High-flow nasal cannula for acute hypoxemic respiratory failure in patients with COVID-19: systematic reviews of effectiveness and its risks of aerosolization, dispersion, and infection transmission.

PURPOSE: We conducted two World Health Organization-commissioned reviews to inform use of high-flow nasal cannula (HFNC) in patients with coronavirus disease (COVID-19). We synthesized the evidence regarding efficacy and safety (review 1), as well as risks of droplet dispersion, aerosol generation, and associated transmission (review 2) of viral products. SOURCE: Literature searches were performed in Ovid MEDLINE, Embase, Web of Science, Chinese databases, and medRxiv. Review 1: we synthesized results from randomized-controlled trials (RCTs) comparing HFNC to conventional oxygen therapy (COT) in critically ill patients with acute hypoxemic respiratory failure. Review 2: we narratively summarized findings from studies evaluating droplet dispersion, aerosol generation, or infection transmission associated with HFNC. For both reviews, paired reviewers independently conducted screening, data extraction, and risk of bias assessment. We evaluated certainty of evidence using GRADE methodology. PRINCIPAL
FINDINGS: No eligible studies included COVID-19 patients. Review 1: 12 RCTs (n = 1,989 patients) provided low-certainty evidence that HFNC may reduce invasive ventilation (relative risk [RR], 0.85; 95% confidence interval [CI], 0.74 to 0.99) and escalation of oxygen therapy (RR, 0.71; 95% CI, 0.51 to 0.98) in patients with respiratory failure. Results provided no support for differences in mortality (moderate certainty), or in-hospital or intensive care length of stay (moderate and low certainty, respectively). Review 2: four studies evaluating droplet dispersion and three evaluating aerosol generation and dispersion provided very low certainty evidence. Two simulation studies and a crossover study showed mixed findings regarding the effect of HFNC on droplet dispersion. Although two simulation studies reported no associated increase in aerosol dispersion, one reported that higher flow rates were associated with increased regions of aerosol density.
CONCLUSIONS: High-flow nasal cannula may reduce the need for invasive ventilation and escalation of therapy compared with COT in COVID-19 patients with acute hypoxemic respiratory failure. This benefit must be balanced against the unknown risk of airborne transmission.

In December 2019, investigators identified a novel coronavirus, subsequently named by the World Health Organization (WHO) as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), as the cause of atypical pneumonia cases in Wuhan, China.1 Since then, the disease caused by SARS-CoV-2 (named COVID-19) has emerged as a global pandemic.2 As of 15 May, SARS-CoV-2 has infected > 4 million people in across 200 countries and has caused more than 290,000 deaths, the majority of which have occurred outside China.3 Although most patients present with mild respiratory symptoms, some have severe pneumonia and a small proportion become critically ill.4,5 The volume of severe cases has created an unprecedented burden on healthcare systems, highlighting the urgency in identifying safe, effective therapies for COVID-19.

Severe COVID-19 often progresses to acute hypoxemic respiratory failure requiring high fractional concentration of inspired oxygen (FO2) and consideration for non-invasive ventilation (NIV) strategies.6–9 High-flow nasal cannula has emerged as a non-invasive strategy improving oxygenation and carbon dioxide clearance by, relative to other NIV strategies, better matching of patients’ inspiratory demands by delivering up to 60 L·min−1 of gas flow with an FO2 up to 1.0, and thus decreasing adverse outcomes.10–14 A recent systematic review found low certainty evidence for a benefit of HFNC in reducing the need for invasive ventilation or escalation of oxygen therapy compared with conventional oxygen therapy (COT), and moderate certainty evidence suggesting no large difference in mortality.15 Nevertheless, HFNC may reduce the need for invasive ventilation and associated adverse events such as ventilator-associated pneumonias, and also alleviate the strain on healthcare systems during the COVID-19 pandemic.

COVID-19 spreads through respiratory droplets and fomites.1,16,17 There is concern, however, that airborne transmission may occur during procedures that generate aerosols.17 Airborne transmission involves smaller particles (droplet nuclei), typically < 5 μm in diameter, which may remain suspended in the air for extended periods of time, transmitted over distances greater than 1 m, and inhaled into the lower airways.17 Reduction of respiratory particles to < 5 µm involves dehydration of larger droplets and their contained organisms, and rehydration after deposition into the airway; therefore, airborne transmission is organism-specific, and requires the organism to survive a process of desiccation and aerosolization in sufficient numbers to cause infection.18

On 29 March 2020, the WHO issued a scientific brief recommending droplet and contact precautions for the care of COVID-19 patients and airborne precautions during aerosol-generating procedures.17 The use of high flow rates raises concerns that HFNC may cause aerosolization of infectious particles. The Surviving Sepsis Campaign COVID-19 guidelines provide a weak recommendation for the preferential use of HFNC over other NIV strategies in patients refractory to COT.19 Nevertheless, the guidelines did not consider how different circumstances may change the balance between risks and benefits of HFNC, and considered only two studies in evaluating risk of disease transmission.20

The severe resource constraints in healthcare settings facing large numbers of COVID-19 patients dictates an urgent need for an updated evidence synthesis and guidance regarding the use of HFNC among these patients. We conducted two rapid systematic reviews commissioned by the WHO to summarize the evidence for the efficacy, safety, and risk of aerosol generation and infection transmission during HFNC use among patients with acute hypoxemic respiratory failure due to COVID-19.

Methods

Prior to beginning, WHO personnel reviewed and approved internal protocols for both systematic reviews; given time constraints of the commissioned reviews (seven days to completion), neither protocol was registered nor published.

Systematic review #1: efficacy and safety of HFNC in acute hypoxemic respiratory failure

Literature search

A previous systematic review searched Ovid MEDLINE, Embase, and Web of Science for eligible randomized-controlled trials (RCTs) from 1 January 2007 to 25 October 2018.15 With input from a health information specialist, we updated this systematic review, searching Ovid MEDLINE, Embase, and Web of Science from 1 October 2018 to 14 May 2020 with no language restrictions (see Appendix 1).

