High-Flow Nasal Cannula Versus Noninvasive Ventilation as Initial Treatment in Acute Hypoxia: A Propensity Score-Matched Study.
Free full text in Europe PMC
Abstract
Importance
Patients presenting to the emergency department (ED) with hypoxemia often have mixed or uncertain causes of respiratory failure. The optimal treatment for such patients is unclear. Both high-flow nasal cannula (HFNC) and noninvasive ventilation (NIV) are used.Objectives
We sought to compare the effectiveness of initial treatment with HFNC versus NIV for acute hypoxemic respiratory failure.Design setting and participants
We conducted a retrospective cohort study of patients with acute hypoxemic respiratory failure treated with HFNC or NIV within 24 hours of arrival to the University of Michigan adult ED from January 2018 to December 2022. We matched patients 1:1 using a propensity score for odds of receiving NIV.Main outcomes and measures
The primary outcome was major adverse pulmonary events (28-d mortality, ventilator-free days, noninvasive respiratory support hours) calculated using a win ratio.Results
A total of 1154 patients were included. Seven hundred twenty-six (62.9%) received HFNC and 428 (37.1%) received NIV. We propensity score matched 668 of 1154 (57.9%) patients. Patients on NIV versus HFNC had lower 28-day mortality (16.5% vs. 23.4%, p = 0.033) and required noninvasive treatment for fewer hours (median 7.5 vs. 13.5, p < 0.001), but had no difference in ventilator-free days (median [interquartile range]: 28 [26, 28] vs. 28 [10.5, 28], p = 0.199). Win ratio for composite major adverse pulmonary events favored NIV (1.38; 95% CI, 1.15-1.65; p < 0.001).Conclusions and relevance
In this observational study of patients with acute hypoxemic respiratory failure, initial treatment with NIV compared with HFNC was associated with lower mortality and fewer composite major pulmonary adverse events calculated using a win ratio. These findings underscore the need for randomized controlled trials to further understand the impact of noninvasive respiratory support strategies.Free full text
High-Flow Nasal Cannula Versus Noninvasive Ventilation as Initial Treatment in Acute Hypoxia: A Propensity Score-Matched Study
Associated Data
Abstract
IMPORTANCE:
Patients presenting to the emergency department (ED) with hypoxemia often have mixed or uncertain causes of respiratory failure. The optimal treatment for such patients is unclear. Both high-flow nasal cannula (HFNC) and noninvasive ventilation (NIV) are used.
OBJECTIVES:
We sought to compare the effectiveness of initial treatment with HFNC versus NIV for acute hypoxemic respiratory failure.
DESIGN, SETTING, AND PARTICIPANTS:
We conducted a retrospective cohort study of patients with acute hypoxemic respiratory failure treated with HFNC or NIV within 24 hours of arrival to the University of Michigan adult ED from January 2018 to December 2022. We matched patients 1:1 using a propensity score for odds of receiving NIV.
MAIN OUTCOMES AND MEASURES:
The primary outcome was major adverse pulmonary events (28-d mortality, ventilator-free days, noninvasive respiratory support hours) calculated using a win ratio.
RESULTS:
A total of 1154 patients were included. Seven hundred twenty-six (62.9%) received HFNC and 428 (37.1%) received NIV. We propensity score matched 668 of 1154 (57.9%) patients. Patients on NIV versus HFNC had lower 28-day mortality (16.5% vs. 23.4%, p = 0.033) and required noninvasive treatment for fewer hours (median 7.5 vs. 13.5, p < 0.001), but had no difference in ventilator-free days (median [interquartile range]: 28 [26, 28] vs. 28 [10.5, 28], p = 0.199). Win ratio for composite major adverse pulmonary events favored NIV (1.38; 95% CI, 1.15–1.65; p < 0.001).
CONCLUSIONS AND RELEVANCE:
In this observational study of patients with acute hypoxemic respiratory failure, initial treatment with NIV compared with HFNC was associated with lower mortality and fewer composite major pulmonary adverse events calculated using a win ratio. These findings underscore the need for randomized controlled trials to further understand the impact of noninvasive respiratory support strategies.
Acute hypoxemic respiratory failure is a major cause of hospitalizations in the United States (1, 2). There is growing evidence that noninvasive respiratory support may help prevent invasive mechanical ventilation in patients with respiratory failure (3). Traditionally, the primary mode of noninvasive respiratory support has been noninvasive positive pressure ventilation (NIV)—continuous positive airway pressure and bilevel positive airway pressure. Over the past decade, accelerated by the COVID-19 pandemic, high-flow nasal cannula (HFNC) has emerged as an alternative (4, 5). NIV and HFNC work through different mechanisms and thus have different benefits and harms. NIV improves oxygenation by increasing mean airway pressure but has the potential to deliver injurious lung volumes which may put patients at risk for patient self-induced lung injury (3, 6, 7). In contrast, HFNC provides less positive pressure ventilatory support than NIV, which may decrease the risk of self-induced lung injury and can help improve patient tolerance (7–10).
