Oxygen therapy for pneumonia in adults

Abstract

Background

Oxygen therapy is widely used in the treatment of lung diseases. However, the effectiveness of oxygen therapy as a treatment for pneumonia is not well known.

Objectives

To determine the effectiveness and safety of oxygen therapy in the treatment of pneumonia in adults older than 18 years.

Search methods

We searched the Cochrane Central Register of Controlled Trials (CENTRAL) 2011, Issue 4, part of The Cochrane Library, www.thecochranelibrary.com (accessed 9 December 2011), which includes the Cochrane Acute Respiratory Infections Group's Specialised Register, MEDLINE (1948 to November week 3, 2011) and EMBASE (1974 to December 2011).

Selection criteria

Randomised controlled trials (RCTs) of oxygen therapy for adults with community‐acquired pneumonia (CAP) and nosocomial (hospital‐acquired) pneumonia (HAP or NP) in intensive care units (ICU).

Data collection and analysis

Two review authors independently reviewed abstracts and assessed data for methodological quality.

Main results

Three RCTs met our inclusion criteria. The studies enrolled 151 participants with CAP or immunosuppressed patients with pulmonary infiltrates. Overall, we found that non‐invasive ventilation can reduce the risk of death in the ICU, odd ratio (OR) 0.28, 95% confidence interval (CI) 0.09 to 0.88; endotracheal intubation, OR 0.26, 95% CI 0.11 to 0.61; complications, OR 0.23, 95% CI 0.08 to 0.70; and shorten ICU length of stay, mean duration (MD) ‐3.28, 95% CI ‐5.41 to ‐1.61.

Non‐invasive ventilation and standard oxygen supplementation via a Venturi mask were similar when measuring mortality in hospital, OR 0.54, 95% CI 0.11 to 2.68; two‐month survival, OR 1.67, 95% CI 0.53 to 5.28; duration of hospital stay, MD ‐1.00, 95% CI ‐2.05 to 0.05; and duration of mechanical ventilation, standard MD ‐0.26, 95% CI ‐0.66 to 0.14. Some outcomes and complications of non‐invasive ventilation were varied according to different participant populations. We also found that some subgroups had a high level of heterogeneity when conducting pooled analyses.

Authors' conclusions

Non‐invasive ventilation can reduce the risk of death in the ICU, endotracheal intubation, shorten ICU stay and length of intubation. Some outcomes and complications of non‐invasive ventilation were varied according to different participant populations. Other than the oxygen therapy, we must mention the importance of standard treatment by physicians. The evidence is weak and we did not include participants with pulmonary tuberculosis and cystic fibrosis. More RCTs are required to answer these clinical questions. However, the review indicates that non‐invasive ventilation may be more beneficial than standard oxygen supplementation via a Venturi mask for pneumonia.

Plain language summary

The effectiveness of oxygen for adult patients with pneumonia

Pneumonia is an inflammatory condition of the lungs. Treatment for pneumonia includes antibiotics, rest, fluids, management of complications and professional home care. Oxygen supplementation is one way to help patients who cannot breathe adequately on their own. Management of oxygen supplementation is divided into nasal cannula and mechanical ventilation. Mechanical ventilation is life‐supporting ventilation that involves the use of a machine called a ventilator, or respirator. There are two main types of mechanical ventilation: non‐invasive ventilation (NIV) and invasive ventilation. The former provides ventilatory support to a patient through a tightly fitted facial or nasal mask and the latter through a tube inserted into the windpipe through the mouth or the nose or a hole made in the windpipe through the front of the throat. At present, oxygen therapy for individuals with pneumonia is commonly prescribed. However, inconsistent results on the effects of oxygen therapy on pneumonia have been reported and no systematic review has been conducted in patients with pneumonia to determine which delivery system of oxygen therapy leads to the best clinical outcomes.

We searched the related literature and included three randomised controlled trials involving 151 adults with pneumonia aged around 60 years. We did not include patients with pulmonary tuberculosis or cystic fibrosis. We found that NIV can reduce the risk of death in the intensive care unit (ICU) and the need for endotracheal intubation, shorten ICU stay and length of intubation. Some outcomes and complications of oxygen therapy depended upon the delivery system and primary diseases. The most common complications of invasive ventilation are ventilator‐associated pneumonia. However, we must be aware that oxygen therapy is just one of the treatments for pneumonia and the other standard treatments used by physicians are of equal importance.

The evidence is weak and it is limited by the small number of studies and the small number of study participants.

Background

Description of the condition

Pneumonia is an inflammatory condition of the lungs. It is often characterised as inflammation and fluid collection in the alveoli. Common clinical symptoms of pneumonia include cough, sputum production, fever, chills, fatigue, shortness of breath, night sweats and pleuritic chest pain. Although pneumonia is most common in young children and the elderly, it can affect people of any age. Pneumonia can be caused by bacteria, viruses, fungi or parasites and almost 100 species of pathogens have been identified as causative agents (Donowitz 2000; File 2003). Pneumonia can be classified in several ways including by anatomic changes, microbial cause and radiological classification. However, the disease is commonly classified as community‐acquired pneumonia (CAP) and nosocomial (hospital‐acquired) pneumonia (HAP or NP) according to a combined clinical classification. The advantage of this classification is that it can help guide the selection of appropriate initial treatment before the microbiologic cause of the pneumonia is known. However, the clinical presentations are often similar in CAP and HAP. In adults, the incidence of CAP increases with age. For people aged 65 to 69 years, the overall rate of CAP ranges from 18.2 cases per 1000 person‐years to 52.3 cases per 1000 person‐years in those aged over 85 years. (Jackson 2004). Hospital admission is needed in 20% to 40% of patients with CAP (BTS 2001; Macfarlane 2004).

The prevalence of HAP increased threefold from those aged < 35 to those aged > 85 years (Humphreys 2010). In low‐income countries, pneumonia causes approximately two million deaths annually among children under five years of age (Graham 1990; Jones 2003; Rudan 2004). In the USA, pneumonia is the sixth most common cause of death and the most common cause of infection‐related mortality (Donowitz 2000). Both CAP and HAP are associated with a significant mortality (Fagon 1993) and high cost. Although 80% of patients with CAP are treated as outpatients, mortality for these patients is usually less than 1%. The other 20% of patients require inpatient management. Approximately 10% to 20% of hospitalised patients with CAP have to be admitted to the intensive care unit (ICU), where 20% to 50% of them will ultimately die (Fine 1996; Kaplan 2002). The overall mean cost per case of CAP management in the United States was calculated as USD 493 for successful treatment and USD 3019 for successful treatment failure (Monte 2008; Sergio 2009). HAP is the leading cause of mortality in hospital‐acquired infections (Flanders 2006). It is associated with crude mortality rates of up to 70% and attributable mortality rates as high as 33% to 50% (Mandell 1998). An episode of HAP adds an additional five to six days to the length of hospital stay and thousands of dollars in cost to medical care (Raghavendran 2000).

