that involved standard ICU ventilators.

Data Source
The COVID-ICU multicenter registry is an international data repository that started on March 1 st , 2020 and is currently ongoing. It includes de-identi ed daily data relative to critically ill patients with con rmed COVID-19 admitted to the ICU. The anonymized data collection strategy utilizes a secure cloud-based online platform (Studytrax, Macon, GA). 9 The present study received approval by the Institutional Review Board of the coordinating institution (Massachusetts General Hospital, Boston, MA, USA, Protocol #2020P000760) with a waiver of informed consent. Patients included in this study were all admitted to Niguarda Hospital, Milan, Italy (Approval No.183-15042020).

Identi cation of the Study Cohort
In the rst week of March, due to an overwhelming need for ICU beds, Niguarda Hospital (Milan, Italy) converted a large postoperative 27-bed ICU and an OR in two COVID-19 speci c ICUs. The former unit was a 27-bed standard ICU, fully equipped with state-of-the-art ICU ventilators and was used entirely for COVID-19 from March 12 th to April 15 th , 2020. The second unit was emergently opened (from March 5 th to May 15 th , 2020) in an adjacent OR area. This OR ICU consisted of a 15-bed temporary ICU, including ve beds from a PACU space and 10 beds placed in ve separate ORs (two beds per single OR). This provisional unit was equipped with seven Anesthesia Machines, alongside eight standard ICU ventilators.
These two units were equipped with the same medical, nursing, and support staff who usually worked in the standard postoperative ICU. Physicians and nurses within both the standard ICU and the OR ICU were highly trained individuals in the eld of anesthesia and critical care. Particularly, the OR ICU was equipped with nurses skilled in the use of standard ICU ventilators and Anesthesia Machines. To further improve the Anesthesia Machines management in the OR-ICU, we planned weekly training sessions on their use by an expert anesthesiologist and two expert nurses of anesthesia. Moreover, every morning an expert nurse of anesthesia and an Anesthesiologist revised the Anesthesia Machines in function ( lters, circuits and ventilator) in order to improve patient's safety.
During the study period, patients were admitted to either unit and assigned to an ICU ventilator, or an Anesthesia Machine, based on bed availability. Patients assigned to Anesthesia Machines were never switched from ICU ventilators and likewise switching did not occur for patients assigned to ICU ventilators.

Study Design
We designed a retrospective study comparing intubated COVID-19 patients whose care involved the use of Anesthesia Machine (AM group) to a cohort receiving care involving the use of standard ICU ventilators (ICU-VENT group) admitted during the same period. The "Strengthening the Reporting of Observational studies in Epidemiology" (STROBE) guidelines were followed. 10 Patients 18 years-or-older admitted to either the standard ICU or the OR ICU were included. A con rmed diagnosis of SARS-CoV-2 infection was required. Patients receiving less than 48-hours of mechanical ventilation were excluded. The primary endpoint was to assess a difference in 60-day survival between the two groups. Secondary endpoints included evaluation of differences between AM and ICU-VENT groups in terms of ICU or hospital length of stay, ventilator-free days, ICU and hospital free days, need for ECMO, need for tracheostomy, incidence of barotrauma (de ned as spontaneous pneumothorax or pneumomediastinum) and need for emergency endotracheal tube exchange secondary to airway occlusion. All outcomes and variables included in this analysis were abstracted from the medical record using clinically documented values.

ICU care
Patients were treated according to internationally recognized standards of care. 11 Details regarding mechanical ventilation settings, the use of respiratory failure rescue strategies, and COVID-19 speci c therapies are reported in the online supplement (See "ICU care and COVID-19 speci c therapies" in Supplemental Digital Content 1).

