Scolaris Content Display Scolaris Content Display

Non‐invasive positive pressure ventilation (CPAP or bilevel NPPV) for cardiogenic pulmonary oedema

Collapse all Expand all

Background

Non‐invasive positive pressure ventilation (NPPV) has been used to treat respiratory distress due to acute cardiogenic pulmonary oedema (ACPE). We performed a systematic review and meta‐analysis update on NPPV for adults presenting with ACPE.

Objectives

To evaluate the safety and effectiveness of NPPV compared to standard medical care (SMC) for adults with ACPE. The primary outcome was hospital mortality. Important secondary outcomes were endotracheal intubation, treatment intolerance, hospital and intensive care unit length of stay, rates of acute myocardial infarction, and adverse event rates.

Search methods

We searched CENTRAL (CRS Web, 20 September 2018), MEDLINE (Ovid, 1946 to 19 September 2018), Embase (Ovid, 1974 to 19 September 2018), CINAHL Plus (EBSCO, 1937 to 19 September 2018), LILACS, WHO ICTRP, and clinicaltrials.gov. We also reviewed reference lists of included studies. We applied no language restrictions.

Selection criteria

We included blinded or unblinded randomised controlled trials in adults with ACPE. Participants had to be randomised to NPPV (continuous positive airway pressure (CPAP) or bilevel NPPV) plus standard medical care (SMC) compared with SMC alone.

Data collection and analysis

Two review authors independently screened and selected articles for inclusion. We extracted data with a standardised data collection form. We evaluated the risks of bias of each study using the Cochrane 'Risk of bias' tool. We assessed evidence quality for each outcome using the GRADE recommendations.

Main results

We included 24 studies (2664 participants) of adult participants (older than 18 years of age) with respiratory distress due to ACPE, not requiring immediate mechanical ventilation. People with ACPE presented either to an Emergency Department or were inpatients. ACPE treatment was provided in an intensive care or Emergency Department setting. There was a median follow‐up of 13 days for hospital mortality, one day for endotracheal intubation, and three days for acute myocardial infarction. Compared with SMC, NPPV may reduce hospital mortality (risk ratio (RR) 0.65, 95% confidence interval (CI) 0.51 to 0.82; participants = 2484; studies = 21; I2 = 6%; low quality of evidence) with a number needed to treat for an additional beneficial outcome (NNTB) of 17 (NNTB 12 to 32). NPPV probably reduces endotracheal intubation rates (RR 0.49, 95% CI 0.38 to 0.62; participants = 2449; studies = 20; I2 = 0%; moderate quality of evidence) with a NNTB of 13 (NNTB 11 to 18). There is probably little or no difference in acute myocardial infarction (AMI) incidence with NPPV compared to SMC for ACPE (RR 1.03, 95% CI 0.91 to 1.16; participants = 1313; studies = 5; I2 = 0%; moderate quality of evidence). We are uncertain as to whether NPPV increases hospital length of stay (mean difference (MD) −0.31 days, 95% CI −1.23 to 0.61; participants = 1714; studies = 11; I2 = 55%; very low quality of evidence). Adverse events were generally similar between NPPV and SMC groups, but evidence was of low quality.

Authors' conclusions

Our review provides support for continued clinical application of NPPV for ACPE, to improve outcomes such as hospital mortality and intubation rates. NPPV is a safe intervention with similar adverse event rates to SMC alone. Additional research is needed to determine if specific subgroups of people with ACPE have greater benefit of NPPV compared to SMC. Future research should explore the benefit of NPPV for ACPE patients with hypercapnia.

PICOs

Population
Intervention
Comparison
Outcome

The PICO model is widely used and taught in evidence-based health care as a strategy for formulating questions and search strategies and for characterizing clinical studies or meta-analyses. PICO stands for four different potential components of a clinical question: Patient, Population or Problem; Intervention; Comparison; Outcome.

See more on using PICO in the Cochrane Handbook.

A breathing intervention for shortness of breath due to heart failure

Background

Heart failure is one of the leading causes of hospital admission in the world. People with heart failure often experience shortness of breath and leg swelling. These symptoms may develop over hours to weeks, or rapidly over a few hours. Such rapid deterioration is called acute cardiogenic pulmonary oedema.

Providing air under pressure through a face or nose mask can treat shortness of breath. This treatment is called non‐invasive ventilation and its use in heart failure is controversial.

Study characteristics

Randomised controlled studies compare treatments to find out if they are truly effective. We searched for randomised studies comparing non‐invasive ventilation to routine care for adults with acute cardiogenic pulmonary oedema. We compared studies treating people with non‐invasive ventilation versus medical care. Medical care includes therapies such as providing extra oxygen and water pills to patients. The evidence upon which this review is based is current to September 2018.

Review question

We sought to address if non‐invasive ventilation in adults with acute cardiogenic pulmonary oedema reduces rates of deaths, the need for a breathing tube, and heart attacks.

Key results and quality of evidence

We found 24 studies with 2664 participants comparing non‐invasive ventilation to medical care alone. Non‐invasive ventilation may decrease the chances of dying in hospital. The quality of results for studies reporting death in hospital was low. Studies were poorly conducted, and results were not similar across studies. In addition, non‐invasive ventilation probably reduces the chances of needing a breathing tube. The quality of results for studies reporting breathing tube rates was moderate. Studies evaluating breathing tube rates were poorly conducted. Non‐invasive ventilation probably has little or no effect on getting a heart attack. The quality of results for studies reporting heart attack rates was moderate, and studies had inconsistent results for this outcome. We are unsure if the length of hospital stay is improved with non‐invasive ventilation. The quality of results for studies reporting hospital length of stay was very low, which was due to poor study conduct and inconsistent results. Finally, non‐invasive ventilation may make little or no difference to adverse events (complications), compared to medical care. The quality of results for studies reporting adverse events was low. Studies evaluating adverse events were poorly conducted and had inconsistent results.

Authors' conclusions

Implications for practice

Meta‐analysis of the best available evidence for NPPV versus SMC to treat ACPE demonstrates that NPPV may reduce hospital mortality and probably reduces ETI. NPPV probably has little or no influence on rates of AMI compared to SMC. We are uncertain whether NPPV reduces hospital length of stay compared to SMC. When considering the type of NPPV, the included studies demonstrated no significant differences between CPAP or bilevel NPPV compared to SMC. Thus, it is reasonable to consider the initiation of NPPV in people with ACPE who are already receiving optimal SMC. Limitations in the current evidence did not allow us to identify patient subgroups (e.g., eucapnic people or people treated in the ER or ICU) that derived more (or less) benefit from the initiation of NPPV.

Implications for research

Current evidence has allowed us to draw conclusions about the efficacy of NPPV (versus SMC) for treating ACPE. Nevertheless, certain shortcomings of the included studies have left some clinical questions unanswered. First, the optimal dosing of diuretics as part of SMC was not clear within the included studies. It is possible that the benefits of NPPV identified herein may be different (e.g. reduced or null), based on the clinical experience of a clinician providing SMC for ACPE. Second, it remains to be seen if certain types of people with ACPE (e.g. those with more severe ACC/AHA heart failure grade) would derive more benefit from the initiation of NPPV (in addition to SMC). Future studies should be judicious about explicitly defining intubation criteria, characterising the aetiology of ACPE, the underlying severity of heart failure, and guideline adherence for home heart‐failure therapies prior to admission. Finally, individual patient‐level data for blood gases with prespecified subgroups for hypercapnia or eucapnia can better characterise whether NPPV influences clinically important outcomes to a greater degree in people with hypercapnia compared to eucapnia.

Summary of findings

Open in table viewer
Summary of findings for the main comparison. NPPV compared to SMC for cardiogenic pulmonary oedema

NPPV compared to standard medical care for cardiogenic pulmonary oedema

Patient or population: People with cardiogenic pulmonary oedema
Setting: Pre‐hospital intensive care, emergency department, coronary care unit, or intensive care unit
Intervention: NPPV
Comparison: Standard medical care (SMC)

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with SMC

Risk with NPPV

HOSPITAL MORTALITY
follow‐up: median 13 days; range 1 day ‐ 41 days

Study population

RR 0.65
(0.51 to 0.82)

2484
(21 RCTs)

⊕⊕⊝⊝
LOWa,b

176 per 1000

114 per 1000
(90 to 144)

ETI RATE
follow‐up: median 1 day;

range 0.1 day ‐ 30 days

Study population

RR 0.49
(0.38 to 0.62)

2449
(20 RCTs)

⊕⊕⊕⊝
MODERATEc

154 per 1000

75 per 1000
(58 to 95)

ACUTE MI INCIDENCE
follow‐up: median 3 days; range 1 day ‐ 41 days

Study population

RR 1.03
(0.91 to 1.16)

1313
(5 RCTs)

⊕⊕⊕⊝
MODERATEd

421 per 1000

433 per 1000
(383 to 488)

HOSPITAL LENGTH OF STAY

The mean HOSPITAL LENGTH OF STAY was 9.65 days

MD 0.31 days lower
(1.23 lower to 0.61 higher)

1714
(11 RCTs)

⊕⊝⊝⊝
VERY LOWe,f,g

ICU LENGTH OF STAY

This outcome could not be pooled due to high heterogeneity. There was no evidence of a difference between NPPV and SMC in 4 RCTs, and 2 RCTs reported a shorter length of stay for NPPV (1 day shorter (95% CI −1.79 to −0.21); n = 30; 4 days shorter (95% CI −4.36 to −3.64); n = 120)

587
(6 RCTs)

⊕⊝⊝⊝
VERY LOWh,i,j

Data were not pooled due to high heterogeneity with an I2 of 99%

INTOLERANCE TO ALLOCATED TREATMENT ‐ not reported

Outcome was not reported

ADVERSE EVENTS

Reported adverse events included skin damage, pneumonia, gastrointestinal bleeding, gastric distention, vomiting, pneumothorax, sinusitis, mask discomfort, hypotension, arrhythmia, cardiorespiratory arrest, gastric aspiration, stroke, seizures, claustrophobia, and hypercapnia.

There was no evidence of a difference between groups for most of these events. However, there was an increase in discomfort with mask reported with NPPV (35/658) compared with SMC (0/442); RR 12.59 (95% CI 2.39 to 66.28).

One small study (n = 83) reported a higher incidence of gastric distention in the NPPV group (13/56 in NPPV group, 0/27 in SMC group; RR 13.26 (95% CI 0.82 to 215.12)), but the number of events was low and the confidence interval wide. There was also some weak evidence of a lower incidence of cardiorespiratory arrest in the NPPV group (24/836 (NPPV) vs 26/516 (SMC); RR 0.60 (95% CI 0.34 to 1.05)

2038
(11 RCTs)

⊕⊕⊝⊝
LOWk,l

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

aDowngraded by one level for risk of bias. Most trials providing hospital mortality evidence were at unclear or high risk of bias for key domains, including: randomisation sequence, allocation concealment, selective reporting bias, and other significant bias.
bDowngraded by one level for imprecision. Most trials had few participants and few mortality events, with a confidence interval crossing a RR of 0.75.
cDowngraded by one level for risk of bias. Most trials providing ETI evidence were rated as having unclear or high risk of bias for key domains, which included randomisation sequence generation, allocation concealment, selective reporting bias, and other significant bias.
dDowngraded by one level for imprecision. Most trials had few participants and few acute MI events, with a confidence interval crossing a RR of 1.0. In addition, the definition of acute MI varied between trials, with some trials using creatinine kinase and others using troponin.
eDowngraded by one level for risk of bias. Most trials were at low or unclear risk of bias. Potential limitations from lack of blinding are likely to lower confidence in hospital length of stay.
fDowngraded by one level for inconsistency. Hospital length of stay was heterogeneous, with an I2 of 58% and visually‐evident heterogeneity.
gDowngraded by one level for imprecision. Most trials had few participants, with a confidence interval crossing a MD of 0.
hDowngraded by one level for risk of bias. Most information was from studies at low or unclear risk of bias. Potential limitations from lack of blinding are likely to lower confidence in ICU length of stay.
iDowngraded by one level for imprecision. Most trials had few participants, with a confidence interval crossing a MD of 0.
jDowngraded by one level for inconsistency. ICU length of stay was heterogeneous, with an I2 of 99% and visually‐evident heterogeneity.
kDowngraded by one level for risk of bias. Most studies had at least one domain at high or unclear risk of bias.
lDowngraded by one level for imprecision, as confidence intervals included both directions of effect for many of the adverse events.

Background

Description of the condition

Abnormal heart function in heart failure (HF) can produce signs and symptoms of reduced cardiac output (Ezekowitz 2017). A rapid deterioration in HF symptoms is called congestive heart failure (CHF) or acute cardiogenic pulmonary oedema (ACPE). Symptoms of ACPE can include dyspnoea, orthopnoea, peripheral oedema, cough, fatigue, and weight gain (Wang 2005). Signs of ACPE can include the presence of a third heart sound, jugular venous distension, rales, lower extremity oedema, wheezing, and ascites (Wang 2005). Approximately one million people a year are discharged from hospitals in the United States with HF (Benjamin 2017). ACPE severity is variable, but the presence of hypoxia, respiratory failure, and hypotension can indicate a higher risk presentation (Ponikowski 2016). Many conditions can trigger ACPE, such as acute coronary syndrome (ACS), tachyarrhythmia, valvular heart disease, and hypertension (Ponikowski 2016). Importantly, ACPE is also associated with an in‐hospital mortality of approximately 10% and one‐year mortality of 30% (Rudiger 2005).

ACPE can be conceptualised as left ventricular failure with elevated left ventricular filling pressures. Elevated filling pressures produce higher pulmonary capillary pressures and fluid extravasation into alveoli due to overwhelmed lymphatic vessel absorption capacity (Allison 1991; Packer 1993). Pulmonary oedema fluid can dilute surfactant and neutralise its lubricating properties, causing a reduction in lung compliance and increased work of breathing (Allison 1991; Park 2001). In the upright position, oedema accumulates at the lung bases, causing a ventilation‐perfusion (V‐Q) mismatch, which can cause hypoxia (Allison 1991).

Description of the intervention

ACPE treatment guidelines have been prepared in several countries, including Canada (Ezekowitz 2017) and Europe (Ponikowski 2016). Treatment options include: loop diuretics (e.g. furosemide), vasodilators (e.g. nitroglycerin), supplemental oxygen, non‐invasive positive pressure ventilation (NPPV), and endotracheal intubation (Ezekowitz 2017; Ponikowski 2016). Treating the underlying cause of ACPE is also necessary (e.g. antihypertensives for hypertension, coronary angiography for ST elevation myocardial infarction) (Ezekowitz 2017; Ponikowski 2016; Yancy 2013).

Endotracheal intubation raises the risk for adverse events, such as nosocomial infections (e.g. pneumonia), tracheal injury, and prolonged hospital length of stay (Gay 2009). In contrast to the invasiveness of endotracheal intubation, NPPV can be provided in the form of continuous positive airway pressure (CPAP) and bilevel NPPV (BiPAP® ‐ Respironics, Inc, Murrysville, PA) using face or nasal masks. CPAP maintains a constant positive airway pressure throughout the respiratory cycle. In contrast, bilevel NPPV provides additional inspiratory positive airway pressure and positive end‐expiratory pressure (Nava 2009).