Description of the intervention

The treatment for pneumonia includes antibiotics, rest, fluids, management of complications and proper home care. Oxygen supplementation is one of the most important ways to help to improve oxygen saturation in arterial blood gases. Oxygen supplementation may be described both in terms of volume (dosage) and delivery system. Low‐volume systems include nasal cannulae (NC), shields and masks. Drawbacks to low‐volume systems include drying of the nasal mucosa and eyes, discomfort, as masks can pinch and cause skin irritation, and the fact that their passive delivery system is dependent on the patient's ability to breathe. If the lungs cannot keep the body oxygenated using these methods, mechanical ventilation may be required. The most common methods of respiratory support are non‐invasive positive pressure ventilation (NPPV or NIPPV), continuous positive airway pressure (CPAP) and invasive positive pressure ventilation (IPPV). The state of respiratory function can be improved more so with mechanical ventilation than with low‐volume systems but the disadvantages of the former are twofold: chest muscles can weaken through disuse and bacteria may access the lungs more easily. Cochrane Reviews have found that oxygen may relieve dyspnoea in mildly and non‐hypoxaemic people with chronic obstructive pulmonary disease (COPD) and relieve breathlessness in end‐stage illness in adults; for patients with interstitial lung disease (ILD) it may have a beneficial survival effect (Cranston 2008; Crockett 2001; Uronis 2011). The evidence was limited by the methods used in the studies, the small number of studies or the small number of study participants. In 1885, George E Holtzapple discovered that oxygen could be used to treat pneumonia. In 1919, following his discovery, Stadie reported that the mortality in patients with pneumonia was 39% with oxygen and 74% without oxygen but data were adjusted for the severity of the illness.

How the intervention might work

Clinical treatment for pneumonia includes oxygen saturation monitoring and oxygen delivery, both of which can significantly reduce length of hospital stay (Loeb 2006). There is consensus among scientists and clinicians on the life‐saving benefits of oxygen therapy for patients with severe pneumonia and signs of decreased oxygen saturation. Mortality from pneumonia has also been shown to be related to arterial blood oxygen saturation (Onyango 1993). Even without pre‐existing co‐morbid medical conditions, patients who have an oxygen saturation < 94% should be considered for oxygen supplementation. Moderate levels of supplemental oxygen (nasal cannulae at 2 to 6 l/min or simple mask at 5 to 10 l/min unless otherwise stated) are required if the patient with pneumonia is hypoxaemic and the initial oxygen saturation target range is 94% to 98%. The recommended grade of therapy is 'C', based on the guidelines of the British Thoracic Society (Levy 2010a). Oxygen therapy is often only necessary in severe cases of pneumonia (severe is defined as a the patient being confused, urea > 7 mmol/l, respiratory rate > = 30/min, low systolic (< 90 mmHg) or diastolic (<= 60 mmHg) blood pressure, or age >= 65 years (or 'CURB‐65 score') of three or more) (Ewig 2006; Lim 2003).

Why it is important to do this review

To date, inconsistent results on the effects of oxygen therapy on pneumonia have been reported. Several studies have reported that supplementing oxygen in NIPPV patients with severe CAP can lead to initial improvement (Antonelli 2001; Domenighetti 2002; Jolliet 2001). Using either CPAP or bi‐level pressure support can improve patients' oxygenation and tachypnoea in patients with severe acute respiratory syndrome (Cheung 2004; Chen 2003; Li 2003). A study from Hong Kong has verified the benefits of either intervention (Yam 2005). In contrast, a study indicated that patients with CAP who have severe hypoxaemia (partial pressure of oxygen in arterial blood (PaO₂), the fraction of inspired oxygen (FiO₂), PaO₂/FiO₂ ratio < 150), are poor candidates for NIPPV (Antonelli 2001; Niederman 2001a) and NIPPV should only be administered in an appropriate critical care setting for CAP patients because the health of over half of these patients will deteriorate and required intubation (BTS 2002). Despite early physiologic improvements, CPAP neither reduced the need for intubation nor improved outcomes in patients with acute hypoxaemia and non‐hypercapnia respiratory insufficiency primarily due to acute lung injury, including changes from pneumonia (Delclaux 2000).

At present, oxygen therapy for individuals with pneumonia is commonly prescribed. However, no systematic review has previously been conducted on patients with non‐hypoxaemic pneumonia to determine which method of oxygen therapy leads to the best clinical outcome.

Objectives

The primary objective is to identify the effectiveness and safety of oxygen therapy in the treatment of adults with pneumonia.

The secondary objective is to compare the therapeutic differences of different methods of oxygen supplementation, including oxygen inhalation via mask or nasal catheter, non‐invasive mechanical ventilation and invasive mechanical ventilation, in different types of pneumonia with or without complications.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs). This was defined as studies that were described by the authors as randomised anywhere in the manuscript. All identified trials, published or unpublished, were eligible.

Types of participants

Adult patients (18 years or older) with CAP or HAP, regardless of severity and complications and mechanically ventilated, with or without lung cancer or immune deficiencies (drug or systemic disease‐induced immunosuppression).

We excluded patients with pulmonary tuberculosis and cystic fibrosis because of their varying prognoses. We considered trials for inclusion if the indication for treatment consisted of multiple diagnoses, most commonly exacerbation of COPD, acute bronchitis, exacerbation of chronic bronchitis and pneumonia, or complications such as active respiratory failure and active lung injury and the results were reported separately for each diagnostic group.

We also excluded studies if they did not have a pneumonia subgroup and pneumonia was not diagnosed by a standard criteria.

Types of interventions

Oxygen therapy, either administered alone or as an adjunctive therapy (co‐interventions include antibiotics, mucolytic agents and fluid supplements), compared with a placebo. In addition, multiple forms of oxygen therapy were compared with each other (for example, at different dosages or delivery methods).

Types of outcome measures

Primary outcomes

1. Mortality (death in hospital, the ICU, by day 90, at two months, etc.) or survival rate.

2. Need for invasive mechanical ventilation.

3. Clinical response based on improvement or deterioration in terms of signs and symptoms.

Secondary outcomes

1. Length of hospital stay.

2. Ventilator‐free days (unassisted breathing continued >= 48 hours).

3. Complications not present on admission (such as ventilator‐associated pneumonia (VAP)).

4. Duration of ICU stay.

5. Duration of ventilatory assistance.

6. Any adverse events, including drying of the nasal mucosa and eyes, discomfort (as masks can pinch and cause skin irritation), chest muscle weakness and bacterial lung infections.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Central Register of Controlled Trials (CENTRAL) 2011, Issue 4, part of The Cochrane Library, www.thecochranelibrary.com (accessed 9 December 2011), which includes the Cochrane Acute Respiratory Infections Group's Specialised Register, MEDLINE (1948 to November week 3, 2011) and EMBASE (1974 to December 2011).

We searched CENTRAL and MEDLINE using the following search strategy. We combined the MEDLINE search strategy with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐ and precision‐maximising version (2008 revision); Ovid format (Lefebvre 2011). We adapted the search terms to search EMBASE (see Appendix 1).