Selection criteria

We included RCTs that compared HFNC with COT in critically ill patients with acute hypoxemic respiratory failure. We defined COT as inhaled oxygen via nasal prongs, simple face masks, face masks with reservoir bags, or Venturi masks. Eligible studies reported one or more of the following pre-specified outcomes: mortality (using the longest follow-up available), invasive ventilation, escalation of therapy (crossover to HFNC in the control group, or initiation of NIV or invasive ventilation in either group), intensive care unit (ICU) and hospital lengths of stay (LOS), patient-reported comfort and dyspnea, and treatment-related complications.

To identify potential additional eligible RCTs, we reviewed relevant systematic reviews. We excluded case reports, case series, and observational studies as well as studies that (i) used NIV or invasive ventilation as a sole comparator with no COT arm, (ii) evaluated the role of HFNC peri-intubation, or (iii) evaluated the role of HFNC for post-extubation respiratory support. These exclusions aligned with the most common indication for HFNC—i.e., as an alternative to COT in a patient with hypoxemic respiratory failure not requiring immediate intubation.

Study selection

Paired reviewers (D.C.1, D.G., M.H., D.C.2) screened the title and abstract of identified citations, followed by full-text review of potentially eligible studies. A third reviewer (A.A.) resolved disagreements. We captured reasons for exclusion at the full-text review stage. Single reviewers (D.C.1, M.H., D.G.) screened the reference lists of relevant systematic reviews to identify additional RCTs meeting eligibility criteria.

Data extraction and quality assessment

Paired reviewers (D.G., D.C.2) performed data extraction independently and in duplicate using pre-designed forms consistent with those used for the original review. We abstracted data regarding study characteristics, demographic data, intervention and control details, outcome data, and risk of bias (RoB) evaluations using the modified Cochrane RoB tool.21 A third reviewer (A.A.) resolved disagreements.

Risk of bias was classified as “low”, “probably low”, “probably high”, or “high” for the following domains: sequence generation, allocation sequence concealment, blinding, selective outcome reporting, and other bias. We rated the overall RoB as the highest risk attributed to any criterion. We rated the overall certainty in evidence for each outcome using the GRADE framework,22 including the following domains: RoB, imprecision, inconsistency, indirectness, and publication bias. Overall certainty of evidence was “very low”, “low”, “moderate”, or “high”. We considered rating down the certainty of evidence for RoB based on lack of blinding for subjective outcomes. Certainty in evidence was not rated down for indirectness if patients with acute hypoxemic respiratory failure meeting eligibility criteria other than SARS-CoV-2 infection were included. Assessors resolved disagreements regarding RoB and GRADE ratings by discussion.

Data analysis

DerSimonian and Laird random effects models were used to conduct the meta-analyses. All analyses were performed in RevMan 5.3 (Cochrane Collaboration, Oxford). Study weights were generated using the inverse variance method. Dichotomous outcomes were presented as risk ratios, and continuous outcomes were presented as mean differences or standardized mean differences, all with 95% confidence intervals (CIs). We assumed a normal distribution for continuous outcomes and converted interquartile ranges to standard deviations (SD) as per Cochrane Collaboration guidance.23 An online plot digitizer was used to obtain estimates for studies in which continuous outcomes were reported graphically only (plotdigitizer.sourceforge.net).

We assessed for heterogeneity between studies using the χ2 test for homogeneity, the I2 measure, and visual inspection of the forest plots.23 We evaluated inconsistency based on magnitude and direction of heterogeneity. Based on limited yield of numerous subgroup analyses performed in the previous systematic review, we restricted subgroup analysis to high RoB studies vs low RoB studies (hypothesizing that HFNC would be more beneficial in high RoB studies).

Systematic review #2: risk of aerosol generation associated with HFNC

With the assistance of a health information specialist (R.C.) and using a combination of subject headings and keywords related to COVID-19, other coronaviruses, and HFNC, we conducted a comprehensive search of Ovid MEDLINE and Embase from inception to 14 May 2020. We supplemented this with a search in the same databases using a combination of subject headings and keywords related to HFNC, aerosol generation, and infection transmission. We did not limit the search to COVID-19 or coronavirus infections (see Appendix 2). We limited the search to literature published between 1 January 2007 and 14 May 2020. To identify Chinese studies or rapid reviews addressing the research question, we also searched the China National Knowledge Infrastructure (CNKI) and Chinese Medical Journal Network (CMJN) using the same search strategy up to 28 March 2020. To identify eligible pre-prints, we searched medRxiv from inception to 14 May 2020 with search terms related to HFNC, aerosol generation, and droplet dispersion. We did not apply any language or quality restrictions.

We included all comparative and non-comparative studies that evaluated droplet dispersion or aerosolization of viable airborne organisms or transmission of infection associated with HFNC use. Anticipating the paucity of direct evidence from COVID-19 and hospitalized patients, we included all study designs and populations evaluating aerosol generation or dispersion associated with HFNC. We included studies of hospitalized and non-hospitalized patients with or without microbiologically confirmed SARS-CoV-2 infection, simulation studies without human participants, and studies describing dispersion of non-infectious air particles or liquid droplets. We included studies that evaluated the following outcomes: detection of droplets or viable airborne organisms through sample analysis, or documented transmission of infection associated with exposure to infected individuals receiving HFNC, with or without comparison with an alternate ventilation modality.

Paired reviewers (J.B., F.M.) screened all identified citations, conducted full-text review of potentially eligible studies and screened the reference lists of reviews to identify additional eligible studies. A third reviewer (A.A.) resolved disagreements. Paired reviewers (X.Y., N.Y., X.L.) screened citations identified from the CNKI and CMJN and resolved disagreements by discussion.

Paired reviewers (J.B., F.M.) abstracted data (study characteristics, participant characteristics, description of the intervention and control, outcomes, and general limitations in study design and conduct) independently and in duplicate using standardized data abstraction forms. A third reviewer (A.A.) resolved disagreements as necessary.