The optimal mode of noninvasive respiratory support in acute hypoxemic respiratory failure remains unclear (3, 11). Although guidelines recommend NIV for patients with acute decompensated heart failure and chronic obstructive pulmonary disease (COPD) exacerbations, these recommendations are based on comparisons of NIV to low-flow oxygen, not HFNC (3, 11). Guidelines further conclude that there is not enough evidence to make a recommendation for HFNC versus NIV in other causes of hypoxemic respiratory failure (3). The few trials that have directly compared HFNC to NIV have focused on specific populations in the ICU (4, 12–15). The largest trial found that for ICU patients with pure hypoxemic respiratory failure, HFNC decreased mortality compared with NIV (13). Other trials have found no difference between treatment with HFNC and forms of NIV in specific ICU populations, for example, COVID-19 and immunocompromised patients, although given the small sizes of these trials, this does not rule out a clinically important difference in outcomes (4, 14, 15). Furthermore, the generalizability of these trials to undifferentiated hypoxemic respiratory failure in the emergency department (ED) may be limited given patients often require initiation of respiratory support before a clinical diagnosis can be made. Additionally, many of the primary outcomes used in trials of noninvasive respiratory support, such as intubation and ventilator-free days, hinge on subjective practice decisions that limit their interpretation (16).
The goal of this study was to compare the effectiveness of initial treatment of acute hypoxemic respiratory failure with HFNC versus NIV on a hierarchical composite outcome of major adverse pulmonary events, calculated using a win ratio.
MATERIALS AND METHODS
Study Design and Setting
This is a single-center, propensity score-matched, retrospective cohort study comparing initial treatment with HFNC versus NIV in patients with acute hypoxemic respiratory failure. This study was approved by the University of Michigan institutional review board on March 14, 2023 (HUM00232776) as a secondary data use, exempt study.
Data Source
We extracted data from the electronic health record (EPIC; Epic Systems, Verona, WI) via queries to our health system’s data warehouses (Clarity and Caboodle) and, for respiratory support mode data, directly from clinical flowsheets. Each variable used in our analysis, its definition, and source are listed in eTable 1 (http://links.lww.com/CCX/B339). There were no missing data for the exposure or outcome variables. Encounters with missing data on covariates in the propensity score model were excluded.
Inclusion and Exclusion Criteria
We included patients 18 years old or older presenting to the University of Michigan adult ED between January 2018 and December 2022 who required at least 6L/min of supplemental oxygen and received HFNC or NIV within 24 hours of ED arrival. To capture acute respiratory failure, we excluded patients who were on chronic home ventilatory support, had a tracheostomy, or received noninvasive respiratory support only after extubation. We also excluded patients who had both positive SARS-CoV-2 antigen test and clinically suspected COVID-19 infection. At our institution, HFNC was the predominant respiratory support mode used for patients with COVID-19 based on concerns about aerosolization and virus transmission with NIV. For patients who had multiple encounters during the study period, only the first encounter was included.
Data Collection and Definition of Respiratory Support Mode
We identified patients who received 6L/min or more of supplemental oxygen and received HFNC or NIV within 24 hours of presentation to the ED based on clinical flowsheets. At our institution, respiratory therapists and nurses record respiratory support mode and settings in a flowsheet at least once per hour and whenever mode or setting changes are made. Choice of respiratory support mode and initial treatment parameters were determined by the treating clinician. We defined HFNC as any delivery of greater than 20
L/min oxygen from a heated, humidified oxygen system via nasal cannula. We defined NIV as any noninvasive ventilation mode (continuous positive airway pressure or bilevel positive airway pressure). We classified patients who received more than one noninvasive respiratory support mode during a given hour according to the dominant mode (mode used most frequently) for the hour. For the primary analysis, we classified patients as receiving initial HFNC or NIV if they received those modes exclusively for the first 2 hours after initiation, to ensure stable initial group assignment.