How the intervention might work

The cardiovascular and pulmonary systems work together to maintain homeostasis. During normal inhalation contraction of the diaphragm and intercostal muscles produces a more negative pleural pressure compared to the lungs at rest (Alviar 2018). This negative pleural pressure reduces left ventricular pressure relative to systemic arterial pressure (Alviar 2018). In contrast, during positive pressure ventilation, inhalation involves a positive pleural pressure which raises left ventricular pressure relative to systemic arterial pressure and reduces left ventricular afterload due to a baroreceptor reflex (Alviar 2018; Buda 1979). Intrapleural pressure influences right ventricular inflow and left ventricular outflow. The difference between airway pressure and intrapleural pressure (transpulmonary pressure gradient) affects pulmonary vasculature, which in turn influences left ventricular inflow and right ventricular outflow (Alviar 2018). NPPV reduces pulmonary workload, improves cardiac output, and enhances lung compliance (Baratz 1992; Lenique 1997). During ACPE interstitial and alveolar oedema reduce lung compliance. Application of positive pressure at end expiration can force fluid out of alveoli, improving pulmonary vascular resistance and enhancing gas exchange (Alviar 2018). In addition, positive pressure ventilation could prevent alveolar collapse and enhance alveolar expansion with inspiration improving airway compliance (Alviar 2018). This mechanism could explain how NPPV applied during ACPE can lead to improved oxygenation (Räsänen 1985). During respiratory distress without NPPV the intrapleural pressure becomes more negative and left ventricular afterload is increased, because left ventricular systolic pressure is more negative compared to the systemic circulation (Alviar 2018; Magder 1983). In terms of left ventricular function, NPPV reduces preload, afterload, myocardial oxygen demand, and enhances hydrostatic displacement of alveolar oedema (Alviar 2018). In left ventricular failure, NPPV can improve cardiac output by reducing left ventricular afterload. In contrast, in right ventricular failure, NPPV reduces right ventricular preload and can reduce right ventricular cardiac output (Alviar 2018). The net benefit of NPPV in heart failure will depend on the relative left and right ventricular functions, and afterload (Alviar 2018). Furthermore, NPPV use in ACPE may prevent endotracheal intubation without significant adverse events (Nava 2009). Mechanical ventilation can be life‐saving for people with respiratory failure needing airway protection and with severe hypoxia associated with failed NPPV therapy (Alviar 2018). However, mechanical ventilation is associated with additional adverse events, such as endotracheal tube complications (Tobin 1994), ventilator‐associated lung injury (Slutsky 2013), ventilator‐associated pneumonia (Spalding 2017), barotrauma (e.g. pneumothorax) (Anzueto 2004), and ventilator‐induced diaphragmatic dysfunction (Levine 2008).

Why it is important to do this review

Since the increased clinical use of NPPV in the 1980s, many studies have evaluated NPPV's effectiveness for ACPE, and have reported mixed results (Nava 2009). Furthermore, current heart failure guidelines differ in their NPPV recommendations. Canadian guidelines recommend against routine use of NPPV, and suggest that NPPV could be used for ACPE with persistent hypoxia despite standard medical care (Ezekowitz 2017). In addition, Canadian guidelines warn of the clinical risks of NPPV, including worsening right heart failure, worsening hypercapnia, aspiration, and pneumothorax (Ezekowitz 2017). In contrast, European guidelines (Ponikowski 2016) suggest NPPV be considered and started quickly in people with ACPE who present with tachypnoea (respiratory rate greater than 25 breaths a minute) and hypoxia (SpO2 less than 90%). Furthermore, they are of the opinion that NPPV can reduce respiratory distress and endotracheal intubation. The main risk of NPPV emphasised by the European guidelines is hypotension (Ponikowski 2016). American guidelines do not provide treatment guidance for ACPE (Yancy 2013). Given the differences in opinion between these two major heart failure guidelines on the use of NPPV for ACPE, we have updated our systematic review and meta‐analysis.

Objectives

To evaluate the safety and effectiveness of NPPV compared to standard medical care (SMC) for adults with ACPE. The primary outcome was hospital mortality. Important secondary outcomes were endotracheal intubation, treatment intolerance, hospital and intensive care unit length of stay, rates of acute myocardial infarction, and adverse event rates.

Methods

Criteria for considering studies for this review

Types of studies

For inclusion, studies had to be randomised controlled trials (RCTs). We excluded studies that were cluster‐randomised or used a cross‐over design. We ruled out cluster‐randomised studies to avoid heterogeneity in study design, heterogeneity from the unit of randomisation (e.g. intensive care unit (ICU) versus emergency room (ER)), and variation in eligibility criteria at the individual level compared to a cluster level. We ruled out cross‐over studies because we did not consider them an appropriate design for acute cardiogenic pulmonary oedema (ACPE). In a cross‐over study, a carry‐over effect from the initial therapy (e.g. medical therapy or NPPV) could influence the second intervention (e.g. medical therapy or NPPV), such that participants entering the second intervention differ from those during the first intervention. To avoid publication bias, we included studies irrespective of final publication status. We also included studies reported as full text, abstract only, and unpublished data.

Types of participants

We included trials reporting on adults (18 years and older) with ACPE. A diagnosis of ACPE can have symptoms and clinical signs of hypoperfusion or congestion, or both, such as dyspnoea, pulmonary congestion, jugular venous distension, congestive hepatomegaly, peripheral oedema, confusion, oliguria, and cool extremities (Wang 2005; Yancy 2013). The clinical diagnosis could be supported by a chest radiograph, electrocardiograms, serum biomarkers (e.g. troponin for acute myocardial infarction or Brain natriuretic peptide / N‐terminal‐pro hormone Brain natriuretic peptide for HF), or echocardiography. We excluded trials investigating NPPV for people with a primary diagnosis of pneumonia, alternative aetiologies of respiratory failure (e.g. endocarditis, cardiac surgery patients, unknown cause), or its use as a weaning strategy.

Types of interventions

For inclusion, the intervention group had to have received nasal or face mask NPPV (CPAP, or bilevel NPPV, or both) with standard medical care (SMC). In contrast, the control group had to have received the same SMC alone (Ponikowski 2016; Yancy 2013).

Types of outcome measures

Primary outcomes

  • Hospital mortality

Secondary outcomes

  • Endotracheal intubation

  • Incidence of acute myocardial infarction (AMI) during hospitalisation, after starting treatment, (e.g. cardiac biomarker elevation creatinine kinase (CK) or troponin) with or without electrocardiographic (ECG) changes or symptoms of myocardial ischaemia

  • Intolerance to allocated treatment (e.g. early treatment discontinuation in people not meeting criteria for endotracheal intubation)

  • Treatment failure (the combination of mortality, intubation, and intolerance to the allocated treatment)

  • Hospital length of stay (from hospital admission to hospital discharge or death)

  • ICU length of stay (from ICU admission to ICU discharge or death)

  • Vital signs: blood pressure one hour post‐intervention, respiratory rate

  • Arterial blood gases (PaO2) one hour post‐intervention

  • Adverse events

For inclusion, studies had to report one or more of the clinical outcomes of interest. We sought additional information from principal investigators as required, including for outcomes that were measured but not reported in the final publication.

Search methods for identification of studies

Electronic searches

We identified trials through systematic searches of the following databases:

  • Cochrane Central Register of Controlled Trials (CENTRAL) (CRS Web, 20 September 2018).

  • Database of Abstracts of Reviews Effectiveness (DARE) in the Cochrane Library (Issue 2 of 4, 2015).

  • MEDLINE (Ovid, 1946 to 19 September 2018).

  • Embase (Ovid, 1974 to 19 September 2018).

  • Cumulative Index to Nursing and Allied Health Literature (CINAHL) Plus (EBSCO, 1937 to 20 September 2018).

  • LILACS (1982 to 20 September 2018).

  • WHO International Clinical Trials Registry Platform (ICTRP; apps.who.int/trialsearch/) (searched 20 September 2018).

  • Clinicaltrials.gov (searched 20 September 2018)

The RCT filter for MEDLINE is the Cochrane sensitivity‐maximising RCT filter and, for Embase search terms as recommended in the Cochrane Handbook for Systematic Reviews of Interventions were applied. (Lefebvre 2011). We searched all databases from their inception to the present, without language restriction or consideration of final publication status. We present the search strategies for each database in Appendix 1. Finally, we also considered studies that were ongoing or awaiting classification at the time of the last version of this review.

Searching other resources

We checked the bibliographies of retrieved articles to identify related published and unpublished studies.

Data collection and analysis

We followed the recommendations outlined in the Cochrane Handbook for Systematic Reviews of Interventions in preparing this review (Higgins 2011; Higgins 2017).

Selection of studies

In our updated search, two review authors (NB, MC) independently screened citations and selected trials that met the inclusion criteria. We resolved disagreements by consensus with a third review author (YW). In the previous version of this review, two review authors (FV and ML) independently screened citations and selected trials meeting the inclusion criteria with disagreements resolved by a third review author (AA) (Vital 2013).

Data extraction and management

We used a standardised and piloted data collection form to record data on study characteristics, risks of bias, and outcomes. Two review authors (NB, YW) independently extracted:

  • characteristics of the study (design, methods of randomisation, withdrawals and dropouts, intention‐to‐treat analysis (ITT), informed consent, place and multicentre study, funding, conflicts of interest, study dates);

  • participants (age, gender, number, diagnostic criteria, inclusion and exclusion criteria);

  • interventions (type of NPPV, timing and duration of therapy, co‐interventions, SMC (intervention and dose); and

  • outcomes reported.

We requested unpublished data from primary authors to supplement outcomes where needed. Two review authors (NB, YW) performed data extraction for all selected study reports independently, with partial extraction by an additional two review authors (CG and MA). We resolved disagreements by consensus with a third senior review author (MC).

Assessment of risk of bias in included studies

Two review authors (NB, YW) independently assessed risks of bias for each study, using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved any disagreements by consensus with a third review author (MC). We assessed risks of bias according to the following domains, using the Cochrane 'Risk of bias' tool:

  • Random sequence generation.

  • Allocation concealment.

  • Blinding of participants and personnel.

  • Blinding of outcome assessment.

  • Completeness of outcome data.

  • Selective reporting bias.

  • Other bias (any other sources of potential bias, e.g. trial prematurely stopped)

We graded each potential source of bias as high, low, or unclear and provide a justification for our judgement in the 'Risk of bias' table. We summarise the 'Risk of bias' judgements across different studies for each of the domains listed. If unpublished data were used in the 'Risk of bias' assessment, we noted this in the 'Risk of bias' table. When we considered treatment effects, we took into account the risks of bias for studies that contributed to that outcome.

Assessment of bias in conducting the systematic review
We conducted the review according to our published protocol and report any deviations from it in the Differences between protocol and review section of the review.

Measures of treatment effect

We analysed dichotomous data as risk ratios (RRs) with 95% confidence intervals (CIs), and continuous data as mean difference (MDs) with 95% CIs.

Unit of analysis issues

We included RCTs with a parallel design. If the same outcome was measured at repeated time points, we selected the longest time point for inclusion in the review, to avoid double counting. In trials with multiple intervention arms, we included the control group with SMC and combined the treatment groups if they had a similar intervention (NPPV: CPAP or BiPAP). For subgroup analysis in studies comparing CPAP, bilevel NPPV, and SMC we divided the control group evenly between groups, as suggested in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011).

Dealing with missing data

We contacted investigators or study sponsors to obtain missing numerical outcome data where required (e.g. when a study was reported as an abstract only). Where this was not possible and the missing data were thought to introduce serious bias, we explored the impact of study inclusion in the overall assessment of results by performing a sensitivity analysis.

Assessment of heterogeneity

We quantified the impact of statistical heterogeneity using the I2 statistic (Higgins 2011). An I2 of 0 to 40% may represent low heterogeneity, 30% to 60% may represent moderate heterogeneity, 50% to 90% may represent substantial heterogeneity, and 75% to 100% may represent considerable heterogeneity. In interpreting the heterogeneity values, we also considered the magnitude and direction of effects and strength of evidence for heterogeneity (e.g. P value from the Chi2 test or a confidence interval for I2). In addition, we inspected forest plots for signs of heterogeneity. If substantial heterogeneity was present, we reported it and conducted exploratory analyses to identify sources of heterogeneity (e.g. participants, treatments and study quality). We hypothesised that age, gender, and co‐morbidities may represent potential sources of heterogeneity among participants. Furthermore, heterogeneity may be related to the initial treatment(s) used, the levels of pressure applied with NPPV, or treatment duration.

Assessment of reporting biases

Funnel plots allowed us to examine and explore small‐study biases and publication bias (Egger 1997). We generated funnel plots for each outcome with at least 10 studies.

Data synthesis

We undertook meta‐analysis only where studies were sufficiently similar for results to be clinically meaningful and if we identified more than two trials reporting data for that outcome. For these studies, we pooled dichotomous and continuous variables using a random‐effects model. We chose a random‐effects model, as we anticipated heterogeneity in study participants, interventions, and outcomes of included studies. We presented the number needed to treat for an additional beneficial outcome (NNTB) when an effect was indicated by meta‐analysis. We report summary estimates of treatment effect with their associated 95% CIs. We conducted all analyses using Cochrane statistical software (Review Manager 2014).

'Summary of findings' table

We created a 'Summary of findings' table for the main outcomes of the review: hospital mortality, endotracheal intubation (ETI), acute myocardial infarction (AMI), intolerance to allocated treatment, hospital length of stay, ICU length of stay, and adverse events. We generated these tables using GRADEpro software (gradepro.org/) and imported them into Review Manager 5.

GRADE

We used the five GRADE domains (study limitations, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of a body of evidence as it relates to the studies. For this process, we used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011; Higgins 2017). We justified all decisions to downgrade the quality of evidence using footnotes provided in the 'Summary of findings' table (Guyatt 2008; Higgins 2011).

Two review authors (NB, YW) working independently, made judgements about evidence quality, with disagreements resolved by consensus with a third senior review author (MC).

Subgroup analysis and investigation of heterogeneity

We performed subgroup analyses for the following outcomes: hospital mortality, ETI, AMI, intolerance to allocated treatment, and hospital length of stay. Subgroups of interest included: type of NPPV (CPAP or bilevel NPPV), location of therapy (ER versus ICU), baseline hypercapnia status. In addition, for AMI we planned a subgroup analysis by time of event relative to treatment (AMI before or after initiation of treatment). Unfortunately, our planned AMI subgroup analysis was not possible, due to inconsistent definitions of AMI between studies. Please see Differences between protocol and review for further details. For mean, systolic, and diastolic blood pressure after one hour of therapy we performed a subgroup analysis by NPPV type (CPAP or bilevel NPPV). We used the test for subgroup differences to detect whether the effect estimate differed between groups of studies. We used a P value of less than 0.05 for this test, as suggesting a true difference between subgroups, but acknowledge that this test has limited power when there are few studies.

Sensitivity analysis

We decided a priori to perform sensitivity analyses on hospital mortality, ETI, and adverse events. These sensitivity analyses covered:

  • including studies with only low risk of bias.

  • including studies without missing data.

  • including only studies with a final diagnosis of cardiogenic pulmonary oedema being present in 50% or more of included participants.

For mortality, we considered a trial to be at low risk of bias if it met the criteria for low risk of bias in the following domains: random sequence generation and incomplete outcome data. For ETI and adverse events, we rated trials at low risk of bias if they met the low‐risk‐of‐bias criteria for the following domains: random sequence generation and allocation concealment.

Reaching conclusions

We base our conclusions only on findings from the quantitative or narrative synthesis of studies included in this review. In addition, we outline the gaps in the evidence for NPPV in ACPE, and suggest future directions for research.

Results

Description of studies

Results of the search

Our updated search identified 3791 references (CENTRAL 905, DARE 12, MEDLINE 1266, Embase 878, LILACS 67, CINAHL Plus 397, WHO ICTRP 168, and clinicaltrials.gov 98). After de‐duplication and reviewing reference lists from included articles, we screened 2807 references for inclusion. After screening, we obtained full‐text articles for 94 references. UItimately, we have included in this review update 24 studies (37 references) with a total of 2664 participants treated with NPPV versus SMC for ACPE. We excluded 43 studies (57 references). Our review update includes seven new studies since May 2013 (Austin 2013; Ducros 2011; El‐Refay 2016; Hao 2002; Li 2005; Moritz 2003; Zokaei 2016) and excludes 15 previously included studies (Bautin 2005;Bellone 2004; Bellone 2005; Bersten 1991; Delclaux 2000; Ferrer 2003; Ferrari 2007; Ferrari 2010; Fontanella 2010; Liesching 2014; Martin‐Bermudez 2002; Mehta 1997; Moritz 2007; Sharon 2000; Weitz 2007). Our PRISMA flowchart summarises the study selection process (Figure 1). We provide additional information on included and excluded studies in the following sections: Characteristics of included studies; Characteristics of excluded studies.


PRISMA statement flow diagram for 2019 review update.

PRISMA statement flow diagram for 2019 review update.