Informed by GRADE guidance, we assessed the overall certainty of the evidence based on imprecision, indirectness, and inconsistency.24

Given anticipated differences in included study designs, we summarized our findings narratively.

Results

Search results and study characteristics

Of the 2,439 citations identified in our search, 1,814 were screened for eligibility after removing duplicates. Full-text review of 38 potentially eligible studies identified 20 eligible studies. Of these, 17 were systematic reviews (SRs)15,25–40 and three were RCTs41–43 (Fig. 1). We did not identify additional RCTs from reference lists of eligible SRs. Therefore, 12 RCTs with 1,989 patients were included in analyses, including nine RCTs from the original review and three RCTs from the updated search.41–52 No trial directly evaluated HFNC in patients with COVID-19 or other coronavirus infections.

Fig. 1

PRISMA flow diagram for systematic review 1 on efficacy and safety of HFNC in acute hypoxemic respiratory failure. SR = systematic reviews; RCT = randomized-controlled trial

Table 1 summarizes study characteristics. Trials randomized between 14 and 776 patients; two used a crossover design.43,52 Five trials were performed in the ICU,45,46,48,50,52 six were performed in the emergency department (ED),42–44,47,49,51 and one was performed in mixed ICU/ED settings.41 One trial included patients with cardiogenic pulmonary edema only,44 two included immunocompromised patients only,45,46 and one included palliative patients only.43 Criteria for hypoxemia varied, including peripheral oxygen saturation (SpO2) thresholds (primarily < 90–95%), arterial partial pressure of oxygen (PaO2) thresholds (< 55–60 mmHg), P/F ratio (< 300), or a combination of criteria including elevated respiratory rate (most commonly > 22–25/min). All eligible studies initiated gas flows at 35 L·min−1 or higher in the HFNC group with one exception42 that used initial flow rates of 19.5 to 30 L·min−1 (Table 1).

Table 1

Characteristics of included studies for review on HFNC for acute hypoxemic respiratory failure

Study	Country	Number of patients randomized	Population	Intervention details	Comparator details	Duration of follow-up for mortality	Outcomes
Azoulay, 201846	France	778	Inclusion: ICU patients, PaO2 < 60 mmHg or SpO2 <90% on R/A, immune-suppression Exclusion: ↑ CO2, CPE, recent surgery	(Fisher and Paykel Healthcare) Initial settings: Flow: 50 L·min−1 FiO2: 100% Duration: not specified	NP or mask. Initial settings: Flow: set to achieve SpO2 ≥ 95%	28 days	Mortality (primary), need for IMV, escalation, ICU and hospital LOS, comfort and dyspnea
Bell, 201547	Australia	100	Inclusion: ED patients, RR ≥ 25 breaths·min−1, SpO2 ≤ 93% Exclusion: Patients requiring immediate NIV or IMV	(AIRVO2, Optiflow, Fisher & Paykel) Initial settings: Flow: 50 L·min−1 FiO2: 30% Duration: 2 hr	NP or face mask Initial settings: O2 in both groups was titrated over a 2-hr period	N/A	Need for IMV, escalation, comfort
Frat, 201548	France and Belgium	313	Inclusion: ICU patients, ARF with RR > 25 breaths·min−1, PF ≤ 300, on ≥ 10 L·min−1 O2 for ≥ 15 min Exclusion: asthma, chronic lung dz, ↑ CO2, CPE, CV instability, need for IMV	(Optiflow, MR850, Fisher and Paykel Healthcare) Initial settings: Flow: 50 L·min−1 FiO2: 100% Duration: not specified	NRB mask Initial settings: Flow: ≥ 10 L·min−1	90 days	Mortality (primary), need for IMV, escalation, ICU LOS, comfort
Geng, 202041	China	36	Inclusion: ICU, ED, or ward patients, acute asthma, PaO2 < 60 mmHg on R/A with or without ↑ CO2 Exclusion: urgent IMV, CV instability, aLOC, RR > 45 breaths·min−1, pH < 7.30	(AIRVO2, Fisher & Paykel) Initial settings: Flow: 30-40 L·min−1 FiO2: N/A Duration: not specified	NP, Venturi mask, or storage balloon mask Initial settings: Flow: 2–6 L·min−1, set to achieve SpO2 92–96%	N/A	Need for IMV, escalation, hospital LOS
Jones, 201649	New Zealand	322	Inclusion: ED patients, SpO2 ≤ 92% on R/A, RR ≥ 22 breaths·min−1 Exclusion: Urgent NIV or IMV required	(Optiflow, Fisher & Paykel Healthcare) Initial settings: Flow: 40 L·min−1 FiO2: 28% Duration: not specified	Mask or NP Initial settings: N/A	90 days	Mortality, need for IMV, escalation (primary), hospital LOS
Lemiale, 201545	France	102	Inclusion: ICU patients, immune-compromised, > 6 L·min−1 O2 to maintain SpO2 > 95% or respiratory distress Exclusion: ↑ CO2, urgent NIV or IMV required	(Fisher & Paykel Healthcare) Initial settings: Flow: 40–50 L·min−1 FiO2: 100% Duration: 2 hr	Venturi mask Initial settings: Flow: 15 L·min−1 FiO2: 60%	N/A	Need for IMV, escalation (primary), dyspnea, comfort
Makdee, 201744	Thailand	136	Inclusion: ED patients, CPE, SpO2 < 95% on R/A, RR >24 breaths·min−1 Exclusion: Urgent NIV or IMV required, CV instability, RR >35 breaths·min−1, SpO2 < 90%, ESRD	(Optiflow, Fisher & Paykel Healthcare) Initial settings: Flow: 35 L·min−1 FiO2: N/A Duration: 1 hr	NP or NRB Initial settings: N/A	7 days	Mortality, need for IMV, escalation, hospital LOS, dyspnea, comfort
Parke, 201150	New Zealand	60	Inclusion: ICU patients, ≥ 4 L·min−1 O2 via NP for > 4 hr or ≥ 6 L·min−1 O2 via a face mask for > 2 hr and/or RR ≥ 25 breaths·min−1 and/or ↑ WOB Exclusion: Urgent NIV or IMV required	(Optiflow, Fisher & Paykel Healthcare) Initial settings: Flow: 35 L·min−1 FiO2: N/A Duration: not specified	Face mask Initial settings: N/A	N/A	Escalation
Raeisi, 201942	Iran	40	Inclusion: ED or ward patients, moderate to severe asthma exacerbation Exclusion: Pregnancy with history of smoking and occupational asthma, ↑ CO2, Infiltrates on chest x-ray	Device not specified Initial settings: Flow: 19.5–30 L·min−1 FiO2: N/A Duration: not specified	NP Initial settings: Flow: 2–5 L·min−1	N/A	Dyspnea
Rittayamai, 201551	Thailand	40	Inclusion: ED patients, RR > 24 breaths·min−1, SpO2 < 94% on R/A Exclusion: Need for IMV, CV instability, CRF	(Optiflow, Fisher & Paykel Healthcare) Initial settings: Flow: 35 L·min−1 FiO2: N/A Duration: 1 hr	NP or NRB Initial settings: N/A	N/A	Need for IMV, escalation, dyspnea (primary), comfort
Ruangsomboon, 201943	Thailand	48	Inclusion: ED palliative patients, SpO2 < 90%, RR ≥ 30 breaths·min−1, dyspneic Exclusion: aLOC, unable to communicate, positive pressure devices contraindicated	(AIRVO2, Optiflow, Fisher & Paykel) Initial settings: Flow: 35 L·min−1 FiO2: N/A Duration: 1 hr	NP or NRB Initial settings: Set to achieve SpO2 > 95%	N/A	Dyspnea
Schwabbauer, 201452 (crossover)	Germany	14	Inclusion: ICU patients, PaO2 < 55 mmHg on R/A Exclusion: CPE, CV instability	(OptiFlow, Fisher & Paykel Healthcare) Initial settings: Flow: 55 L·min−1 FiO2: 60% Duration: 30 min	Venturi mask Initial settings: Flow: 15 L·min−1 FiO2: 60%	N/A	Dyspnea and comfort