MEDLINE (Ovid)

1 exp Pneumonia/ 
 2 pneumon*.tw. 
 3 cap.tw. 
 4 hap.tw. 
 5 or/1‐4 
 6 exp Oxygen Inhalation Therapy/ 
 7 ((oxygen* or o2*) adj5 (therap* or inhal* or respirat* or deliver*)).tw. 
 8 ((oxygen* or o2*) adj5 (mask* or hudson* or ventur* or nrm*)).tw. 
 9 (nasal adj2 (cannula* or prong*)).tw. 
 10 exp Positive‐Pressure Respiration/ 
 11 (positive pressure ventilation or positive pressure respiration).tw. 
 12 continuous positive airway* pressure.tw. 
 13 Respiration, Artificial/ 
 14 mechanical ventilation.tw. 
 15 (cpap or ncpap or nppv).tw. 
 16 or/6‐15 
 17 5 and 16

Searching other resources

We reviewed the references of all included trials to identify other relevant trials. In addition, we identified all other reviews and meta‐analyses on the topic. We searched appropriate journals. We contacted ventilator‐manufacturing companies and authors of relevant unpublished articles to identify any additional unpublished studies. We did not apply any language or publication restrictions. We searched for ongoing trials in the following registers: ISRCTN (http://www.controlled‐trials.com/isrctn/) and WHO ICTRP (http://www.who.int/ictrp/). The latest search was conducted on 13 December 2011 using the search strategy described in Appendix 2 .

Data collection and analysis

Selection of studies

Two review authors (YZ, CF) independently selected relevant articles and assessed their eligibility according to the inclusion and exclusion criteria. We resolved disagreements by discussion. The other two review authors (BD, TW) resolved disagreements when necessary. We obtained the full text for those articles identified as either relevant or ambiguous from their titles and abstracts. We searched the full text and analysed data to verify the content of articles that were not mentioned in the Methods section but were described in the results of the included trials.

Data extraction and management

Two review authors (YZ, CF) independently extracted and cross‐checked data for each trial. We resolved discrepancies by discussion. We contacted the trial authors for clarification when necessary.

Assessment of risk of bias in included studies

We assessed the methodological quality of each trial using the Cochrane Collaboration’s ‘Risk of bias’ tool (Higgins 2011). We assessed each trial in terms of random sequence generation, allocation concealment, blinding of participants, personnel and outcome assessment, selective reporting and loss to follow‐up. We then classified each trial as 'low risk of bias', 'high risk of bias' or 'unclear risk of bias' according to the Cochrane Handbook of Systematic Reviews of Interventions(Higgins 2011) and described by Wu 2007. We resolved differences by discussion among the review authors. We assessed the following characteristics:

• Random sequence generation

Low risk of bias: adequate generation of allocation sequence (for example, table of random numbers, computer‐generated random numbers, or similar). 
 High risk of bias: inadequate generation of allocation sequence (for example, date of birth, case record number, day, month or year of admission, or allocation by judgement of the clinician, the participant, laboratory test or a series of tests, availability of the intervention). 
 Unclear risk of bias: the generation was unclear.

• Allocation concealment

Low risk of bias: adequate concealment of allocation (for example, non‐translucent sealed envelopes, central independent unit, or similar). 
 High risk of bias: inadequate concealment of allocation (any procedure which is transparent before allocation (for example, dates of birth, alternation, open table of random numbers, the use of case record numbers, or similar). 
 Unclear risk of bias: unclear concealment of allocation (for example, only specifying that sealed envelopes were used or not reporting any concealment approach) or inadequate.

• Blinding: blinding of participants and personnel and blinding of outcome assessment

Low risk of bias: masking of both the participants and results assessor (for example, identical placebo tablets or similar). Blinding was not considered necessary for mortality or other outcomes (that were not influenced by blinding). 
 High risk of bias: not used or non‐blinding for detection of outcomes (for example, not performed or tablets versus fluids or similar). 
 Unclear risk of bias: insufficient information provided to judge 'low risk of bias' or 'high risk of bias', single‐blinding of the results assessor and blinding was performed on the participants but not the results assessor.

• Incomplete outcome data: assessment for potential bias of exclusions and attrition

Low risk of bias: trials had no missing outcome data or a few exclusions (lower than 5%), attrition is noted; or if exclusions were 5% to 10% and an intention‐to‐treat (ITT) analysis was used. 
 High risk of bias: exclusions were wide differences in exclusions between intervention group and control group or the rate of exclusion and/or attrition is higher than 15%, whatever ITT analysis was used. 
 Unclear risk of bias: the rate of attrition and/or exclusions is higher than 10%.

Low risk of bias ‐ all quality criteria met. 
 Unclear risk of bias ‐ one or more of the quality criteria partly met. 
 High risk of bias ‐ one or more criteria not met.

Measures of treatment effect

We analysed the data using Review Manager 5.1 (RevMan 2011) software. We used the Mantel‐Haenszel odds ratio with a 95% confidence interval (CI) for dichotomous outcomes and mean differences (MD) or standardised mean difference (SMD) for continuous outcomes. A level of P < 0.05 was considered statistically significant.

Unit of analysis issues

We did not include cross‐over trials in this review. We carefully assessed the trial to avoid double‐counting of participants where multiple interventions were used in the same trial.

Dealing with missing data

If we do find studies with missing data, we will analyse the outcome measures based on an ITT population (i.e. we will consider participants who drop out of the study along with those who continued in the study).

Assessment of heterogeneity

We used the Chi² test with significance being set at P value < 0.1 to test for heterogeneity. We also used the I² statistic to estimate the total variation across studies: I² statistic < 25%: low level of heterogeneity; 25% to 50%: moderate level; > 50%: high level of heterogeneity (Higgins 2003).

Assessment of reporting biases

We did not assess the presence of publication bias in a funnel plot because it is difficult to detect with small numbers of studies (i.e. fewer than 10) in a systematic review. If we identify more studies in the future, we will use a funnel plot for all included trials to assess the presence of publication bias.

Data synthesis

We used a random‐effects model to synthesise all data due to potential heterogeneity between the included trials.

Subgroup analysis and investigation of heterogeneity

We analysed subgroups according to the different interventions in some outcomes of the review. In future updates we plan to analyse separately patients with or without COPD.

Sensitivity analysis

We did not perform a sensitivity analysis because there were only three trials studying different interventions.

Results

Description of studies

Results of the search

We initially identified 1219 results with duplicates removed. This total was composed of 600 results from MEDLINE, 932 results from EMBASE and 487 results from CENTRAL. We updated our search in June 2011 and identified a further 42 results in MEDLINE, 126 results in EMBASE and 27 search results in CENTRAL. Our search was again updated in December 2011 which identified a further 87 records when duplicates were removed, 26 of these records were identified in MEDLINE, 80 records in EMBASE and 17 records in CENTRAL. We did not find any ongoing study suitable for the review in ISRCTN (http://www.controlled‐trials.com/isrctn/) and WHO ICTRP. A preliminary screening according to titles and abstracts ruled out the obviously unrelated articles and we merged the duplicates. We identified 361 trials. We obtained the full text for those articles that were ambiguous from their titles and abstracts so that we could determine whether to exclude them from the review.

Included studies

We included three trials (Confalonieri 1999; Cosentini 2010; Hilbert 2001) in our review.

Design

All the included RCTs used a parallel design.

Participants

The three trials focused on adults aged about 60 years. Forty‐one percent of participants in one trial (Confalonieri 1999) had COPD. Two trials were conducted in Italy (Confalonieri 1999; Cosentini 2010) and it was unclear in which countries the other trial were conducted (Hilbert 2001). One study (Hilbert 2001) focused on immunosuppressed patients. Baseline data were stated and the comparability was analysed in all three trials (Confalonieri 1999; Cosentini 2010; Hilbert 2001).

