Scolaris Content Display Scolaris Content Display

Cochrane Database of Systematic Reviews Review - Intervention

Chronic non‐invasive ventilation for chronic obstructive pulmonary disease

Authors' declarations of interest

Version published: 09 August 2021 Version history

https://doi.org/10.1002/14651858.CD002878.pub3
Collapse all Expand all

Background

Chronic non‐invasive ventilation (NIV) is increasingly being used to treat people with COPD who have respiratory failure, but the evidence supporting this treatment has been conflicting.

Objectives

To assess the effects of chronic non‐invasive ventilation at home via a facial mask in people with COPD, using a pooled analysis of IPD and meta‐analysis.

Search methods

We searched the Cochrane Airways Register of Trials, MEDLINE, Embase, PsycINFO, CINAHL, AMED, proceedings of respiratory conferences, clinical trial registries and bibliographies of relevant studies. We conducted the latest search on 21 December 2020.

Selection criteria

We included randomised controlled trials (RCTs) comparing chronic NIV for at least five hours per night for three consecutive weeks or more (in addition to standard care) versus standard care alone, in people with COPD. Studies investigating people initiated on NIV in a stable phase and studies investigating NIV commenced after a severe COPD exacerbation were eligible, but we reported and analysed them separately. The primary outcomes were arterial blood gases, health‐related quality of life (HRQL), exercise capacity (stable COPD) and admission‐free survival (post‐exacerbation COPD). Secondary outcomes for both populations were: lung function, COPD exacerbations and admissions, and all‐cause mortality. For stable COPD, we also reported respiratory muscle strength, dyspnoea and sleep efficiency.

Data collection and analysis

We used standard methodological procedures expected by Cochrane. After inclusion of a study, we requested the IPD. We analysed continuous and time‐to‐event data using linear‐ and cox‐regression mixed‐effect models with a random effect on study level. We analysed dichotomous IPD using generalised estimating equations. We adjusted all models for age and sex. We assessed changes in outcomes after three and 12 months.  We also conducted a meta‐analysis on aggregated trial data.

Main results

We included 14 new RCTs in this review update, in addition to the seven previously included. Seventeen studies investigated chronic NIV in stable COPD and four studies investigated chronic NIV commenced after a severe COPD exacerbation. Three studies compared NIV to sham continuous positive airway pressure (2 to 4 cmH2O). Seven studies used a nasal mask, one study used an oronasal mask and eight studies used both interfaces. Five studies did not report the interface. The majority of trials (20/21) were at high risk of performance bias due to an unblinded design. We considered 11 studies to have a low risk of selection bias and 13 to have a low risk of attrition bias. We collected and analysed the IPD from 13 stable COPD studies (n = 778, 68% of the participants included) and from three post‐exacerbation studies (n = 364, 96% of the participants included).

In the stable COPD group, NIV probably results in a minor benefit on the arterial partial pressure of oxygen (PaO2) after three months (adjusted mean difference (AMD) 0.27 kPa, 95% CI 0.04 to 0.49; 9 studies, 271 participants; moderate‐certainty evidence), but there was little to no benefit at 12 months (AMD 0.09 kPa, 95% CI ‐0.23 to 0.42; 3 studies, 171 participants; low‐certainty evidence). The arterial partial pressure of carbon dioxide (PaCO2) was reduced in participants allocated to NIV after three months (AMD ‐0.61 kPa, 95% CI ‐0.77 to ‐0.45; 11 studies, 475 participants; high‐certainty evidence) and persisted up to 12 months (AMD ‐0.42 kPa, 95% CI ‐0.68 to ‐0.16; 4 studies, 232 participants; high‐certainty evidence). 