aLOC = altered level of consciousness; ARF = acute respiratory failure; CPE = cardiogenic pulmonary edema; CRF = chronic respiratory failure; CV = cardiovascular; Dz = disease; ESRD = end stage renal disease; ED = emergency department; ESRD = end-stage renal disease; ICU = intensive care unit; IMV = invasive mechanical ventilation; LOS = length of stay; NIV = non-invasive ventilation; NP = nasal prongs; NRB = non-rebreathe mask; PaO2 = partial pressure of oxygen; PF = PaO2:FO2 ratio; R/A = room air; RR = respiratory rate; SpO2 = peripheral oxygen saturation; WOB = work of breathing

Initial settings: Flow: 50 L·min−1 FO2: 100%

Flow: 50 L·min−1 FO2: 30%

FO2: 100%

FO2: N/A

FO2: 28%

FO2: 100%

FO2: 60%

FO2: N/A

FO2: N/A

FO2: N/A

FO2: N/A

FO2: N/A

FO2: 60%

FO2: 60%

Quality assessment

No RCT was blinded. Most were judged to be at low RoB for random sequence generation, allocation concealment, incomplete data, selective reporting, and other sources of bias. Apart from blinding, seven41,43,44,46,48,50,52 of 12 included trials were deemed to be at low overall RoB (Table 2).

Table 2

Individual study risk of bias for review on HFNC for acute hypoxemic respiratory failure

	Random sequence generation	Allocation concealment	Blinding	Incomplete outcome data	Selective reporting	Other bias	Overall RoB
Azoulay, 201846	Low	Low	High	Low	Low	Low	Low
Bell, 201547	Low	Low	High	High	Low	Low	High
Frat, 201548	Low	Low	High	Low	Low	Low	Low
Geng, 202041	Low	Low	High	Low	Low	Low	Low
Jones, 201649	Low	Low	High	High	Low	Low	High
Lemiale, 201545	Probably low	Low	High	Low	Probably high	Low	High
Makdee, 201744	Probably low	Low	High	Low	Low	Probably low	Low
Parke, 201150	Low	Low	High	Probably low	Probably low	Low	Low
Raeisi, 201942	Probably low	Probably high	High	High	High	Low	High
Rittayamai, 201551	Probably low	Low	High	High	Probably low	Low	High
Ruangsomboon, 201943	Low	Low	High	Probably low	Low	Low	Low
Schwabbauer, 201452	Probably low	Probably low	High	Probably low	Probably low	Low	Low

HFNC = high-flow nasal cannula; RoB = risk of bias

Outcomes

Table 3 presents the GRADE summary of findings for all pre-specified outcomes except treatment-related complications (summarized below), with anticipated effects of HFNC and evidence certainty when applied to patients with acute respiratory failure.

Table 3

Summary of findings table for review on HFNC for acute hypoxemic respiratory failure

