Adjuvants to Mechanical Ventilation for Acute Respiratory Failure

Transcription

1 Adjuvants to Mechanical Ventilation for Acute Respiratory Failure by Laveena Munshi A thesis submitted in conformity with the requirements for the degree of Masters of Science Institute of Health Policy, Management and Evaluation University of Toronto Copyright by Laveena Munshi 2016

2 Adjuvants to Mechanical Ventilation for Acute Respiratory Failure Abstract Laveena Munshi Masters of Science Institute of Health Policy, Management and Evaluation University of Toronto 2016 OBJECTIVES: To evaluate use of adjuvants to mechanical ventilation for acute respiratory failure (ARF; ) and the impact of landmark publications on adoption and de-adoption. METHODS: Adult patients with ARF who underwent mechanical ventilation were evaluated using a large US quality improvement database. Adjuvant use was identified using International Classification of Disease, 9 th Edition codes and billing data. RESULTS: Among 514,913 ARF patients, 11,567 (2.3%) were treated with adjuvants. There was an increase in recent years of extracorporeal membrane oxygenation and inhaled epoprostenol but no change in inhaled nitric oxide or continuous paralysis over time. Segmented regression analysis used to evaluate whether clinical practice was in accordance with evidence from landmark studies, did not reveal any change in use following publication with the exception of inhaled epoprostenol. CONCLUSIONS: The use of adjuvants is infrequent across our population. Published evidence to support current adoption/de-adoption appear to have had limited effect on practice. ii

3 Acknowledgments I would like to acknowledge the support and guidance I received from my thesis committee Drs. Gordon Rubenfeld (supervisor) Eddy Fan, Hannah Wunsch, Niall Ferguson and Therese Stukel. Their patience, wisdom and mentorship have had a great impact on the career path I hope to pursue. I would also like to acknowledge Dr. Hayley Gershengorn for her contribution in assisting with the database and for facilitating access to the data in order for me to perform the data analysis at Montefiore Medical Center. iii

4 Table of Contents Contents Acknowledgments... iii Table of Contents... iiv List of Tables... vi List of Figures... vii List of Appendices... viii Chapter 1 Adjuvants to Mechanical Ventilation for Acute Respiratory Failure Background Rationale, Evidence and Current Use of Adjuvant Therapy Acute Respiratory Failure Management Pharmacologic Adjuvants Non-Pharmacologic Adjuvants Factors Affecting Adoption and De-adoption Conclusions Adjuvant Therapy to Mechanical Ventilation Changes in Use and Factors Associated with Use Introduction Methods Study Population Outcomes Statistical Analysis.12. iv

5 2.2.4 Sensitivity Analysis Results Cohort Assembly Changes Over Time Factors Associated with Use Discussion Conclusions and Future Direction...20 References 44 Appendix..48 v

6 List of Tables Table 1: Reported Use for Common Adjuvants in Acute Respiratory Failure Table 2: ICD9-CM Inclusion and Exclusion Criteria Codes Table 3: Landmark Studies Evaluated Table 4: Baseline Demographic Characteristics Table 5: Adjuvant Frequencies Table 6: Patient and Hospital Characteristics According to Adjuvant Table 7: Sensitivity Analyses vi

7 List of Figures Figure 1. Adjuvants for Acute Respiratory Failure Figure 2: Factors that Drive Decision Making Figure 3: Cohort Creation Figure 4: Changes in Use of Adjuvant Therapy Figure 5: Impact of Landmark Publications on Changes in Use Over Time Figure 6: Changes in Use of Pulmonary Artery Catheter Figure 7: Factors Associated with Adjuvant Use vii

8 List of Appendices Table 1: Neuromuscular Blockading Agent Dose Determination viii

9 Adjuvants to Mechanical Ventilation for Acute Respiratory Failure 1 Background 1.1 Rationale, Evidence and Current Use of Adjuvant Therapy Acute Respiratory Failure Management Acute respiratory failure (ARF) is a common reason for admission to an intensive care unit (ICU) and the need for mechanical ventilation. Acute respiratory distress syndrome (ARDS) is a severe form of hypoxemic ARF, with bilateral infiltrates consistent with pulmonary edema on chest radiography that is not primarily due to a cardiogenic etiology (1). Mortality from ARF and ARDS is high and similar across both categories (30-40%) with most deaths resulting from multi-organ failure and sepsis (2). Limited pharmacologic therapy has proven effective in ARF and management is focused primarily on supportive care with mechanical ventilation (1, 3-5). Currently, lung protective pressure- and volume-limited ventilatory strategies aimed at mitigating ventilator-associated lung injury (VALI) have become the standard of care (5). Despite the use of lung protective ventilation, a number of patients may develop refractory hypoxemia and/or hypercapnia, and may not be able to achieve adequate gas exchange without using injurious levels of ventilatory support. An adjuvant for ARF is any intervention, in addition to or instead of mechanical ventilation, that is used to facilitate gas exchange or enhance compliance with lung protective ventilation (Figure 1). Pharmacologic adjuvants have been the focus of many studies for years and include diuretics, corticosteroids, neuromuscular blocking agents and inhaled pulmonary vasodilators. Nonpharmacologic agents include prone positioning, high frequency oscillatory ventilation and extracorporeal life support. These non-pharmacologic options have been the focus of many large trials in recent years (Table 1). A series of proposal algorithms for adjuvant application have been proposed over recent years (Figure 1). While most of the evidence surrounding adjuvant use has focused on patients with ARDS, recent international observational data has demonstrated 1

10 2 the application of adjuvants in patients with ARF even if formal criteria for ARDS is not met or in patients in whom ARDS is unrecognized (3) Pharmacologic Adjuvants Neuromuscular blocking agents may minimize ventilator associated lung injury by preventing large spontaneous tidal volumes, reducing ventilator dys-synchrony, and possibly decreasing the inflammatory response associated with ARF and ARDS (6). In addition, paralysis may stop any subclinical evidence of muscle activity, potentially improving oxygenation through minimization of oxygen consumption. A large retrospective observational study of data from , evaluated the frequency of use and impact of early administration of paralysis in patients with severe sepsis who were mechanically ventilated and found that 23% of patients received early paralysis (7). Early paralysis was associated with a lower mortality (32% vs. 38%) In a randomized controlled trial (RCT) in 2010, the continuous, early administration of cisatracurium in moderate-severe ARDS (PaO 2 /FiO 2 <150) was associated with an improvement in 90-day mortality (8). The trial found that the beneficial effect of cisatracurium was confined to patients presenting with a PaO 2 /FiO 2 < 120 suggesting that patients with severe ARDS might receive the greatest benefit from this intervention. Following this publication, only 15% of UK physicians report using neuromuscular blocking agents routinely (9). Most recently, in the Large Observational Study to Understand the Global Impact of Severe ARF (LUNG-SAFE) study, a prospective observational study of ARF and ARDS across 50 countries, paralysis was used in 22% of all ARDS patients and in 38% of the severe ARDS subgroup (3). Non-hydrostatic pulmonary edema is one of the hallmarks of ARDS and can complicate all causes of ARF. Additionally, excess fluid administered during the resuscitation phase of septic shock could contribute to the development of abdominal compartment syndrome, further restricting lung expansion. The role of a conservative fluid management strategy combined with diuretic administration during ARF management has been evaluated as a mechanism by which one could improve lung compliance and oxygenation and has been demonstrated to reduce duration of mechanical ventilation. The Fluid and Catheter Treatment Trial (FACTT) assessed a conservative fluid management strategy combined with diuretic administration as a mechanism to improve lung compliance and oxygenation, finding an increased number of ventilator free days (10). However, the complex algorithm, lack of mortality benefit and possible risk of