Included studies

Study design

We included 24 parallel‐design RCTs in this review update (Agmy 2008; Austin 2013; Crane 2004; Ducros 2011; El‐Refay 2016; Frontin 2011; Gray 2008; Hao 2002; Kelly 2002; L'Her 2004; Levitt 2001; Li 2005; Lin 1991; Lin 1995; Masip 2000; Moritz 2003; Nava 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Thys 2002; Zokaei 2016). The funding sources for seven studies was public (Agmy 2008; Austin 2013; Ducros 2011; Frontin 2011; Gray 2008; Masip 2000; Thys 2002). One study reported receiving funds from a nonprofit organisation (Kelly 2002). Two studies were funded by device manufacturers (Crane 2004; L'Her 2004) and 14 studies provided no details on funding sources (El‐Refay 2016; Hao 2002; Levitt 2001; Li 2005; Lin 1991; Lin 1995; Moritz 2003; Nava 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Zokaei 2016). We summarise each study in our Characteristics of included studies tables.

Population

The mean participant age in our review was 73.3 ± 9.0 years. Our review includes studies from 14 countries: Australia, Belgium, Brazil, China, Egypt, Finland, France, Iran, Italy, Japan, Spain, Taiwan, the UK, and the USA. Furthermore, five studies were multicentre (Crane 2004; Ducros 2011; Gray 2008; L'Her 2004; Nava 2003). Studies varied in size from eight to 1069 participants, with a median and mean study size of 55 and 114 participants respectively. Ten studies were conducted in an emergency department (ED or ER) setting (Crane 2004; El‐Refay 2016; Gray 2008; Kelly 2002; Levitt 2001; L'Her 2004; Moritz 2003; Nava 2003; Park 2004; Thys 2002), eight studies were conducted in an ICU setting (Agmy 2008; Lin 1991; Lin 1995; Masip 2000; Räsänen 1985; Takeda 1997; Takeda 1998; Zokaei 2016), and three studies did not reference a specific study location (Hao 2002; Li 2005; Park 2001). Three studies were started in the pre‐hospital setting with further care in the ER and the ICU if required (Austin 2013; Ducros 2011; Frontin 2011).

Intervention

Six studies compared all three interventions: CPAP, bilevel NPPV, and SMC (Agmy 2008; Crane 2004; El‐Refay 2016; Gray 2008; Park 2001; Park 2004). Twelve studies compared CPAP against SMC (Austin 2013; Ducros 2011; Frontin 2011; Hao 2002; Kelly 2002; L'Her 2004; Lin 1991; Lin 1995; Moritz 2003; Räsänen 1985; Takeda 1997; Takeda 1998). Six studies compared bilevel NPPV against SMC (Levitt 2001; Li 2005; Masip 2000; Nava 2003; Thys 2002; Zokaei 2016). We summarise the NPPV settings including mask type, inspiratory positive airway pressure (IPAP), expiratory positive airway pressure (EPAP), positive end‐expiratory pressure (PEEP), and duration of NPPV in Table 1. The NPPV patient‐ventilator interface varied between exclusive use of nasal masks (Takeda 1997; Takeda 1998), a choice between a nasal or face mask (Levitt 2001; Park 2001; Zokaei 2016), and exclusive use of face masks. We have summarised the SMC provided in each trial in Table 2. SMC included supplemental oxygen and pharmacologic treatments. SMC pharmacologic treatments included loop diuretics (furosemide), nitrates (e.g. nitroglycerin, isosorbide dinitrate), opioids (e.g. morphine), and ionotropes if required.

Open in table viewer
Table 1. NPPV intervention summary: EPAP, IPAP, and PEEP settings, duration of therapy.

Study

Mask type

IPAP level (cmH2O) ± SD

EPAP in bilevel NPPV (cmH2O) ± SD

PEEP in CPAP (cmH2O) ± SD

Time of bilevel NPPV (h) ± SD

Time of CPAP (h) ± SD

Agmy 2008

Face mask

NA

NA

NA

NA

NA

Austin 2013

Face mask

NA

NA

10

NA

0.583333

Crane 2004

Face mask

15

5

10

NA

NA

Ducros 2011

Face mask

NA

NA

10

NA

3

El‐Refay 2016

Face mask

15

10

10

1

Frontin 2011

Face mask

NA

NA

10

NA

NA

Gray 2008

Face mask

14±5

7 ± 3

10 ± 4

2 ± 1.3

2.2 ± 1.5

Hao 2002

Face mask

NA

NA

6 to 10

NA

4.6 ± 2.8

Kelly 2002

Face mask

NA

NA

7.5

NA

NA

L'Her 2004

Face mask

NA

NA

7.5

NA

NA

Levitt 2001

Face mask selected for mouth breathers Nasal available

NA

NA

NA

2

2

Li 2005

Face mask

15 to 18

5 to 8

NA

2

NA

Lin 1991

Face mask

NA

NA

12.5

NA

6

Lin 1995

Face mask

NA

NA

12.5

NA

6

Masip 2000

Face mask

15.2 ± 2.4

5

NA

4.2 ± 1.5

NA

Moritz 2003

Face mask

NA

NA

9.3 ± 0.3

0.5

NA

Nava 2003

Face mask

14.5 ± 21.1

6.1 ± 3.2

NA

11.4 ± 3.6

NA

Park 2001

Nasal bilevel and face mask for CPAP

12

4

7.5

2.6 ± 0.6

2.8 ± 1.5

Park 2004

Face mask

17 ± 2

11 ± 2

11 ± 2

2.1 ± 1

1.7 ± 0.7

Räsänen 1985

Face mask

NA

NA

10

NA

NA

Takeda 1997

Nasal

NA

NA

4 to 10

11.9 ± 8.4

NA

Takeda 1998

Nasal

NA

NA

4 to 10

NA

NA

Thys 2002

Face mask

16.5 ± 3.3

6.1 ± 1.5

NA

1.3 ± 0.3

NA

Zokaei 2016

Face mask 1st, nasal 2nd

10 to 20

4 to 7

NA

1

NA

Not all papers provided a mean and standard deviation. Where unavailable, the range represents the IPAP/EPAP/CPAP settings described in the Methods. NA: not applicable

Open in table viewer
Table 2. Standard medical therapy for each trial

Study

Lasix

Lasix dose

Nitrate

FiO2 mask and %FiO2

Other

Additional differences between SMC received in treatment group

Agmy 2008

Unclear

Unclear

Unclear

Unclear

Unclear

Unclear

Austin 2013

yes

40 mg

Sublingual 400 ‐ 1600 mcg q5min

Yes, 8 ‐ 15 Lpm

Morphine 1 ‐ 2 mg IV

None

Crane 2004

yes

Unclear

nitrates

Yes, 10 Lpm

diamorphine

None

Ducros 2011

yes

Lasix 40 mg ‐ 120 mg IV or 1 ‐ 3 mg bumetanide

nitroglycerin (1 mg per 3 min) and a continuous IV infusion unless systolic BP < 110 mmHg

15 Lpm

NA

None

El‐Refay 2016

yes

Lasix 40 mg IV

nitroglycerin 0.4 mg sublingual

15 Lpm

Morphine 2 mg IV

None

Frontin 2011

yes

1 mg/kg IV

isosorbide dinitrate IV 2 mg if SBP > 180 mmHg

15 Lpm

NA

None

Gray 2008

yes

50 mg above usual dose to max 100 mg IV

Buccal nitrates 2 ‐ 5 mg

15 Lpm

Opiates

None

Hao 2002

yes

20 ‐ 40 mg IV

Buccal nitrates 0.5 mg to 1 mg IV

6 Lpm

Morphine 2 mg IV

None

Kelly 2002

yes

50 ‐ 100 mg IV

buccal nitrate 5 mg if systolic BP > 90 mmHg

oxygen 60% by venturi mask

morphine IV 2.5 ‐ 10 mg

None

Li 2005

yes

Unclear

nitroglycerin

Mask high concentration

Unclear

None

Levitt 2001

yes

Unclear

nitroglycerin IV

Mask high flow

Morphine

None

L'Her 2004

yes

80 mg IV

nitroglycerin IV infusion 1 mg/hour

8 Lpm

Morphine 2 ‐ 10 mg IV

None

Lin 1991

Unclear

Unclear

Unclear

FiO2 100%

Unclear

Unclear

Lin 1995

yes

40 mg IV or double home dose

nitroglycerin sublingual 0.6 mg or isosorbide 10 ‐ 20 mg, or nitro infusion 10 ‐ 50 mcg/min

FiO2 100%

Morphine 2 ‐ 10 mg IV

None

Masip 2000

yes

40 mg IV

IV glyceryl trinitrate 1mg if systolic BP > 180 mmHg

FiO2 50%

Morphine 4 mg IV

None

Moritz 2003

yes

40 mg IV or double home dose

IV nitroglycerin infusion 0.125 ‐ 0.25 mcg/kg/min

FiO2 100%

Unclear

None

Nava 2003

yes

40 mg IV or double usual dose, repeated, if necessary, every 20 mins, up to 320 mg

continuous glyceryl trinitrate at an initial rate of 1.5 mg/hour. A bolus dose of 1 mg IV was added if systolic BP > 180 mmHg

10 Lpm

morphine sulphate up to 4 mg

None

Park 2001

Unclear

Unclear

5 mg isosorbide dinitrite if systolic BP > 100 mm Hg

15 Lpm

Unclear

Unclear

Park 2004

Unclear

Unclear

5 mg isosorbide dinitrate was given sublingually and if necessary titrated up to 15 mg

FiO2 50%

Unclear

Unclear

Räsänen 1985

yes

40 ‐ 80 mg IV

Nitroprusside, nitroglycerine

FiO2 30%

Morphine

Unclear

Takeda 1997

yes

Unclear

Nitroglycerin infusion

FiO2 50%

Morphine

Unclear

Takeda 1998

yes

Unclear

Nitroglycerin infusion

FiO2 70%

Morphine

Unclear

Thys 2002

yes

40 mg IV

isosorbide dinitrate 2 mg/hour

Unclear

Unclear

Unclear

Zokaei 2016

yes

40 ‐ 320 mg

IV nitroglycerin 5 ‐ 50 mcg/kg/min

10 Lpm

morphine 5 ‐ 10 mg

Unclear

BP: blood pressure; FiO2: fraction of inspired oxygen; IV: intravenous; Lpm: litres per minute; NA: not available; mcg: micrograms; mcg/kg/min: micrograms per kilogram per minute; mg: milligrams; min: minutes; mmHg = millimetres of mercury

Outcomes

Commonly‐reported outcomes included hospital mortality, endotracheal intubation (ETI), acute myocardial infarction (AMI), hospital length of stay, ICU length of stay, change in systolic blood pressure (SBP) and respiratory rate (RR) during treatment, and change in arterial blood glasses (PaO2). In reviewing the methodology for each study, we found that only three studies (Austin 2013; Ducros 2011; Gray 2008) provided a power calculation to detect a mortality difference. In addition, hospital mortality was reported at variable time periods. We summarise the time point at which hospital mortality was measured in Table 3. The median length of follow‐up for hospital mortality was 13 days. AMI during the study period was reported by five studies (Crane 2004; Gray 2008; Levitt 2001; Nava 2003; Park 2001). AMI definitions were inconsistent, varying between new ST segment elevation on ECG, to myocardial enzyme elevation (e.g. CK, troponin) with ECG changes, and myocardial enzyme elevation alone. Treatment intolerance or early treatment discontinuation was reported by Agmy 2008. Explicit definitions for treatment failure as an indicator for potential endotracheal intubation are summarised in Table 4.

Open in table viewer
Table 3. In‐hospital mortality duration of follow‐up

Study

Follow‐up (days)

Agmy 2008

Unclear

Austin 2013

7.2a

Crane 2004

41b

Ducros 2011

> 7

El‐Refay 2016

1

Frontin 2011

30

Gray 2008

30

Hao 2002

NA

Kelly 2002

15a

Li 2005

NA

Levitt 2001

38b

L'Her 2004

12a

Lin 1991

1

Lin 1995

9

Masip 2000

14a

Moritz 2003

NA

Nava 2003

5.4a

Park 2001

3b

Park 2004

15

Räsänen 1985

Unclear

Takeda 1997

7.7a

Takeda 1998

Unclear

Thys 2002

17.6a

Zokaei 2016

7

aEstimated based on longest reported mean hospital length of stay; bBased on latest death reported in trial

Open in table viewer
Table 4. Intubation criteria for trials reporting endotracheal intubation rates

Study

Intubation Criteria

Agmy 2008

Unclear

Austin 2013

Unclear

Crane 2004

RR > 40, RR < 10, altered level of consciousness, arterial pH < 7.2

Ducros 2011

SpO2 < 85% after 30 mins of 15 Lpm or FiO2 60% CPAP, respiratory arrest, psychomotor agitation, SBP < 80 mmHg

El‐Refay 2016

Unclear

Frontin 2011

Worsening SpO2 despite effective treatment, loss of airway protective reflexes (cough, swallow), decreased level of consciousness, haemodynamic instability, intolerance/poor fit of face mask for the CPAP group, medical clinical impression of deterioration, or participant request

Gray 2008

At discretion of treating clinician

L'Her 2004

Cardiac or respiratory arrest, severe haemodynamic instability (SBP < 80), administration of epinephrine/norepinephrine, refractory hypoxaemia SaO2 < 92% despite face mask, clinical signs of respiratory exhaustion (accessory muscle use with paradoxical abdominal motion), coma or seizures, agitation requiring sedation

Levitt 2001

Clinical or arterial blood gas findings. Severe respiratory distress, deterioration in mental status, or further deterioration in vital signs with increasing HR, RR, or a decrease in BP. Decrease in pO2 < 60 or rise in pCO2 > 50 along with signs of clinical deterioration warranted intubation

Lin 1991

Clinician discretion. Loss of consciousness or airway obstruction were indications for intubation

Lin 1995

Cardiopulmonary resuscitation or clinical deterioration with any 2 of the following: (1) a rise in arterial carbon dioxide tension to more than 55 mm Hg; (2) arterial blood oxygen partial pressure divided by fraction of inspired oxygen content less than 200 mm Hg; and (3) respiratory rate over 35 breaths/min.

Masip 2000

Patients were intubated when the criteria for treatment failure were met. Details not specified.

Nava 2003

Unclear

Park 2001

Clinical judgement

Park 2004

Clinical judgement

Räsänen 1985

Clinical judgement. Loss of consciousness or airway obstruction were indications for intubation at any time.

Takeda 1997

Clinical deterioration and either a decrease in the PaO2/FiO2 < 100 with an FIO2 ≥ 70% or a increase in the PaCO2 > 55 mmHg

Takeda 1998

Clinical deterioration and either a fall in PaO2/FiO2 < 100 with an FiO2 of 70% or a rise in PaCO2 to > 55 mmHg.

Thys 2002

Deterioration in clinical status including all of the following: dyspnoea, respiratory or heart frequency or both, sweating and agitation or deterioration in blood gases or in haemodynamic status or both

Zokaei 2016

RR after 1 hour of bilevel NPPV > 30, persistent hypoxaemia, haemodynamic instability (systolic BP < 70), agitation, or worsened neurologic status, inability to tolerate mask or aspiration of gastric content

BP: blood pressure; CPAP: continuous positive airway pressure; FiO2: fraction of inspired oxygen; HR: heart rate; Lpm: litres per minute; PaCO2: arterial partial pressure of carbon dioxide; PaO2: arterial partial pressure of oxygen; RR: respiratory rate; SBP: systolic blood pressure

Several outcomes, including systolic blood pressure, mean blood pressure, diastolic blood pressure, PaO2, and respiratory rate, are reported as continuous outcomes. For these outcomes, we used the reported mean and standard deviations. Certain data were only available in graphical format and we extracted them by measuring the graph: PaO2 (Lin 1995). We converted arterial blood gas data from kilopascals (kPa) to millimetres of mercury (mmHg) for three studies (Crane 2004; Gray 2008; Kelly 2002), using the conversion of 1 kPA to 7.50 mmHg (Zumdahl 2002).