Certainty assessment							№ of patients		Effect		Certainty	Importance
№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	HFNC	Usual care	Relative(95% CI)	Absolute(95% CI)
Mortality (follow-up: range 7 days to 90 days; assessed with longest available)
4a	Randomized trials	Not serious	Not serious	Not serious	Seriousb	None	187/722 (25.9%)	186/685 (27.2%)	RR, 0.94 (0.67 to 1.31)	16 fewer per 1,000 (from 90 fewer to 84 more)	⊕⊕⊕ MODERATE	CRITICAL
Invasive ventilation (follow-up range: 2 days to 28 days)
8c	Randomized trials	Seriousd	Not serious	Not serious	Seriouse	None	206/862 (23.9%)	236/825 (28.6%)	RR, 0.85 (0.74 to 0.99)	44 fewer per 1,000 (from 76 fewer to 3 fewer)	⊕⊕ LOW	CRITICAL
Escalation of therapy (follow-up: range 2 days to 28 days; assessed with any escalation to HFNC, NIV, or invasive ventilation)
8f	Randomized trials	Seriousd	Not serious	Not serious	Seriouse	None	219/871 (25.1%)	266/832 (32.0%)	RR, 0.71 (0.51 to 0.98)	93 fewer per 1,000 (from 157 fewer to 6 fewer)	⊕⊕ LOW	CRITICAL
ICU length of stay (assessed with days)
2g	Randomized trials	Not serious	Serioush	Not serious	Seriousi	None	494	482	–	MD, 1.38 days fewer (0.90 fewer to 3.66 more)	⊕⊕ LOW	CRITICAL
Hospital length of stay (assessed with days)
4j	Randomized trials	Not serious	Not serious	Not serious	Seriousi	None	636	611	–	MD, 0.67 days fewer (1.41 fewer to 0.08 more)	⊕⊕⊕ MODERATE	CRITICAL
Patient-reported dyspnea (assessed with variable score)
7k	Randomized trials	Seriousl	Not Seriousm	Not Serious	Seriouse	None	458	436	–	SMD, 1.17 SD lower (2.60 lower to 0.25 higher)	⊕⊕ LOW	CRITICAL
Patient-reported comfort (assessed with variable score)
7n	Randomized trials	Seriousl	Seriouso	Not serious	Seriousb	None	624	607	–	SMD, 0.12 SD lower (0.61 lower to 0.37 higher)	⊕ VERY LOW	CRITICAL

CI = confidence interval; ICU = intensive care unit; HFNC = high-flow nasal cannula; RR = risk ratio; MD = mean difference; SD = standard deviation; SMD = standardized mean difference

Explanations

aAzoulay 201846, Frat 201548, Jones 201649, Makdee 201744

bAlthough point estimate suggests no effect, confidence intervals do not exclude important benefit and important harm

cAzoulay 201846, Bell 201547, Frat 201548, Geng 202041, Jones 201649, Lemiale 201545, Makdee 201744, Rittayamai 201551

dNone of the included trials were at low risk of bias for blinding and decision to escalate therapy or intubate may be subjective

eUpper end of 95% confidence interval does not exclude no effect

fAzoulay 201846, Bell 201547, Frat 201548, Jones 201649, Lemiale 201545, Makdee 201744, Parke 201150, Rittayamai 201551

gAzoulay 201846, Frat 201548

hHigh I2 and of two studies reporting this outcome, results are discrepant

iLower end of the 95% confidence interval does not exclude benefit with HFNC

jAzoulay 201846, Geng 202041, Jones 201649, Makdee 201744

kAzoulay 201846, Lemiale 201545, Makdee 201744, Raeisi 201942, Rittayamai 201551, Ruangsomboom 201943, Schwabbauer 201452

lSubjective outcome in unblinded trials. Also other risk of bias issues in the trials reporting this outcome

mHigh I2 however, studies showed benefit with HFNC so did not downgrade

nAzoulay 201846, Bell 201547, Frat 201548, Lemiale 201545, Makdee 201744, Rittayamai 201551, Schwabbauer 201452

oHigh I2 with variable effect across included studies

⊕⊕⊕

⊕⊕

⊕⊕

⊕⊕

⊕⊕⊕

⊕⊕

⊕

The use of HFNC may reduce the need for invasive ventilation compared with COT (eight RCTs; relative risk [RR], 0.85; 95% CI, 0.74 to 0.99; risk difference [RD], 4.4%; 95% CI, 0.3 to 7.6; number needed to treat [NNT], 23; 95% CI, 13 to 333; low certainty, rated down for RoB and imprecision; I2, 0%) (Fig. 2). There was no credible subgroup effect comparing high vs low RoB studies (Fig. 2).

Fig. 2

Need for invasive ventilation forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula; RoB = risk of bias

The use of HFNC may also reduce the need for escalation of therapy (i.e., other NIV or intubation) compared with COT (eight RCTs; RR, 0.71; 95% CI, 0.51 to 0.98; RD, 9.3%; 95% CI, 0.6 to 15.7; NNT, 11; 95% CI, 6 to 167; low certainty, rated down for RoB and imprecision; I2, 52%) (Fig. 2). There was no credible subgroup effect based on individual study RoB (Fig. 3). Results provided no support for differences in mortality (moderate certainty), in-hospital or intensive care LOS (moderate and low certainty, respectively), and patient-reported dyspnea or comfort (low and very low certainty, respectively) (Table 3, Figs. 4, 5, 6, 7, 8).

Fig. 3

Escalation of therapy forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula; RoB = risk of bias

Fig. 4

Mortality forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula

Fig. 5

Intensive care unit length of stay forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula

Fig. 6

Hospital length of stay forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula

Fig. 7

Patient-reported dyspnea forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula

Fig. 8

Patient-reported comfort forest plot for review on HFNC for acute hypoxemic respiratory failure. CI = confidence interval; HFNC = high-flow nasal cannula

Eligible studies reported treatment-related complications variably, precluding pooled analyses. Among reported complications with HFNC, thoraco-cervical discomfort, heat-related discomfort, and mild altered level of consciousness were most common. One trial reported serious complications, including cardiac dysrhythmias, septic shock, cardio-respiratory arrest, and nosocomial pneumonias; the incidence of these complications were either similar or lower than HFNC compared with COT.48 Studies generally did not suggest a significantly increased risk of complications with HFNC compared with COT (Table 4).