Interventions

The included three studies (Confalonieri 1999; Cosentini 2010; Hilbert 2001) compared NPPV and standard treatment with oxygen supplementation delivered by a Venturi mask. All the included studies used a level of pressure support and two studies (Cosentini 2010; Hilbert 2001) used a positive end‐expiratory pressure (PEEP).

Outcome measures

The included studies reported different outcome measures. Two trials (Confalonieri 1999; Hilbert 2001) reported mortality in hospital, participants requiring intubation, duration of ICU stay and duration of use of mechanical ventilation. One trial (Confalonieri 1999) reported mortality in the ICU, two‐month survival rate, duration of hospital stay and duration of intubation. Both trials reported side effects.

See the Characteristics of included studies table for more details.

Excluded studies

We excluded 16 trials (Antonelli 1998; Antonelli 2000; Antonelli 2001; Antonelli 2002; Antonelli 2007; Brett 1993; Brochard 2010; CRGSINV 2006; Delclaux 2000; Eisner 2001; Hilbert 1997; Ozsancak 2011; Prevedoros 1991; Terragni 2010; Wysocki 1995; Wysocki 2001). See Characteristics of excluded studies table for more information.

Risk of bias in included studies

The overall risk of bias is presented graphically in Figure 1 and summarised in Figure 2.

1.

'Risk of bias' graph: review authors' judgements about each methodological quality item, presented as percentages across all included studies

2.

'Risk of bias' summary: review authors' judgements about each methodological quality item for each included study

Allocation

Two trials clearly described adequate sequence generation (Confalonieri 1999; Cosentini 2010). The other trial (Hilbert 2001) did not describe the method of randomised sequence generation. All three trials described adequate allocation concealment.

Blinding

All included trials did not report adequate blinding, because of the type of intervention. One study (Cosentini 2010) reported that the block size of the randomisation list was known only to the study statistician. Blinding was not considered necessary for mortality or survival which was not influenced by blinding.

Incomplete outcome data

Two included trials (Confalonieri 1999; Hilbert 2001) provided complete outcome data. The other trial (Cosentini 2010) reported that two participants had to withdraw due to intolerance of the device; one in the control group and one in the CPAP group.

Selective reporting

There was not enough information to assess selective reporting bias because we did not have access to the protocols of the included studies.

Other potential sources of bias

All the included trials had small sample sizes and this might have led to other potential sources of bias.

Effects of interventions

We included three trials with a total of 151 participants.

Primary outcome measures

Mortality (death in hospital and intensive care unit (ICU)) and two‐month survival rate

Mortality in hospital

The three included studies reported mortality in the hospital. However, we could only extract these data from two studies (Confalonieri 1999; Hilbert 2001) because in the Cosentini 2010 study there was no report of participants dying. There were 54 participants in the non‐invasive ventilation group and 54 participants in the standard treatment. One study (Confalonieri 1999) looked at severe community‐acquired pneumonia; the results did not show any benefit in the non‐invasive ventilation group in mortality in hospital, odds ratio (OR) 1.22, 95% confidence interval (CI) ‐0.35 to 4.24 (Analysis 1.1). However, in the other study conducted among immunosuppressed participants (Hilbert 2001), the results show that the non‐invasive ventilation was better (OR 0.24, 95% CI 0.07 to 0.82) (Analysis 1.1). Pooled analyses showed that the mortality in hospital after treatment was not statistically significant (OR 0.54, 95% CI 0.11 to 2.68). The level of heterogeneity between these studies was high (Chi² test = 3.33; df = 1, P = 0.07; I² statistic = 70%) (Analysis 1.1).

1.1. Analysis.

Comparison 1 Death in the hospital, Outcome 1 Non‐invasive ventilation versus standard treatment.

Mortality in ICU and two‐month survival rate

Only one study (Hilbert 2001) reported mortality in the ICU and one study (Confalonieri 1999) reported the two‐month survival rate. The results showed that the non‐invasive ventilation was statistically significantly better than standard treatment in mortality in ICU and the 95% CI did not cross 1.0 (OR 0.28, 95% CI 0.09 to 0.88) (Analysis 2.1). However, the results were not statistically significant for the two‐month survival rate (OR 1.67, 95% CI 0.53 to 5.28) (Analysis 3.1).

2.1. Analysis.

Comparison 2 Death in the ICU, Outcome 1 Non‐invasive ventilation versus standard treatment.

3.1. Analysis.

Comparison 3 2‐month survival, Outcome 1 Non‐invasive ventilation versus standard treatment.

Need for invasive mechanical ventilation

Participants requiring intubation

The three included studies reported participants requiring intubation. However, we could only extract these data from two studies (Confalonieri 1999; Hilbert 2001) because in the Cosentini 2010 study there was no report of participants requiring intubation. There were 54 participants in the non‐invasive ventilation group and 54 participants in the standard treatment. Two studies (Confalonieri 1999; Hilbert 2001) were conducted in severe community‐acquired pneumonia and immunosuppressed participants respectively; their results both showed that non‐invasive ventilation was better than standard treatment and the difference between the two compared groups was statistically significant (OR 0.27, 95% CI 0.08 to 0.88; OR 0.26, 95% CI 0.08 to 0.85, respectively) (Analysis 4.1). Pooled analyses also showed that the participants requiring intubation after treatment was statistically significant (OR 0.26, 95% CI 0.11 to 0.61). The level of heterogeneity between these studies was low (Chi² test = 0; df = 1, P = 0.94; I² statistic = 0%) (Analysis 4.1).

4.1. Analysis.

Comparison 4 Participants required intubation, Outcome 1 Non‐invasive ventilation versus standard treatment.

Duration of intubation

Only one study (Confalonieri 1999) reported the duration of intubation; the result showed that non‐invasive ventilation can decease the duration of intubation more than standard treatment and the difference between the two compared groups was statistically significant (mean difference (MD) ‐3.00, 95% CI ‐4.57 to ‐1.43) (Analysis 5.1).

5.1. Analysis.

Comparison 5 Duration of intubation, Outcome 1 Non‐invasive ventilation versus standard treatment.

Clinical response based on improvement or deterioration in terms of signs and symptoms

Neither of the included trials reported on clinical response based on signs and symptoms. This outcome will be included when the review is updated if data are available.

Secondary outcome measures

Duration of hospital stay

One study (Confalonieri 1999) reported the duration of hospital stay. The result showed that the duration of hospital stay after treatment was not statistically significant (MD ‐1.00, 95% CI ‐2.05 to 0.05) (Analysis 6.1).

6.1. Analysis.

Comparison 6 Duration of hospital stay (d), Outcome 1 Non‐invasive ventilation versus standard treatment.

Duration of ICU stay

Two studies (Confalonieri 1999; Hilbert 2001) reported on the duration of ICU stay. Their results both showed that non‐invasive ventilation was better than standard treatment and the difference between the two compared groups was statistically significant (MD ‐4.20, 95% CI ‐4.98 to ‐3.42; MD ‐2.00, 95% CI ‐3.92 to ‐0.08, respectively) (Analysis 7.1). Pooled analyses also showed that participants requiring ICU stay after treatment was statistically significant (MD ‐3.28, 95% CI ‐5.41 to ‐1.16). The level of heterogeneity between these studies was high (Chi² test = 4.31; df = 1, P = 0.04; I² statistic = 77%) (Analysis 7.1).