Excluded studies

We excluded citations if they were duplicates of the same study (N = 27), an ineligible intervention (N = 14), an ineligible population (N = 5), a quasi‐randomised controlled trial (N = 4), an ineligible study type (N = 17), and if the study was withdrawn after study registration (N = 3). We have excluded 15 studies included in the previous version of this review. Of these, we excluded three studies conducted on an inappropriate population (Bautin 2005; Delclaux 2000; Ferrer 2003). We excluded four quasi‐randomised controlled trials (Bersten 1991; Moritz 2007; Sharon 2000; Weitz 2007). Finally, we excluded eight studies which compared bilevel NPPV against CPAP, without an additional comparison against SMC (Bellone 2004; Bellone 2005; Ferrari 2007; Ferrari 2010; Fontanella 2010; Liesching 2014; Martin‐Bermudez 2002; Mehta 1997). We provide additional details explaining each excluded study in Characteristics of excluded studies tables.

Risk of bias in included studies

We present our 'Risk of bias' assessments in Characteristics of included studies tables. We provide a graphic summary of our assessments (Figure 2; Figure 3).


Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.


Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

All included studies had randomly‐allocated treatments. Randomisation sequence generation was at low risk of bias in eight studies (Austin 2013; Crane 2004; Frontin 2011; Gray 2008; L'Her 2004; Levitt 2001; Masip 2000; Nava 2003) and at unclear risk of bias in 16 studies (Agmy 2008; Ducros 2011; El‐Refay 2016; Hao 2002; Kelly 2002; Li 2005; Lin 1991; Lin 1995; Moritz 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Thys 2002; Zokaei 2016).

Allocation concealment was at low risk of bias in seven studies (Austin 2013; Crane 2004; Frontin 2011; Gray 2008; L'Her 2004; Nava 2003; Thys 2002), at unclear risk of bias in 16 studies (Agmy 2008; Ducros 2011; El‐Refay 2016; Hao 2002; Kelly 2002; Levitt 2001; Li 2005; Lin 1991; Lin 1995; Moritz 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Zokaei 2016), and at high risk of bias in one study (Masip 2000).

Blinding

Blinding of participants and personnel was not possible for most studies, given the nature of the intervention. Twelve studies were at unclear risk of bias, as they did not state whether or not they had blinded participants and personnel (Hao 2002; Kelly 2002; Levitt 2001; Li 2005; Moritz 2003; Nava 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Zokaei 2016). Twelve studies did not blind participants and personnel (Agmy 2008; Austin 2013; Crane 2004; Ducros 2011; El‐Refay 2016; Frontin 2011; Gray 2008; L'Her 2004; Lin 1991; Lin 1995; Masip 2000; Thys 2002).

Blinding of outcome assessments was not explicitly stated for most studies. Three studies were at low risk of bias for outcome assessment blinding (Austin 2013; Ducros 2011; El‐Refay 2016), 19 studies were at unclear risk of bias (Agmy 2008; Crane 2004; Frontin 2011; Hao 2002; Kelly 2002; Levitt 2001; Li 2005; Lin 1991; Lin 1995; Masip 2000; Moritz 2003; Nava 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Thys 2002; Zokaei 2016), and two studies were at high risk of bias (Gray 2008; L'Her 2004).

Incomplete outcome data

Three studies had high risk of bias due to incomplete outcome data (El‐Refay 2016; Lin 1991; Park 2004). These studies excluded a significant number of participants after randomisation. Seven studies did not present enough information to judge the completeness of outcome data and were rated at unclear risk (Hao 2002; Levitt 2001; Li 2005; Moritz 2003; Park 2001; Takeda 1997; Zokaei 2016). Fourteen studies presented complete outcome data and were at low risk of bias (Agmy 2008; Austin 2013; Crane 2004; Ducros 2011; Frontin 2011; Gray 2008; Kelly 2002; L'Her 2004; Lin 1995; Masip 2000; Nava 2003; Räsänen 1985; Takeda 1998; Thys 2002).

Selective reporting

Most studies did not provide enough information to judge whether they included all planned outcomes. Three studies (Austin 2013; Frontin 2011; Gray 2008) were at low risk of bias for selective reporting, given that the authors performed prospective trial registration and reported all their prespecified outcomes. Eighteen studies did not present enough information to determine if the outcomes presented were prespecified, and were at unclear risk of bias (Crane 2004; El‐Refay 2016; Hao 2002; Kelly 2002; L'Her 2004; Levitt 2001; Li 2005; Lin 1991; Lin 1995; Masip 2000; Moritz 2003; Nava 2003; Park 2001; Park 2004; Räsänen 1985; Takeda 1997; Takeda 1998; Thys 2002). Three studies were at high risk of bias, as they failed to report prespecified outcomes or the outcomes reported were not in an extractable format (Agmy 2008; Ducros 2011; Zokaei 2016).

Other potential sources of bias

Two studies were reported in abstract form only (Agmy 2008; Austin 2013). Seven studies had high risk of other potential sources of bias (Agmy 2008; Ducros 2011; El‐Refay 2016; Hao 2002; L'Her 2004; Levitt 2001; Li 2005). Agmy 2008 was reported as a personal communication in addition to abstracts which did not have quantitative data. There is a discrepancy between counts reported on clinicaltrials.gov and the personal communication data. We have used the personal communication data in our analysis. Other potential sources of bias included study termination due to poor recruitment, low likelihood of significant findings, or presenting data without a measure of variance (e.g. standard deviation). Our funnel plot of hospital mortality for NPPV versus SMC demonstrates some asymmetry for studies of intermediate sample size, which may indicate the presence of publication bias (see Figure 4).


Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.1 HOSPITAL MORTALITY.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.1 HOSPITAL MORTALITY.

Effects of interventions

See: Summary of findings for the main comparison NPPV compared to SMC for cardiogenic pulmonary oedema

Hospital mortality

NPPV may reduce hospital mortality compared to SMC alone (RR 0.65, 95% CI 0.51 to 0.82; participants = 2484; studies = 21; I2 = 6%; Analysis 1.1; low quality of evidence), with a number needed to treat for an additional beneficial outcome (NNTB) of 17 (NNTB 12 to NNTB 32). We summarise our findings in summary of findings Table for the main comparison. We downgraded the evidence for hospital mortality by one level due to serious risk of bias (unclear or high risk of bias for randomisation sequence, allocation concealment, and other significant bias), and by one level due to imprecision (most trials had few participants and events, with a wide confidence interval).

We performed subgroup analysis by NPPV type (CPAP or bilevel), and found no significant difference between CPAP or bilevel NPPV (Analysis 1.2), based on the test for differences between subgroups (P = 0.64). Park 2001 reported no deaths in the bilevel NPPV arm and the SMC arm. El‐Refay 2016 reported no deaths in the CPAP or bilevel NPPV arms. CPAP may reduce hospital mortality (RR 0.65, 95% CI 0.48 to 0.88; participants = 1454; studies = 16; I2 = 9%) and bilevel NPPV may reduce hospital mortality (RR 0.72, 95% CI 0.53 to 0.98; participants = 1030; studies = 11; I2 = 0%).

Our treatment location subgroup analysis (Analysis 1.3) revealed no significant difference between participants treated with NPPV in the ER compared to the ICU, based on the test for differences between subgroups (P = 0.51). People treated with NPPV in the ICU may have reduced hospital mortality compared to SMC (RR 0.52, 95% CI 0.35 to 0.77; participants = 862; studies = 10; I2 = 0%). Similarly, people treated with NPPV in the ER may have reduced hospital mortality compared to SMC (RR 0.66, 95% CI 0.47 to 0.93; participants = 1596; studies = 10; I2 = 18%). Baseline PaCO2 subgroup analysis (Analysis 1.4) revealed a significant subgroup difference between studies with baseline eucapnia or hypercapnia (test for differences between subgroups P = 0.005). People with eucapnia treated with NPPV compared to SMC may have reduced hospital mortality (RR 0.41, 95% CI 0.27 to 0.63; participants = 581; studies = 10; I2 = 0%). In contrast, hypercapnic people may not benefit from NPPV compared to SMC (RR 0.82, 95% CI 0.65 to 1.03; participants = 1903; studies = 11; I2 = 0%).

Low‐risk‐of‐bias sensitivity analysis eliminated the observed reduction in hospital mortality with NPPV compared to SMC (RR 0.81, 95% CI 0.61 to 1.06; participants = 1505; studies = 7; Analysis 1.5; I2 = 6%). A sensitivity analysis that excluded studies with missing data (RR 0.68, 95% CI 0.54 to 0.86; participants = 2283; studies = 18; Analysis 1.6; I2 = 5%) indicates that NPPV may reduce hospital mortality. Finally, changing our statistical model from random‐effects to fixed‐effect also indicates that NPPV may reduce hospital mortality compared to SMC alone (RR 0.67, 95% CI 0.55 to 0.81; participants = 2484; studies = 21; Analysis 1.7; I2 = 6%).

Endotracheal intubation (ETI)

NPPV compared to SMC probably reduces ETI (RR 0.49, 95% CI 0.38 to 0.62; participants = 2449; studies = 20; Analysis 1.8; I2 = 0%; moderate quality of evidence) with a NNTB of 13 (NNTB 11 to NNTB 18). Two studies contributed data from only one of their three study arms: Kelly 2002 reported no ETI events for CPAP or SMC arms, and El‐Refay 2016 reported no ETI events for CPAP or bilevel NPPV.

Our subgroup analysis by NPPV type indicates no significant difference between CPAP or bilevel NPPV compared to SMC (test for differences between subgroups P = 0.75). Both CPAP compared to SMC (RR 0.46, 95% CI 0.34 to 0.62; participants = 1413; studies = 15; I2 = 0%) and bilevel NPPV compared to SMC (RR 0.50, 95% CI 0.31 to 0.81; participants = 1036; studies = 11; I2 = 23%) probably reduce ETI (Analysis 1.9). Treatment location subgroup analysis (ER or ICU, Analysis 1.10) revealed no significant difference between participants treated in the ER or ICU with NPPV compared to SMC (test for differences between subgroups P = 0.40). NPPV use in the ER setting probably reduces ETI (RR 0.60, 95% CI 0.37 to 0.96; participants = 1561; studies = 9; I2 = 21%). Participants treated with NPPV in the ICU setting probably have reduced ETI (RR 0.40, 95% CI 0.29 to 0.56; participants = 862; studies = 10; I2 = 0%). Our PaCO2 subgroup analysis (Analysis 1.11) revealed a significant difference between baseline eucapnic and hypercapnic studies (test for differences between subgroups P = 0.03). Both eucapnic participants treated with NPPV (RR 0.37, 95% CI 0.26 to 0.52; participants = 523; studies = 9; I2 = 0%) and hypercapnic participants treated with NPPV (RR 0.64, 95% CI 0.46 to 0.91; participants = 1926; studies = 11; I2 = 0%) probably have reduced ETI rates compared to SMC, but the effect was larger in those with PaCO2 ≤ 45 mmHg.

Low‐risk‐of‐bias sensitivity analysis indicates there was probably no ETI rate reduction with NPPV compared to SMC (RR 0.85, 95% CI 0.55 to 1.32; participants = 1491; studies = 6; Analysis 1.12; I2 = 0%). In contrast, sensitivity analysis that excluded studies with missing data (Analysis 1.13) indicates that NPPV probably reduces ETI rates compared to SMC (RR 0.52, 95% CI 0.40 to 0.69; participants = 2248; studies = 17; I2 = 0%). We performed additional post hoc analysis exploring the effect of exclusive use of face masks to deliver NPPV compared to studies with permissive use of nasal or face masks (Analysis 1.14). We found no significant subgroup differences. Exclusive face mask use (RR 0.52, 95% CI 0.40 to 0.69; participants = 2213; studies = 15; I2 = 0%) or permissive use of nasal masks (RR 0.35, 95% CI 0.20 to 0.62; participants = 236; studies = 5; I2 = 0%) probably reduced ETI.

Incidence of acute myocardial infarction (AMI)

There is probably little or no difference in AMI incidence with NPPV compared to SMC for ACPE (RR 1.03, 95% CI 0.91 to 1.16; participants = 1313; studies = 5; Analysis 1.15; I2 = 0%; moderate quality of evidence). The median length of follow‐up for AMI was three days. Subgroup analysis by NPPV type revealed no significant differences between CPAP or bilevel NPPV (Analysis 1.16; test for differences between subgroups P = 0.30). Both CPAP (RR 0.96, 95% CI 0.80 to 1.14; participants = 569; studies = 3; I2 = 0%) and bilevel NPPV (RR 1.09, 95% CI 0.92 to 1.29; participants = 744; studies = 5; I2 = 0%) probably have little or no difference in AMI incidence compared to SMC. Several trials reported no AMI events in either treatment arm: CPAP versus bilevel versus SMC (Park 2004), and bilevel NPPV versus SMC (Park 2001; Thys 2002). Study location subgroup analysis was not possible, as only one study (Park 2001) was conducted in an unknown location while the remaining studies were conducted in the ER. Baseline PaCO2 subgroup analysis was not possible, as only one study (Park 2001) had baseline eucapnia while the remaining studies had baseline hypercapnia. Our planned subgroup analysis comparing AMI rates before and after NPPV intervention was not possible, due to inconsistent AMI definitions (e.g. STEMI, NSTEMI, angina).

Intolerance to allocated treatment

We were interested in obtaining information on treatment intolerance of NPPV compared to SMC. Unfortunately, this outcome was inconsistently defined. Potential outcomes providing similar information included:

  • Treatment failure. This outcome was more commonly reported. Treatment failure could indicate the need for endotracheal intubation. This outcome was inconsistently defined and was often a clinical diagnosis. Criteria for endotracheal intubation are summarised in Table 4.

    • Crane 2004 reported on treatment failure, which was defined as worsening respiratory function or reduced level of consciousness. Treatment failure occurred in one SMC participant, four CPAP participants, and one bilevel NPPV participant. Only one CPAP participant and one bilevel NPPV participant were intubated.

    • Ducros 2011 reported on the presence of intubation criteria, which they defined as medically refractory hypoxaemia, loss of consciousness, psychomotor agitation, or haemodynamic instability. They reported 13 SMC participants and four CPAP participants meeting intubation criteria. Only six SMC participants and three CPAP participants were intubated.

    • Kelly 2002 reported no treatment failures in the CPAP group and two treatment failures in the SMC group.

    • L'Her 2004 reported no treatment failures or changes to the assigned treated after one hour of therapy. After 12 hours of treatment there were 11 SMC participants with coma and one CPAP participant with coma.

    • Lin 1991 reported treatment failure in five CPAP participants and 10 SMC participants during the first three hours of treatment, leading to endotracheal intubation.

    • Lin 1995 reported nine CPAP participants and 17 SMC participants meeting treatment failure criteria. Only six CPAP participants and 12 SMC participants were ultimately intubated, as the final decision was based on clinical judgement.

    • Räsänen 1985 reported on treatment failure after three hours of therapy in 13 SMC participants and seven CPAP participants.

    • Thys 2002 defined treatment failure based on clinical deterioration assessed by the clinician. Thys 2002 reported that all five participants assigned to SMC had treatment failure with placebo NPPV, and were switched over to active NPPV therapy.

  • Treatment intolerance was inconsistently defined.

    • Agmy 2008 reported one event in each treatment group (CPAP, bilevel NPPV, and SMC) of NPPV intolerance leading to ETI.

    • Gray 2008 reported rates of participants changing to a new treatment, i.e. 65 of 363 SMC participants, 55 of 340 CPAP participants, and 85 of 352 participants for bilevel NPPV.

Given the heterogeneity in clinical definitions of treatment failure and our desired outcome being intolerance to allocated treatment, we are unable to draw any conclusions on the effect of NPPV compared to SMC for ACPE.