11 3 neurocognitive complications might impact widespread adoption (11). Evidence of adoption of diuretic administration has not been extensively evaluated following this publication. Variability in use of diuretics for ARDS has been reported across different centers. In a recent survey administered to intensivists in Australia and New Zealand evaluating diuretic use, 74% of intensivists indicated that they would administer loop diuretics for ARDS; however, approximately 20% reported that they would not (12). An observational study evaluating prescribing patterns across 150 ARDS patients demonstrated that loop diuretics were only actually prescribed in 39% of patients (13). Despite an increasing focus on the harms associated with a positive fluid balance in patients, there is a lack of evidence on how physicians implement this in practice. Non-invasive hemodynamic monitoring devices leading potentially to more precise evaluations of volume status or a focus on de-resuscitation in sepsis has perhaps led to changes in diuretic administration. More insight into current practice, particularly in light of the recent publication of FACTT lite which provides a simpler approach to a conservative fluid management strategy, is needed to highlight whether areas for improved compliance with diuresis exist (14). The early administration of corticosteroids as a mechanism to combat septic shock in order to down regulate the systemic inflammation as well as the late administration of steroids in the fibroproliferative phase of ARDS, has been evaluated in a number of clinical trials (15, 16). However, beyond steroid-responsive precipitants for ARDS (e.g. systemic vasculitis), a role for corticosteroids in routine care of ARDS patients has not been established. Promising preliminary evidence currently exists for its potential role with ARF in order to prevent ARDS in the setting of severe community acquired pneumonia (17). Despite these findings, current use of corticosteroid for ARDS patients remains variable but high. One study reports use in 41% of ARDS cases higher than the use of diuretics or neuromuscular blocking agents (13). In a 2010 questionnaire in German ICUs regarding ARDS management practices, corticosteroids were used in 52% of hospitals (18). In a separate study, 70% of UK physicians surveyed in 2012 endorsed the use of corticosteroids; however, only 6% used them routinely (9). The LUNG- SAFE study found that high dose corticosteroids (equivalent to >1mg/kg methylprednisone) was used in 17.9% of patients with ARDS and 23.3% in the subset with severe ARDS (3). Inhaled nitric oxide (NO) is a selective pulmonary vasodilator, which acts by preferentially diffusing to capillary beds of less inflamed alveoli leading to a reduction in ventilation/perfusion

12 4 mismatch and pulmonary vascular pressures in addition to having anti-inflammatory properties (19). The use of inhaled nitric oxide as a rescue therapy was characterized across 6 ARDS Network trials between (20). Of the patients who received rescue therapy, inhaled nitric oxide was the second most commonly employed agent during that time period (28% of patients receiving rescue therapies). The most recent meta-analyses of ARDS patients has since demonstrated no mortality benefit associated with NO use regardless of severity (21). Moreover, the use of NO was associated with an increase in the incidence of renal failure(22). In surveys, 29-44% of intensivists from the UK and Germany report administering NO in ARDS (9, 18). In the LUNG-SAFE study, the frequency of use of any type of inhaled vasodilator was found to be much lower: 7.7% in all ARDS and 13.0% in the severe ARDS subgroup (3). However, the impact of this more recent evidence from the 2014 meta-analysis on frequency of use of NO has not been described Non-Pharmacologic Agents Theoretically, prone positioning may prevent lung injury by recruiting non-dependent lung, improving respiratory mechanics and clearing pulmonary secretions. The creation of more homogeneous chest wall compliance, offloading the weight of the heart and minimizing the weight of the abdominal contents on the diaphragm are mechanisms by which prone positioning may enhance respiratory mechanics and lead to an increase in recruitable lung units (4, 23). Prior to 2013, studies applied prone positioning to patients with a range of severity of ARDS. While these trials consistently demonstrated an improvement in oxygenation with prone positioning, a reduction in mortality was only seen in post hoc subgroup analyses of the most severe ARDS cohorts (24). In 2010, 60% of German centers reported that they proned ARDS patients and 84% of surveyed UK intensivists in 2012 reported that they would employ prone positioning as a rescue strategy (9, 18). An RCT in 2013, which focused on patients with moderate/severe ARDS (PaO 2 /FiO 2 <150), demonstrated an absolute mortality risk reduction of 17% with prolonged intermittent prone positioning (25). Since this trial; however, the LUNG-SAFE study revealed relatively low rates of proning: 7.9% of all ARDS (and 16.3% of severe ARDS patients) (3). High frequency oscillatory ventilation (HFOV) is an open lung ventilatory strategy that attempts to recruit the lung using a high mean airway pressure and minimizes cyclic tidal reopening and closing. Although early randomized-controlled trials of HFOV in adults suggested the possibility

13 5 of benefit (26), two recent, high quality, large-scale trials failed to show any mortality benefit and one trial suggested possible harm when employed in moderate to severe ARDS (27, 28). One possible explanation of the lack of benefit was more hemodynamic instability in the HFOV arm, possibly attributable to a decrease in venous return, or impairment of right ventricular afterload with higher mean airway pressure. Occult barotrauma and an increase in sedative requirements are additional plausible mechanisms (27). Frequency of use as a rescue strategy ranged from 7%- 50% in earlier literature (9, 20). Most recently, much lower use has been reported from LUNG- SAFE (1.2% across all ARDS and 1.5% in the severe ARDS subgroup) (3). Venovenous extracorporeal membrane oxygenation (ECMO) is a form of partial cardiopulmonary bypass that acts as an oxygenating and ventilatory shunt and can allow a reduction in the intensity of invasive mechanical ventilation or complete lung rest. An RCT in 2009, the Conventional Ventilator Support vs. Extracorporeal Membrane Oxygenation for Severe Adult Respiratory Failure (CESAR) trial, evaluated the impact of transport to an ECMO capable center in patients with severe ARDS and demonstrated an improvement in disability-free survival (29). However, some unanswered questions included whether the improved outcome was due to ECMO itself or being managed at an expert center, as not all patients received ECMO. In addition, the lack of compliance with lung protective ventilation in the nonprotocolized control arm might have contributed to the difference in outcomes (29). Given some conflicting recent evidence regarding its benefit in severe ARDS (30), an international multicenter trial is currently underway to evaluate its use for severe ARDS patients (NCT ). The creation of modern extracorporeal circuitry technology in combination with the publication of the CESAR trial and the subsequent H 1 N 1 influenza outbreak has led to an exponential increase in use of ECMO; according to the Extracorporeal Life Support Organization: 117 cases of adult respiratory ECMO reported in 2004 to 1,497 cases reported in 2014 (31). In the LUNG- SAFE study, 6.6% of patients with severe ARDS across the 50 countries received ECMO (3). The adjuvants reviewed here are only a few of many potential pharmacologic (e.g. aspirin, statins) and non-pharmacologic (e.g. non-invasive ventilation) adjuvants that continue to be assessed.