Study Outcomes
The primary outcome was a hierarchical composite outcome of major adverse pulmonary events: 28-day mortality, 28-day ventilator-free days, and hours spent on noninvasive respiratory support from initiation through hour 72 (NIRS hours). NIRS hours included all time spent on any device (HFNC or NIV), counting cross-over and repeat periods of use, and were censored at death, intubation, or 72 hours, whichever came first. The primary outcome was calculated using a win ratio (see Data Analysis below for details). We also evaluated the components of the composite major adverse pulmonary events outcome individually: mortality and time to death (28-d and in-hospital), intubation rate, ventilator-free days, and noninvasive respiratory support hours.
Data Analysis
We used a logistic regression model to calculate a propensity score for the odds of receiving NIV using prespecified patient characteristics typically available at the time of initiation of HFNC or NIV: age, sex, body mass index, Charlson comorbidity index, history of congestive heart failure (CHF), COPD, or OSA, comparison of measured Pco2 by first blood gas in the ED (venous or arterial) to expected Pco2 by Winter formula, time from ED arrival to HFNC or NIV initiation, initial lactate, initial Glasgow Coma Score, and highest day 1 Sequential Organ Failure Assessment (SOFA) score. Although SOFA score is not available on ED arrival, it was included in the propensity score model because it provides a surrogate measure for the severity of illness. International Classification of Diseases, 10th Edition diagnosis codes were recorded to understand causes of respiratory failure but were not included in the propensity score model, as discharge-level diagnoses are not available at the time of HFNC or NIV initiation.
Using the propensity score, we matched patients 1:1 using greedy nearest neighbor matching with calipers set at 0.2 sds (17). We used standard descriptive statistics, including standard mean differences (SMD) and Kolmogorov-Smirnov statistics, to assess for covariate balance and distribution before and after propensity score matching. SMD less than 0.1 after matching was considered an acceptable balance.
To calculate the primary outcome of major pulmonary adverse events, we used a win ratio, a statistical method for combining variables into a hierarchical composite outcome. Unlike typical composite outcomes that assign similar weight to outcomes regardless of their severity (i.e., including both all-cause mortality and need for repeat intervention) or take an “all-or-nothing” approach to evaluating the presence versus absence of an event, the win ratio accommodates mixed variable types (e.g., binary, ordinal, continuous, time-to-event) and ranks variables by clinical relevance and patient priorities. Prior literature has defined the calculation of the win ratio in detail (18). In brief, we generated all possible pairs of patients on NIV and HFNC. Pairs were sequentially compared on outcomes based on predetermined importance: 1) 28-day mortality as a survival event, 2) 28-day ventilator-free days, and 3) noninvasive respiratory support hours from initiation to hour 72. We incorporated selective tie-breaking to best parallel clinical reasoning. For example, for two individuals who tied on mortality, we did not compare ventilator-free days or noninvasive respiratory support hours. Total wins and losses were added up across all three outcome tiers and compared to generate the win ratio, which was calculated using the WinRatio package (version 1.0) in R (R Foundation for Statistical Computing, Vienna, Austria) (19).
Individual secondary outcomes were compared using Mann-Whitney U test for continuous variables and chi-square test for categorical variables. Time to death and time to intubation were analyzed using a Kaplan-Meier survival analysis. Hourly respiratory support mode and daily patient status were visualized using stacked histogram plots.
We conducted sensitivity analyses using an intention-to-treat framework (inclusion of patients who crossed over between respiratory support modes or had discontinued noninvasive respiratory support in the first 2hr) and including all ED encounters for patients who had multiple encounters during the study period. We also used alternative approaches to calculating the win ratio.
Initial data cleaning, data exploration, and visualization were performed using Tableau (Tableau Software, LLC, Seattle, WA). Final data cleaning and all statistical analyses were performed using R (version 4.2.1) in R Studio (version 2022.2.3).
RESULTS
Between January 2018 and December 2022, 2,208 of 361,459 (0.6%) adult ED encounters required noninvasive respiratory support within 24 hours of ED arrival. Of these, 1154 of 2208 (57.3%) patients met the study inclusion criteria (Fig. Fig.11). HFNC was the predominant mode, used in 726 of 1154 (62.9%) encounters. Patients treated with NIV and HFNC had large differences in baseline CHF (77.3% vs. 51.9%, SMD 0.551), Pco2 above expected based on Winter formula (67.5% vs. 33.9%, SMD 0.729), shock index (heart rate/systolic blood pressure: 0.7 vs. 0.8, SMD 0.356) and lactate greater than 4 mmol/L (10.0% vs. 15.2%, SMD 0.236) (Table Table11). The most common encounter diagnosis codes were respiratory failure not otherwise specified (89.3% vs. 85.0%), followed by volume overload (53.4%) and pneumonia (49.0%) in the HFNC group and volume overload (77.8%) and COPD/asthma exacerbation (50.7%) in NIV group (eTable 2, http://links.lww.com/CCX/B339).