Anesthesia Machine Setup
The Primus TM workstation (Dräger, Lubeck, Germany) was the only Anesthesia Machine used. A heat and moisture exchanger with a lter (HMEF) and lter exclusive to the airway were used for every patient. The HMEF was placed at the endotracheal tube mouthpiece (DAR™ Adult-Pediatric Electrostatic Filter HME, Small, Medtronic, Minneapolis). The airway lter without HME (DAR™ Electrostatic Filter, Large, Medtronic, Minneapolis) was placed at the end of the expiratory limb of the circuit. 8 Both devices were routinely changed every 24 hours or if there were signs of obstruction. For every Anesthesia Machine, a successful startup test was performed at baseline and at least every 72-hours. A rebreathing circuit was in place, with a soda-lime scavenger to adsorb carbon dioxide. During approximately the rst month of the study, the total fresh gas ow rate was maintained at 50-60% of the patient's minute ventilation, with the intent to spare sevo urane and oxygen. Following the publication of consensus recommendations, the fresh gas ow was increased to around 80% of minute ventilation in patients receiving halogenates, and to over 100% in patients without inhaled anesthetic. 8,12

ICU Ventilators
The SERVO-i Mechanical Ventilator (Getinge, Gothenburg, Sweden) is the primary ICU ventilator in use at the study institution. A similar ltering strategy to that used in the AM group was initially implemented, with an HMEF and an exclusive airway lter placed at the endotracheal tube and expiratory inlet on the ventilator, respectively. However, as soon as adequate supplies of personal protective equipment and proper isolation logistics could be guaranteed, active humidi cation was preferred to HMEF for most patients in the ICU-VENT group.

Statistical Analysis
The statistical plan was written after the data were accessed and no statistical power calculation was performed prior to the start of the study. Instead, the sample size was based on all available data from the time period in which both units were functioning as COVID ICUs.
Baseline characteristics are presented as median and interquartile range for continuous covariates, and proportions for categorical variables. The Mann-Whitney U test or Fisher's exact test were used for differences between the two groups (AM group vs. ICU-VENT group), and survivors vs. non-survivors, as appropriate. Kaplan-Meier survival analysis was used to compare 60-day survival between the two cohorts. Signi cance was assessed using a Log-Rank test. There were no censored survival data in this study.
In order to evaluate the impact of receiving care with an Anesthesia Machine on patient survival, we performed an adjusted and multivariable analysis. After con rming the proportional hazards assumption was met (p=0.86), Cox Regression models were performed, in which we assessed the relationship with 60-day mortality. In the adjusted model only variables with p< 0.10 on the univariate screen were considered candidate variables for inclusion. Using this list of variables, backwards selection was then performed (considering p < 0.1 for exclusion at each step) in order to elucidate a nal model. Hazards ratios (HR) and their associated 95% con dence intervals (CI) from the nal model are presented. SPSS software v26 (Microsoft Corporation -Redmond, USA) and SAS 9.4 (SAS Institute Inc., Cary, NC) was used for data analysis. Two-sided p-values < 0.05 were considered statistically signi cant.
During the peer review process a sensitivity analysis was performed in order to address the possibility of biased estimates of the predictors by using a data-driven variable selection process. In this analysis 100 bootstrapped samples were obtained from the original dataset with replacement, and backward selection was similarly performed as above for each of the bootstrapped samples. Variables in the nal sensitivity model were based on frequency and included the four most common variables in order to maintain model parsimony. Full results of the model selection process are detailed in Supplementary Digital Content 2 Table 4.
Rates of missing data are reported in Supplementary Digital Content 2 Table 3. No imputation was performed for missing data.