Hospital and ICU length of stay

We are uncertain whether NPPV reduces hospital length of stay compared to SMC for ACPE (MD −0.31 days, 95% CI −1.23 to 0.61; participants = 1714; studies = 11; Analysis 1.17; I2 = 55%; very low quality of evidence). The mean hospital length of stay was 9.65 days. Subgroup analysis by NPPV type found no significant subgroup differences (P = 0.22) between participants treated with CPAP (MD −0.52 days, 95% CI −1.77 to 0.72; participants = 943; studies = 7; I2 = 59%) or with bilevel NPPV (MD 0.39 days, 95% CI −0.35 to 1.13; participants = 771; studies = 6; I2 = 0%) compared to SMC for hospital length of stay (Analysis 1.18). Treatment location subgroup analysis identified no significant difference (P = 0.84) between ACPE participants treated with NPPV in the ER (MD −0.38 days, 95% CI −1.70 to 0.93; participants = 1455; studies = 8; I2 = 68%) compared to the ICU (MD −0.21 days, 95% CI −1.30 to 0.89; participants = 259; studies = 3; Analysis 1.19; I2 = 0%). Subgroup analysis by baseline PaCO2 (Analysis 1.20) found a significant difference (P = 0.01) in hospital length of stay between eucapnic ACPE participants (MD −1.18 days, 95% CI −2.33 to −0.04; participants = 397; studies = 5; I2 = 44%) and hypercapnic ACPE participants (MD 0.60 days, 95% CI −0.15 to 1.34; participants = 1317; studies = 6; I2 = 7%). However, the evidence quality was very low, and we are uncertain whether eucapnic or hypercapnic ACPE participants had reduced hospital lengths of stay.

We are uncertain whether NPPV reduces ICU length of stay compared to SMC for ACPE. We found significant heterogeneity with ICU length‐of‐stay data which were reported by six studies (Ducros 2011; Frontin 2011; Lin 1995; Takeda 1997; Thys 2002; Zokaei 2016). In addition, the quality of evidence was very low. We did not pool ICU length‐of‐stay data, due to significant heterogeneity (Analysis 1.21). We downgraded the quality of evidence for ICU length of stay by one level due to serious risk of bias. Most information was from studies at low or unclear risk of bias. Furthermore, lack of blinding was likely to lower confidence in our ICU length‐of‐stay findings. We also downgraded the evidence quality by one level due to data inconsistency. ICU length‐of stay‐data were heterogeneous with an I2 of 99% and visually‐evident heterogeneity. Finally, we downgraded by one level for imprecision, because most of the trials had few participants, with the confidence interval crossing a mean difference of 0.

Vital signs one hour after intervention

Systolic blood pressure (SBP)

There is probably little or no difference in SBP after one hour in people with ACPE treated with NPPV compared to SMC (MD −1.72 mmHg, 95% CI −5.03 to 1.58; participants = 1408; studies = 7; Analysis 1.22; I2 = 0%; moderate quality of evidence). The mean SBP was 128.4 mmHg in the SMC group. We downgraded the quality of evidence for SBP by one level for imprecision, as the included studies had few participants and wide confidence intervals. Subgroup analysis by NPPV type (Analysis 1.23) found no significant subgroup difference (P = 0.95) between SBP in people with ACPE treated with CPAP (MD −1.65 mmHg, 95% CI −5.58 to 2.28; participants = 866; studies = 7; I2 = 0%) or bilevel NPPV (MD −1.89 mmHg, 95% CI −8.01 to 4.23; participants = 542; studies = 3; I2 = 0%) compared to SMC.

Diastolic blood pressure (DBP)

People with ACPE treated with NPPV compared to SMC probably have little or no difference in DBP after one hour of treatment (MD 1.46 mmHg, 95% CI −1.86 to 4.78; participants = 1361; studies = 6; Analysis 1.24; I2 = 42%; moderate quality of evidence). The mean DBP was 71.9 mmHg in the SMC group. We downgraded DBP quality of evidence by one level for imprecision, because the included studies had few participants and wide confidence intervals. Subgroup analysis by NPPV type (Analysis 1.25) demonstrated no significant subgroup difference (P = 0.96) between DBP in people with ACPE treated with CPAP (MD 0.92 mmHg, 95% CI −3.92 to 5.75; participants = 823; studies = 6; I2 = 64%) or bilevel NPPV (MD 1.08, 95% CI −2.88 to 5.04; participants = 538; studies = 3; I2 = 0%) compared to SMC.

Mean blood pressure (MBP)

There is probably little or no difference in MBP after one hour for people with ACPE treated with NPPV compared to SMC (MD ‐2.50 mmHg, 95% CI ‐8.29 to 3.30; participants = 251; studies = 3; I2 = 20%; Analysis 1.26; moderate quality of evidence). The mean MBP was 101 mmHg. We downgraded MBP quality of evidence by one level for imprecision, because the included studies had few participants and wide confidence intervals. We did not perform subgroup analysis by NPPV type, due to the small number of studies.

Respiratory rate (RR)

NPPV for ACPE probably reduces the RR after one hour of treatment compared to SMC (MD −1.87 breaths a minute, 95% CI −2.70 to −1.03; participants = 1636; studies = 10; Analysis 1.27; I2 = 17%; moderate quality of evidence). The mean RR after one hour of therapy was 27.4 breaths a minute in the SMC group. We downgraded RR quality of evidence by one level for imprecision, because the included studies had few participants and wide confidence intervals. Subgroup analysis by NPPV type (Analysis 1.28) demonstrated no significant subgroup differences (P = 0.70) in RR after one hour between CPAP compared to SMC (MD −1.64 breaths a minute, 95% CI −2.41 to −0.87; participants = 1107; studies = 10; I2 = 0%) and bilevel NPPV compared to SMC (MD −2.17 breaths a minute, 95% CI −4.69 to 0.35; participants = 529; studies = 3; I2 = 38%).

Arterial oxygen concentration (PaO2) after one hour of intervention

NPPV compared to SMC for ACPE may improve PaO2 after one hour of therapy (MD 16.19 mmHg, 95% CI 3.54 to 28.84; participants = 1428; studies = 10; Analysis 1.29; I2 = 91%; low quality of evidence). The mean PaO2 after one hour of therapy was 133.9 mmHg in the SMC group. We downgraded PaO2 quality of evidence by one level due to imprecision. Most trials had few participants and the confidence interval crossed the line of no effect. We downgraded by an additional level for inconsistency, because the PaO2 data were visually heterogeneous, with an I2 of 91%. Our subgroup analysis by NPPV type (Analysis 1.30) demonstrated no significant subgroup differences (P = 0.42) in PaO2 after one hour between CPAP compared to SMC (MD 9.55 mmHg, 95% CI −9.10 to 28.19; participants = 761; studies = 8; I2 = 88%) and bilevel NPPV compared to SMC (MD 19.50 mmHg, 95% CI 4.29 to 34.71; participants = 667; studies = 6; I2 = 86%). We did not pool PaCO2 and pH after one hour of treatment, as these outcomes require assessment compared to baseline values. In addition, we were unable to pool the treatment failure outcome, as this composite outcome of mortality, ETI rate, and treatment intolerance was not reported.

Adverse events

Eleven trials (Agmy 2008; Crane 2004; Ducros 2011; Frontin 2011; Gray 2008; L'Her 2004; Lin 1995; Masip 2000; Nava 2003; Park 2004; Räsänen 1985) compared NPPV against SMC for ACPE and reported adverse events (Analysis 1.31). There were 228 adverse events reported in 1230 NPPV participants compared to 116 adverse events for 808 SMC participants. Adverse events included skin damage, pneumonia, gastrointestinal bleeding, gastric distention, vomiting, pneumothorax, sinusitis, mask discomfort, hypotension, arrhythmia, cardiorespiratory arrest, gastric aspiration, stroke, seizures, claustrophobia, and hypercapnia.

Three trials (Gray 2008; Masip 2000; Nava 2003) reported that participants treated with NPPV probably have increased mask discomfort compared to SMC (RR 12.59, 95% CI 2.39 to 66.28; participants = 1100; studies = 3; I2 = 0%; moderate quality of evidence). We downgraded the evidence for mask discomfort by one level for imprecision, because there were few events and wide confidence intervals.

Three trials (Agmy 2008; Nava 2003; Räsänen 1985) reported worsening hypercapnia with treatment. Participants treated with NPPV or SMC for ACPE probably have little or no difference in rates of worsening hypercapnia. We downgraded the evidence quality for hypercapnia by one level for imprecision, due to few events and wide confidence intervals.

Park 2004 (n = 83) reported a higher incidence of gastric distention in the NPPV group (13/56 in NPPV group, 0/27 in SMC group; RR 13.26, 95% CI 0.82 to 215.12), but the number of events was low and the confidence interval wide. There was some weak evidence of a lower incidence of cardiorespiratory arrest in the NPPV group (24/836 (NPPV) versus 26/516 (SMC); RR 0.60, 95% CI 0.34 to 1.05). There was no evidence of a difference between groups for the other adverse events.

Overall, we assessed the evidence for adverse events to be of low quality. We downgraded the evidence by one level for serious risk of bias (unclear or high risk of bias in at least one domain), and by one additional level for imprecision (the included studies had few participants and wide confidence intervals).

Discussion

Summary of main results

Our updated systematic review and meta‐analysis includes 24 RCTs with 2664 participants. Compared to standard medical care (SMC), the use of non‐invasive positive pressure ventilation (NPPV) for acute cardiogenic pulmonary oedema (ACPE) may reduce hospital mortality and probably reduces endotracheal intubation (ETI) rates. In addition, there was probably little or no difference between NPPV and SMC for rates of acute myocardial infarction (AMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), or mean blood pressure (MBP) after one hour of therapy. We are uncertain if NPPV reduces hospital length of stay. However, NPPV (compared to SMC) probably improves respiratory rate after one hour of treatment, may slightly improve PaO2 after one hour of therapy, and probably increases mask discomfort.

NPPV use for ACPE may reduce hospital mortality compared to SMC alone, with a number needed to treat for an additional beneficial outcome (NNTB) of 17 (NNTB 12 to NNTB 32) (summary of findings Table for the main comparison). Our subgroup analysis by NPPV type identified no significant difference between continuous positive airway pressure (CPAP) and bilevel NPPV subgroups. Both NPPV forms may reduce hospital mortality compared to SMC. We found no significant difference between people with ACPE treated in the Emergency Room (ER) or intensive care unit (ICU) in our subgroup analysis. NPPV may reduce hospital mortality compared to SMC in both treatment locations. Our baseline PaCO2 subgroup analysis for people with ACPE treated with NPPV compared to SMC revealed a significant difference between studies with baseline average eucapnia and hypercapnia. Eucapnic people with ACPE treated with NPPV may have reduced hospital mortality, while hypercapnic people may have little or no difference in hospital mortality. The non‐significant hospital mortality benefit with hypercapnia may have been due to the greater weight of the trial by Gray 2008 (conducted in the ER setting with baseline hypercapnia), which did not show any hospital mortality benefit with NPPV. Prior smaller studies have suggested enhanced benefit for people with baseline hypercapnia, due to their higher risk of complications and NPPV potentially preventing endotracheal intubation (Masip 2000; Moritz 2007; Nava 2003). Our baseline PaCO2 subgroup analysis did not support the idea that hypercapnic people with ACPE would benefit more compared to eucapnic people with ACPE. One possible reason for this discrepancy was that mean PaCO2 values could hide hypercapnic people who derived significant benefit from NPPV or clinical worsening with SMC. Finally, our sensitivity analysis limited to low‐risk‐of‐bias studies found NPPV may make little or no difference in hospital mortality compared to SMC. In contrast, our sensitivity analysis limited to studies without missing data indicated that NPPV for ACPE may reduce hospital mortality compared to SMC alone.

NPPV in ACPE probably reduces ETI compared to SMC alone, with a NNTB of 13 (NNTB 11 to 18). We did not find a significant difference between trials comparing CPAP or bilevel NPPV in our subgroup analysis. Both forms of NPPV probably reduce ETI compared to SMC. We found no significant difference between trials comparing NPPV in the ER or ICU setting in our subgroup analysis. People with ACPE treated with NPPV in the ER or ICU probably have reduced ETI compared to SMC. Baseline PaCO2 subgroup analysis revealed a significant difference between baseline eucapnic and hypercapnic people with ACPE treated with NPPV compared to SMC. Both eucapnic and hypercapnic people with ACPE probably have reduced ETI compared to SMC. However, eucapnic people probably get more benefit compared to hypercapnic people. Sensitivity analysis limited to low‐risk‐of‐bias studies found NPPV use probably makes little or no difference to ETI compared with SMC. In contrast, sensitivity analysis limited to studies without missing data indicated NPPV for ACPE probably reduces ETI compared to SMC alone. Our post hoc subgroup analysis which compared exclusive use of face masks to permissive use of nasal masks revealed no significant subgroup differences, and both NPPV subgroups probably reduce ETI compared to SMC alone.

Several trials (Mehta 1997; Rusterholtz 1999; Sharon 2000) reported increased AMI incidence with NPPV compared to SMC. We explored the safety of NPPV by identifying studies that reported AMI during or after NPPV initiation, compared to SMC. We found there was probably little or no difference in AMI incidence with NPPV. Subgroup analysis by NPPV type did not identify a difference between CPAP or bilevel NPPV studies. An important limitation of our AMI analysis was the inconsistent AMI definitions used across studies (e.g. ST elevation MI, ECG changes, biomarker positive, symptoms) and many studies only provided baseline AMI rates (Frontin 2011, L'Her 2004, Lin 1991, Lin 1995, Masip 2000, Räsänen 1985, Takeda 1997).

We were unable to perform an analysis of NPPV treatment intolerance compared to SMC in ACPE, due to the outcome not being reported. Furthermore, treatment failure which was reported had inconsistent definitions. Treatment failure data suggest that not all participants meeting treatment failure criteria were intubated. Possible reasons include changes to the underlying medical therapy, participant goals of care, and the difference between a quantitative description of failure based on vital signs and blood gas values and the clinical evaluation of treatment failure.

NPPV could reduce hospital and ICU length of stay for ACPE due to reduced ETI and avoidance of its complications (e.g. pneumonia and tracheal injury) (Gay 2009). Ideally, studies would have provided data on ICU or hospital‐free days as participants who die early during the trial would have a short hospital or ICU length of stay, despite not benefiting from treatment. Unfortunately, hospital‐free days were not reported. We are uncertain whether NPPV reduces hospital length of stay compared to SMC for ACPE. We found no subgroup differences between participants treated with CPAP or bilevel NPPV. In addition, we found no subgroup differences between studies comparing NPPV to SMC in the ER or ICU setting. Our subgroup analysis by baseline PaCO2 indicated a significant difference between eucapnic and hypercapnic studies. However, given the very low quality of evidence we are uncertain whether NPPV reduces hospital length of stay in eucapnic people with ACPE. We were unable to pool ICU length‐of‐stay data due to significant heterogeneity. Based on the available data, we are uncertain whether NPPV reduces ICU length of stay.

We pooled mean blood pressure values obtained after one hour of therapy to identify differences between NPPV and SMC. In our comparison of NPPV to SMC in people with ACPE we found there is probably little or no difference between SBP, DBP, and MBP after one hour of therapy. These non‐significant differences may represent a true lack of difference between therapies for reducing blood pressure, or the inability of mean blood pressure values to convey any potential benefit or harm with NPPV over SMC in individuals with significant hypertension or hypotension. Our analysis of mean respiratory rate data after one hour of therapy with NPPV or SMC found that NPPV probably reduces the respiratory rate compared to SMC in ACPE. Furthermore, our analysis of PaO2 after one hour of therapy demonstrated that NPPV may improve PaO2 slightly compared to SMC. Finally, 16 adverse event types were reported in 11 studies. We found higher rates of face discomfort when using NPPV compared to SMC.

Overall completeness and applicability of evidence

Our objective in performing this systematic review and meta‐analysis update was to evaluate the impact of new literature published since the last version of this review (May 2013, Vital 2013) on the safety and effectiveness of NPPV compared to SMC for ACPE. Furthermore, we have updated the protocol and conduct of the review to recently‐revised Cochrane standards (Higgins 2018). We reassessed the literature and performed a new assessment for study eligibility, which identified 24 RCTs for inclusion. Of these, seven are new studies published since the last version of this review (Austin 2013; Ducros 2011; El‐Refay 2016; Hao 2002; Li 2005; Moritz 2003; Zokaei 2016). Three newly‐included studies are based on a reassessment of previously excluded studies (Hao 2002; Li 2005; Moritz 2003), which lacked extractable outcomes. The outcomes in these trials were measured on different time scales for our secondary outcomes: Li 2005 (PaO2, respiratory rate), Hao 2002 (PaO2, respiratory rate, SBP), and Moritz 2003 (SBP, DBP). Only Hao 2002 provided data on ETI as a primary outcome.