14 Factors affecting adoption and de-adoption Despite many intervention studies with no difference in outcomes and given the heterogeneity of patients with ARF, decision making about the use of adjuvant therapy in specific subgroups of patients, is complex. For example, in advance of the positive proning trial described above, many clinicians advocated for the use of prone ventilation based on the compelling physiologic rationale, subgroup evidence of benefit for severe ARDS and limitations of the existing trials (24). Since statistically negative trials may not be able to prove lack of efficacy in specific subgroups, arguments can be offered about the use of treatments from these trials in different patient subsets or with different protocols than those studied. Information gained from subgroup analyses or further insight on adjuvant administration has led to new trials re-evaluating their use. However, despite this, trial results will continue to remain only one component of medical decision-making. Factors that impact adoption or de-adoption often extend beyond evidence. More importantly, when the evidence base is weak or inconsistent, as it is in much of medicine, factors other than evidence drive adoption and de-adoption (32-34). Furthermore, in the literature available, variability exists in the real world with respect to adoption/de-adoption and it does not always follow the evidence. For example, many of the severe ARDS adjuvants improve oxygenation. Despite the lack of association between oxygenation improvement and mortality (5, 22), physicians are likely to reach for these adjuvants for the reassurance provided by improving oxygenation. Other factors that may also drive adoption and de-adoption include experience with the treatment, cost, availability, perceived risk, and patient comfort (32) (Figure 2). Lung protective ventilation, which has consistently demonstrated a mortality benefit across a series of trials, has not yet achieved widespread adoption in critical care (5, 34, 35). De-adoption of tight glycemic control, has also been slow despite evidence of harm (33). However, this is not always the pattern. There are other examples in critical care of rapid de-adoption as was the story of the pulmonary arterial catheter following the observational study by Connors et al (36). Provocative editorials, associated press and possibly the increase in non-invasive hemodynamic monitoring potentially contributed to the rapid drop in use noted. Non evidence-based factors that impact adoption/de-adoption include benefits/risk, feasibility, applicability and physician preferences (32). When confidence in the evidence is lacking or if inconsistencies arise, these non-evidence

15 7 factors drive patterns of use to a greater degree. Furthermore, the role of evidence in rescue therapy becomes less clear and factors that drive adoption/de-adoption likely focus more on physician preferences Conclusions This section highlights adjuvants to standard mechanical ventilation for ARF and specifically ARDS patients, the limited evidence base for the use of many of these adjuvants, and the available data regarding how they are deployed in current practice. The current state of data surrounding use has predominantly been cross sectional data (LUNG-SAFE) or self-reported data on stated practice. Given the high costs in terms of equipment and personnel associated with the use of many of these adjuvants, more research surrounding trends in use, how practice changes in response to evidence, and factors that may influence adoption and de-adoption of these adjuvants are needed. Arguably, the adjuvants that have been subject to the greatest debate and scientific evaluation over the past decade include the use of ECMO, inhaled pulmonary vasodilators, continuous neuromuscular blockading agents, high frequency oscillatory ventilation and prone positioning, and therefore will be the focus of this analysis.

16 8 2 Adjuvants to Mechanical Ventilation: Changes in Use and Factors Associated with Use 2.1 Introduction Acute respiratory failure (ARF) is associated with high morbidity and mortality and optimal treatment strategies are incompletely defined. Interventions evaluated have predominantly focused on patients with acute respiratory distress syndrome (ARDS); however, observational data suggests they are used in patients with ARF even if the formal ARDS criteria are not met (3). While lung protective ventilation and prone positioning have robust evidence supporting their use, extracorporeal membrane oxygenation (ECMO), inhaled pulmonary vasodilators, continuous neuromuscular blockading agents (cnmba), and high frequency oscillatory ventilation (HFOV) are used despite inconclusive evidence of benefit or possible harm (7, 21, 22, 25, 27-29, 37-39). As debates on adjuvant use continue and ongoing trials attempt to establish effectiveness (NCT , NCT ),(40) clinicians face daily decisions on when to initiate these treatments. The LUNG-SAFE study provides insight into ARF incidence, ARDS under-recognition, and current management (3). However, LUNG-SAFE focused on academic medical centers and was not designed to evaluate the use of treatments over time. A better understanding of factors associated with adjuvant use across a broad range of hospitals, how their use is affected by landmark publications, and what drives adoption/de-adoption is important to inform daily clinical decisions and policies until further evidence becomes available. 2.2 Materials and Methods We performed a retrospective cohort study of adults ( 18 years) discharged from a United States (US) hospital between July 2008-June 2013 using the Premier Perspective Hospital Database (Charlotte, NC and Washington, DC) (41). Premier is a voluntary fee-supported database that was developed as a healthcare performance improvement alliance and includes over 500 hospitals. It represents a sample of structurally and geographically diverse hospitals and 8