7.1. Analysis.

Comparison 7 Duration of ICU stay (d), Outcome 1 Non‐invasive ventilation versus standard treatment.

Duration of mechanical ventilation

Two studies (Confalonieri 1999; Hilbert 2001) reported the duration of use of mechanical ventilation. There were 54 participants in the non‐invasive ventilation group and 54 participants in the standard treatment group. One study (Confalonieri 1999) was conducted among severe community‐acquired pneumonia participants; the results did not show any benefit of non‐invasive ventilation in the duration of use of mechanical ventilation among patients with or without COPD (standardised mean difference (SMD) ‐0.74, 95% CI ‐1.60 to 0.11; SMD ‐0.34, 95% CI ‐1.03 to 0.34, respectively) (Analysis 8.1). In the other study (Hilbert 2001), conducted among immunosuppressed participants, the results did not show any benefit of non‐invasive ventilation compared with standard treatment (SMD 0, 95% CI ‐0.54 to 0.54) (Analysis 8.1). Pooled analyses showed that the duration of use of mechanical ventilation after treatment was not statistically significant (SMD ‐0.26, 95% CI ‐0.66 to 0.14). The level of heterogeneity between these studies was low (Chi² test = 2.18; df = 2, P = 0.34; I² statistic = 8%) (Analysis 8.1).

8.1. Analysis.

Comparison 8 Duration of use of mechanical ventilation, Outcome 1 Non‐invasive ventilation versus standard treatment.

Complications and side effects

All included trials reported side effects. One study (Confalonieri 1999) conducted among severe community‐acquired pneumonia participants reported complications including gastric distention, ventilation‐associated pneumonia, otitis media, mastoiditis and pneumothorax. The result showed that between the compared groups the incidence of complications was not significantly different (OR 0.22, 95% CI 0.02 to 2.13) (Analysis 9.1). However, in the Hilbert 2001 trial, conducted among immunosuppressed participants, serious complications were reported, such as severe sepsis or septic shock, cardiogenic shock and ventilator‐associated pneumonia; the results showed that non‐invasive ventilation was better than standard treatment in terms of serious complications and the difference between the two compared groups was statistically significant (OR 0.24, 95% CI 0.07 to 0.82) (Analysis 9.1). Pooled analyses also showed that the participants who had complications after treatment were statistically significant (OR 0.23, 95% CI 0.08 to 0.70). The level of heterogeneity between these studies was low (Chi² test = 0; df = 1, P = 0.96; I² statistic = 0%) (Analysis 9.1).

9.1. Analysis.

Comparison 9 Complications, Outcome 1 Non‐invasive ventilation versus standard treatment.

One study (Cosentini 2010) focusing on moderate hypoxaemic acute respiratory failure (diagnosis of community‐acquired pneumonia (CAP) as the only cause), reported that no participants required intubation or died during treatment and no serious adverse events were seen. Only two participants had to withdraw due to intolerance of the device, one in the control group and one in the continuous positive airway pressure (CPAP) group.

Discussion

Summary of main results

The results for the included outcomes were unsatisfactory and susceptible to bias because in some subgroups the level of heterogeneity between pooled studies was substantial and some of them were extracted from only one or two studies. In addition, the included studies had small sample sizes and the participant populations in the included studies were very different which may result in clinical heterogeneity.

Overall completeness and applicability of evidence

Completeness of evidence

At present, our information regarding treatment of patients with pneumonia requiring oxygen comes from clinical guidelines. No guidelines or studies suggest whether patients without hypoxaemia should be given oxygen supplementation or not. In general, it has been suggested that oxygen saturation should be maintained above 90% in patients with pneumonia (Alvin 2004). Oxygen therapy is often only necessary in severe cases of pneumonia (Ewig 2006; Lim 2003) or in those patients with hypoxaemia (partial pressure of arterial oxygen (PaO₂) less than 8 Kpa) or oxygen saturation (SaO₂) less than 92% (Niederman 2001a). The British Thoracic Society (BTS) recommends that the initial oxygen saturation target range should be 94% to 98% if the patient with pneumonia is hypoxaemic (Levy 2010a). We assumed there would be many randomised controlled trials (RCTs) of pneumonia with hypoxaemia because of the routine nature of oxygen therapy. To our surprise, very few RCTs conducted on pneumonia with definite criteria were identified, although there were a large number of references about oxygen therapy for respiratory failure. We know that the general methods of oxygen supplementation include nasal catheter or mask, non‐invasive mechanical ventilation and invasive mechanical ventilation.

There are also few articles comparing the advantages and disadvantages of different oxygen therapies for severe pneumonia. We identified six articles; three of them were included (Confalonieri 1999; Cosentini 2010; Hilbert 2001) and 16 (Antonelli 1998; Antonelli 2000; Antonelli 2001; Antonelli 2002; Antonelli 2007; Brett 1993; Brochard 2010; CRGSINV 2006; Delclaux 2000; Eisner 2001; Hilbert 1997; Ozsancak 2011; Prevedoros 1991; Terragni 2010; Wysocki 1995; Wysocki 2001) were excluded after re‐evaluation. In the three included studies, there were no comparisons between effects of placebo and oxygen treatment. All of them focused on the impact of non‐invasive ventilation with standard oxygen supplementation via a Venturi mask on participants with severe community‐acquired pneumonia (CAP) (with or without chronic obstructive pulmonary disease (COPD)) and immunosuppressed patients with pulmonary infiltrates.

Clinical interpretation of the data

The outcomes of the analysis show that non‐invasive ventilation was better than standard oxygen supplementation via a Venturi mask in terms of death in the intensive care unit (ICU), the number of participants who required intubation, the duration of intubation, duration of ICU stay and complications. This was also true for death in the hospital, two‐month survival, duration of hospital stay and duration of use of mechanical ventilation, where there was no statistically significant difference observed between the treatment and control groups. The primary outcomes of death in the ICU, two‐month survival and duration of intubation were based only on one study (Confalonieri 1999) in each subgroup conducted among severe CAP patients. In addition to this, there were different kinds of populations of participants in the studies so that heterogeneity cannot be avoided in some outcomes. The outcomes of the subgroup analysis show that non‐invasive ventilation was better than standard oxygen supplementation via a Venturi mask in terms of death in hospital and complications among immunosuppressed patients with pulmonary infiltrates, whereas in the subgroup of severe community‐acquired pneumonia, there was no statistically significant difference observed between the non‐invasive ventilation and standard treatment groups.

Non‐invasive ventilation was safe and adverse effects were minor among moderate hypoxaemic acute respiratory failure participants (diagnosis of CAP as the only cause). Among severe community‐acquired pneumonia populations the incidence of complications including gastric distention, ventilation‐associated pneumonia, otitis media, mastoiditis and pneumothorax between the compared groups was not significantly different. However, among immunosuppressed patients there were serious complications, such as severe sepsis or septic shock, cardiogenic shock and ventilator‐associated pneumonia; the results showed that non‐invasive ventilation was better than standard treatment in terms of serious complications and that the difference between the two compared groups was statistically significant. Pooled analyses also showed that non‐invasive ventilation can decrease the incidence of complications. Furthermore, we did not include patients with pulmonary tuberculosis and cystic fibrosis because of their varying prognoses. We suggest separate studies should be implemented for these patients. So when we need to apply the conclusions of this review to clinical practice, we must carefully evaluate whether our patients are comparable with the participants described in this review.