Exercise capacity was measured with the 6‐minute walking distance (minimal clinical important difference: 26 m). There was no clinically relevant effect of NIV on exercise capacity (3 months: AMD 15.5 m, 95% CI ‐0.8 to 31.7; 8 studies, 330 participants; low‐certainty evidence; 12 months: AMD 26.4 m, 95% CI ‐7.6 to 60.5; 3 studies, 134 participants; very low‐certainty evidence). HRQL was measured with the Severe Respiratory Insufficiency and the St. Georges's Respiratory Questionnaire and may be improved by NIV, but only after three months (3 months: standardised mean difference (SMD) 0.39, 95% CI 0.15 to 0.62; 5 studies, 259 participants; very low‐certainty evidence; 12 months: SMD 0.15, 95% CI ‐0.13 to 0.43; 4 studies, 200 participants; very low‐certainty evidence). Lastly, the risk for all‐cause mortality is likely reduced by NIV (adjusted hazard ratio (AHR) 0.75, 95% CI 0.58 to 0.97; 3 studies, 405 participants; moderate‐certainty evidence).

In the post‐exacerbation COPD group, there was little to no benefit on the PaO2 after three months, but there may be a slight decrease after 12 months (3 months: AMD ‐0.10 kPa, 95% CI ‐0.65 to 0.45; 3 studies, 234 participants; low‐certainty evidence; 12 months: ‐0.27 kPa, 95% CI ‐0.86 to 0.32, 3 studies; 170 participants; low‐certainty evidence). The PaCO2 was reduced by NIV at both three months (AMD ‐0.40 kPa, 95% CI ‐0.70 to ‐0.09; 3 studies, 241 participants; moderate‐certainty evidence) and 12 months (AMD ‐0.52 kPa, 95% CI ‐0.87 to ‐0.18; 3 studies, 175 participants; high‐certainty evidence). NIV may have little to no benefit on HRQL (3 months: SMD 0.25, 95% CI ‐0.01 to 0.51; 2 studies, 219 participants; very low‐certainty evidence; 12 months: SMD 0.25, 95% ‐0.06 to 0.55; 2 studies, 164 participants; very low‐certainty evidence). Admission‐free survival seems improved with NIV (AHR 0.71, 95% CI 0.54 to 0.94; 2 studies, 317 participants; low‐certainty evidence), but the risk for all‐cause mortality does not seem to improve (AHR 0.97, 95% CI 0.74 to 1.28; 2 studies, 317 participants; low‐certainty evidence).

Authors' conclusions

Regardless of the timing of initiation, chronic NIV improves daytime hypercapnia. In addition, in stable COPD, survival seems to be improved and there might be a short term HRQL benefit. In people with persistent hypercapnia after a COPD exacerbation, chronic NIV might prolong admission‐free survival without a beneficial effect on HRQL.

In stable COPD, future RCTs comparing NIV to a control group receiving standard care might no longer be warranted, but research should focus on identifying participant characteristics that would define treatment success. Furthermore, the optimal timing for initiation of NIV after a severe COPD exacerbation is still unknown.

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.

Non‐invasive ventilation (ventilators) used at night by people with chronic obstructive pulmonary disease (COPD)

Background: Non‐invasive ventilation (NIV) is a method to assist or replace spontaneous breathing (or normal breathing) with the aid of a machine called a ventilator, using a mask fitted over the nose or both the nose and mouth. NIV can be used chronically (long‐term) at the person's home if they have levels of carbon dioxide in their blood that are persistently too high. We wanted to discover if using chronic NIV at home during the night alongside standard therapy was better or worse than standard therapy alone in people with chronic obstructive pulmonary disease (COPD) who have raised carbon dioxide levels. In 2002 and 2013, we published our original Cochrane Reviews investigating this. It is important to check if new studies have been done that could be added to the existing studies of the original review and would change the findings.

What is individual participant data: In this review we used individual participant data (IPD). This means that we attempted to collect original research data for each person who had participated in the original studies by requesting these individuals' data from the researcher who performed the study. We used IPD as this offers a greater chance to detect changes between groups of participants and enable the investigation of additional hypotheses. We used the IPD to perform our calculations.