Table 4

Complications from included studies for systematic review 1 on HFNC for acute hypoxemic respiratory failure

	HFNC	Standard O2
Makdee, 201744	n = 63	n = 65
Thoracic and cervical discomfort	2	0
Feeling hot	4	0
Jones, 201649	n = 165	n = 138
Apnea	0	1
Drop in GCS of 2 or more points	1	6
Fall in GCS due to CO2 retention	0	3
Raeisi, 201942	n = 20	n = 20
Device-induced heat	2	0
Nasal irritation	2	0
Refractory asthma/hypoxia	0	1
Rittayamai, 201551	n = 20	n = 20
Unpleasant smell	1	0
Temperature too warm	1	0
Chest discomfort	1	0
Ruangsomboon, 201943	n = 44	n = 44
Discomfort	5	0
Feeling hot	2	0
Could not tolerate HFNC	1	0
Frat, 201548	n = 106	n = 94
Cardiac dysrhythmia	11	16
Septic shock	19	26
Cardio-respiratory arrest	5	7
Nosocomial pneumonia	4	8
Azoulay, 201846	n = 388	n = 388
ICU-acquired infection	39	41

Makdee 2017 included aspiration and nasal ulceration but no events occurred in either group

Jones 2016 included pneumothorax, subcutaneous emphysema, and nasal pressure sore but no events occurred in either group

Bell 2015 reported that no adverse events occurred in either group

ICU = intensive care unit; GCS = Glasgow Coma Scale; HFNC = high-flow nasal cannula

We identified 3,523 unique citations using our electronic searches, 26 pre-prints from medRxiv, and one additional citation suggested by an expert panelist. We completed full-text review of 33 potentially eligible studies and included six studies,53–58 and identified one additional study through reference list screening,59 for a total of seven eligible studies (Fig. 9).53–59

Fig. 9

PRISMA flow diagram for systematic review 2 on aerosol generation associated with HFNC. SR = systematic reviews; RCT = randomized-controlled trial

Of the seven eligible studies, six were simulation studies53–55,57–59 and one was a crossover study.56 No studies directly evaluated risk of aerosol generation or infection transmission associated with HFNC use among patients with COVID-19. Three simulation studies included healthy adult volunteers,54,58,59 and three included a model patient simulator.53,55,57 The crossover study included 19 critically ill adult patients who received supplemental oxygen therapy and crossed over to HFNC.56 Three studies53,58,59 evaluated HFNC at 30 L·min−1, one evaluated HFNC at 40 L·min−1,57 and six studies53–56,58,59 evaluated HFNC at 60 L·min−1. One study compared HFNC with continuous positive airway pressure (CPAP) delivering pressures of 5–20 cmH2O,53 another compared HFNC with COT by face mask,56 two compared HFNC with COT by nasal prongs at 6 L·min−1,57,58 and one compared HFNC with non-rebreather mask with non-humidified air at 15 L·min−1.58 The remaining three studies54,55,59 did not include an alternative oxygen administration or ventilatory support strategy as a comparator. Study outcomes included the number, diameter, evaporation rates, and velocity of exhaled aerosols,58,59 regions of high aerosol density,53 droplet dispersion distance,54,55,57 and microbial colony counts in air and surface samples (Table 5).56

Table 5

Characteristics of included studies for systematic review 2 on aerosol generation associated with HFNC

Study	n	Sample	Intervention/control	Outcome	Results
Roberts, 201559	N/A	Healthy adults	HFNC at 30 and 60 L·min−1 compared with no HFNC at rest and during violent exhalation	Aerosol dispersion of particles 25–250 µm in diameter	HFNC did not increase aerosol dispersion above risk of typical breathing with violent exhalation With and without HFNC, aerosols 25–250 µm travelled up to 4.4 m and remained airborne for up to 43 sec
Hui, 201953	N/A	Patient simulator	HFNC at 10–60 L·min−1 to CPAP at 5–20 cm H2O	Regions of high exhaled aerosol density following injection of smoke into simulator bronchus	Increased regions of high aerosol density were noted with increasing flow rates (maximum dimension 6.5 ± 1.5 cm at 10 L·min−1 to 17.2 ± 3.3 cm at 60 L·min−1; P < 0.001) and increasing positive pressure using CPAP.
Leung, 201856	19	Critically ill patients with Gram-negative pneumonia	HFNC at 60 L·min−1 to O2 mask	Cough-generated droplet dispersion based on degree of environmental bacterial contamination	No difference in GNB count between HFNC and O2 mask for air samples, settle plates at 0.4 m or 1.5 m (P = 0.119–0.500)
Kotoda, 201955	N/A	Mannequin simulator	HFNC at 60 L·min−1 compared with HFNC at 0 L·min−1	Droplet dispersion determined by measuring distance of water on water-sensitive paper and dispersion of live yeast	Water and yeast colony formation were detected on sheet placed at 30 cm from mannequin’s face (3.7 ± 1.2 spots and 2.3 ± 0.5 yeast CFU) during use of HFNC.
Loh, 202054	5	Healthy adults	HFNC at 60 L·min−1 compared with no HFNC	Cough-generated droplet dispersion determined by measuring distance of food colouring droplet	Similar droplet dispersion distance (2.91 ± 1.09 m) with HFNC compared with no HFNC (2.48 ± 1.03 m). Highest cough-generated droplet dispersion distances with simulated coughs were 450 cm and 390 cm with and without HFNC, respectively.
Leonard, 202057	N/A	In silico simulator	HFNC at 40 L·min−1 compared with nasal prongs at 6 L·min−1 and spontaneous breathing (all with face mask)	Intentional mask leak, droplet capture by face mask, droplet dispersion from point of origin.	Greater leak (16.5%) with HFNC compared with nasal prongs (12.6%) and spontaneous breathing (11.6%). Droplets captured by face mask were variable with HFNC (85.9%), nasal prongs (75.9%) and spontaneous breathing (89.9%). Variable proportions of escaped particles travelled greater than 1 metre from point of origin with HFNC (15.9%) compared with nasal prongs (6.9%) and spontaneous breathing (31%).
Iwashyna, 202058	4	Healthy adults	HFNC at 30 L·min−1 and 60 L·min−1 compared with nasal cannula 6 L·min−1 and non-rebreather mask (non-humidified) at 15 L·min−1.	Aerosol levels of particles 10–500 nm in size with spontaneous breathing and intentional coughing, measured at bed rail beside patient’s head and 10 cm from patient’s mouth	Similar aerosol levels with HFNC at 30 L·min−1 and 60 L·min−1, nasal prongs at 6 L·min−1, 15 L·min−1 non-rebreather mask and spontaneous breathing at room air. Similar particle number concentration (across all particle sizes) with HFNC compared with other interventions.