TABLE 1.
Baseline Patient Characteristics Before and After Propensity Score Matching
| Overall Cohort | Matched Cohort | |||||
|---|---|---|---|---|---|---|
| HFNC (n = 726) | NIV (n = 428) | SMD | HFNC (n = 334) | NIV (n = 334) | SMD | |
| Age, yr | 66.9 (55.6, 76.0) | 69.3 (60.2, 77.9) | 0.213 | 69.4 (56.7, 78.5) | 69.0 (59.0, 77.5) | 0.036 |
| Male sex | 427 (58.8) | 225 (52.6) | 0.126 | 179 (53.6) | 186 (55.7) | 0.042 |
| Body mass index (kg/m2) | 26.3 (22.1, 31.2) | 29.1 (23.6, 35.5) | 0.4 | 27.4 (23.1, 33.0) | 27.4 (23.0, 33.3) | 0.094 |
| Charlson Comorbidity Index | 8.0 (3.0, 14.0) | 8.5 (5.0, 14.0) | 0.132 | 8.0 (4.0, 15.0) | 9.0 (5.0, 14.0) | 0.024 |
| Individual comorbidities at baseline | ||||||
| 377 (51.9) | 331 (77.3) | 0.551 | 235 (70.4) | 238 (71.3) | 0.02 | |
| 422 (58.1) | 292 (68.2) | 0.211 | 213 (63.8) | 225 (67.4) | 0.076 | |
| 131 (18.0) | 118 (27.6) | 0.229 | 79 (23.7) | 86 (25.7) | 0.049 | |
| Expected Pco2a | ||||||
| 166 (22.9) | 62 (14.5) | 0.729 | 56 (16.8) | 60 (18.0) | 0.077 | |
| 246 (33.9) | 289 (67.5) | 190 (56.9) | 197 (59.0) | |||
| 314 (43.3) | 77 (18.0) | 88 (26.3) | 77 (23.1) | |||
| First hours of noninvasive respiratory support | 2.0 (1.0, 7.0) | 2.0 (1.0, 5.0) | 0.268 | 2.0 (1.0, 5.0) | 2.0 (1.0, 5.0) | 0.063 |
| Systolic shock index | 0.8 (0.7, 1.0) | 0.7 (0.5, 0.9) | 0.356 | 0.7 (0.6, 0.9) | 0.7 (0.6, 0.9) | 0.006 |
| Lactate group | ||||||
| 375 (51.7) | 269 (62.9) | 0.236 | 193 (57.8) | 202 (60.5) | 0.07 | |
| 241 (33.2) | 116 (27.1) | 96 (28.7) | 94 (28.1) | |||
| 110 (15.2) | 43 (10.0) | 45 (13.5) | 38 (11.4) | |||
| Glasgow Coma Score | 15 (15, 15) | 15 (15, 15) | 0.122 | 15.0 (15.0, 15.0) | 15.0 (15.0, 15.0) | 0.007 |
| Highest day 1 SOFA | 5.0 (4.0, 7.0) | 5.0 (3.0, 6.0) | 0.281 | 5.0 (4.0, 7.0) | 5.0 (3.0, 7.0) | 0.065 |
HFNC = high-flow nasal cannula, NIV = noninvasive positive pressure ventilation, SMD = standard mean difference, systolic shock index = ratio of heart rate/blood pressure, SOFA = Sequential Organ Failure Assessment.
Baseline characteristics in the overall cohort and matched study cohort. Data are presented as median (interquartile range) and n (%). Noninvasive respiratory support refers to both HFNC and NIV.
Study flow diagram. The primary analysis included n = 1154 encounters in the pool for matching. A sensitivity analysis was performed including all encounters where patients received qualifying noninvasive respiratory support (NIRS), including early cross-over and discontinuation (N = 1615, see eTable 1, http://links.lww.com/CCX/B339). ED = emergency department, vent= ventilation, HFNC = high-flow nasal canula, ICD-10 = International Classification of Diseases, 10th Edition, NIV = noninvasive ventilation.
In our propensity model (eFig. 1, http://links.lww.com/CCX/B339) body mass index (odds ratio [OR] 1.03 [per kg/m2 increase]; 95% CI, 1.01–1.05), history of CHF (OR 2.37; 95% CI, 1.74–3.26), and Pco2 above expected by Winter formula (OR 3.54; 95% CI, 2.46–5.15) were associated with increased odds of treatment with NIV, whereas hemodynamic instability (shock index: OR 0.53 [per unit increase]; 95% CI, 0.29–0.95) and higher SOFA (OR 0.91 [per point increase]; 95% CI, 0.86–0.96) were associated with lower odds of receiving NIV.