Results
From March 5 to May 15, 2020, a total of 156 critically ill COVID-19 patients have been treated at Niguarda Hospital. Among these, 93 patients were admitted to the two study units: 52 to the standard ICU and 41 to the OR ICU. Four patients were excluded from the analysis (three patients did not need invasive mechanical ventilation; one patient was transferred within 24 hours). Eighty-nine patients were included in the study. Most of the patients in both groups (AM group and ICU-VENT group) were admitted to the ICUs during the month of March: 16 (94%) for the AM group and 58 (81%) for the ICU-VENT group (p=0.28).
For 17 patients (19%), an Anesthesia Machine was used (AM group), while for 72 (81%), an ICU ventilator was available (ICU-VENT group). Baseline characteristics are reported in Table 1. The only difference of note between groups with respect to sta ng was that patients in the OR ICU had a slightly higher nurseto-patient ratio. (See "ICU Sta ng" in Supplemental Digital Content 1).
Patients in the AM group showed similar COVID-19 severity at admission compared to the ICU-VENT group ( Table 2).
The degree of respiratory impairment and mechanical ventilation requirement were similar between the two cohorts. Although not statistically signi cant, the AM group showed slight lower levels of Mean Arterial Pressure at ICU admission (71 [67-76] mmHg vs. 78 [70-86] mmHg; AM vs. ICU-VENT group, respectively; p=0.050). No signi cant differences in mean arterial pressure were observed after day one between the ICU-VENT and AM groups (See Figure 1 in Supplemental Digital Content 2).
The intensity of treatment and the use of rescue therapies were comparable between the two groups ( Table 2 and 3).
Patients in the AM group were more frequently treated, for one day or more, with inhalation anesthetics (0% vs. 82%, p < 0.001). In contrast, continuous intravenous (IV) sedation was used more often in the ICU-VENT group (84.7% vs. 17.7 %, p < 0.001).
Critical COVID-19 patients in the AM group died more frequently compared to those in the ICU-VENT group (Table 3 and Figure 1).
Comparing the two ICUs we report a 60-day mortality of 51.2% in the OR ICU and of 37.5% in the conventional-ICU (p=0,207). Moreover, the 60-day mortality among patients receiving ICU ventilator care (regardless of ICU setting) was identical (37,5%). Care of patients that involved the use of Anesthesia Machines was independently associated with an increased risk of death, adjusting for potential confounding factors in a multivariable regression model (Table 4).
In a univariate analysis (see table 2 in Supplemental Digital Content 2), 60-day mortality was signi cantly higher in patients who at baseline had the following characteristics: were cared for with Anesthesia Machines, were older, had a higher body mass index, showed higher lactate, higher driving pressure, lower pH, lower hemoglobin, higher bilirubin, higher creatinine level, or those with a history of hypertension, diabetes, COPD or hypercholesterolemia. A higher mean arterial pressure was associated with a limited, although signi cant, protective effect (HR 0.96 per mmHg, 95% CI 0.93-0.99, p=0.008). Of note, we performed an analysis of blood pressure trends over the course of several ICU days (days 1, 2, and 7) among patients receiving volatile anesthetic and found no evidence suggesting lower blood pressures in this subgroup (see Figure 1 in Supplemental Digital Content 2).
In a sensitivity model in which model covariates were based on inclusion frequency in a bootstrapped sample, the nal model included the covariates for creatinine, hypertension and bilirubin in addition to the use of anesthesia machines. In this adjusted model the association between the use of anesthesia machines and 60-day mortality remained robust (HR 3.46, 95% CI 1.57-7.63, p =0.002; Supplementary Digital Content 2 Tables 4 and 5).
The use of Anesthesia Machines for prolonged periods might be associated with the risk of technical failure or airway occlusion (Table 5) During the study period, two cases of sudden Anesthesia Machine failure were observed that required emergent replacement of the workstation. No technical issues were experienced in the ICU-VENT group. Several episodes of mucus plugging of the endotracheal tube occurred in the AM group. In most instances, the obstruction resolved with vigorous suction, or beroptic bronchoscopy. Emergency tube exchange was needed in 3/17 cases (18%), compared to 1/72 (1%) in the ICU-VENT group (p= 0.021). One patient in the AM group died due to sudden complete airway obstruction following the accumulation of secretions at the level of the carina, which could not be effectively and timely relieved.