Review question: What is the effect of chronic NIV in people with COPD on blood gases (oxygen and carbon dioxide), exercise capacity, quality of life, lung function, respiratory muscle function, COPD exacerbations and admissions, and survival?

Study characteristics: The evidence is current to 21 December 2020. This review update identified 14 new studies in addition to those already in the review, so in total we included 21 studies. Ten of the studies looked at people in a stable phase (stable COPD) and four studies looked at people shortly after a COPD hospital admission (post‐exacerbation COPD). All studies included men and women. For our analyses, we used data from 778 people with stable COPD and 364 people with post‐exacerbation COPD.

Results: In all people with COPD who had raised levels of carbon dioxide, chronic NIV for three and 12 months improved blood gases. In stable COPD, chronic NIV also might have improved quality of life, and survival seemed to be better compared to people who were treated with standard care only. There was no relevant benefit of NIV on exercise capacity. People using chronic NIV after a COPD admission experienced less benefit; carbon dioxide levels decreased, the time to the next hospital admission might have been longer when treated with NIV but quality of life and survival were not affected by chronic NIV. 

Certainty of the results: Our confidence in the certainty (according to GRADE criteria) is good when looking at the blood gases. For the other outcomes, the certainty of the evidence is moderate to very low because the participants and the researcher were aware of the treatment the people received, and due to a wide range in the observed effect. This means that further research might change the results.

Authors' conclusions

Implications for practice

The results of this systematic review and individual participant data (IPD) analysis provide sufficient evidence for the routine application of chronic NIV in people with severe chronic obstructive pulmonary disease (COPD) who have chronic hypercapnic respiratory failure (CHRF), while its place remains uncertain for people who remain hypercapnic shortly after a severe COPD exacerbation. The benefits and goals of chronic non‐invasive ventilation (NIV) seem to differ between different COPD populations. For people initiated in a stable clinical state, significant and relevant improvements in diurnal hypercapnia can be achieved, and oxygenation, health‐related quality of life (HRQL) and all‐cause mortality might improve. In contrast, people who remain hypercapnic after a severe COPD exacerbation seem to benefit mostly in terms of a reduction in hypercapnia, and a prolonged time to readmission or death, but without relevant improvements in HRQL. Evidence also suggests that the effect of NIV on gas exchange is largest when the NIV is applied with sufficiently high pressures and for at least five hours of NIV use per night.

Currently, the application of chronic NIV in people with COPD varies greatly between different healthcare providers (Crimi 2016). Considering the results of the participant‐centred outcomes of this review, reluctance towards the initiation of NIV in stable hypercapnic COPD may no longer be warranted, whilst the application of chronic NIV post‐exacerbation can be discussed. Consistent with the guideline recommendations of the European Respiratory Society and the American Thoracic Society, the findings of this review support the application of this treatment in people with severe end‐stage COPD (Ergan 2019; Macrea 2020).

Implications for research

Given that the certainty of the evidence on the participant‐centred outcomes is moderate to very low, there is a need for new randomised controlled trials (RCTs) that investigate chronic NIV in people with stable COPD to strengthen the evidence. For future studies it should be taken into consideration that these people are severely limited by their disease and have poor prognosis and a limited number of other treatment options. Therefore, ethical concerns might arise when withholding NIV from a control group. In contrast, there remains a need for additional RCTs that investigate chronic NIV shortly after a severe exacerbation. Nevertheless, not all people eventually benefit from NIV, and the initiation of this treatment can be challenging and stressful. Therefore, future studies may focus on the identification of the person with COPD who is most likely to benefit from this treatment. Large observational studies may identify different COPD phenotypes based on individual and disease characteristics, and try to identify the underlying cause of chronic hypercapnic respiratory failure in the individual person. Furthermore, studies are needed to further elucidate what pathophysiological mechanisms are improved by NIV (and in what way), so that more goal‐directed therapy can be provided. The evidence regarding respiratory muscle functioning improvement is limited, as is the evidence regarding sleep quality improvement. Treatment targets are needed for the different phenotypes, and optimal criteria to start and continue NIV are to be identified, realising that these targets and criteria might differ substantially between different phenotypes. Lastly, the current aim of chronic NIV is to reduce the partial pressure of carbon dioxide (PaCO2) by improving nocturnal ventilation. Evidence that only this reduction is responsible for the improved outcomes is insufficient, and there is a need to unravel the mechanisms through which this intervention improves physiological and participant‐centred outcomes.