Importantly, we also excluded 15 studies that we had included in the previous version of this review but that no longer meet the revised inclusion criteria. For transparency, these were four quasi‐randomised controlled trials (Bersten 1991; Moritz 2007; Sharon 2000; Weitz 2007), three studies conducted on an ineligible population (Bautin 2005; Delclaux 2000; Ferrer 2003) and eight studies which compared CPAP to bilevel NPPV without a SMC arm (Bellone 2004; Bellone 2005; Ferrari 2007; Ferrari 2010; Fontanella 2010; Liesching 2014; Martin‐Bermudez 2002; Mehta 1997).

Overall, the evidence we identified was able to address our review objective. Studies that we identified focused on a single acute intervention (NPPV versus SMC) for people with ACPE provided during their hospital presentation. Our inclusion criteria for adults with ACPE were met. However, diagnosis of ACPE across studies was not standardised but based on a clinical assessment supported by clinical signs, chest x‐ray, and exclusion of important alternative causes of dyspnoea (e.g. chronic obstructive pulmonary disease (COPD) and pneumonia). An important limitation of the studies we identified was the lack of baseline demographic data that could explain why certain recruited participants were at higher mortality risk than others (e.g. severe untreated valvular heart disease or AHA stage D heart failure). For example, the mortality rates reported by included studies were extremely disparate, ranging from 0 to 27.91% for NPPV (mean 9.64%) and 0 to 63.64% for SMC (mean 21.06%). These large variations in mortality may be influenced by the recruitment of ACPE participants with different underlying risks of mortality, regardless of their aggressive medical treatment with SMC or NPPV or both, and subtle differences in SMC between the different countries in which the included trials were performed.

Regarding the interventions under study, the trials included CPAP or bilevel NPPV or both, that had adjustable settings. We tried to ascertain whether the specific devices and pressure settings were reflective of contemporary practice. However, the NPPV device settings were not completely reported for all included trials. Where available, we have reported the mean inspiratory positive airway pressure (IPAP), expiratory positive airway pressure (EPAP), positive end‐expiratory pressure (PEEP), and duration of NPPV used by each study (Table 1). For studies where this information was available, the devices and settings do reflect contemporary practice.

In addition, the ideal trial would have compared both forms of NPPV against SMC, given the possibility that CPAP versus bilevel NPPV may differ in their treatment efficacy. Only six of 24 included trials used a three‐arm design. One such study (Gray 2008) was the largest RCT performed to date (N = 1069). The other five studies had smaller sample sizes (range 26 to 129; mean 73). The remaining studies compared one type of NPPV (bilevel NPPV or CPAP) against SMC. Subgroup analysis by type of NPPV did not reveal any significant differences for the primary or main secondary outcomes. Current ACPE treatment practices vary from targeting an oxygen saturation of greater than 92% by titration of FiO2, initiation of NPPV, or endotracheal intubation (Ezekowitz 2017). Finally, we considered whether the SMC used in each trial reflected current treatment guidelines (Ezekowitz 2017; Ponikowski 2016). Such contemporary guidelines for acute heart failure recommend therapy with intravenous loop diuretics, supplemental oxygen, and nitrates, while maintaining an adequate blood pressure (Ezekowitz 2017; Ponikowski 2016). Our summary of the SMC used in each trial is listed in Table 2. All studies provided loop diuretics. However, the doses provided were unclear in nine studies. Nitrates were provided in all trials, but the form and dose was unclear in two studies. Similarly, each trial provided supplemental oxygen, but the specific FiO2 used was unclear in two studies. In summary, the NPPV and SMC used by the included studies generally reflected contemporary clinical practice

Quality of the evidence

All outcomes were assessed according to GRADE recommendations (Guyatt 2008; Higgins 2017). For the main outcomes, evidence quality ranged from moderate (ETI, AMI), to low (hospital mortality, adverse events), to very low (hospital length of stay). Hospital mortality was assessed in 21 studies that compared NPPV versus SMC and included 2484 patients. Key methodological limitations of the meta‐analysis include the clinical heterogeneity of included studies, such as variation in the aetiology of cardiogenic pulmonary oedema (e.g. myocardial ischaemia, severe valvular heart disease) and the lack of clinical context for each individual participant (e.g. home therapies, ACC/AHA heart failure grade). Additional limitations were the impracticality of blinding participants, personnel, and outcome assessment, given the nature of NPPV. Lack of blinding may have influenced objective outcomes (e.g. hospital mortality or ETI) because participants treated with SMC could potentially cross over to NPPV if they appeared to be clinically deteriorating, thereby attenuating the benefit of NPPV when using intention‐to‐treat analysis. Another limitation of the included studies was the unclear risk of selective reporting bias, as reflected by the inconsistent reporting of outcomes. Finally, most studies recruited a low number of participants and their primary outcome was not hospital mortality. Interestingly, the well‐powered study by Gray 2008 which had hospital mortality as a primary outcome did not detect a mortality difference between NPPV and SMC.

Our GRADE quality of evidence assessments for our primary outcomes were influenced by a number of factors, including risk of bias, imprecision, and inconsistency. Specifically, we rated the hospital mortality outcome as low quality, due to serious concerns with risks of bias and imprecision. We downgraded evidence quality by one level for risk of bias because most trials were rated as having unclear or high risk of bias for key domains, which included: randomisation sequence generation, allocation concealment, selective reporting bias, and other significant bias. We also downgraded this outcome for imprecision because there were few mortality events and the confidence interval included the appreciable benefit risk ratio of 0.75. The funnel plot for hospital mortality (Figure 4) was also asymmetrical, which suggests the presence of publication bias. However, we did not further downgrade the evidence for this outcome, as the asymmetry could be explained by the poor design of smaller studies included in our review, clinical heterogeneity, and sampling variation. Furthermore, the findings were consistent in a post hoc fixed‐effect sensitivity analysis (Analysis 1.7; Sterne 2011).

We rated the quality of evidence for ETI as moderate. We downgraded this outcome by one level, due to serious concerns with risks of bias. Most trials that provided ETI evidence had unclear or high risk of bias for key domains, which included: randomisation sequence generation, allocation concealment, selective reporting bias, and other significant bias. Our ETI funnel plot (Figure 5) was symmetrical, which suggests the absence of graphical evidence of publication bias. For AMI, the quality of evidence was moderate. We downgraded the evidence by one level for serious concerns with imprecision. We downgraded for imprecision because few trials reported AMI events, the confidence interval crossed a risk ratio of 1, and the AMI definition varied between trials. We rated hospital length‐of‐stay evidence quality as very low, and downgraded it for the following reasons: serious risk of bias due to lack of blinding, visually heterogeneous data, and high imprecision with few participants and a confidence interval crossing a mean difference of 0. Our hospital length‐of‐stay funnel plot (Figure 6) was symmetrical and we therefore did not downgrade the evidence for publication bias. Adverse events evidence quality was low, and downgraded for the following reasons: serious risk of bias due to most studies having at least one domain at unclear or high risk of bias, and imprecision with wide confidence intervals crossing a risk ratio of 1 for many of the reported adverse events. For our secondary outcomes SBP, DBP, MBP, respiratory rate and PaO2, the quality of evidence was moderate. We downgraded SBP, DBP, MBP, and respiratory rate evidence quality by one level for imprecision because the included studies had few participants and wide confidence intervals. We downgraded PaO2 evidence quality by one level due to imprecision, because most trials had few participants and the confidence interval crossed a mean difference of 0. In summary, the quality of evidence was low or very low for most primary outcomes, except for ETI and AMI, which were moderate and have important clinical significance.


Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.8 ETI RATE.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.8 ETI RATE.


Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.17 HOSPITAL LENGTH OF STAY.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.17 HOSPITAL LENGTH OF STAY.

Potential biases in the review process

Our review is strengthened by using prespecified methods to reduce bias. We reduced bias in our search strategy by searching eight databases including clinical trial databases without language restrictions. After obtaining reference lists, we minimised bias through duplicate independent screening, duplicate data abstraction, and prespecified criteria for appraising methodological quality. In addition, the study protocol was established prospectively and followed the Cochrane format for systematic reviews (Higgins 2018; Jadad 2000; McKibbon 2004). Furthermore, two of our review authors (NB, FV) obtained additional data from study authors for outcomes incompletely described in their initial publication (Agmy 2008; Austin 2013; Crane 2004; Gray 2008; Kelly 2002; Masip 2000; Nava 2003; Park 2004; Räsänen 1985; Takeda 1998; Thys 2002). Our meta‐analysis includes the largest number of trials exploring the role of NPPV compared to SMC in ACPE and had increased power in pooling outcomes compared to earlier reviews (Potts 2009; Weng 2010). We performed extensive analysis of the impact of NPPV on a broad range of clinically important outcomes and physiological parameters. Furthermore, our results were robust to the exclusion of studies with missing data. The generalisability of our findings is enhanced by the broad range of treatment locations and countries included, representing international experience with NPPV for ACPE.

The small number of events (e.g. mortality, ETI, AMI) within identified trials is one important limitation of our review and meta‐analysis. Additionally, the underlying heterogeneity of included ACPE participants may have influenced our findings on hospital mortality and ETI. Several important prognostic factors were not clearly reported by trial investigators, such as the participants’ underlying heart failure severity, compliance with guideline‐appropriate home therapy, and presence of heart failure aetiologies requiring surgical interventions (e.g. severe valvular heart disease). Furthermore, there were slight variations in SMC among studies, which suggest that clinical heterogeneity may have influenced some of the treatment effects. Additionally, the definition of AMI was inconsistent across studies. Finally, intubation criteria were not standardised and variations in clinical practice could have influenced the observed differences in ETI.

Agreements and disagreements with other studies or reviews

The first systematic review on NPPV for ACPE included three trials and reported a significant reduction in ETI with CPAP compared to SMC (Pang 1998). Hess 2004 expressed concern about the use of NPPV for ACPE, due to greater AMI rates in two studies and no significant benefit for mortality or ETI. Nadar 2005 identified reduced ETI for CPAP compared to SMC with no significant mortality difference. Several meta‐analyses have reported similar conclusions to our own (Collins 2006; Ho 2006; Masip 2005; Peter 2006; Weng 2010; Winck 2006), despite small differences in methodology. The results of previous systematic reviews were challenged by a large new RCT (Gray 2008), which found no significant differences in hospital mortality or ETI between CPAP, bilevel NPPV, and SMC. Important caveats within Gray 2008 include exclusion of people requiring an emergent intervention who could possibly benefit from NPPV, exclusion of people with AMI undergoing an intervention, and cross‐overs between treatment arms of 19.9% (SMC), 16.2% (CPAP), and 24.1% (bilevel NPPV) (Gray 2008). There was evidence in Gray 2008 that SMC cross‐overs had more severe illness, and their cross‐over to NPPV could have attenuated the benefit with NPPV compared to SMC. Nevertheless, as the largest single trial to date, Gray 2008 is an important trial in the meta‐analysis and its results contrast with those of smaller included studies, which tend to show greater effect estimates favouring NPPV over SMC. The negative results of Gray 2008 make it imperative that we are cautious in interpreting the overall findings of our meta‐analysis, which demonstrated that NPPV may reduce hospital mortality and probably reduces ETI when compared to SMC.

Our main findings are important and add support to current European heart failure guidelines (Ponikowski 2016). This guideline recommends that NPPV be considered and initiated promptly in people with ACPE presenting with tachypnoea (respiratory rate higher than 25 breaths a minute) and hypoxia (SpO2 less than 90%). In contrast, our findings may challenge Canadian heart failure guidelines which recommend against routine use of NPPV in ACPE (Ezekowitz 2017). The Canadian heart failure guideline suggests starting with SMC and considering NPPV if there is persistent hypoxia (Ezekowitz 2017). Based on our findings, their recommendation may be too conservative. Furthermore, NPPV is a relatively benign intervention. The main harms that were evident in our review were increased mask discomfort compared to SMC. Overall, the results of our review suggest that people with ACPE may experience more benefits than harms from the addition of NPPV to SMC for ACPE.

PRISMA statement flow diagram for 2019 review update.
Figures and Tables -
Figure 1

PRISMA statement flow diagram for 2019 review update.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.
Figures and Tables -
Figure 2

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.
Figures and Tables -
Figure 3

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.1 HOSPITAL MORTALITY.
Figures and Tables -
Figure 4

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.1 HOSPITAL MORTALITY.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.8 ETI RATE.
Figures and Tables -
Figure 5

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.8 ETI RATE.

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.17 HOSPITAL LENGTH OF STAY.
Figures and Tables -
Figure 6

Funnel plot of comparison: 1 NPPV vs SMC, outcome: 1.17 HOSPITAL LENGTH OF STAY.

Comparison 1 NPPV vs SMC, Outcome 1 HOSPITAL MORTALITY.
Figures and Tables -
Analysis 1.1

Comparison 1 NPPV vs SMC, Outcome 1 HOSPITAL MORTALITY.

Comparison 1 NPPV vs SMC, Outcome 2 Hospital mortality ‐ by NPPV.
Figures and Tables -
Analysis 1.2

Comparison 1 NPPV vs SMC, Outcome 2 Hospital mortality ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 3 Hospital mortality ‐ by location.
Figures and Tables -
Analysis 1.3

Comparison 1 NPPV vs SMC, Outcome 3 Hospital mortality ‐ by location.

Comparison 1 NPPV vs SMC, Outcome 4 Hospital mortality ‐ by baseline PaCO2.
Figures and Tables -
Analysis 1.4

Comparison 1 NPPV vs SMC, Outcome 4 Hospital mortality ‐ by baseline PaCO2.

Comparison 1 NPPV vs SMC, Outcome 5 Hospital mortality ‐ sensitivity analysis (low risk of bias).
Figures and Tables -
Analysis 1.5

Comparison 1 NPPV vs SMC, Outcome 5 Hospital mortality ‐ sensitivity analysis (low risk of bias).

Comparison 1 NPPV vs SMC, Outcome 6 Hospital mortality ‐ sensitivity analysis (missing data).
Figures and Tables -
Analysis 1.6

Comparison 1 NPPV vs SMC, Outcome 6 Hospital mortality ‐ sensitivity analysis (missing data).

Comparison 1 NPPV vs SMC, Outcome 7 Hospital mortality ‐ sensitivity analysis (fixed‐effect).
Figures and Tables -
Analysis 1.7

Comparison 1 NPPV vs SMC, Outcome 7 Hospital mortality ‐ sensitivity analysis (fixed‐effect).

Comparison 1 NPPV vs SMC, Outcome 8 ETI RATE.
Figures and Tables -
Analysis 1.8

Comparison 1 NPPV vs SMC, Outcome 8 ETI RATE.

Comparison 1 NPPV vs SMC, Outcome 9 ETI rate ‐ by NPPV.
Figures and Tables -
Analysis 1.9

Comparison 1 NPPV vs SMC, Outcome 9 ETI rate ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 10 ETI rate ‐ by location.
Figures and Tables -
Analysis 1.10

Comparison 1 NPPV vs SMC, Outcome 10 ETI rate ‐ by location.

Comparison 1 NPPV vs SMC, Outcome 11 ETI rate ‐ by baseline PaCO2.
Figures and Tables -
Analysis 1.11

Comparison 1 NPPV vs SMC, Outcome 11 ETI rate ‐ by baseline PaCO2.

Comparison 1 NPPV vs SMC, Outcome 12 ETI rate ‐ sensitivity analysis (low risk of bias).
Figures and Tables -
Analysis 1.12

Comparison 1 NPPV vs SMC, Outcome 12 ETI rate ‐ sensitivity analysis (low risk of bias).

Comparison 1 NPPV vs SMC, Outcome 13 ETI rate ‐ sensitivity analysis (missing data).
Figures and Tables -
Analysis 1.13

Comparison 1 NPPV vs SMC, Outcome 13 ETI rate ‐ sensitivity analysis (missing data).

Comparison 1 NPPV vs SMC, Outcome 14 ETI rate ‐ by face mask type.
Figures and Tables -
Analysis 1.14

Comparison 1 NPPV vs SMC, Outcome 14 ETI rate ‐ by face mask type.

Comparison 1 NPPV vs SMC, Outcome 15 ACUTE MI INCIDENCE.
Figures and Tables -
Analysis 1.15

Comparison 1 NPPV vs SMC, Outcome 15 ACUTE MI INCIDENCE.