CFU = colony forming units; CPAPs = continuous positive airway pressure, GNB = gram-negative bacteria; HFNC = high-flow nasal cannula

Study findings

Exhaled aerosol dispersion

Using a human patient simulator programmed to different severities of lung injury in a negative pressure room with 20+ breathing cycles at every given flow rate, Hui et al.53 compared CPAP via nasal pillows and oronasal mask, delivering pressures between 5 and 20 cmH2O to humidified HFNC with flow rates of 10, 30, and 60 L·min−1. Under normal lung conditions, increased HFNC flow rates were associated with a larger distance of high aerosol density (maximum dimension 6.5 ± 1.5 cm at 10 L·min−1 to 17.2 ± 3.3 cm at 60 L·min−1; P < 0.001). Similar, though smaller, increases were noted in simulated mild (4.3 ± 1.0 cm at 10 L·min−1 to 7.2 ± 1.8 cm at 60 L·min−1) and severe lung disease (3.0 ± 0.8 cm at 10 L·min−1 to 4.8 ± 1.6 cm at 60 L·min−1). The region was not uniform, with negligible lateral extension with a well-fitted, well-positioned cannula, although the lateral distance with the cannula loosely positioned in the nose was 62 cm.53

Roberts et al.59 conducted a simulation study including healthy adult volunteers. They compared dispersion of exhaled aerosols with and without nasal HFNC at 30 and 60 L·min−1 during two “violent” (snorting) exhalations and at rest using imaging (number of simulations and imaging methods not described). During violent exhalation, there was less dispersion with HFNC than without, though flow rates of 60 L·min−1 were associated with greater dispersion than flow rates of 30 L·min−1. With and without HFNC, 25–250-µm aerosols travelled up to 4.4 m and remained airborne for up to 43 sec (unclear if at rest or with violent exhalation). The authors concluded that HFNC did not increase the risk of aerosol dispersion more than typical patient breathing with violent exhalation.59

Exhaled aerosol production

Iwashyna et al.58 conducted a simulation study of four healthy adult volunteers. They evaluated variations in exhaled aerosol production with spontaneous breathing with intentional coughing, comparing HFNC at 30 L·min−1 and 60 L·min−1, nasal cannula at 6 L·min−1, and non-rebreather mask with non-humidified air at 15 L·min−1. The study was conducted in a simulated single occupancy hospital room with all equipment, monitors, and computers standard to this setting. Investigators wore standard surgical masks. Measurements were taken in two positions: 10 cm from the simulated patient’s mouth, and attached to a bed rail next to the head. Similar aerosol production levels and particle number concentrations were found with both flow rates of HFNC compared with nasal prongs, non-rebreather mask, and spontaneous breathing.58

Exhaled droplet dispersion distance

Loh et al.54 evaluated cough-generated droplet dispersion distance with two coughs per participant across five healthy volunteers, using gargled water containing coloured dye. The study found similar mean (SD) maximum cough-generated droplet dispersion distances at baseline [248 (103) cm] and with application of HFNC at 60 L·min−1 [291 (109) cm]. Highest cough-generated droplet dispersion distances with simulated coughs were 450 cm and 390 cm with and without application of HFNC, respectively.54

Kotoda et al.55 evaluated thickened liquid dispersion distance with and without HFNC at 60 L·min−1 using an experimental mannequin model with three simulations using water. Water dispersion was detected using 18 sheets of water-sensitive paper positioned at 30 cm intervals. Water was only detected on the first sheet (30 cm) from the mannequin’s face with a mean (SD) of 3.7 (1.2) spots. Manual repositioning of the cannula led to a statistically significant increase in liquid dispersion across the first three sheets (P = 0.0032).55

Leonard et al.57 conducted an in silico simulation using a three-dimensional head, comparing intentional mask leak, droplet capture by face mask, and dispersion from point of origin with HFNC at 40 L·min−1, nasal prongs at 6 L·min−1, and tidal breathing. A level-1 surgical mask was placed over the face for all interventions. The proportion of escaped particles while wearing a surgical face mask that travelled greater than 1 m were higher with HFNC (15.9%) compared with nasal prongs (6.9%), though lower than tidal breathing without a mask (31%). There were significant reductions in exhaled gas flow velocities and particle dispersion with a surgical face mask in place, although there was both greater mask leak and droplet capture by face mask with HFNC compared with nasal prongs. In comparison, tidal breathing had lower mask leak and higher droplet capture.57

Dispersion of viable organisms

Kotoda et al.55 conducted three simulations using fresh yeast (Sacchromyces cerevisiae), evaluating dispersion with and without HFNC at 60 L·min−1 in an experimental mannequin model. Yeast dispersion was evaluated using 18 Petri dishes placed at 30-cm intervals from the mannequin and four dishes placed 5 m away. Colonies were only detected in the closest dish with a mean (SD) of 2.3 (0.5) colony forming units, and there was increased dispersion extending to two dishes in front of and lateral to the mannequin with manual repositioning of the cannula (P = 0.039). The investigators did not observe colony formation on the dishes 5 m away from the mannequin.55

Leung et al.56 conducted a prospective study of 19 critically ill adults receiving COT because of gram-negative bacterial pneumonia. They evaluated the degree of environmental bacterial contamination with HFNC vs simple face mask oxygen. The study measured airborne and surface contaminants using an Andersen-type impactor air sampler and Petri dishes, respectively. No significant differences were found in gram-negative bacterial counts between HFNC and simple face mask oxygen in air samples, Petri dishes at 0.4 m (bedside rails) or 1.5 m (longest distance consistently achievable in the room) from the patient’s nose, or with different air changes per hour (P = 0.119 to 0.500 across comparisons).56

There was concern for substantial RoB in design and conduct across all seven studies. Available evidence was significantly limited by small sample sizes with healthy volunteers or simulations, and in the absence of any studies directly including COVID-19 patients or evaluating aerosolization of similar microbes, by indirectness in applying findings to SARS-CoV-2 aerosolization and COVID-19 management. Based on GRADE guidance, there was very low certainty in estimates due to inconsistency in the magnitude and direction of the association between HFNC and aerosol and droplet dispersion across studies, as well as indirectness and imprecision.