Summary of findings

Open in table viewer
Summary of findings 1. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD

Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD

Patient or population: stable COPD Setting: home treatment Intervention: NIV Comparison: Standard care

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with Standard care

Risk with NIV

Change in arterial partial pressure of oxygen
follow up: 3 months

The mean change in arterial partial pressure of oxygen was ‐0.07 kPa a

MD 0.27 kPa higher
(0.04 higher to 0.49 higher) b

‐

271
(9 RCTs)

⊕⊕⊕⊝
MODERATE c,d

Chronic NIV probably results in a slightly greater increase in arterial partial pressure of oxygen after 3 months.

Change in arterial partial pressure of oxygen
follow up: 12 months

The mean change in arterial partial pressure of oxygen was ‐0.06 kPa

MD 0.09 kPa higher
(0.23 lower to 0.42 higher) b

‐

171
(3 RCTs)

⊕⊕⊝⊝
LOW c,e

Chronic NIV may result in little to no difference in arterial partial pressure of oxygen after 12 months.

Change in arterial partial pressure of carbon dioxide
follow up: 3 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.17 kPa

MD 0.61 kPa lower
(0.77 lower to 0.45 lower) b

‐

475
(11 RCTs)

⊕⊕⊕⊕
HIGH c,f

Chronic NIV results in a reduction in the arterial partial pressure of carbon dioxide after 3 months.

Change in arterial partial pressure of carbon dioxide
follow up: 12 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.05 kPa

MD 0.42 kPa lower
(0.68 lower to 0.16 lower) b

‐

232
(4 RCTs)

⊕⊕⊕⊕
HIGH c

Chronic NIV results in a reduction in arterial partial pressure of carbon dioxide after 12 months.

Change in 6‐minute walking distance
follow up: 3 months

The mean change in 6‐minute walking distance was 16.0 metres

MD 15.5 metres higher
(0.8 lower to 31.7 higher) b

‐

330
(8 RCTs)

⊕⊕⊝⊝
LOW d,g

The evidence suggests that chronic NIV results in little to no difference on the 6‐minute walking distance test. Sensitivity analysis limited to high‐quality studies suggest a significant benefit; however, this treatment estimate does not reach the minimal important difference of 26m.

Change in 6‐minute walking distance
follow up: 12 months

The mean change in 6‐minute walking distance was ‐10.7 metres

MD 26.4 metres higher
(7.6 lower to 60.5 higher) b

‐

134
(3 RCTs)

⊕⊝⊝⊝
VERY LOW g,h

Chronic NIV may increase the 6‐minute walking distance but the evidence is very uncertain. This treatment estimate reaches the clinical relevance of 26m in this specific population.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 3 months i

‐

SMD 0.39 SD higher
(0.15 higher to 0.62 higher) b

‐

259
(5 RCTs)

⊕⊝⊝⊝
VERY LOW d,j

Chronic NIV may increase health‐related quality of life after 3 months but the evidence is very uncertain. Considering the standard deviation of 12 units on the SRI, the observed SMD does seem to reach the minimal important difference of approximately 5 points in this specific population.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 12 months i

‐

SMD 0.15 SD higher
(0.13 lower to 0.43 higher) b

‐

200
(4 RCTs)

⊕⊝⊝⊝
VERY LOW h,j

The evidence is very uncertain about the effect of chronic NIV on health‐related quality of life after 12 months. Considering a standard deviation of 12.2 units on the SRI, the observed SMD does not seem to reach the minimal important difference of approximately 5 points.