Comparison 1 NPPV vs SMC, Outcome 16 Acute MI incidence ‐ by NPPV.
Figures and Tables -
Analysis 1.16

Comparison 1 NPPV vs SMC, Outcome 16 Acute MI incidence ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 17 HOSPITAL LENGTH OF STAY.
Figures and Tables -
Analysis 1.17

Comparison 1 NPPV vs SMC, Outcome 17 HOSPITAL LENGTH OF STAY.

Comparison 1 NPPV vs SMC, Outcome 18 Hospital length of stay ‐ by NPPV.
Figures and Tables -
Analysis 1.18

Comparison 1 NPPV vs SMC, Outcome 18 Hospital length of stay ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 19 Hospital length of stay ‐ by location.
Figures and Tables -
Analysis 1.19

Comparison 1 NPPV vs SMC, Outcome 19 Hospital length of stay ‐ by location.

Comparison 1 NPPV vs SMC, Outcome 20 Hospital length of stay ‐ by baseline PaCO2.
Figures and Tables -
Analysis 1.20

Comparison 1 NPPV vs SMC, Outcome 20 Hospital length of stay ‐ by baseline PaCO2.

Comparison 1 NPPV vs SMC, Outcome 21 ICU LENGTH OF STAY.
Figures and Tables -
Analysis 1.21

Comparison 1 NPPV vs SMC, Outcome 21 ICU LENGTH OF STAY.

Comparison 1 NPPV vs SMC, Outcome 22 SYSTOLIC BP AFTER ONE HOUR.
Figures and Tables -
Analysis 1.22

Comparison 1 NPPV vs SMC, Outcome 22 SYSTOLIC BP AFTER ONE HOUR.

Comparison 1 NPPV vs SMC, Outcome 23 Systolic BP after one hour ‐ by NPPV.
Figures and Tables -
Analysis 1.23

Comparison 1 NPPV vs SMC, Outcome 23 Systolic BP after one hour ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 24 DIASTOLIC BP AFTER ONE HOUR.
Figures and Tables -
Analysis 1.24

Comparison 1 NPPV vs SMC, Outcome 24 DIASTOLIC BP AFTER ONE HOUR.

Comparison 1 NPPV vs SMC, Outcome 25 Diastolic BP after one hour ‐ by NPPV.
Figures and Tables -
Analysis 1.25

Comparison 1 NPPV vs SMC, Outcome 25 Diastolic BP after one hour ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 26 MEAN BP AFTER ONE HOUR.
Figures and Tables -
Analysis 1.26

Comparison 1 NPPV vs SMC, Outcome 26 MEAN BP AFTER ONE HOUR.

Comparison 1 NPPV vs SMC, Outcome 27 RESPIRATORY RATE AFTER ONE HOUR.
Figures and Tables -
Analysis 1.27

Comparison 1 NPPV vs SMC, Outcome 27 RESPIRATORY RATE AFTER ONE HOUR.

Comparison 1 NPPV vs SMC, Outcome 28 Respiratory rate after one hour ‐ by NPPV.
Figures and Tables -
Analysis 1.28

Comparison 1 NPPV vs SMC, Outcome 28 Respiratory rate after one hour ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 29 PaO2 (mmHg) AFTER ONE HOUR.
Figures and Tables -
Analysis 1.29

Comparison 1 NPPV vs SMC, Outcome 29 PaO2 (mmHg) AFTER ONE HOUR.

Comparison 1 NPPV vs SMC, Outcome 30 PaO2 (mmHg) after one hour ‐ by NPPV.
Figures and Tables -
Analysis 1.30

Comparison 1 NPPV vs SMC, Outcome 30 PaO2 (mmHg) after one hour ‐ by NPPV.

Comparison 1 NPPV vs SMC, Outcome 31 ADVERSE EVENTS.
Figures and Tables -
Analysis 1.31

Comparison 1 NPPV vs SMC, Outcome 31 ADVERSE EVENTS.

Summary of findings for the main comparison. NPPV compared to SMC for cardiogenic pulmonary oedema

NPPV compared to standard medical care for cardiogenic pulmonary oedema

Patient or population: People with cardiogenic pulmonary oedema
Setting: Pre‐hospital intensive care, emergency department, coronary care unit, or intensive care unit
Intervention: NPPV
Comparison: Standard medical care (SMC)

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with SMC

Risk with NPPV

HOSPITAL MORTALITY
follow‐up: median 13 days; range 1 day ‐ 41 days

Study population

RR 0.65
(0.51 to 0.82)

2484
(21 RCTs)

⊕⊕⊝⊝
LOWa,b

176 per 1000

114 per 1000
(90 to 144)

ETI RATE
follow‐up: median 1 day;

range 0.1 day ‐ 30 days

Study population

RR 0.49
(0.38 to 0.62)

2449
(20 RCTs)

⊕⊕⊕⊝
MODERATEc

154 per 1000

75 per 1000
(58 to 95)

ACUTE MI INCIDENCE
follow‐up: median 3 days; range 1 day ‐ 41 days

Study population

RR 1.03
(0.91 to 1.16)

1313
(5 RCTs)

⊕⊕⊕⊝
MODERATEd

421 per 1000

433 per 1000
(383 to 488)

HOSPITAL LENGTH OF STAY

The mean HOSPITAL LENGTH OF STAY was 9.65 days

MD 0.31 days lower
(1.23 lower to 0.61 higher)

1714
(11 RCTs)

⊕⊝⊝⊝
VERY LOWe,f,g

ICU LENGTH OF STAY

This outcome could not be pooled due to high heterogeneity. There was no evidence of a difference between NPPV and SMC in 4 RCTs, and 2 RCTs reported a shorter length of stay for NPPV (1 day shorter (95% CI −1.79 to −0.21); n = 30; 4 days shorter (95% CI −4.36 to −3.64); n = 120)

587
(6 RCTs)

⊕⊝⊝⊝
VERY LOWh,i,j

Data were not pooled due to high heterogeneity with an I2 of 99%

INTOLERANCE TO ALLOCATED TREATMENT ‐ not reported

Outcome was not reported

ADVERSE EVENTS

Reported adverse events included skin damage, pneumonia, gastrointestinal bleeding, gastric distention, vomiting, pneumothorax, sinusitis, mask discomfort, hypotension, arrhythmia, cardiorespiratory arrest, gastric aspiration, stroke, seizures, claustrophobia, and hypercapnia.

There was no evidence of a difference between groups for most of these events. However, there was an increase in discomfort with mask reported with NPPV (35/658) compared with SMC (0/442); RR 12.59 (95% CI 2.39 to 66.28).

One small study (n = 83) reported a higher incidence of gastric distention in the NPPV group (13/56 in NPPV group, 0/27 in SMC group; RR 13.26 (95% CI 0.82 to 215.12)), but the number of events was low and the confidence interval wide. There was also some weak evidence of a lower incidence of cardiorespiratory arrest in the NPPV group (24/836 (NPPV) vs 26/516 (SMC); RR 0.60 (95% CI 0.34 to 1.05)

2038
(11 RCTs)

⊕⊕⊝⊝
LOWk,l

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: Confidence interval; RR: Risk ratio

GRADE Working Group grades of evidence
High certainty: We are very confident that the true effect lies close to that of the estimate of the effect
Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different
Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect
Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect

aDowngraded by one level for risk of bias. Most trials providing hospital mortality evidence were at unclear or high risk of bias for key domains, including: randomisation sequence, allocation concealment, selective reporting bias, and other significant bias.
bDowngraded by one level for imprecision. Most trials had few participants and few mortality events, with a confidence interval crossing a RR of 0.75.
cDowngraded by one level for risk of bias. Most trials providing ETI evidence were rated as having unclear or high risk of bias for key domains, which included randomisation sequence generation, allocation concealment, selective reporting bias, and other significant bias.
dDowngraded by one level for imprecision. Most trials had few participants and few acute MI events, with a confidence interval crossing a RR of 1.0. In addition, the definition of acute MI varied between trials, with some trials using creatinine kinase and others using troponin.
eDowngraded by one level for risk of bias. Most trials were at low or unclear risk of bias. Potential limitations from lack of blinding are likely to lower confidence in hospital length of stay.
fDowngraded by one level for inconsistency. Hospital length of stay was heterogeneous, with an I2 of 58% and visually‐evident heterogeneity.
gDowngraded by one level for imprecision. Most trials had few participants, with a confidence interval crossing a MD of 0.
hDowngraded by one level for risk of bias. Most information was from studies at low or unclear risk of bias. Potential limitations from lack of blinding are likely to lower confidence in ICU length of stay.
iDowngraded by one level for imprecision. Most trials had few participants, with a confidence interval crossing a MD of 0.
jDowngraded by one level for inconsistency. ICU length of stay was heterogeneous, with an I2 of 99% and visually‐evident heterogeneity.
kDowngraded by one level for risk of bias. Most studies had at least one domain at high or unclear risk of bias.
lDowngraded by one level for imprecision, as confidence intervals included both directions of effect for many of the adverse events.

Figures and Tables -
Summary of findings for the main comparison. NPPV compared to SMC for cardiogenic pulmonary oedema
Table 1. NPPV intervention summary: EPAP, IPAP, and PEEP settings, duration of therapy.

Study

Mask type

IPAP level (cmH2O) ± SD

EPAP in bilevel NPPV (cmH2O) ± SD

PEEP in CPAP (cmH2O) ± SD

Time of bilevel NPPV (h) ± SD

Time of CPAP (h) ± SD

Agmy 2008

Face mask

NA

NA

NA

NA

NA

Austin 2013

Face mask

NA

NA

10

NA

0.583333

Crane 2004

Face mask

15

5

10

NA

NA

Ducros 2011

Face mask

NA

NA

10

NA

3

El‐Refay 2016

Face mask

15

10

10

1

Frontin 2011

Face mask

NA

NA

10

NA

NA

Gray 2008

Face mask

14±5

7 ± 3

10 ± 4

2 ± 1.3

2.2 ± 1.5

Hao 2002

Face mask

NA

NA

6 to 10

NA

4.6 ± 2.8

Kelly 2002

Face mask

NA

NA

7.5

NA

NA

L'Her 2004

Face mask

NA

NA

7.5

NA

NA

Levitt 2001

Face mask selected for mouth breathers Nasal available

NA

NA

NA

2

2

Li 2005

Face mask

15 to 18

5 to 8

NA

2

NA

Lin 1991

Face mask

NA

NA

12.5

NA

6

Lin 1995

Face mask

NA

NA

12.5

NA

6

Masip 2000

Face mask

15.2 ± 2.4

5

NA

4.2 ± 1.5

NA

Moritz 2003

Face mask

NA

NA

9.3 ± 0.3

0.5

NA

Nava 2003

Face mask

14.5 ± 21.1

6.1 ± 3.2

NA

11.4 ± 3.6

NA

Park 2001

Nasal bilevel and face mask for CPAP

12

4

7.5

2.6 ± 0.6

2.8 ± 1.5

Park 2004

Face mask

17 ± 2

11 ± 2

11 ± 2

2.1 ± 1

1.7 ± 0.7

Räsänen 1985

Face mask

NA

NA

10

NA

NA

Takeda 1997

Nasal

NA

NA

4 to 10

11.9 ± 8.4

NA

Takeda 1998

Nasal

NA

NA

4 to 10

NA

NA

Thys 2002

Face mask

16.5 ± 3.3

6.1 ± 1.5

NA

1.3 ± 0.3

NA

Zokaei 2016

Face mask 1st, nasal 2nd

10 to 20

4 to 7

NA

1

NA

Not all papers provided a mean and standard deviation. Where unavailable, the range represents the IPAP/EPAP/CPAP settings described in the Methods. NA: not applicable

Figures and Tables -
Table 1. NPPV intervention summary: EPAP, IPAP, and PEEP settings, duration of therapy.
Table 2. Standard medical therapy for each trial

Study

Lasix

Lasix dose

Nitrate

FiO2 mask and %FiO2

Other

Additional differences between SMC received in treatment group

Agmy 2008

Unclear

Unclear

Unclear

Unclear

Unclear

Unclear

Austin 2013

yes

40 mg

Sublingual 400 ‐ 1600 mcg q5min

Yes, 8 ‐ 15 Lpm

Morphine 1 ‐ 2 mg IV

None

Crane 2004

yes

Unclear

nitrates

Yes, 10 Lpm

diamorphine

None

Ducros 2011

yes

Lasix 40 mg ‐ 120 mg IV or 1 ‐ 3 mg bumetanide

nitroglycerin (1 mg per 3 min) and a continuous IV infusion unless systolic BP < 110 mmHg

15 Lpm

NA

None

El‐Refay 2016

yes

Lasix 40 mg IV

nitroglycerin 0.4 mg sublingual

15 Lpm

Morphine 2 mg IV

None

Frontin 2011

yes

1 mg/kg IV

isosorbide dinitrate IV 2 mg if SBP > 180 mmHg

15 Lpm

NA

None

Gray 2008

yes

50 mg above usual dose to max 100 mg IV

Buccal nitrates 2 ‐ 5 mg

15 Lpm

Opiates

None

Hao 2002

yes

20 ‐ 40 mg IV

Buccal nitrates 0.5 mg to 1 mg IV

6 Lpm

Morphine 2 mg IV

None

Kelly 2002

yes

50 ‐ 100 mg IV

buccal nitrate 5 mg if systolic BP > 90 mmHg

oxygen 60% by venturi mask

morphine IV 2.5 ‐ 10 mg

None

Li 2005

yes

Unclear

nitroglycerin

Mask high concentration

Unclear

None

Levitt 2001

yes

Unclear

nitroglycerin IV

Mask high flow

Morphine

None

L'Her 2004

yes

80 mg IV

nitroglycerin IV infusion 1 mg/hour

8 Lpm

Morphine 2 ‐ 10 mg IV

None

Lin 1991

Unclear

Unclear

Unclear

FiO2 100%

Unclear

Unclear

Lin 1995

yes

40 mg IV or double home dose

nitroglycerin sublingual 0.6 mg or isosorbide 10 ‐ 20 mg, or nitro infusion 10 ‐ 50 mcg/min

FiO2 100%

Morphine 2 ‐ 10 mg IV

None

Masip 2000

yes

40 mg IV

IV glyceryl trinitrate 1mg if systolic BP > 180 mmHg

FiO2 50%

Morphine 4 mg IV

None

Moritz 2003

yes

40 mg IV or double home dose

IV nitroglycerin infusion 0.125 ‐ 0.25 mcg/kg/min

FiO2 100%

Unclear

None

Nava 2003

yes

40 mg IV or double usual dose, repeated, if necessary, every 20 mins, up to 320 mg

continuous glyceryl trinitrate at an initial rate of 1.5 mg/hour. A bolus dose of 1 mg IV was added if systolic BP > 180 mmHg

10 Lpm

morphine sulphate up to 4 mg

None

Park 2001

Unclear

Unclear

5 mg isosorbide dinitrite if systolic BP > 100 mm Hg

15 Lpm

Unclear

Unclear

Park 2004

Unclear

Unclear

5 mg isosorbide dinitrate was given sublingually and if necessary titrated up to 15 mg

FiO2 50%

Unclear

Unclear

Räsänen 1985

yes

40 ‐ 80 mg IV

Nitroprusside, nitroglycerine

FiO2 30%

Morphine

Unclear

Takeda 1997

yes

Unclear

Nitroglycerin infusion

FiO2 50%

Morphine

Unclear

Takeda 1998

yes

Unclear

Nitroglycerin infusion

FiO2 70%

Morphine

Unclear

Thys 2002

yes

40 mg IV

isosorbide dinitrate 2 mg/hour

Unclear

Unclear

Unclear

Zokaei 2016

yes

40 ‐ 320 mg

IV nitroglycerin 5 ‐ 50 mcg/kg/min

10 Lpm

morphine 5 ‐ 10 mg

Unclear

BP: blood pressure; FiO2: fraction of inspired oxygen; IV: intravenous; Lpm: litres per minute; NA: not available; mcg: micrograms; mcg/kg/min: micrograms per kilogram per minute; mg: milligrams; min: minutes; mmHg = millimetres of mercury