Discussion
Italy was the rst country outside of China to suffer a major outbreak of SARS-CoV-2 infection. 1,13,14 The registry population provided unique advantages for studying the impact of Anesthesia Machines use in the care of COVID-19 patients compared to standard ICU-VENT use. Our results indicate that during the emergency response to the initial peak of COVID-19, the care of critically ill patients with repurposed Anesthesia Machines was associated with an increased rate of complications and mortality.
The population we described in this study had similar characteristics to that of other published case series. 13,15-17 The 70% mortality for patients in the AM group however is remarkably high. The overall 60-day mortality though of the analyzed cohort (43.8%) is similar to what has recently been reported by the COVID-19 Lombardy ICU Network in the largest currently available outcome study of Italian cases (48.7%). 14 While greatly varying across the multiple available reports, ICU mortality from COVID-19 is primarily driven by the development of ARDS, with 50% mortality among patients with COVID-19 ARDS generally considered an accepted estimate. 18 Our overall mortality ndings are, therefore, in agreement with the currently available literature, supporting the quality and external validity of our data.
Given the dramatic increase in mortality that we observed in patients whose care involved the use of Anesthesia Machine, we aimed to quantify factors associated with their care contributing to lethality. We wish to be explicit in stating that our registry analysis does not allow us to conclude that the Anesthesia Machine ventilator itself is the exclusive culprit. Instead, we believe that our study demonstrates that the clinical care scenarios associated with using Anesthesia Machines are linked to increased mortality. There are several considerations regarding possible changes in clinical care and unique Anesthesia Machine-related challenges that should be discussed.
First, the correct setup of audible alarms on Anesthesia Machines and the ability to respond with a prompt corrective action might prove challenging for any operator, particularly when clinicians are trying to limit proximity to patients and must don and doff personal protective equipment. 5,8 We believe that several non-quanti able factors related to Anesthesia Machines not being a normal standard of ICU care -even for clinicians comfortable with their operation in the OR setting -could have led to a higher degree of mortality.
Second, clogging of HMEF and lters due to excess moisture or secretion burden is a major problem in patients recieving prolonged ventilation on an AM without a heat source and active humidi cation, particularly when low fresh gas ow is used 19,20 . HMEs are a passive form of humidi cation. The device stores heat and moisture from the patient's own exhaled gas which is released during inhalation of fresh gas, which would otherwise be dry and at ambient temperature 21,22 . In our study, frequent HME replacements were required in the AM group to prevent occlusion. The use of a higher fresh gas ow rates reduced this complication and is currently recommended by the APSF/ASA guidelines. 8 Third, COVID-19 patients often show tenacious and abundant tracheal secretions, whose inspissation might lead to an even higher risk of ETT occlusion. Given the lower temperature in the Anestehsia Machine circuit (without a dedicated heating system or active humidi cation), we believe Anesthesia Machines could make this risk even higher. In COVID-19 patients ventilated with repurposed Anesthesia Machines, Panchami, KR, et al. reported a 29% incidence of critical airway obstruction requiring emergency Fiber Optic Bronchoscopy or tube exchange. 23 In our study, subtotal tube obstruction due to mucus accumulation was also a common occurrence with Anesthesia Machines. In one case, an exceptionally large mucus plug caused a sudden complete airway obstruction at the level of the carina leading to hypoxia and cardiac arrest. It reasonable to presume that inadequate heating and humidi cation could also have led to increased secretion burden and decreased secretion clearance in more distal airways.
Four, we estimate that each patient on an Anesthesia Machine had to be disconnected roughly twice per day on average, to either change lters, or perform startup self-tests. On these occasions, we had no other choice than to ventilate the patient with a manual resuscitator. Disconnections from the mechanical ventilator in ARDS might result in loss of PEEP and lung collapse and should be avoided at all costs. 24,25 The use of manual bag ventilation might lead to hyperventilation with excessive rate, pressure, and tidal volume, all critical determinants of VILI. 26,27 Finally, one must theoretically consider that ventilator-induced lung injury of increased severity is also a potential explanation for the decreased survival in patients in the AM group. The accumulation of excess condensation in the circuit of AMs often hindered the accuracy of ow sensors and increased resistance, leading to inconsistently delivered tidal volumes. The importance of an accurately set tidal volume within a lung-protective ventilatory strategy is a mainstay of ARDS treatment. 28,29 Our registry demonstrated a signi cantly increased of volatile anesthetic in patients receiving Anesthesia Machines. While there is limited literature suggesting bene ts of halogenate use in ARDS, we were more concerned that there could be a potential connection between lower blood pressures and inhaled anesthetic use. However, an analysis of blood pressures amongst patients receiving inhaled anesthetics during the rst week of ventilation did not reveal concerning trends.
Although our initial experience using Anesthesia Machines in the COVID-19 pandemic saw an increased incidence of mortality, we do hope that there were learned lessons that we can share with other clinicians currently experiencing COVID-19 surges. In the event that Anesthesia Machines are required to keep hospital capacity a oat, a summary of the issues we encountered using Anesthesia Machines and related proposed solutions is provided in Table 5.