All‐cause mortality
follow up: median 30 months

Study population

HR 0.75
(0.58 to 0.97)
[All‐cause mortality] k

405
(3 RCTs)

⊕⊕⊕⊝
MODERATE c,d

Chronic NIV likely reduces all‐cause mortality (number needed to treat for an additional beneficial outcome 14, 95% CI 8 to 120).

655 per 1000

550 per 1000
(461 to 644)

*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; MD: Mean difference; SMD: Standardised mean difference; HR: Hazard 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

See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_417569665332318026.

a. For the risk with standard care, we calculated an absolute difference in the outcome between the baseline measurement and both follow‐up measurements (3 and 12 months)
b. The mean difference was obtained from the pooled individual participant data, and was adjusted for age and sex.
c. Not downgraded; although the participants and investigators were not blinded in the majority of the trials, this is unlikely to affect blood gases and survival.
d. Downgraded by one for serious imprecision; the CI includes the possibility of no (relevant) benefit and of a substantial benefit.
e. Downgraded by one due to serious imprecision: the CI includes the possibility of harm and of a large substantial benefit.
f. Inconsistency not downgraded; although substantial heterogeneity was observed (P = 0.001, I2 = 65%), this could be explained by the subgroup analysis.
g. Downgraded by one due to high risk of bias; the majority of the trials did not blind participants and investigators/personnel to the treatment allocation.
h. Downgraded by two due to serious imprecision: the CI includes the possibility of harm and of a large substantial benefit.
i. SRI: Severe Respiratory Insufficiency questionnaire; SGRQ: St. George's Respiratory Questionnaire
j. Downgraded by one due to high risk of bias; no trials blinded participants and investigators/personnel to the treatment allocation.
k. The hazard ratio was obtained from the pooled individual participant data, and was adjusted for age, sex and the baseline arterial pressure of carbon dioxide.

Open in table viewer
Summary of findings 2. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD

Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD

Patient or population: post‐exacerbation COPD Setting: home treatment Intervention: NIV Comparison: Standard care

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with Standard care

Risk with NIV

Change in arterial partial pressure of oxygen
follow up: 3 months

The mean change in arterial partial pressure of oxygen was 0.19 kPa a

MD 0.1 kPa lower
(0.65 lower to 0.45 higher) b

‐

234
(3 RCTs)

⊕⊕⊝⊝
LOW c,d

The evidence suggests that chronic NIV results in little to no difference in arterial partial pressure of oxygen after 3 months.

Change in arterial partial pressure of oxygen
follow up: 12 months

The mean change in arterial partial pressure of oxygen was 0.48 kPa

MD 0.27 kPa lower
(0.86 lower to 0.32 higher) b

‐

170
(3 RCTs)

⊕⊕⊝⊝
LOW c,d

Chronic NIV may result in a slight reduction in arterial partial pressure of oxygen after 12 months.

Change in arterial partial pressure of carbon dioxide
follow up: 3 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.59 kPa

MD 0.4 kPa lower
(0.7 lower to 0.09 lower) b

‐

241
(3 RCTs)

⊕⊕⊕⊝
MODERATE c,e

Chronic NIV likely reduces the arterial partial pressure of carbon dioxide after 3 months.

Change in arterial partial pressure of carbon dioxide
follow up: 12 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.61 kPa

MD 0.52 kPa lower
(0.87 lower to 0.18 lower) b

‐

175
(3 RCTs)

⊕⊕⊕⊕
HIGH c

Chronic NIV results in a reduction in the arterial partial pressure of carbon dioxide after 12 months.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 3 months f

‐

SMD 0.25 SD higher
(0.01 lower to 0.51 higher) b

‐

219
(2 RCTs)

⊕⊝⊝⊝
VERY LOW e,g

Chronic NIV may increase/have little to no effect on health‐related quality of life after 3 months but the evidence is very uncertain. Considering a standard deviation of 11.2 on the SRI, the SMD does not seem to reach the minimal important difference.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 12 months f

‐

SMD 0.25 SD higher
(0.06 lower to 0.55 higher) b

‐

164
(2 RCTs)

⊕⊝⊝⊝
VERY LOW e,g

Chronic NIV may increase/have little to no effect on health‐related quality of life after 12 months but the evidence is very uncertain. Considering a standard deviation of 11.9 on the SRI, the SMD does not seem to reach the minimal important difference.