Quality of the evidence

One trial (Hilbert 2001) did not describe the methods of randomised sequence generation but the study clearly described adequate allocation concealment. The quality of the other two trials (Confalonieri 1999; Cosentini 2010) was higher and clearly described adequate sequence generation methods and adequate allocation concealment.

None of the included trials documented the methods for blinding due to the specific intervention which is very difficult to blind. This did not influence outcomes such as hospital mortality and two‐month survival rate. However, for other outcomes in this review, this could have potentially biased the results in favour of treatment.

Potential biases in the review process

All outcomes measured in this review come from one or two included studies and all were small trials. Any real effects of interventions may have remained undetected because of the small sample sizes. Furthermore, substantial statistical heterogeneity and clinical heterogeneity due to different kinds of participants must be considered.

Agreements and disagreements with other studies or reviews

COPD is a very common underlying disease in patients admitted with severe CAP (Ewig 1997; Moine 1994) but is not recognised as an independent risk factor for mortality (Ewig 1997). Even so, outcomes for participants with or without COPD were analysed in a subgroup analysis. In this review, we did not find a difference between patients with and without COPD in duration of use of mechanical ventilation. One included study (Confalonieri 1999) found that in patients with COPD, non‐invasive ventilation required a lower workload for nursing staff over three days, had a lower rate of endotracheal intubation, a shorter duration of ICU stay and a reduced duration of intubation. These differences were statistically significant. Conversely, the differences between outcomes were not found in participants without COPD. In fact, non‐invasive ventilation itself could result in decreased mortality, decreased need for intubation and a reduction in the treatment failure rate in patients with COPD (Felix 2004). The differences may be due to an improvement in outcomes of patients with COPD rather than an improvement due to non‐invasive ventilation use in participants with pneumonia.

A study showed non‐invasive ventilation could reduce the rate of endotracheal intubation and length of ICU stay in patients with a PaCO₂ greater than 45 mmHg but the advantage could not be found in those with PaCO₂ less than 45 mm Hg (Wysocki 1995). The most common complications associated with non‐invasive ventilation are gastric distention, difficulty in clearing secretions and facial skin necrosis. In this review, complications associated with conventional mechanical ventilation, influenced by population of participants, ranged from very minor to very serious.

Authors' conclusions

Implications for practice.

Current available evidence shows that non‐invasive ventilation can reduce the risk of death in the intensive care unit (ICU) and endotracheal intubation, and shorten ICU length of stay and intubation in patients with severe community‐acquired pneumonia; non‐invasive ventilation can reduce the risk of death in hospital, shorten ICU length of stay and reduce the incidence of complications in immunosuppressed patients with pulmonary infiltrates. Overall, non‐invasive ventilation can reduce the risk of death in ICU and endotracheal intubation, and shorten ICU length of stay and length of intubation. Non‐invasive ventilation and standard oxygen supplementation via a Venturi mask were similar when measuring mortality in hospital, two‐month survival, duration of hospital stay and duration of mechanical ventilation. Other than oxygen therapy, we must mention the importance of standard treatment by physicians. The evidence is weak. However, the review indicates that non‐invasive ventilation may be more beneficial than standard oxygen supplementation via a Venturi mask for pneumonia.

Implications for research.

Few studies of oxygen therapy have been performed in adults with pneumonia. Well‐designed randomised controlled trials are required with larger numbers of participants to answer these clinical questions. We did not include patients with pulmonary tuberculosis and cystic fibrosis because of their varying prognoses. We suggest separate studies should be implemented for these patients.

History

Protocol first published: Issue 3, 2007
 Review first published: Issue 3, 2012

Date	Event	Description
29 July 2010	Amended	Converted to new review format.

Notes

This protocol was withdrawn from The Cochrane Library in Issue 9, 2010, as the review authors were not able to write the review. A new team have subsequently updated this review.

Appendices

Appendix 1. Embase.com search strategy

EMBASE

#1 'Pneumonia' /exp 
 #2 pneumon* 
 #3 cap 
 #4 hap 
 #5 #1 OR #2 OR #3 OR #4 
 #6 'Oxygen Inhalation Therapy' /exp 
 #7 (oxygen* or o2*) AND (therap* or inhal* or respirat* or deliver*) 
 #8 (oxygen* or o2*) AND (mask* or hudson* or venture* or nrm*) 
 #9 nasal AND (cannula* or prong*) 
 #10 'Positive‐Pressure Respiration' /exp 
 #11 'non invasive positive pressure ventilation' 
 #12 nasal continuous positive airway* pressure 
 #13 'artificial ventilation' /de 
 #14 'mechanical ventilation' 
 #15 cpap or ncpap or nppv 
 #16 #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 
 #17 #5 and #16 
 #18 #17 AND [randomized controlled trial]/lim

Appendix 2. Trials registry search terms

WHO ICTRP

1 (oxygen* OR o2* OR mask* OR hudson* OR ventur* OR nrm* OR nasal OR cannula* OR positive pressure OR continuous positive airway* OR pressure mechanical ventilation OR cpap OR CPAP OR NPPV)  in the Title  
 2 (pneumon* OR lung infection OR CAP OR HAP) in the Condition 
 3 1 AND 2

ISRCTN

(pneumon* AND oxyge*)

Data and analyses

Comparison 1. Death in the hospital.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	2	108	Odds Ratio (M‐H, Random, 95% CI)	0.54 [0.11, 2.68]
1.1 Severe community‐acquired pneumonia	1	56	Odds Ratio (M‐H, Random, 95% CI)	1.22 [0.35, 4.24]
1.2 Immunosuppressed patients with pulmonary infiltrates	1	52	Odds Ratio (M‐H, Random, 95% CI)	0.24 [0.07, 0.82]

Comparison 2. Death in the ICU.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	1		Odds Ratio (M‐H, Random, 95% CI)	Totals not selected

Comparison 3. 2‐month survival.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	1		Odds Ratio (M‐H, Random, 95% CI)	Totals not selected

Comparison 4. Participants required intubation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	2	108	Odds Ratio (M‐H, Random, 95% CI)	0.26 [0.11, 0.61]
1.1 Severe community‐acquired pneumonia	1	56	Odds Ratio (M‐H, Random, 95% CI)	0.27 [0.08, 0.88]
1.2 Immunosuppressed patients with pulmonary infiltrates	1	52	Odds Ratio (M‐H, Random, 95% CI)	0.26 [0.08, 0.85]

Comparison 5. Duration of intubation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 6. Duration of hospital stay (d).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	1		Mean Difference (IV, Random, 95% CI)	Totals not selected

Comparison 7. Duration of ICU stay (d).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	2	108	Mean Difference (IV, Random, 95% CI)	‐3.28 [‐5.41, ‐1.16]
1.1 Severe community‐acquired pneumonia	1	56	Mean Difference (IV, Random, 95% CI)	‐4.2 [‐4.98, ‐3.42]
1.2 Immunosuppressed patients with pulmonary infiltrates	1	52	Mean Difference (IV, Random, 95% CI)	‐2.0 [‐3.92, ‐0.08]

Comparison 8. Duration of use of mechanical ventilation.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	2	108	Std. Mean Difference (IV, Random, 95% CI)	‐0.26 [‐0.66, 0.14]
1.1 Severe community‐acquired pneumonia: patients with COPD	1	23	Std. Mean Difference (IV, Random, 95% CI)	‐0.74 [‐1.60, 0.11]
1.2 Severe community‐acquired pneumonia: patients without COPD	1	33	Std. Mean Difference (IV, Random, 95% CI)	‐0.34 [‐1.03, 0.34]
1.3 Immunosuppressed patients with pulmonary infiltrates	1	52	Std. Mean Difference (IV, Random, 95% CI)	0.0 [‐0.54, 0.54]

Comparison 9. Complications.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Non‐invasive ventilation versus standard treatment	2	108	Odds Ratio (M‐H, Random, 95% CI)	0.23 [0.08, 0.70]
1.1 Severe community‐acquired pneumonia	1	56	Odds Ratio (M‐H, Random, 95% CI)	0.22 [0.02, 2.13]
1.2 Immunosuppressed patients with pulmonary infiltrates	1	52	Odds Ratio (M‐H, Random, 95% CI)	0.24 [0.07, 0.82]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Confalonieri 1999.