Limitations
Our study presents several limitations. Our analysis sought to limit confounding factors and to study the use of Anesthesia Machines as the only difference between groups. However, our sample size is relatively small. It is possible because of the small group size there are differences that persist between groups that were not identi ed as statistically signi cant in our analysis, but may be clinically meaningful. As in any retrospective study there is the possibility of residual confounding or bias in our interpretation. It should be noted however that patients who received AMs did so because of bed availability, not patient acuity or other factors, thus these are not likely biasing our results. Detailed data on bed assignments are not available. Our study took place during the very early stages of the pandemic. At the time, we witnessed a shifting emphasis on using certain drugs (e.g., antivirals, hydroxychloroquine, or immunomodulators). Their use in our cohort has been fragmented and based on institutional indications and local availability, rather than supporting evidence. Globally, a better understanding of the disease became available in the summer months of 2020. Our group gained growing experience in the management of COVID-19 patients as time went by, raising the possibility that the results were in uenced by secular trends. Finally given that the nature of the study is an analysis of a registry, we are not able to draw causal interpretations to our results. Rather, our results are suggestive of a pattern increased mortality associated with the clinical care of COVID-19 patients whose management involved Anesthesia Machines.

Conclusions
This is the rst analysis of a relatively large registry investigating mortality in patients who received care with Anesthesia Machines versus standard ICU ventilators for COVID-19 respiratory failure during a peak surge. There is a widespread dramatic change in the care pro le involved in managing a patient on prolonged ventilation with an Anesthesia Machines, including differences in humidi cation, volatile anesthetic use, ventilator disconnections, and centralized alarm systems. Our analysis demonstrated a signi cantly increased risk of death in those patients whose care involved the use of Anesthesia Machines during a pandemic surge.

Declarations
Ethics approval and consent to participate: The present study received approval by the Institutional Review Board of the coordinating institution (Massachusetts General Hospital, Boston, MA, USA, Protocol #2020P000760) with a waiver of informed consent. Patients included in this study were all admitted to Niguarda Hospital, Milan, Italy (Approval No.183-15042020).

Consent for publication: Not applicable
Availability of data and materials: The dataset used during the current study is available from the corresponding author on reasonable request.
LB receives salary support from K23 HL128882/NHLBI NIH as principal investigator for his work on hemolysis and nitric oxide. LB receives technologies and devices from iNO Therapeutics LLC, Praxair Inc., Masimo Corp. LB receives a grant from iNO Therapeutics LLC. The other authors declare no competing interests.
Funding: Support was provided solely from institutional and/or departmental sources.      · Maximize staff proximity to the workstation.
Condensed water accumulation in the circuit causing obstruction of HMEF or lters Reduced reliability of ow sensors · Use of high fresh gas ow (dryer gas mixture); · HME perpendicularly positioned above the endotracheal tube to reduce the back ow of excess moisture into the circuit; · Use of heated breathing circuits, condensers and water traps to limit water accumulation.
Endotracheal tube obstruction · Dedicated endotracheal tube cleaning devices.
Frequent disconnection due to lter change and machine selftests · Temporary use of a portable ventilator during disconnection to maintain protective ventilation and PEEP settings.
Limited functionality for the assessment of respiratory mechanics · Prioritize the use of newer AMs in more complicated patients considering the possibility to perform measurements of respiratory mechanics (e.g. end-inspiratory and end-expiratory pauses).