Admission‐free survival
follow up: 1 year

Study population

HR 0.71
(0.54 to 0.94)
[readmission or death] h

317
(2 RCTs)

⊕⊕⊝⊝
LOW e,i,j

The evidence suggests chronic NIV improves admission‐free survival (number needed to treat for an additional beneficial outcome 12, 95% CI 7 to 61).

333 per 1000

458 per 1000
(356 to 552)

All‐cause mortality
follow up: median 21 months

Study population

HR 0.97
(0.74 to 1.28)
[All‐cause mortality] h

318
(2 RCTs)

⊕⊕⊝⊝
LOW c,d

The evidence suggests that chronic NIV does not improve all‐cause mortality.

642 per 1000

631 per 1000
(532 to 731)

*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; MD: Mean difference; SMD: Standardised mean difference; HR: Hazard 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

See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_417570213828006491.

a. For the risk with standard care, we calculated an absolute difference in the outcome between the baseline measurement and both follow‐up measurements (3 and 12 months).
b. The mean difference was obtained from the pooled individual participant data, and was adjusted for age and sex.
c. Not downgraded; although the participants and investigators were not blinded in the majority of the trials, this is unlikely to affect blood gases and survival.
d. Downgraded by two due to serious imprecision: the CI includes the possibility of significant harm and of a substantial benefit.
e. Downgraded by one due to serious imprecision: the CI includes the possibility of no (relevant) benefit and of a substantial benefit.
f. SRI: Severe Respiratory Insufficiency questionnaire; SGRQ: St. George's Respiratory Questionnaire
g. Downgraded by two due to high risk of bias; no trials blinded participants and investigators/personnel to the treatment allocation.
h. The hazard ratio was obtained from the pooled individual participant data, and was adjusted for age sex, baseline arterial partial pressure of carbon dioxide and number of COPD hospitalisations.
i. Downgraded by one; the majority of the participants were unblinded which could affect the likelyhood of attending the emergency department. Other domains low risk of bias
j. Inconsistency not downgraded; although substantial heterogeneity was observed (P = 0.05, I2 = 75%), this heterogeneity might be explained by differences in participant selection and the timing of NIV.

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is a common, preventable and treatable but incurable disease that is characterised by chronic airflow obstruction due to airway disease and lung parenchymal destruction (GOLD 2021). In 2016, COPD affected over 251 million people worldwide, and is currently the third leading cause of death, claiming approximately 3.2 million lives in 2015 (WHO 2020). In general, the treatment of COPD is directed at slowing down the progression of its natural history, symptom alleviation, improvement in daily functioning and health‐related quality of life (HRQL). Many patients experience exacerbations of their COPD, defined as an acute worsening of symptoms and requiring additional treatment (GOLD 2021). Exacerbations are a major problem as patients who experience frequent exacerbations, especially those requiring hospitalisation, experience accelerated deterioration of their condition, a faster decline in lung function, a high risk of rehospitalisation and a higher mortality risk (Donaldson 2002; Hurst 2010). Therefore, treatment options that reduce the risk of exacerbations are particularly attractive. As the disease progresses to an advanced stage, people experience severe symptoms of dyspnoea and are usually severely disabled in activities of daily living. In some people with advanced disease, chronic hypercapnic respiratory failure (CHRF) may follow with related symptoms, such as morning headache, and fragmented and un‐refreshing sleep resulting in chronic fatigue and daytime sleepiness (Windisch 2015). At this advanced stage, treatment options are urgently needed but unfortunately still limited (van Dijk 2020).