We matched 668 of 1154 (57.9%) eligible encounters at a ratio of 1:1, including 334 of 726 (46.0%) receiving HFNC and 334 of 428 (78.0%) receiving NIV. Patient characteristics were well-balanced between groups after matching (Table (Table1,1, eFig. 2, http://links.lww.com/CCX/B339). Matched NIV patients were similar to overall NIV patients, whereas matched HFNC patients had higher rates of CHF, higher body mass index, and more Pco2 above expected by Winter formula than overall HFNC patients (Table (Table11).
The primary outcome was major adverse pulmonary events. We first evaluated individual components of the major adverse pulmonary events outcome (Table Table22). There was no difference between patients treated with NIV versus HFNC in ventilator-free days (median [interquartile range 28 [26, 28] vs. 28 [10.5, 28]; p = 0.199). However, patients treated with NIV had significantly lower 28-day mortality (16.5% vs. 23.4%, p = 0.033) and spent fewer hours on noninvasive respiratory support within the first 72 hours of initiation (median 7.5 vs. 13.5hr, p < 0.001). Hourly respiratory support modes up to 72 hours post-HFNC or NIV initiation and daily patient status out to 28 days by treatment group are presented in the supplement (eFigs. 3 and 4, http://links.lww.com/CCX/B339). Time to death by 28 days was also similar between patients treated with NIV versus HFNC (eFig. 5, http://links.lww.com/CCX/B339).
TABLE 2.
Individual Outcomes for Patients Receiving High-Flow Nasal Cannula Versus Noninvasive Positive Pressure Ventilation in the Overall and Matched Cohort
| Overall Cohort | Matched | |||||
|---|---|---|---|---|---|---|
| HFNC (n = 726) | NIV (n = 428) | p | HFNC (n = 334) | NIV (n = 334) | p | |
| 28-d mortality | 197 (27.1) | 66 (15.4) | < 0.001 | 78 (23.4) | 55 (16.5) | 0.033 |
| Time to death in 28 d | 6.6 (3.1, 13.7) | 6.0 (2.7, 12.9) | 0.57 | 6.3 (3.1, 13.1) | 6.1 (2.7, 13.0) | 0.661 |
| In-hospital mortality | 159 (21.9) | 54 (12.6) | < 0.001 | 62 (18.6) | 45 (13.5) | 0.091 |
| Time to death in hospital (d) | 5.3 (2.4, 12.8) | 4.4 (2.3, 9.7) | 0.544 | 5.6 (2.4, 12.6) | 4.5 (2.5, 8.9) | 0.641 |
| Intubation within 72 | 154 (21.2) | 65 (15.2) | 0.015 | 66 (19.8) | 53 (15.9) | 0.225 |
| Time to intubation (hr) | 10.0 (4.0, 21.8) | 8.0 (4.0, 26.0) | 0.908 | 11.5 (6.2, 24.5) | 8.0 (4.0, 23.0) | 0.374 |
| Ventilator-free days | 28.0 (0.0, 28.0) | 28.0 (26.0, 28.0) | < 0.001 | 28.0 (10.5, 28.0) | 28.0 (26.0, 28.0) | 0.199 |
| Noninvasive respiratory support hours | 14.0 (6.0, 34.0) | 7.0 (4.0, 19.0) | < 0.001 | 13.5 (6.0, 33.0) | 7.5 (4.0, 19.0) | < 0.001 |
HFNC = high-flow nasal cannula, NIV = noninvasive positive pressure ventilation.
Individual patient outcomes before and after propensity score matching. Data are presented as n (%) and median (interquartile range). p values were calculated using Mann-Whitney U for continuous variables and χ2 for categorical variables. Time to death in 28 d, ventilator-free days and noninvasive respiratory support hours contribute to the primary composite major adverse pulmonary events outcome. Ventilator-free days were calculated from admission through day 28. Noninvasive respiratory support hours were hours spent on HFNC or NIV calculated from initiation through hour 72.