Methods	Design: multi centre, prospective, randomised controlled trial (n = 56) Method of randomisation: computer‐generated random assignments (software RND; Istituto Nazionale Ricerca sul) The random sequence was concealed in sealed envelopes Blinding: double‐blinding not clearly stated
Participants	Country: Italy Setting: intensive care unit (ICU) No. of participants: 56; 28 in non‐invasive ventilation group, 43 in standard treatment group Age: about 60 years Inclusion criteria: participants with severe community‐acquired pneumonia (CAP) and active respiratory failure (ARF) were enrolled. The criteria for severe CAP were 1 or more of the American Thoracic Society (ATS) non‐respiratory criteria and the criteria for severe ARF were 2 or more of the following criteria: 1) acute respiratory distress including severe dyspnoea at rest and a respiratory rate (RR) > 35 breaths/min and/or active contraction of the accessory muscles of respiration or paradoxical abdominal motion 2) PaO2 < 68 mmHg while receiving a fraction of inspired oxygen (FiO2) >= 0.4, or a ratio of the partial pressure of arterial oxygen to the fraction of inspired oxygen (PaO2:FiO2) < 250, while receiving a FiO2 > 0.5 3) hypercapnia (PaCO2 > 50 mmHg) with respiratory acidosis (pH < 7.33) Exclusion criteria: a requirement for emergency intubation for cardiopulmonary resuscitation, respiratory arrest, severe haemodynamic instability, encephalopathy, severe neurologic disease, concomitant severe disease with an expectation of life less than 4 months (for example, advanced cancer), long‐term oxygen therapy or home mechanical ventilation, contraindications for using masks (tracheostomy or facial deformities), or inability to expectorate
Interventions	Participants received non‐invasive ventilation through a face mask and standard oxygen supplementation through a Venturi mask. Medical management was similar in the 2 groups and included initial antibiotics following the ATS guidelines
Outcomes	The primary outcomes: 1) Developing preselected objective for endotracheal intubation 2) The need for endotracheal intubation and mechanical ventilation at any time during the study Secondary end points: 1) Length of hospital stay 2) Complications not present on admission (such as VAP) 3) Duration of ventilatory assistance 4) Duration of ICU and hospital stay 5) In‐hospital survival 6) 2‐month (from study entry) survival
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated random assignments
Allocation concealment (selection bias)	Low risk	Concealed in sealed envelopes
Blinding (performance bias and detection bias) All outcomes except mortality	High risk	No information provided
Incomplete outcome data (attrition bias) All outcomes except mortality and survival	Low risk	No participants lost to follow‐up
Blinding of participants and personnel (performance bias) All outcomes	High risk	The study may not have used blinded participants
Blinding of outcome assessment (detection bias) All outcomes	High risk	The study may not have used blinding

Cosentini 2010.

Methods	Design: multi centre, prospective, open‐label, controlled trial in parallel groups Method of randomisation: computer‐generated randomisation list, unique for each centre The random sequence was concealed in opaque sealed envelopes Blinding: open‐label
Participants	Country: Italy Setting: Emergency Department (ED) between January 2006 and February 2008 No. of participants: 47; 20 in non‐invasive continuous positive airway pressure (CPAP) group, 27 in standard oxygen therapy group Age: ≥ 18 years Inclusion criteria: diagnosis of CAP as the only cause of ARF, respiratory rate (RR) 35 breaths/min and PaO2 /FiO2 ratio ≥ 210 and ≤ 285 evaluated during oxygen therapy supplied for at least 15 min through a Venturi mask (FiO2 of 0.50, flow around 40 L/m) Exclusion criteria: at least 1 of the following was present: diagnosis of healthcare‐associated pneumonia, immunosuppressive state, acute cardiogenic pulmonary oedema, unstable angina or acute myocardial infarction, respiratory acidosis, failure of 3 or more organs, systolic BP, 90 < mm Hg despite fluid resuscitation or vasopressors, severe arrhythmias, contraindications to CPAP treatment or pregnancy
Interventions	CPAP was delivered through a high‐flow generator (90 to 140 L/min; VitalSigns Inc; Totowa, NJ) using a helmet (StarMed, Italy) as interface with a PEEP valve (VitalSigns). CPAP was applied with an initial PEEP of 10 cm H2O and with a FiO2 set to maintain a pulse oximetry ≥ 92%. PEEP value was maintained at 10 cm H2O until CPAP removal Standard oxygen therapy was supplied through a Venturi mask with an FiO2 delivered to maintain a pulse oximetry 92% Participants could withdraw from the study at any time and be removed by the attending physician if any of the following occurred: PaO2 /FiO2 ratio < 200 with an RR > 35 breaths/min, acute failure of 3 or more organs, or criteria for endotracheal intubation (at least 1 among: respiratory arrest, severe haemodynamic instability; or at least 2 among: RR > 35 breaths/min, PaO2 /FiO2 ratio < 150, an increase of the PaCO2 > 20% in comparison with the previous ABG, mental status change due to respiratory failure)
Outcomes	1. Primary end point: the time to reach an improvement in terms of gas exchange, defined as a PaO2 /FiO2 ratio ≥ 315. Participants who did not reach this threshold level before the last planned arterial blood gas (ABG) measurement at 48 hours were considered as failures 2. The following data were monitored: the proportion of participants who could maintain, once reached, a PaO2 /FiO2 ratio ≥ 315 at 1 and 24 hours after the qualifying measurement, the frequency of adverse events and in‐hospital mortality
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Using a computer‐generated randomisation list, unique for each centre
Allocation concealment (selection bias)	Low risk	Using opaque sealed envelopes to conduct the allocation concealment
Blinding (performance bias and detection bias) All outcomes except mortality	Unclear risk	The block size of randomisation was known only to the study statistician
Incomplete outcome data (attrition bias) All outcomes except mortality and survival	Low risk	Only 2 participants (4.3%) lost to follow‐up
Blinding of participants and personnel (performance bias) All outcomes	High risk	An open‐label design
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	The block size of randomisation was known only to the study statistician

Hilbert 2001.