Description of the intervention

Non‐invasive ventilation (NIV) is the application of ventilatory support through a non‐invasive interface, usually a nasal or oronasal mask. NIV is currently applied as evidence‐based therapy for COPD patients admitted to hospital with acute hypercapnic respiratory failure (AHRF) due to an exacerbation. In this situation, NIV reduces the likelihood of endotracheal intubation, complications associated with treatment, duration of hospital stay and in‐hospital mortality (Osadnik 2017). During an exacerbation, NIV is often applied intermittently or continuously for a few days to overcome the acute life‐threatening respiratory failure, after which the person is weaned from ventilation and NIV can be stopped. Chronic NIV, however, entails the long‐term use of nocturnal NIV. There has been much discussion about the need for chronic NIV in stable COPD, mainly because the earlier trials published conflicting results (Rossi 2000). The previous version of this Cochrane Review, which investigated chronic NIV in people with stable COPD, did not find any significant benefits on gas exchange, exercise capacity, lung function or HRQL (Struik 2013). However, it did find that a subgroup of people with more severe hypercapnia was more likely to benefit, especially when treated with higher inspiratory pressures and when NIV was used for more than five hours per day (Struik 2014).

How the intervention might work

Several theories exist as to why chronic NIV may be beneficial in stable COPD. First, chronic nocturnal NIV ameliorates nocturnal hypoventilation. This effect is intended to persist during the daytime; improved lung mechanics, improved respiratory muscle function and resetting of the chemoresponsiveness might all contribute to a more advantageous breathing pattern, enabling the person to maintain the improved gas exchange during the day (Elliott 1991). Several studies have indicated that NIV probably stabilises forced expiratory volume in one second (FEV1) (Duiverman 2008; Köhnlein 2014); some observations indicate that NIV might also decrease residual volume (Diaz 1999). The observation regarding FEV1 is interesting as this was also observed without changes in static lung volume (Duiverman 2008), so might be caused by a different mechanism, possibly through the reduction of airway inflammation and oedema. It has also been postulated that NIV might rest chronically fatigued respiratory muscles. Periods of rest may lead to recovery of the inspiratory muscle function, thereby leading to an increased muscle strength and endurance capacity of the respiratory muscles during the daytime (Ambrosino 1990). Finally, NIV has been shown to improve sleep time and efficiency (Meecham Jones 1995), an effect that might be especially important for improving daily functioning and HRQL. 

Why it is important to do this review

Despite all these theoretical benefits, the added value of chronic NIV in people with stable COPD remains unclear and needs further investigation. This is the second update of a Cochrane Review on this topic, the first being published in 2002 and subsequently updated in 2013 (Struik 2013; Wijkstra 2002). Since the 2013 update, more studies have been reported, including the large 2014 randomised controlled trial (RCT) by Köhnlein 2014, which showed improved gas exchange and HRQL, and a survival benefit with NIV compared to standard care alone. Also, the role of chronic NIV in the treatment of people with severe stable COPD has become more established with the publication of the European Respiratory Society (ERS) and American Thoracic Society (ATS) guidelines on long‐term NIV in COPD (Ergan 2019; Macrea 2020). Additionally, since the last review update, important RCTs with conflicting results have been published that investigated the initiation of chronic NIV shortly after a severe COPD exacerbation with AHRF (Murphy 2017; Struik 2014). It has been suggested that these conflicting results are due to differences in participant selection and timing of NIV initiation. Despite these conflicting results, it is common practice in many countries to initiate chronic NIV in this specific COPD population. These new trials justify the current update of the 2013 review, which in addition to stable COPD also includes trials which investigated chronic NIV following an exacerbation of COPD, as we realise this has become part of regular clinical care by many healthcare providers.