We then calculated the composite major adverse pulmonary events using the win ratio. There were 111,556 potential matched patient pairs (334×
334). In total, NIV won in 63,553 (57.0%) pairs whereas HFNC won in 46,039 (41.3%), resulting in a win ratio for NIV of 1.38 (95% CI, 1.15–1.465; p < 0.001). NIV won over HFNC for 28-day mortality (wins, as percent of all pairs: 21.3% vs. 14.6%) and NIRS hours (wins: 25.0% vs. 14.1%), but not ventilator-free days (wins: 10.6% vs. 12.6%). Only 1964 (1.8%) pairs tied on all tiers (Table Table33 and Fig. Fig.22). Results were robust to sensitivity analyses including patients with multiple encounters and with early respiratory mode cross-over or discontinuation (eTable 3, http://links.lww.com/CCX/B339) and alternative approaches to win ratio calculation (eTable 4, http://links.lww.com/CCX/B339). Cross-over between modes occurred in both groups (eTable 5, http://links.lww.com/CCX/B339).
TABLE 3.
Win Ratio of Composite Major Adverse Pulmonary Events for Patients Receiving High-Flow Nasal Cannula Versus Noninvasive Positive Pressure Ventilation
| Tier | Outcome | Overall Cohort | Matched Cohort | ||||
|---|---|---|---|---|---|---|---|
| NIV Wins | HFNC Wins | Ties | NIV Wins | HFNC Wins | Ties | ||
| 1 | Time to death (28-d mortality) | 77,511 (24.9%) | 41,718 (13.4%) | 191,499 (61.6%) | 23,811 (21.3%) | 16,321 (14.6%) | 71,424 (64.0%) |
| 2 | Ventilator-free days | 39,220 (12.6%) | 35,714 (11.5%) | 116,565 (37.5%) | 11,833 (10.6%) | 14,006 (12.6%) | 45,585 (40.9%) |
| 3 | Noninvasive respiratory support hours | 73,207 (23.6%) | 38,171 (12.3%) | 5,187 (1.7%) | 27,909 (25.0%) | 15,712 (14.1%) | 1,964 (1.8%) |
| Totals, by category | 189,938 (61.1%) | 115,603 (37.2%) | 5,187 (1.7%) | 63,553 (57.0%) | 46,039 (41.3%) | 1,964 (1.8%) | |
| Total possible pairs | 310,728 | 111,556 | |||||
| Win ratio (95% CI), p | 1.64 (1.42–1.90), p < 0.001 | 1.38 (1.15–1.65), p < 0.001 | |||||
HFNC = high-flow nasal cannula, NIV= noninvasive positive pressure ventilation.
Details of the win ratio used to calculate the composite major adverse pulmonary events in the overall cohort (n = 1154) and in the matched study cohort (n = 668). The win ratio in the matched cohort is the primary outcome, which is also displayed visually in Figure Figure2.2. Ventilator-free days were calculated from admission through day 28. Noninvasive respiratory support hours were hours spent on HFNC or NIV calculated from initiation through hour 72. The win ratio is the ratio of overall “wins for NIV” over “wins for HFNC.” A positive win ratio suggests NIV results in a better composite outcome compared with HFNC. Percentages represent the percent of pairs out of the total possible pairs. On tier 1 (time to death) and overall, the percentages sum to 100% because all patient pairs are compared at these levels. However, tiers 2 and 3 only compare patients who tied on the previous tier. For example, in tier 2 (ventilator-free days), the percentages add to the number of patients who tied on the previous tier.
Comparison of major pulmonary adverse events between patients receiving noninvasive ventilation (NIV) versus high-flow nasal cannula (HFNC) in the matched cohort, using a win ratio. A win ratio is the ratio of overall “wins for NIV” over “wins for HFNC.” A positive win ratio suggests NIV results in a better composite outcome compared with HFNC. Percentages represent the percent of pairs out of the total possible pairs. On tier 1 (time to death) and overall, the percentages sum to 100% because all patient pairs are compared at these levels. However, tiers 2 and 3 only compare patients who tied on the previous tier. For example, in tier 2 (ventilator-free days), the percentages add to the number of patients who tied on the previous tier (64.0%). Noninvasive respiratory support hours = time spent on noninvasive respiratory support (NIV or HFNC) in hours, from initiation through hour 72.
DISCUSSION
In this propensity-matched retrospective study of patients with acute hypoxemic respiratory failure, initial treatment with NIV decreased the individual outcomes of mortality and time spent on noninvasive respiratory support in the first 72 hours but did not have a significant association with ventilator-free days. Treatment with NIV was associated with lower composite major adverse pulmonary events, calculated using a win ratio.