17 9 approximately 15-20% of hospital discharges nationwide. In addition to containing discharge data, it contains complete billing information with date-stamped logs of all charges (40) Study Population and Cohort Creation After removing hospitals with missing billing data, we included any adult patient who received invasive mechanical ventilation and had a primary diagnosis of ARF defined by the presence of an International Classification of Disease, 9 th Edition, Clinical Modification (ICD9-CM) code for ARF or ARDS (518.81, , , 518.5, 799.1, ) (7, 42-44). Given the potential for low sensitivity using the ARDS codes alone as well as high reports of use of adjuvants across the ARF population (3, 44), we also included those identified as having a direct infectious or direct non-infectious cause of ARF using ICD-9-CM codes (Table 2) (1). To avoid capturing indications for adjuvant use outside of ARF, patients with a code for pulmonary hypertension, cardiac surgery or transplantation were excluded (Table 2). Patient s extubated alive within 24 hours of mechanical ventilation were also excluded. Research ethics approval was obtained from the institutional review boards at Sunnybrook Health Sciences Center and Albert Einstein School of Medicine Outcomes We identified each adjuvant of interest using the hospital charges and/or ICD9-CM codes. ECMO was identified using ICD9-CM codes (ECMO) and (percutaneous ECMO) (45). Pulmonary vasodilators including inhaled NO and inhaled epoprostenol (off label use) were identified using hospital charge terms. Continuous NMBA (atracurium, cisatracurium, doxacurium, mivacurium, pancuronium, rocuronium, or vecuronium) was defined by charges for at least two days at a dose over 24 hours that could achieve continuous paralysis (intermittent bolus or continuous infusion) (46). We selected this dose to replicate the use of NMBA for ARF as opposed to short-term use for procedures (Appendix Table 1). In sensitivity analyses, we evaluated 1 day of cnmba, continuous cisatracurium and any use of NMBA. We set out to capture HFOV using the hospital billing term for the circuit specific to the SensorMedics 3100A/3100B (Viasys Healthcare, Yorba Linda, California USA); but concerns of undercoding led us to exclude HFOV. Prone positioning was also not identifiable through this dataset. 9

18 10 Patient and Hospital Data We collected data on patient demographic, insurance status, Elixhauser comorbidities (47), and clinical variables (e.g. dialysis, vasopressors). Hospital data included geographic location, bed size, urban/rural location, and academic status Statistical Analysis Patient and hospital level characteristics were summarized across the entire cohort and for patients who received each adjuvant. Proportions were used for categorical and means (standard deviation) or medians (interquartile ranges) for continuous variables as appropriate. First, we evaluated rates of use of each adjuvant per 1000 persons per year by dividing the number of patients with adjuvants by the total number of patients with ARF that year (x1000). For each adjuvant, we used Poisson regression (48) to estimate the annual incidence rate ratio and 95% confidence interval over the study period (reference year 2008), adjusted for age and gender and clustered by hospital (49, 50). We hypothesized that ECMO and cnmba use would increase and that inhaled NO would be replaced with inhaled epoprostenol over time. Second, we evaluated the impact of landmark studies on rates of use over time using segmented regression analysis of an interrupted time series. A segmented regression analysis allows the evaluation, in statistical terms, of how much a publication affects monthly rates of use. Segments were defined by major landmark publications for each adjuvant (51) and the time series was the monthly rates of use. A sufficient number of time points before and after the publication is needed to conduct segmented regression analysis and generally 12 data points before and after is recommended and was present for each of these analyses (51). Segmented regression answers two different questions. First, if publication of the landmark study causes an immediate change in use, it will be detected as a change in the intercept of the fitted regression line. Second, if publication causes a change in use over time, the slope of the regression line will be different from that time point compared to prior to that time point. Landmark studies published over this time period were determined, a priori, by journal impact factor and evidence quality (Table 3): for (1) ECMO the Conventional Ventilator Support vs. Extracorporeal Membrane Oxygenation for Severe Adult Respiratory Failure (CESAR) trial published on October 2009 (29) 10

19 11 demonstrating transportation to a specialized center for ECMO consideration was associated with improved disability free-survival at 6 months; (2) inhaled pulmonary vasodilators the Cochrane Collaboration meta-analysis published on July, 2010 (52) showing NO had no statistically significant effect on mortality and appeared to increase renal injury risk; and (3) cnmba the randomized trial published September 2010 demonstrating a survival benefit from the early continuous cisatracurium administration in moderate-severe ARDS (8). We evaluated patient and hospital level factors, determined a priori, associated with each adjuvant using logistic regression. One of the fundamental assumptions of logistic regression, that each observation is independent, may not hold true in this cohort due to potential clustering at the hospital level. Given this, the analysis was fitted with generalized estimating equations to properly account for the violation of the independence assumption due to the correlation between outcomes within the cluster. Generalized estimating equations as opposed to multi-level modeling was chosen to account for clustering given our interest in adjusting for the average impact of hospitals on rates of use; while the specific relationship between detailed hospital variation and the outcomes seen was not the primary focus. Physician level data was not available in this dataset. Adjuvant specific analyses were conducted only in hospitals capable of using that adjuvant defined as those hospitals that used the adjuvant at least once over the study period. It was assumed that all hospitals could administer NMBAs. Statistical analyses were performed using Stata 13.0 (StataCorp LP, College Station, TX). Results were considered statistically significant at p< Sensitivity Analyses We conducted 5 sensitivity analyses were conducted to assess the robustness of the results. First, to address the potential for hospitals leaving/entering the database, we repeated the primary analysis restricted to hospitals contributing data in all study years. Second, we repeated the segmented regression analysis excluding data from the H 1 N 1 influenza epidemic year of 2009 to ensure that temporal trends were not confounded by this unusual event. Third, since severity of illness data are limited in the Premier Database, we restricted the analysis to patients who died assuming that this stratifies the population to those with a high severity of illness and confirmed 11

20 12 that the changes noted were similarly robust in this analysis (36, 53). Fourth, to further focus on a population more representative of ARDS, we limited the analysis to those patients who were intubated within the first two days of hospitalization and had the combined diagnostic codes for severe sepsis, pneumonia and respiratory failure (7). We did not restrict this analysis to ARDS codes alone given the high proportion of ARF patients who undergo adjuvant use in unrecognized ARDS (3). For our primary analysis, we modeled year as a categorical variable to evaluate changes over time; however, we also evaluated it as a continuous variable in order to evaluate a trend over this time period. In an additional analysis, we evaluated a tracer condition, the use of pulmonary artery catheters, whose temporal trends we anticipated would be decreasing to validate the observed trends (36, 54). 2.3 Results Cohort Assembly and Overview Over the study period, 514,913 patients across 543 hospitals met the definition of ARF (Table 4 and Figure 3). The majority of patients were from urban centers (89%, n = 458,272) with 44% (n=226,561) admitted to academic centers and 50% (n=257,457) to large hospitals ( 400 beds). Pneumonia was the most frequent cause of ARF; 73% (n=375,886) received invasive mechanical ventilation within the first 2 days of admission and 56% (n=288,351) received vasopressors. Hospital mortality was 33% (n=169,921). There were 12,146 adjuvants used among 11,567 patients (2.3% of patients), with most receiving only one adjuvant (95%, n=11,006) (Table 5). cnmbas were the most frequently used adjuvant (n=10,073 cases, 2%), followed by inhaled pulmonary vasodilators (n=1,878, 1% in capable hospitals, 0.4% entire cohort), and ECMO (n=195, 0.2% in capable hospitals, 0.04% entire cohort). When two adjuvants were used, cnmbas were used in 95% (n=474) and inhaled pulmonary vasodilators were used in 93%. (n=465). When all three adjuvants were used in combination, cnmbas were most frequently used first (median (IQR) first day 1 (1-2)) followed by inhaled pulmonary vasodilators (median (IQR) first day 3 (1-8)) and ECMO (median (IQR) first day 4 (1-10)). 12