Methods	Design: prospective, controlled trial in parallel groups Method of randomisation: no information provided The random sequence was concealed in sealed envelopes Blinding: no information provided
Participants	Country: no information provided Setting: intensive care unit (ICU) No. of participants: 52; 26 in non‐invasive ventilation group, 26 in standard treatment group Age: about 50 years Inclusion criteria: a clinical history of pulmonary infiltrates and fever, as evidenced by a temperature of more than 38.3 °C, the finding of persistent pulmonary infiltrates on radiographs and a deterioration in pulmonary gas exchange; severe dyspnoea at rest; a respiratory rate of more than 30 breaths per minute; and a ratio of the partial pressure of arterial oxygen to the fraction of inspired oxygen (PaO2:FiO2) of less than 200 while the patient was breathing oxygen through a Venturi mask Exclusion criteria: a requirement for emergency intubation for cardiopulmonary resuscitation or as a result of respiratory arrest or a rapid deterioration in neurologic status; haemodynamic instability, defined as a systolic blood pressure of less than 80 mm Hg or evidence on electrocardiography of ischaemia or clinically significant ventricular arrhythmias; chronic obstructive pulmonary disease; respiratory failure of cardiac origin; a partial pressure of arterial carbon dioxide of more than 55 mm Hg, with acidosis; uncorrected bleeding diathesis; and tracheotomy, a facial deformity, or a recent history of oral, oesophageal or gastric surgery
Interventions	In all participants, the heart rate and respiratory rate were monitored continuously, and arterial oxygen saturation was monitored continuously with a bedside pulse oximeter. The head of the bed was kept elevated at a 45‐degree angle. Medications included antimicrobial agents, diuretics, bronchodilators, immunosuppressive agents and subcutaneous heparin. Parenteral nutrition was given, fluids were administered to maintain volume and electrolyte abnormalities were corrected Non‐invasive ventilation group: non‐invasive ventilation was delivered to the patient through a full face mask. The mask was adjusted and connected to a ventilator set in the pressure‐support mode. Ventilator settings were adjusted on the basis of continuous monitoring of arterial oxygen saturation, clinical data and measurements of arterial blood gases Standard treatment group: participants who were assigned to the standard treatment group received oxygen through a Venturi mask. The rate of administration of oxygen was adjusted to achieve a level of arterial oxygen saturation above 90%. Criteria for intubation: the predetermined criteria were as follows: failure to maintain a PaO2:FiO2 of more than 85, the development of conditions necessitating endotracheal intubation to protect the airways; the development of copious tracheal secretions; an increase in the partial pressure of arterial carbon dioxide accompanied by a pH of 7.30 or less; agitation requiring sedation; severe haemodynamic instability, defined as a systolic blood pressure of less than 70 mm Hg or evidence on electrocardiography of ischaemia or clinically significant ventricular arrhythmias; and inability on the part of a patient who was randomly assigned to receive non‐invasive ventilation to tolerate the face mask
Outcomes	Primary outcome variable: the need for endotracheal intubation and mechanical ventilation at any time during the study Secondary outcome variables: the development of complications not present on admission; the length of stay in the intensive care unit; the duration of ventilatory assistance; death in the intensive care unit; and death in the hospital. In addition, arterial blood gas levels, gas exchange and diagnostic microbiologic tests were conducted
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No information provided
Allocation concealment (selection bias)	Low risk	Concealed in sealed envelopes
Blinding (performance bias and detection bias) All outcomes except mortality	Unclear risk	No information provided
Incomplete outcome data (attrition bias) All outcomes except mortality and survival	Low risk	No participants lost to follow‐up
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	No information provided
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	No information provided

ABG: arterial blood gas 
 ALI: acute lung injury 
 ARDS: acute respiratory distress syndrome 
 ARF: active respiratory failure 
 ATS: American Thoracic Society 
 BP: blood pressure 
 CAP: community‐acquired pneumonia 
 CPAP: continuous positive airway pressure 
 ED: Emergency department 
 FiO₂: fraction of inspired oxygen 
 ICU: intensive care unit 
 mmHg: mm of mercury 
 PaO₂: partial pressure of arterial oxygen 
 PEEP: positive end‐expiratory pressure 
 RR: respiratory rate 
 VAP: ventilator‐associated pneumonia

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Antonelli 1998	This study did not assess the role of 2 interventions in pneumonia participants. It only reported that for the 5 pneumonia participants in the non‐invasive group, 4 of them did not need intubation and in the 4 participants who received invasive ventilation, 2 died and 2 survived
Antonelli 2000	The study was not a RCT
Antonelli 2001	This study was a cohort study not a RCT
Antonelli 2002	The participants in this study were not diagnosed with pneumonia and the study did not have a pneumonia subgroup
Antonelli 2007	The study focuses on participants with acute respiratory distress syndrome and there was no pneumonia subgroup
Brett 1993	This study is 3 case reports rather than a RCT
Brochard 2010	The study focus on participants with COPD rather than pneumonia
CRGSINV 2006	The study was to evaluate the feasibility and the efficacy of early extubation and sequential non‐invasive mechanical ventilation in COPD patients with exacerbated hypercapnic respiratory failure which was exacerbated by pulmonary infection not by pneumonia
Delclaux 2000	The study focuses on patients with acute hypoxaemic respiratory insufficiency which was caused by pneumonia and cardiac disease and there was no subgroup of pneumonia
Eisner 2001	The participants in this study had acute lung injury and acute respiratory distress syndrome. Pneumonia is only a risk of acute lung injury and acute respiratory distress syndrome
Hilbert 1997	This study was a cohort study not a RCT
Ozsancak 2011	Only 23% participants had pneumonia and the study did not have a pneumonia subgroup
Prevedoros 1991	This study is a retrospective study rather than a RCT
Terragni 2010	The enrolled participants in this study were not diagnosed with pneumonia and the study did not have a pneumonia subgroup
Wysocki 1995	The study did not separately statistically analyse of outcomes of patients with pneumonia. It just reported that in 7 participants assigned to receive non‐invasive ventilation, all of them received endotracheal intubation because of encephalopathy (N = 3), deterioration in respiratory rate and in gas exchange (N = 2), deterioration in gas exchange (N = 1) and shock (N= 1). But for the other 9 participants who used conventional therapy, including drugs and oxygen supplement via a nasal catheter, their endotracheal intubation rate was not reported
Wysocki 2001	This study was not a RCT

CPAP: continuous positive airway pressure 
 RCT: randomised controlled trial

Characteristics of studies awaiting assessment [ordered by study ID]

Maraffi 2009.

Methods	We cannot find the details of the study
Participants
Interventions
Outcomes
Notes

Wijesinghe 2009.

Methods	We cannot find the details of the study
Participants
Interventions
Outcomes
Notes

Differences between protocol and review

According to the new version of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), we have replaced the 'Quality assessment of included studies' in the original version with 'Assessment of risk of bias in included studies'.

Contributions of authors

Yanling Zhang (YZ) contributed to the review proposal, wrote the draft, finished full text and the search strategy section. 
 Chengli Fang (CF) contributed to search strategy section, data extraction and drafting the review. 
 Juelin Deng contributed to modify the draft. 
 Birong Dong (BD) and Taixiang Wu (TW) contributed to drafting the review and advised on changes and the quality assessment.

Sources of support

Internal sources

• Chinese Cochrane Centre, West China Hospital, Sichuan University, China.

External sources

• Chinese Medical Board of New York, USA.

Declarations of interest

None known.

New