Our finding that initial treatment with NIV may decrease overall major adverse pulmonary events compared with HFNC contrasts with the findings of the prominent Clinical Effect of the Association of Noninvasive Ventilation and High Flow Nasal Oxygen Therapy in Resuscitation of Patietns with Acute Lung Injury (FLORALI) trial. In that trial, HFNC improved mortality and ventilator-free days compared with NIV (13). The results of the FLORALI trial have not been replicated consistently in the few other trials comparing HFNC and NIV in acute hypoxemic respiratory failure, perhaps because of small sample sizes or the focus on different conditions (e.g., COVID-19, immunocompromise) (3, 4, 14, 15, 20). Furthermore, most existing trials, including FLORALI, have been based in the ICU and have included specific patient populations. For example, although our study included all patients initially treated with HFNC or NIV regardless of diagnosis, the FLORALI trial enrolled a much narrower population of ICU patients with pure hypoxemic respiratory failure, excluding patients with hypercarbia, cardiogenic pulmonary edema, or COPD exacerbations (13). HFNC and NIV have not been directly compared in these other conditions (i.e., heart failure and COPD exacerbations), where recommendations to use NIV are based on comparisons of NIV to low-flow oxygen, not HFNC (3, 11).
There has only been one trial comparing HFNC versus NIV in a broad population of patients (12). Similar to our study, that trial by Doshi et al (12) enrolled ED patients with acute hypoxemic respiratory failure and found HFNC was noninferior to NIV based on intubation rates, though the trial was small and not powered to assess other outcomes. Therefore, the optimal noninvasive respiratory support mode for treating early, undifferentiated hypoxemic respiratory failure remains unclear. Our results add to this clinical equipoise by suggesting that initial use of NIV may improve outcomes compared with HFNC in the broad population of patients with early, undifferentiated hypoxemic respiratory failure.
Understanding the optimal initial treatment for undifferentiated hypoxemic respiratory failure is critical, particularly in the ED where the choice between HFNC and NIV often must be made before information about a patient’s diagnosis is fully available. Indeed, our findings suggest that in practice patients frequently have multiple risk factors for respiratory failure. For example, even among patients who received HFNC, over half had a documented history of heart failure and/or COPD. Although the frequency of these comorbidities may be high due to our institution’s role as a tertiary referral center, this finding reflects the challenge of identifying patients with specific causes of respiratory failure in real-time. This is further complicated by the fact that many patients have multiple causes of respiratory failure and do not fit into single diagnostic categories, as seen in the mix of encounter diagnosis codes in our cohort.
Our results underscore the need for a randomized controlled trial to further understand the impact of NIV versus HFNC in patients with early, undifferentiated hypoxemic respiratory failure. To best inform practice decisions, such trials must use novel design approaches, such as pragmatic designs (21), to capture patients early in their disease course, ideally in the ED at the time of NIV or HFNC initiation. Future clinical trials must also select appropriate outcomes to overcome the limitations of potentially subjective primary outcomes, such as intubation or ventilator-free days.
Our study suggests that using a win ratio may be a feasible approach to evaluating composite outcomes in future studies of respiratory support modes. In our study, calculating the composite outcome of major adverse pulmonary events using a win ratio allowed a more nuanced understanding of the impact of NIV and HFNC on patient outcomes than individual outcomes alone. In particular, one challenge in studies of respiratory failure is the limitation of single outcomes such as mortality or intubation, which can obscure other clinically important outcomes particularly when only a subset of patients experience death or intubation. For example, in our study, a majority of patients survived and were never intubated, but patients treated initially with NIV spent less total time on noninvasive respiratory support. Although HFNC may be more comfortable than NIV and may not always require ICU-level support, both devices require close monitoring and high oxygen flows (8, 10, 22–24). Therefore, spending less time on noninvasive respiratory support may be a potentially important outcome in patients who survive and are not intubated. The win ratio offers a mechanism to compare multiple relevant patient outcomes while maintaining a hierarchical approach that reflects the relative importance of outcomes to patients and their families: first comparing mortality, then comparing ventilator-free days among survivors, and finally comparing noninvasive respiratory support hours among survivors who were not intubated or had the same number of ventilator-free days. Our results suggest that a similar win ratio could be used as a primary outcome for future clinical trials comparing respiratory support modalities.