21 Changes Over Time Extracorporeal Membrane Oxygenation One-hundred ninety-five patients (0.2%) received ECMO. Only 62 (11%) hospitals had at least one ECMO case over the study period. ECMO-capable hospitals were predominantly large (61%; 38/62), urban (97%; 60/62), academic (61%; 38/62) centers. The mean (SD) age of ECMO patients was 46 (±17). Hospital mortality was 54% (n=106) (Table 6). ECMO use increased significantly over the time period (Figure 4a); there was no difference in the rate of use before and after the publication of the CESAR trial (29) (Figure 5a). These results did not change after eliminating the 2009 data. Inhaled Pulmonary Vasodilators Inhaled pulmonary vasodilators were used in 1,878 patients (1% of patients in inhaled vasodilator-capable hospitals). Twenty-two percent of hospitals (119) had at least one case of inhaled pulmonary vasodilator use (inhaled NO or inhaled epoprostenol). These patients were predominantly admitted to larger ( 400 beds, 73%, n = 1,371), urban (97%, n=1,821), and academic (72%, n=1,352) centers. Their mean (SD) age was 55 (± 18) with 58% (n=1,087) receiving inhaled NO. Hospital mortality was 59% (n=1,108) (Table 6). Inhaled NO use did not change over the study time period (Figure 4b); there was no change in use after publication of the Cochrane review (Figure 5b) (52). The results did not change after eliminating the 2009 data. The use of inhaled epoprostenol increased over the time period (Figure 4c). However, following the Cochrane publication which focused on NO (52), the rate of growth in epoprostenol use diminished significantly (Figure 5c). The results did not change after eliminating the 2009 data. Continuous Neuromuscular Blockading Agents Continuous NMBAs ( 2 days) were used in 10,073 patients (2% of all patients in the cohort) with a mean (SD) age of 51 (± 17). The majority of patients were admitted to urban hospitals 13

22 14 (94%, n = 9,469) and only 50% (n=5,037) were admitted to academic hospitals. Cisatracurium was the most commonly used agent (52%, n = 5,238). The median (IQR) duration of continuous paralysis was 3 days (2-5) with a median first day of initiation being day 2 (1-4) of mechanical ventilation for all agents. Hospital mortality was 43% (n=4,331) (Table 6). There was an isolated increase in use of cnbma in 2009; however, following that year, there was no consistent increase in use over time (Figure 4d). Upon evaluating the impact of the cnmba trial, no change was noted in the yearly cohort and after the 2009 data was removed (Figure 5d) (8). These results were robust when restricted to 1 day of cnmba, any use of NMBA, and cisatracurium specifically (Table 7). Sensitivity Analyses The results were similar when limited to hospitals who contributed data for the entire study duration and amongst patients who died (Table 7). In our ARF cohort designed to replicate an ARDS population, the overall number of patients and adjuvants used was significantly reduced. Because of this, there was some variability compared to the above results, however, none of the rates of use of the adjuvants decreased overtime (Table 7). When year was modeled as a continuous variable, there was no time trend overall in ECMO use seen (Table 7); however, it is possible that this absence of a statistically significant linear time trend may be attributable to the absence of an ordered increase given the H 1 N 1 outlier. As expected, we observed a decrease in the rate of pulmonary artery catheter use (Figure 6 ) Factors Associated with Adjuvant Use A diagnosis of severe sepsis and vasopressor use were associated with higher likelihood of receiving each adjuvant (Figure 7). Patient factors consistently associated with a lower likelihood of receipt of each adjuvant included older age, non-private insurance status and undergoing surgery during admission (Figure 7). Mechanical ventilation within the first two days of hospitalization was associated with a higher likelihood of receiving ECMO, whereas obesity and dialysis were more likely to be associated 14

23 15 with inhaled NO and epoprostenol. An admission diagnosis of chronic renal failure was associated with a decrease in use of inhaled NO (Figure 7). Among ECMO-capable hospitals, patients were more likely to receive ECMO at larger (odds ratio [OR] 1.01; 95% confidence interval [CI] ) and non-academic hospitals (23 centers) (OR for academic hospital 0.39; 95% CI ) (Figure 7a). In inhaled pulmonary vasodilator-capable hospitals, one was more likely to receive epoprostenol at academic (OR 5.86; 95%CI ) but less likely in larger hospitals (OR 0.99; 95% CI ) (Figure 7d). No hospital specific characteristics were associated with NO use (Figure 7c). Finally, cnmbas were more likely administered in urban (OR 1.49; 95% CI ) and academic hospitals (OR 1.16; 95% CI ) (Figure 7b). 2.4 Discussion We investigated the use of four adjuvants for ARF and found variable rates of change in use over the study period. As hypothesized, ECMO and inhaled epoprostenol use increased. However, contrary to our hypothesis, rates of inhaled NO did not decrease and rates of cnmba use were unchanged. There was no immediate impact of landmark publications with the exception of inhaled epoprostenol. Patient- and hospital-level factors associated with use vary considerably by adjuvant. Previous studies have evaluated self-reported use of adjuvants (9, 18) and historic utilization patterns (20). The LUNG-SAFE study was a cross sectional study that characterized management of 2,377 ARDS patients (3). The higher use of adjuvants reported may be attributed, in part, to the LUNG-SAFE cohort being clearly defined as ARDS and the study s ability to stratify by hypoxemia. The data were also collected during the winter months when the number of severe ARDS cases, and thus adjuvant use, may have been higher. Recruitment for LUNG-SAFE was achieved through announcements by ICU societies, which could bias results towards centers with greater enthusiasm surrounding ARDS and the use of adjuvant therapies. To our knowledge, this study is the largest to characterize changes in use and factors associated with adjuvant use for all forms of ARF. This study is unique in that it captures a recent period during which there was a lot of scholarly activity surrounding adjuvants thus enabling us to 15