Discussion

Our SRs—neither of which identified studies with direct evidence on COVID-19—provide limited but nevertheless the best current synthesis of the evidence on the benefits, harms, and risks of SARS-CoV-2 transmission through HFNC. Whereas HFNC applied in acute hypoxemic respiratory failure may substantially reduce the need for invasive ventilation and escalation of therapy to other NIV or intubation, we found no apparent differences in mortality, ICU/hospital LOS, patient-reported dyspnea and comfort, or differences in treatment-related complications. In the second systematic review on aerosol generation associated with HFNC, we found no studies directly related to COVID-19. Very low certainty experimental and observational data suggested mixed findings in terms of significant droplet dispersion and aerosol generation with HFNC.

Our findings bear direct relevance for all countries and healthcare systems and hospitals affected by the COVID-19 pandemic, of which many are now forced to consider the use of HFNC in patients with acute respiratory failure due to COVID-19 in the face of limited access to invasive ventilation strategies.

The studies identified by our search do not provide data that can be extrapolated to the risk of airborne transmission of SARS-CoV-2. Among included studies, four examined dispersion of particles of droplet or larger size.54,55,57,59 Two studies were unable to show dispersion of live bacteria and yeast to a distance compatible with airborne dispersion55,56; however, this may reflect an inability of these organisms to survive the process of dehydration and rehydration,18 whereas SARS-CoV2 is known to survive aerosolization.60 One study identified a smaller region of high aerosol density around HFNC than nasal CPAP but did not quantify the total amount of aerosol generated by HFNC53; another showed comparable aerosol production levels with HFNC compared with COT strategies at distances close to the head, but the testing environment included multiple potential sources of aerosol generation, which may have obscured any increase due to HFNC.58

In terms of droplet dispersion, one study54 showed coughing while receiving HFNC may result in the dispersion of droplets further than 2 metres (i.e., beyond the distance typically considered the extent of droplet dispersion),61 suggesting that the area around a patient in which droplet precautions are applied may need to be increased when HFNC is used. The applicability of findings among healthy adults during forceful exhalation to critically ill patients is uncertain.

The burden of COVID-related respiratory failure is straining ICU resources,62 and anecdotal evidence suggests mechanical ventilators may be insufficient for the patients that require them. In this context, a significant reduction in the need for invasive ventilation with HFNC may be of substantial benefit. Against this benefit, however, is the unknown risk of nosocomial transmission of SARS-CoV-2, and therefore any strong recommendation regarding the use of HFNC is clearly inappropriate. Instead, decisions should be context specific, taking into account the availability of invasive ventilation and the presence of other factors that decrease the risk of infection transmission. These include adequate room ventilation, limiting healthcare personnel exposure to the patient, viral load, and use of high-filtration fit-tested respirators (e.g., N95, FFP2) for healthcare workers.63 Use of a surgical face mask on patients receiving HFNC may also provide benefit.57 Ongoing field experiments and clinical studies during the current pandemic may provide additional information.

The risks and benefits of HFNC must also be balanced against the risks and benefits of alternatives, when available. A recent guideline made no recommendation regarding use of NIV for de novo hypoxemic respiratory failure or pandemic viral illness.64 Both NIV and invasive ventilation were associated with nosocomial transmission of SARS,65,66 although some simulation data suggest that NIV is not an aerosol-generating procedure.67 Therefore, optimal management may differ across settings, depending on the availability of ventilators and other NIV modalities.

The strengths of the first of our two SRs reported herein include a comprehensive literature search with the inclusion of the most recent trial evidence. We systematically and transparently assessed the certainty and relevance of the identified evidence through the use of GRADE. Inherent limitations in the available evidence include lack of sufficient data to explore certain subgroup effects, imprecision, and high RoB due to lack of blinding. No eligible RCTs included COVID-19 patients; however, we did not rate down for indirectness, given the likelihood that similar principles of management are applicable to COVID-19 patients with acute hypoxemic respiratory failure. Despite clinical heterogeneity in study populations and definitions of hypoxemia, most outcomes did not show statistical heterogeneity (i.e., consistency). One trial46 contributed approximately one-third of the data for most major outcomes and may affect generalizability of the findings. Finally, because of reporting variability, treatment-related complications could not be pooled for quantitative analyses.

Strengths of the second systematic review include a comprehensive literature search incorporating English and Chinese studies and pre-prints, and inclusion of clinical expert input regarding aerosol generation and HFNC. Limitations include substantial RoB issues, indirectness in applying findings to SARS-CoV-2 aerosolization and COVID-19 management, imprecision with limited studies involving small samples of healthy individuals or simulations, and inconsistent experimental conditions and effects observed across studies.

Conclusions

We found that HFNC applied to patients with respiratory failure may substantially reduce the need for invasive ventilation and escalation of therapy to NIV or intubation (low certainty), with no apparent effect on mortality or patient-reported symptoms. Complications of therapy were comparable to COT modalities. Very low certainty evidence showed uncertain findings with regards to droplet dispersion and aerosol generation with HFNC. No direct evidence applicable to COVID-19 was available for either efficacy or infection-related risks.

Taken together, the benefits of HFNC in the face of the COVID-19 pandemic must be carefully balanced against the unknown risk of airborne transmission of infection to healthcare workers and other patients. As a result, and until further data specific to COVID are available, guidance and subsequent care decisions will need to be based on the specific context, including considerations around availability of personal protective equipment, a safe environment for HFNC delivery, ventilator resources, and individual patient values and preferences. Studies of COVID-19 with application of HFNC to appropriate patients are required to adequately assess this risk of infection transmission using viral samplers, reverse transcriptase polymerase chain reaction testing, and viral cultures.