Figures and Tables -
Table 2. Standard medical therapy for each trial
Table 3. In‐hospital mortality duration of follow‐up

Study

Follow‐up (days)

Agmy 2008

Unclear

Austin 2013

7.2a

Crane 2004

41b

Ducros 2011

> 7

El‐Refay 2016

1

Frontin 2011

30

Gray 2008

30

Hao 2002

NA

Kelly 2002

15a

Li 2005

NA

Levitt 2001

38b

L'Her 2004

12a

Lin 1991

1

Lin 1995

9

Masip 2000

14a

Moritz 2003

NA

Nava 2003

5.4a

Park 2001

3b

Park 2004

15

Räsänen 1985

Unclear

Takeda 1997

7.7a

Takeda 1998

Unclear

Thys 2002

17.6a

Zokaei 2016

7

aEstimated based on longest reported mean hospital length of stay; bBased on latest death reported in trial

Figures and Tables -
Table 3. In‐hospital mortality duration of follow‐up
Table 4. Intubation criteria for trials reporting endotracheal intubation rates

Study

Intubation Criteria

Agmy 2008

Unclear

Austin 2013

Unclear

Crane 2004

RR > 40, RR < 10, altered level of consciousness, arterial pH < 7.2

Ducros 2011

SpO2 < 85% after 30 mins of 15 Lpm or FiO2 60% CPAP, respiratory arrest, psychomotor agitation, SBP < 80 mmHg

El‐Refay 2016

Unclear

Frontin 2011

Worsening SpO2 despite effective treatment, loss of airway protective reflexes (cough, swallow), decreased level of consciousness, haemodynamic instability, intolerance/poor fit of face mask for the CPAP group, medical clinical impression of deterioration, or participant request

Gray 2008

At discretion of treating clinician

L'Her 2004

Cardiac or respiratory arrest, severe haemodynamic instability (SBP < 80), administration of epinephrine/norepinephrine, refractory hypoxaemia SaO2 < 92% despite face mask, clinical signs of respiratory exhaustion (accessory muscle use with paradoxical abdominal motion), coma or seizures, agitation requiring sedation

Levitt 2001

Clinical or arterial blood gas findings. Severe respiratory distress, deterioration in mental status, or further deterioration in vital signs with increasing HR, RR, or a decrease in BP. Decrease in pO2 < 60 or rise in pCO2 > 50 along with signs of clinical deterioration warranted intubation

Lin 1991

Clinician discretion. Loss of consciousness or airway obstruction were indications for intubation

Lin 1995

Cardiopulmonary resuscitation or clinical deterioration with any 2 of the following: (1) a rise in arterial carbon dioxide tension to more than 55 mm Hg; (2) arterial blood oxygen partial pressure divided by fraction of inspired oxygen content less than 200 mm Hg; and (3) respiratory rate over 35 breaths/min.

Masip 2000

Patients were intubated when the criteria for treatment failure were met. Details not specified.

Nava 2003

Unclear

Park 2001

Clinical judgement

Park 2004

Clinical judgement

Räsänen 1985

Clinical judgement. Loss of consciousness or airway obstruction were indications for intubation at any time.

Takeda 1997

Clinical deterioration and either a decrease in the PaO2/FiO2 < 100 with an FIO2 ≥ 70% or a increase in the PaCO2 > 55 mmHg

Takeda 1998

Clinical deterioration and either a fall in PaO2/FiO2 < 100 with an FiO2 of 70% or a rise in PaCO2 to > 55 mmHg.

Thys 2002

Deterioration in clinical status including all of the following: dyspnoea, respiratory or heart frequency or both, sweating and agitation or deterioration in blood gases or in haemodynamic status or both

Zokaei 2016

RR after 1 hour of bilevel NPPV > 30, persistent hypoxaemia, haemodynamic instability (systolic BP < 70), agitation, or worsened neurologic status, inability to tolerate mask or aspiration of gastric content

BP: blood pressure; CPAP: continuous positive airway pressure; FiO2: fraction of inspired oxygen; HR: heart rate; Lpm: litres per minute; PaCO2: arterial partial pressure of carbon dioxide; PaO2: arterial partial pressure of oxygen; RR: respiratory rate; SBP: systolic blood pressure

Figures and Tables -
Table 4. Intubation criteria for trials reporting endotracheal intubation rates
Comparison 1. NPPV vs SMC

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1 HOSPITAL MORTALITY Show forest plot

21

2484

Risk Ratio (M‐H, Random, 95% CI)

0.65 [0.51, 0.82]

2 Hospital mortality ‐ by NPPV Show forest plot

21

2484

Risk Ratio (M‐H, Random, 95% CI)

0.71 [0.58, 0.87]

2.1 CPAP vs SMC

16

1454

Risk Ratio (M‐H, Random, 95% CI)

0.65 [0.48, 0.88]

2.2 bilevel NPPV vs SMC

11

1030

Risk Ratio (M‐H, Random, 95% CI)

0.72 [0.53, 0.98]

3 Hospital mortality ‐ by location Show forest plot

21

2484

Risk Ratio (M‐H, Random, 95% CI)

0.65 [0.51, 0.82]

3.1 ER

10

1596

Risk Ratio (M‐H, Random, 95% CI)

0.66 [0.47, 0.93]

3.2 ICU

10

862

Risk Ratio (M‐H, Random, 95% CI)

0.52 [0.35, 0.77]

3.3 unclear

1

26

Risk Ratio (M‐H, Random, 95% CI)

1.94 [0.09, 43.50]

4 Hospital mortality ‐ by baseline PaCO2 Show forest plot

21

2484

Risk Ratio (M‐H, Random, 95% CI)

0.65 [0.51, 0.82]

4.1 PaCO2 <= 45mmHg

10

581

Risk Ratio (M‐H, Random, 95% CI)

0.41 [0.27, 0.63]

4.2 PaCO2 > 45mmHg

11

1903

Risk Ratio (M‐H, Random, 95% CI)

0.82 [0.65, 1.03]

5 Hospital mortality ‐ sensitivity analysis (low risk of bias) Show forest plot

7

1505

Risk Ratio (M‐H, Random, 95% CI)

0.81 [0.61, 1.06]

6 Hospital mortality ‐ sensitivity analysis (missing data) Show forest plot

18

2283

Risk Ratio (M‐H, Random, 95% CI)

0.68 [0.54, 0.86]

7 Hospital mortality ‐ sensitivity analysis (fixed‐effect) Show forest plot

21

2484

Risk Ratio (M‐H, Fixed, 95% CI)

0.67 [0.55, 0.81]

8 ETI RATE Show forest plot

20

2449

Risk Ratio (M‐H, Random, 95% CI)

0.49 [0.38, 0.62]

9 ETI rate ‐ by NPPV Show forest plot

20

2449

Risk Ratio (M‐H, Random, 95% CI)

0.49 [0.39, 0.63]

9.1 CPAP vs SMC

15

1413

Risk Ratio (M‐H, Random, 95% CI)

0.46 [0.34, 0.62]

9.2 Bilevel NPPV vs SMC

11

1036

Risk Ratio (M‐H, Random, 95% CI)

0.50 [0.31, 0.81]

10 ETI rate ‐ by location Show forest plot

20

2449

Risk Ratio (M‐H, Random, 95% CI)

0.49 [0.38, 0.62]

10.1 ER

9

1561

Risk Ratio (M‐H, Random, 95% CI)

0.60 [0.37, 0.96]

10.2 ICU

10

862

Risk Ratio (M‐H, Random, 95% CI)

0.40 [0.29, 0.56]

10.3 unclear

1

26

Risk Ratio (M‐H, Random, 95% CI)

0.47 [0.13, 1.67]

11 ETI rate ‐ by baseline PaCO2 Show forest plot

20

2449

Risk Ratio (M‐H, Random, 95% CI)

0.49 [0.38, 0.62]

11.1 PaCO2 <= 45mmHg

9

523

Risk Ratio (M‐H, Random, 95% CI)

0.37 [0.26, 0.52]

11.2 PaCO2 > 45mmHg

11

1926

Risk Ratio (M‐H, Random, 95% CI)

0.64 [0.46, 0.91]

12 ETI rate ‐ sensitivity analysis (low risk of bias) Show forest plot

6

1491

Risk Ratio (M‐H, Random, 95% CI)

0.85 [0.55, 1.32]

13 ETI rate ‐ sensitivity analysis (missing data) Show forest plot

17

2248

Risk Ratio (M‐H, Random, 95% CI)

0.52 [0.40, 0.69]

14 ETI rate ‐ by face mask type Show forest plot

20

2449

Risk Ratio (M‐H, Random, 95% CI)

0.49 [0.38, 0.62]

14.1 Exclusive full face mask use

15

2213

Risk Ratio (M‐H, Random, 95% CI)

0.52 [0.40, 0.69]

14.2 Exclusive or permissive nasal mask use

5

236

Risk Ratio (M‐H, Random, 95% CI)

0.35 [0.20, 0.62]

15 ACUTE MI INCIDENCE Show forest plot

5

1313

Risk Ratio (M‐H, Random, 95% CI)

1.03 [0.91, 1.16]

16 Acute MI incidence ‐ by NPPV Show forest plot

5

1313

Risk Ratio (M‐H, Random, 95% CI)

1.03 [0.91, 1.16]

16.1 CPAP vs SMC

3

569

Risk Ratio (M‐H, Random, 95% CI)

0.96 [0.80, 1.14]

16.2 bilevel NPPV vs SMC

5

744

Risk Ratio (M‐H, Random, 95% CI)

1.09 [0.92, 1.29]

17 HOSPITAL LENGTH OF STAY Show forest plot

11

1714

Mean Difference (IV, Random, 95% CI)

‐0.31 [‐1.23, 0.61]

18 Hospital length of stay ‐ by NPPV Show forest plot

11

1714

Mean Difference (IV, Random, 95% CI)

‐0.19 [‐1.02, 0.64]

18.1 CPAP vs SMC

7

943

Mean Difference (IV, Random, 95% CI)

‐0.52 [‐1.77, 0.72]

18.2 bilevel NPPV vs SMC

6

771

Mean Difference (IV, Random, 95% CI)

0.39 [‐0.35, 1.13]

19 Hospital length of stay ‐ by location Show forest plot

11

1714

Mean Difference (IV, Random, 95% CI)

‐0.30 [‐1.21, 0.61]

19.1 ER

8

1455

Mean Difference (IV, Random, 95% CI)

‐0.38 [‐1.70, 0.93]

19.2 ICU

3

259

Mean Difference (IV, Random, 95% CI)

‐0.21 [‐1.30, 0.89]

20 Hospital length of stay ‐ by baseline PaCO2 Show forest plot

11

1714

Mean Difference (IV, Random, 95% CI)

‐0.30 [‐1.21, 0.61]

20.1 PaCO2 <= 45mmHg

5

397

Mean Difference (IV, Random, 95% CI)

‐1.18 [‐2.33, ‐0.04]

20.2 PaCO2 > 45mmHg

6

1317

Mean Difference (IV, Random, 95% CI)

0.60 [‐0.15, 1.34]

21 ICU LENGTH OF STAY Show forest plot

6

Mean Difference (IV, Random, 95% CI)

Totals not selected

22 SYSTOLIC BP AFTER ONE HOUR Show forest plot

7

1408

Mean Difference (IV, Random, 95% CI)

‐1.72 [‐5.03, 1.58]

23 Systolic BP after one hour ‐ by NPPV Show forest plot

7

1408

Mean Difference (IV, Random, 95% CI)

‐1.72 [‐5.03, 1.59]

23.1 CPAP vs SMC

7

866

Mean Difference (IV, Random, 95% CI)

‐1.65 [‐5.58, 2.28]

23.2 bilevel NPPV vs SMC

3

542

Mean Difference (IV, Random, 95% CI)

‐1.89 [‐8.01, 4.23]

24 DIASTOLIC BP AFTER ONE HOUR Show forest plot

6

1361

Mean Difference (IV, Random, 95% CI)

1.46 [‐1.86, 4.78]

25 Diastolic BP after one hour ‐ by NPPV Show forest plot

6

1361

Mean Difference (IV, Random, 95% CI)

1.05 [‐2.15, 4.25]

25.1 CPAP vs SMC

6

823

Mean Difference (IV, Random, 95% CI)

0.92 [‐3.92, 5.75]

25.2 bilevel NPPV vs SMC

3

538

Mean Difference (IV, Random, 95% CI)

1.08 [‐2.88, 5.04]

26 MEAN BP AFTER ONE HOUR Show forest plot

3

251

Mean Difference (IV, Random, 95% CI)

‐2.50 [‐8.29, 3.30]

27 RESPIRATORY RATE AFTER ONE HOUR Show forest plot

10

1636

Mean Difference (IV, Random, 95% CI)

‐1.87 [‐2.70, ‐1.03]

28 Respiratory rate after one hour ‐ by NPPV Show forest plot

10

1636

Mean Difference (IV, Random, 95% CI)

‐1.58 [‐2.22, ‐0.94]

28.1 CPAP vs SMC

10

1107

Mean Difference (IV, Random, 95% CI)

‐1.64 [‐2.41, ‐0.87]

28.2 bilevel NPPV vs SMC

3

529

Mean Difference (IV, Random, 95% CI)

‐2.17 [‐4.69, 0.35]

29 PaO2 (mmHg) AFTER ONE HOUR Show forest plot

10

1428

Mean Difference (IV, Random, 95% CI)

16.19 [3.54, 28.84]

30 PaO2 (mmHg) after one hour ‐ by NPPV Show forest plot

10

1428

Mean Difference (IV, Random, 95% CI)

13.74 [2.72, 24.76]

30.1 CPAP vs SMC

8

761

Mean Difference (IV, Random, 95% CI)

9.55 [‐9.10, 28.19]

30.2 bilevel NPPV vs SMC

6

667

Mean Difference (IV, Random, 95% CI)

19.50 [4.29, 34.71]

31 ADVERSE EVENTS Show forest plot

11

9570

Risk Ratio (M‐H, Random, 95% CI)

1.04 [0.73, 1.50]

31.1 Skin damage

2

190

Risk Ratio (M‐H, Random, 95% CI)

9.09 [0.74, 111.09]

31.2 Pneumonia

1

130

Risk Ratio (M‐H, Random, 95% CI)

0.5 [0.05, 5.38]

31.3 Gastrointestinal bleeding

2

170

Risk Ratio (M‐H, Random, 95% CI)

2.32 [0.35, 15.42]

31.4 Gastric distention

1

83

Risk Ratio (M‐H, Random, 95% CI)

13.26 [0.82, 215.12]

31.5 Vomiting

5

1467

Risk Ratio (M‐H, Random, 95% CI)

1.03 [0.49, 2.17]

31.6 Pneumothorax

2

1165

Risk Ratio (M‐H, Random, 95% CI)

0.72 [0.08, 6.89]

31.7 Sinusitis

1

130

Risk Ratio (M‐H, Random, 95% CI)

3.0 [0.12, 72.31]

31.8 Discomfort with mask

3

1100

Risk Ratio (M‐H, Random, 95% CI)

12.59 [2.39, 66.28]

31.9 Hypotension

1

1030

Risk Ratio (M‐H, Random, 95% CI)

0.82 [0.58, 1.16]

31.10 Arrhythmia

1

1027

Risk Ratio (M‐H, Random, 95% CI)

0.83 [0.50, 1.38]

31.11 Cardiorespiratory arrest

4

1352

Risk Ratio (M‐H, Random, 95% CI)

0.60 [0.34, 1.05]

31.12 Gastric aspiration

1

1037

Risk Ratio (M‐H, Random, 95% CI)

1.58 [0.06, 38.61]

31.13 Stroke

1

130

Risk Ratio (M‐H, Random, 95% CI)

0.2 [0.01, 4.09]

31.14 Seizures

1

130

Risk Ratio (M‐H, Random, 95% CI)

0.33 [0.01, 8.03]

31.15 Claustrophobia

1

130

Risk Ratio (M‐H, Random, 95% CI)

3.0 [0.12, 72.31]

31.16 Hypercapnia

3

299

Risk Ratio (M‐H, Random, 95% CI)

0.76 [0.16, 3.62]

Figures and Tables -
Comparison 1. NPPV vs SMC