24 16 explore adoption/de-adoption. We found the adoption of a complex and costly adjuvant for which there is some evolving evidence (ECMO), no adoption of a seemingly simple, widely available adjuvant for which there exists some evidence of benefit (cnmba), no de-adoption of an adjuvant for which there has been consistent evidence of no benefit and some signal for harm (NO), and adoption of an adjuvant for which there has been minimal evidence (inhaled epoprostenol). Understanding physician preferences and what drives them is complex yet critical to influencing care decisions (33). Evidence does not always drive adoption; non-evidence factors such as patient characteristics, perceived benefits/risks, feasibility, and physician preference may play a role (32). Slow adoption for seemingly useful interventions has been seen with lung protective ventilation (34). While de-adoption of ineffective interventions tends to be slow (33) as the epidemiology of pulmonary artery catheter use shows, it is possible (36, 54). The appeal of a novel intervention or the observed improvements in oxygenation (despite lack of association with mortality), likely contribute to ECMO use and the lack of de-adoption of NO, respectively. For cnmbas, absence of adoption could be attributed to the uncertainty surrounding benefit, (55) the absence of instant patient improvement, or concerns about harm. It is likely that rescue is an important driver of decisions in managing patients with ARF and severe hypoxemia. In cases where physicians perceive that patients are at imminent risk of death, cost and lack of evidence may play a smaller role in decision-making (56). Strengths and Limitations This study has several strengths. We used a large, well-defined dataset reflecting real world practice across a diverse group of hospitals. The results were robust to a number of sensitivity analyses that tested the effect of an outlier H 1 N 1 year and severity of illness. Finally, we were able to replicate expected changes in practice over time: a reduction in use of pulmonary artery catheters and an increased use of all adjuvants in 2009 coincident with the H 1 N 1 influenza epidemic. These findings are reassuring that we were capturing true trends rather than changes in documentation or coding. There are several limitations to this study. First, some ICD-9-CM validation studies demonstrate low sensitivity but high specificity of claims data thus potentially underestimating the frequency of diseases/interventions (7, 36). The ECMO numbers, in particular, appear small; this could be 16

25 17 due to under-coding or capturing a subset of low-volume ECMO centers. We think this is unlikely as ECMO has important billing implications in the US. However, if either were true, this would potentially underestimate the rate of rise and likely not negate the positive trend noted. The discrepancy noted between year evaluated as a categorical and continuous variable may be attributable to the H 1 N 1 outlier year, which eliminates an ordered increase in use when year is evaluated continuously. Alternatively, there may be no incremental linear increase over the years; however, there may exist changes over time that were more pronounced in the most recent years as was found in the categorical analysis. Second, our ability to identify the severity of ARF was limited given the lack of ICU level physiology variables. This precluded us from performing stratification across strata of varying severities of illnesses. This would bias the rates of use of adjuvants to a lower estimate, but we feel it would not affect our exploration of trends over time and factors associated with use. The decedent analysis, which restricts the analysis to those sick enough to die, demonstrated similar results. Thirdly, it is plausible that misclassification could have arisen with regard to indications for the adjuvants of interest. We attempted to minimize this by excluding any alternative indications for the adjuvants of interest. While this minimizes its generalizability (i.e., excluded transplant and cardiac surgery), it increases our confidence that we were capturing its use primarily for respiratory failure. Fourthly, selection bias may exist in the subset of hospitals that contribute to the Premier database (i.e., more prone to use evidence based therapies or new innovations). However, we felt that this would be minimized given the large number of hospitals. After reviewing the distribution of hospitals that contribute data to Premier, it is possible that the database may not be generalizable. It favored urban centers and centers from the Southern US; however, even within this region, there was variability with respect to hospital size and academic status. In addition, patterns of adoption and de-adoption, may be different in hospitals outside of the US given that physician preferences appears to be a large driver of use. Finally, for evaluating changes in use, even if there existed a true of change in response to the evidence, it is possible that the landmark publication date may not result in an immediate change. 17

26 18 It is possible that the time period being evaluated was not sufficiently long to see a change where a change may exist particularly if the change was a more gradual one in response to the evidence. Often times, landmark publications are affiliated with presentations that international meetings which occur at a different time point. This could have also led to a different time point that was not captured in this analysis. We also do not know how adoption and de-adoption changes in extreme clinical circumstances where death may be imminent such as the application of adjuvant strategies. 3.0 Conclusions and Future Directions This study highlights real world variability in adjuvant use; variability which likely reflects, in part, the limits of available evidence. However, even when the evidence base is strong, adoption and de-adoption practices are not solely driven by evidence with many factors affecting bedside decision-making. Extreme clinical circumstances may also have an impact on decision making regarding use and has not been evaluated. In conjunction with trying to develop a stronger body of evidence for the adjuvants evaluated, developing a better understanding of factors associated with adjuvant use is necessary. Understanding what drives adoption and de-adoption at the patient, physician and hospital level will be important to drive knowledge translation activities that could have an impact on bedside application. Only then can we safely encourage adoption of effective interventions and promote de-adoption of ineffective interventions. 18

27 19 Table 1. Reported Use for Common Adjuvants in Acute Respiratory Failure Adjuvant Recent Evidence* Reported Use in Acute Respiratory Failure Pharmacologic Continuous Neuromuscular Blocking Agents Papazian L, et al. (2010) Neuromuscular blockers in early acute respiratory distress syndrome. N Engl J Med. 2010;363(12): (Pre-Papazian Trial) 15-23% use in ARDS (7, 9) (Post-Papazian Trial) LUNG-SAFE: 37.8 % reported use in severe ARDS (3) Diuretics Wiedemann HP, et al (2006) Comparison of two fluid management strategies in acute lung injury. N Engl J Med 354: Grissom CK, et al (2015) Fluid management with a simplified conservative protocol for acute respiratory distress syndrome. Crit Care Med. 43 (2): Corticosteroids Meduri G, et al. (2007) Methylprednisolone infusion in early severe ARDS: results of a randomized controlled trial. Chest. 131: Steinberg K, et al. (2006) Efficacy and safety of corticosteroids for persistent acute respiratory distress syndrome N Eng J Med. 354: Administered in 39% of patients with ARDS (single center retrospective study) (12) Self-reported use 70% of patients with ARDS (survey of intensivists) (9) Of 70% of UK physicians surveyed who used corticosteroids in ARDS: 30% reported initiating early in ARDS ( 7 days), 53% reported initiating late in ARDS (>7 days) (9) LUNG-SAFE ** : 17.3% reported use across all ARDS, 23.3% severe ARDS (3) Inhaled Nitric Oxide Non-pharmacologic Adhikari NK, et al (2014) Inhaled nitric oxide does not reduced mortality in patients with ARDS regardless of severity: systematic review and meta-analysis. Crit Care Med. 42 (2): % (9, 18, 20) LUNG-SAFE: 7.7 % reported use across all ARDS, 13.0% severe ARDS (3) Prone Positioning Guerin C, et al (2013) Prone positioning in severe acute respiratory distress syndrome. New Engl J Med. 368: (Post- Guerin Trial ) LUNG-SAFE: 7.9% across all ARDS, 163% severe ARDS (3) High Frequency Oscillatory Ventilation Ferguson N, et al (2013) High frequency oscillation in early acute respiratory distress syndrome. 19 (Pre-Ferguson/Young Trials) 7-50% (rescue therapy) (9, 20)