Limitations
This study has several limitations common to retrospective studies. First, there is a risk that our findings are the result of residual confounding. The use of propensity-score matching limits that risk, but the presence of confounding on unmeasured variables remains possible and a prospective clinical trial is needed to verify these findings. Further, as a retrospective study, respiratory support modes were set by the clinical team rather than protocols, which may limit interpretability of the findings. The assignment of the exposure group also required that patients receive HFNC or NIV exclusively for the first two hours after initiation, which may limit generalizability to patients who switch support modalities or discontinue them within the first two hours. However, in sensitivity analyses where the exposure group was assigned by first modality, results were similar. Additionally, although respiratory support modes are documented at least hourly at our institution, there were limitations in some of the other data available for matching. Notably, a majority of first available blood gases were venous, which may be less accurate than arterial blood gases for measuring Pco2 (25, 26). Dividing patients into categories based on the comparison of measured to expected Pco2, rather than using exact Pco2 values, may help decrease inaccurate judgments based on venous blood gases (27). Also, given some patients were treated with HFNC or NIV immediately on ED arrival, it was not always possible to isolate the pretreatment severity of illness. The inclusion of the highest day 1 SOFA score in the propensity score model allowed us to capture each patient’s worst day 1 illness severity, thereby minimizing the effects of improvement with treatment. Finally, as a single-center study performed at a tertiary care hospital, generalizability may be limited as practice patterns regarding HFNC and NIV use may differ at other centers. Although indications for HFNC and NIV selection may vary across institutions, propensity matching helps ensure that the patients included in our study were similar and might be candidates for either HFNC or NIV more broadly.
CONCLUSIONS
In this propensity-matched retrospective study of patients with early acute hypoxemic respiratory failure, initial treatment with NIV was superior to HFNC for major adverse pulmonary events, calculated using a win ratio. These results underscore the need for novel randomized controlled trials to definitively determine the merits of each noninvasive respiratory support strategy.
Footnotes
Drs. Munroe, Prevalska, and Fung contributed to the conceptualization and design of the study, data analysis and interpretation, and drafting of this article. Drs. Hyer, Meurer, Mosier, Tidswell, Wei, and Wang contributed to the conceptualization of the study, data interpretation, and the writing of the article, including substantive revisions. Dr. Prescott contributed to data interpretation and the writing of the article, including substantive revisions. All authors have read and approved the final article.
Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s website (http://journals.lww.com/ccejournal).
Dr. Munroe was supported by grant number T32 HL 007749 (Multidisciplinary Training Program in Lung Disease), grant number F32 HL 172463 and grant number L30 HL 170379 (Loan Repayment Award) from the National Institutes of Health (NIH) and the National Heart, Lung, and Blood Institute (NHLBI). This work was also supported in part by a grant from NIH (NINDS and NHLBI) for infrastructure for the Clinical Coordinating Center for the Strategies to Innovate EmeRgENcy Care Clinical Trials Network—2U24NS100659. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. This article does not represent the views of the Department of Veterans Affairs or the U.S. government. Dr. Mosier has received travel support from Fisher & Paykel. The remaining authors have disclosed that they do not have any potential conflicts of interest.
This study was approved by the University of Michigan institutional review board on March 14, 2023 (HUM00232776) as a secondary data use, exempt study.
This material is the result of work supported with resources and the use of facilities at the Ann Arbor VA Medical Center.
REFERENCES
Articles from Critical Care Explorations are provided here courtesy of Wolters Kluwer Health
Full text links
Read article at publisher's site: https://doi.org/10.1097/cce.0000000000001092
Read article for free, from open access legal sources, via Unpaywall:
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11081605
Similar Articles
To arrive at the top five similar articles we use a word-weighted algorithm to compare words from the Title and Abstract of each citation.
High-flow Nasal Cannula Versus Noninvasive Ventilation for Treatment of Acute Hypoxemic Respiratory Failure in Renal Transplant Recipients.
Transplant Proc, 49(6):1325-1330, 01 Jul 2017
Cited by: 15 articles | PMID: 28736002
Efficacy of High-Flow Nasal Cannula Therapy in Acute Hypoxemic Respiratory Failure: Decreased Use of Mechanical Ventilation.
Respir Care, 60(10):1390-1396, 23 Jun 2015
Cited by: 31 articles | PMID: 26106206
Noninvasive ventilation and high-flow nasal cannula in patients with acute hypoxemic respiratory failure by covid-19: A retrospective study of the feasibility, safety and outcomes.
Respir Physiol Neurobiol, 298:103842, 10 Jan 2022
Cited by: 11 articles | PMID: 35026479 | PMCID: PMC8744300
High flow nasal cannula versus conventional oxygen therapy and non-invasive ventilation in adults with acute hypoxemic respiratory failure: A systematic review.
Respir Med, 121:100-108, 03 Nov 2016
Cited by: 56 articles | PMID: 27888983
Review
Funding
Funders who supported this work.
NHLBI NIH HHS (3)
Grant ID: F32 HL172463
Grant ID: T32 HL007749
Grant ID: L30 HL170379
NINDS NIH HHS (1)
Grant ID: U24 NS100659