28 20 Extracorporeal Membrane Oxygenation New Engl J Med 368: Young D, et al (2013) High frequency oscillation for acute respiratory distress syndrome. New Engl J Med 386: Peek G, et al (2009) Efficacy and economic assessment of convention ventilator support vs. extracorporeal membrane oxygenation for severe adult respiratory failure. Lancet.374 (9698): (Post-Ferguson/Young Trials) LUNG-SAFE: 1.5% in severe ARDS (3) 12 fold increase in rate of use over the past decade ( ) (31) LUNG-SAFE: 3.2 % across all ARDS, 6.6% severe ARDS (3) ARDS Acute Respiratory Distress Syndrome LUNG SAFE The Large Observational Study to Understand the Global Impact of Severe Acute Respiratory Failure *Evidence: Trials or systematic reviews/meta analyses over the past 10 years ** high corticosteroids dose defined as equivalent to 1 mg/kg methylprednisone all inhaled vasodilators (inhaled nitric oxide) 20

29 21 Table 2: Inclusion and Exclusion ICD-9 CM codes Inclusion ICD9 CM Codes Mechanical Ventilation 96.70, 96.71, Respiratory Failure , , , DIRECT INFECTIOUS CAUSES OF ARDS , 481, , 483.0, 483.1, 483.8, 484, 485, 486, 487.0, 011, 012, 018, 114, 115, 116, 510, 513 DIRECT NON-INFECTIOUS CAUSES OF ARDS , 986, 507.0, , 022.1, E910.0,1,3,4,8,9, , 958.0, 958.1, 417.8, Exclusion ICD9 CM Codes Pulmonary Hypertension 416.0, Cardiac Surgery 39.61, , , , 37.41, Transplant V , V42.83, 37.51, , 50.5, 55.61, 55.69, 52.8, ARDS Acute Respiratory Distress Syndrome ICD-9-CM International Classification of Disease, 9 th Edition, Clinical Modification 21

30 22 Table 3: Landmark Studies Evaluated Adjuvant Reference Publication Date Results Comments ECMO Peek G et al. Efficacy and economic assessment of conventional ventilatory support versus extracorporeal membrane oxygenation for severe adult respiratory failure (CESAR): a multicenter randomized controlled trial. Lancet. 2009;374: October 2009 Transport to an ECMO capable center was associated with improved 6 month disability free survival Less compliance with lung protective ventilation in control arm. Not all patients who were transported to the specialized center for ECMO underwent ECMO Inhaled Pulmonary Vasodilators Afshari A et al. Inhaled nitric oxide for acute respiratory distress syndrme (ARDS) and acute lung injury in children and adults. The Cochrane Collaboration (7). July 2010 After pooling the data from 14 randomized controlled trials, inhaled nitric oxide had no statistically significant effect on mortality and appeared to increase the risk of renal injury amongst adult patients with ARDS Unclear of its role as a rescue therapy. In a later meta-analysis, evaluating its impact on varying severities of illness, there was no benefit across more severe ARDS cohorts (21) cnmba Papazian L, et al. Neuromuscular blockers in early acute respiratory distress syndrome. N Engl J Med. 2010;363(12): ECMO: extracorporeal membrane oxygenation cnmba: continuous neuromuscular blockading agents September 2010 The early, continuous application of cisatracurium for 48 hours in moderate to severe ARDS was associated with an improved adjusted 90 day mortality Some uncertainty followed with respect to the mechanism by which paralysis lead to the improved late mortality 22

31 23 Table 4: Baseline Demographic Characteristics ( ) BASELINE DEMOGRAPHIC CHARACTERISTICS N = 514, 913 Patients Age (mean ± SD) 62 ((± 17) Gender (Female, n (%)) 247,158 (48) Race (White, n (%)) 334,693 (65) BASELINE CLINICAL VARIABLES No. Elixhauser Comorbidities, median (IQR) 0 (0-0) ARDS risk factor, n (%): Pneumonia 278,053 (54) Non-Pulmonary Sepsis 72,087 (14) Non-Infectious SIRS 25,745 (5) Trauma 56,640 (11) Mechanical Vent in first 2 days (n (%)) 375,886 (73) Vasopressors (n (%)) 288,351 (56) In-Hospital Dialysis (n (%)) 56,640 (11) ADJUVANT CHARACTERISTICS Extracorporeal Membrane Oxygenation (n (%))* 195 (0.2)* Inhaled Pulmonary Vasodilators (n (%))* 1,878 (1.0)* Inhaled Nitric Oxide (n (%))* 1,087 (0.7)* Neuromuscular Blockading Agents** (1 day) (n (%)) 20,077 (3.8) Neuromuscular Blockading Agents** (2 days) (n (%)) 10,073 (2.0) OUTCOMES Duration of Mechanical Vent (survivors, days, median IQR) 4 (3-9) Length of ICU Admission (survivors, days, median IQR) 6 (3-12) Hospital Mortality (n(%)) 169,921 (33) Cost (mean ±SD) $27,979 ($14,468-$51,980) PATIENT DISTRIBUTION ACROSS HOSPITALS Urban (n (%)) 458,272 (89) Teaching (n (%)) 226,562 (44) Bed size (n (%)) Large 400 beds 257,457 (50) Medium beds 190,518 (37) Small < 200 beds 66,939 (13) Location (n (%)) Midwest 92,684 (18) Northeast 62,684 (18) South 242,009 (47) West 87,535 (17) IQR interquartile range; No. number; SD standard deviation, * % reported across hospitals capable of employing that adjuvant, **estimated minimum dosage that can attain continuous paralysis/24 hours, 28 day mortality assumes that those discharged before 28 days were alive at 28 days, total patient cost for hospitalization, all patients included 23

32 24 Table 5: Adjuvant Frequencies Total Adjuvants (n (%)) 12, ,006 (2) (0.1) 3 40 (<0.01) One Adjuvant (n (%)) Continuous neuromuscular blockading agents 9537 (86) Inhaled pulmonary vasodilators 1373 (12) Extracorporeal membrane oxygenation 96 (1) Two Adjuvants (n (%)) Continuous neuromuscular blockading agents 474 (95) Inhaled pulmonary vasodilators 465 (89) Extracorporeal membrane oxygenation 59 (11) 24

33 25 Table 6: Patient and Hospital Characteristics According to Adjuvant ECMO IPV cnmba PATIENT CHARACTERISTICS n=195 n=1,878 n=10,073 Age (mean ± SD) 46 (17) 55 (18) 51 (18) Race (n (%) Caucasian) 127 (64) 1,052 (56) 6,346 (63) Gender (n (%) Female) 86 (44) 789 (42) 3,727 (37) Insurance Status (n (%)) Private 96 (49) 657 (35) 3,123 (31) Medicare 41 (21) 695 (37) 3,324 (33) Medicaid 49 (25) 338 (18) 2,015 (20) Elixhauser Comorbidities (median 0 (0-0) 0 (0-0) 0 (0-1) IQR) Immunocompromised (n (%)) 2 (1) 94 (5) 403 (4) No. Hospitals Vasopressor Use (n (%)) 167 (87) 1,615 (86) 8,159 (81) In-Hospital Dialysis (n (%)) 45 (23) 488 (26) 1,813 (18) In-Hospital Mortality (n (%)) 105 (54) 1,108 (59) 4,331 (43) Cost per hospitalization (mean USD) 97,501 74,292 61,579 HOSPTIAL CHARACTERISTICS* Bed size (mean ± SD) 626 (255) 528 (210) 475 (225) Teaching (n (%)) 140 (72) 1,352 (72) 5,037 (50) Urban (n (%)) 193 (99) 1,821 (97) 9,469 (94) cnmba continuous neuromuscular blockading agents ECMO extracorporeal membrane oxygenation IPV inhaled pulmonary vasodilators IQR interquartile range No. - number SD standard deviation USD United States Dollars; *number of patients across hospital characteristics 25

34 26 Table 7: Changes Over Time Sensitivity Analyses ECMO Hospitals Present Entire Cohort 100% Mortality Cohort Sensitivity Analysis Year RR (95% CI) Year RR (95% CI) Year RR (95% CI) 2008 (reference 1.0) 2008 (reference 1.0) 2008 (reference 1.0) ( ) ( ) ( )) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Analysis using year as a continuous variable: RR % CI Hospitals Present Entire Cohort Inhaled Nitric Oxide 100% Mortality Cohort Sensitivity Analysis Year RR (95% CI) Year RR (95% CI) Year RR (95% CI) 2008 (reference 1.0) 2008 (reference 1.0) 2008 (reference 1.0) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Analysis using year as a continuous variable: RR % CI Hospitals Present Entire Cohort Inhaled Epoprostenol 100% Mortality Cohort Sensitivity Analysis Year RR (95% CI) Year RR (95% CI) Year RR (95% CI) 2008 (reference 1.0) 2008 (reference 1.0) 2008 (reference 1.0) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Analysis using year as a continuous variable: RR % CI cnmba Hospitals Present Entire Cohort 100% Mortality Cohort Sensitivity Analysis Year RR (95% CI) Year RR (95% CI) Year RR (95% CI) 2008 (reference 1.0) 2008 (reference 1.0) 2008 (reference 1.0) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Analysis using year as a continuous variable: RR % CI

35 27 cnmba 1 day* Cisatracurium 2 days** Any dose/use NMBA cnmba SENSITIVITY Year RR (95% CI) Year RR (95% CI) Year RR (95% CI) 2008 (reference 1.0) 2008 (reference 1.0) 2008 (reference 1.0) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) *dose to achieve continuous paralysis for 24 hours for 1 day only ** dose to achieve continuous paralysis for 24 hours for 2 days for cisatracurium only 27

36 28 Figure 1: Adjuvants for Acute Respiratory Failure Figure 1 depicts the spectrum of adjuvants to mechanical ventilation 28 for acute respiratory failure across different severities of illness

37 29 Figure 2: Factors that Drive Decision Making Figure 2 depicts factors that may influence clinician decision-making (32) 29

38 30 Figure 3: Cohort Creation Adult in-patients undergoing mechanical ventilation n = 736,488 EXCLUDE No Mechanical Ventilation Billing Code (16,453) Non-Respiratory Failure (56,732) Adult in-patients undergoing mechanical ventilation with respiratory failure n = 663,303 EXCLUDE Pulmonary hypertension, transplant, cardiac surgery (102,494) Extubated alive after 24 hours of mechanical ventilation and discharged alive (45,896) Final Cohort for Analysis n = 514,913 Figure 3 depicts the cohort assembly based upon inclusion and exclusion criteria 30

39 Figure 4: Age and Sex Adjusted Changes in Use of Adjuvants Over Time Figure 4a depicts annual ECMO rates per 1000 persons per year using Poisson regression analysis expressed as the incidence rate ratio. 31

40 Figure 4: Age and Sex Adjusted Changes in Use of Adjuvants Over Time Figure 4b depicts annual ino rates per 1000 persons per year using Poisson regression analysis expressed as the incidence rate ratio. Test for trend p= 0.88 evaluating year as a continuous variable. 32

41 Figure 4: Age and Sex Adjusted Changes in Use of Adjuvants Over Time 33 Figure 4c depicts annual iepoprostenol rates per 1000 persons per year using Poisson regression analysis expressed as the incidence rate ratio. 33

42 Figure 4: Age and Sex Adjusted Changes in Use of Adjuvants Over Time Figure 4d depicts annual cnmba rates 1000 persons per year using Poisson regression analysis expressed as the incidence rate ratio Figure 5: Impact of Landmark Publications on Changes of Use Over Time 34

43 35 a) Extracorporeal Membrane Oxygenation Figure 5a depicts the segmented regression analysis of changes in monthly rates in use of ECMO in response to landmark publications (arrow specifies time point of publication) 35

44 36 Figure 5: Impact of Landmark Publications on Changes of Use Over Time b) Inhaled Nitric Oxide Figure 5b depicts the segmented regression analysis of changes in monthly rates in use of ino in response to landmark publications (arrow specifies time point of publication) 36

45 37 Figure 5: Impact of Landmark Publications on Changes of Use Over Time c) Inhaled Epoprostenol Figure 5c depicts the segmented regression analysis of changes in monthly rates in use of iepoprostenol in response to landmark publications (arrow specifies time point of publication) 37

46 38 Figure 5: Impact of Landmark Publications on Changes of Use Over Time d) Continuous Neuromuscular Blockading Agents Figure 5d depicts the segmented regression analysis of changes in monthly rates in use of cnmba in response to landmark publications (arrow specifies time point of publication) 38

47 39 Figure 6: Changes in Use of Pulmonary Artery Catheter Figure 6 depicts annual rates per 1000 persons per year over time and the incident rate ratio using Poisson regression analysis 39

48 40 Figure 7a: Factors Associated With Adjuvant Use Figure 7a depicts patient and hospital factors associated with ECMO adjuvant use using logistic regression fitted with generalized estimating equations 40

49 41 Figure 7b: Factors Associated With Adjuvant Use Figure 7b depicts patient and hospital factors associated with cnmba adjuvant use using logistic regression fitted with generalized estimating equations 41

50 42 Figure 7c: Factors Associated With Adjuvant Use Figure 7c depicts patient and hospital factors associated ino adjuvant use using logistic regression fitted with generalized estimating equations 42

51 43 Figure 7d: Factors Associated With Adjuvant Use Figure 7d depicts patient and hospital factors associated iepoprostenol adjuvant use using logistic regression fitted with generalized estimating equations 43