As in the earlier versions of this review, we conducted a systematic review and a pooled analysis of individual participant data (IPD). We aimed to collect the original data of each participant from the trial authors for data checking, validation and re‐analysis. Compared to conventional meta‐analysis based on aggregated trial data, pooled analyses of IPD yields greater power and enables the investigation of additional hypotheses related to individual characteristics, for example within subgroups or treatment across trials. Although IPD reviews are regarded as the highest standard in systematic reviews, these reviews rely on the willingness of trial authors to share the IPD, which poses a threat to the review process. Potential risks might arise if trial authors do not agree to share the IPD as this could reduce the body of evidence. This could also introduce bias as the trials that do agree to share the IPD could differ from trials that did not agree, e.g. differ on the population or methodological approaches. To minimise this threat, we also conducted a meta‐analysis of aggregate trial data in addition to the analysis of IPD.

Objectives

To assess the effects of chronic non‐invasive ventilation at home via a facial mask in people with COPD, using a pooled analysis of IPD and meta‐analysis.

Methods

Criteria for considering studies for this review

Types of studies

We included RCTs in participants with COPD comparing chronic NIV at home plus standard therapy with standard therapy alone.

We differentiated between studies investigating chronic NIV in the following COPD populations:

  1. stable COPD: concerns studies of people with COPD initiated on chronic NIV in a stable clinical state;

  2. post‐exacerbation COPD: concerns studies of people with COPD initiated on chronic NIV following an exacerbation of COPD with acute respiratory failure.

Types of participants

We included people with COPD (a ratio of FEV1 to forced vital capacity (FVC) < 0.70), according to the most recent guideline of the Global initiative for chronic Obstructive Lung Disease at the time of writing (GOLD 2021).

Types of interventions

The intervention of interest was NIV, applied through a facial mask, prescribed for at least five hours during the night and for at least three consecutive weeks. Participants also received their usual standard COPD therapy, which comprised supplementary oxygen, optimised medication or pulmonary rehabilitation or exercise programs.

The intervention in the control group was standard therapy alone. The control group did not receive chronic NIV; sham treatment in the form of continuous positive airway pressure (without pressure support) was permitted.

Types of outcome measures

We analysed the following outcome measures with a short‐term follow‐up (up to three months) and a long‐term follow‐up (12 months).

Primary outcomes

Stable COPD

  1. Change in daytime arterial blood gas tensions (partial pressure of oxygen (PaO2), partial pressure of carbon dioxide (PaCO2))

  2. Change in exercise capacity, measured with the six‐minute walking distance (6MWD)

  3. Change in HRQL, measured with the Severe Respiratory Insufficiency questionnaire (SRI) (Windisch 2003) or the St George's Respiratory Questionnaire (SGRQ) (Jones 1991)

Post‐exacerbation COPD

  1. Change in daytime arterial blood gas tensions (PaO2 and PaCO2)

  2. Change in HRQL, measured with the SRI or the SGRQ

  3. Admission‐free survival, time to readmission or death of any cause

Secondary outcomes

Stable COPD

  1. Change in lung function (FEV1, FVC and residual volume/total lung capacity (RV/%TLC))

  2. Change in respiratory muscle function (maximal inspiratory pressure (PImax), maximal expiratory pressure (PEmax))

  3. Change in dyspnoea (measured with the Transitional Dyspnoea Index (TDI) (Mahler 1984), or the (modified) Medical Research Council (MRC) dyspnoea scale) (Bestall 1999)

  4. Change in sleep efficiency (percentage time asleep of total time in bed)

  5. Number of COPD exacerbations and hospital admissions, either self‐reported or documented

  6. All‐cause mortality, time to death of any cause

Post exacerbation COPD

  1. Change in lung function (FEV1 and FVC)

  2. Number of COPD exacerbations and hospital admissions, either self‐reported or documented

  3. All‐cause mortality, time to death of any cause

Search methods for identification of studies

Electronic searches

We identified trials from the Cochrane Airways Register of Trials, which is maintained by the Information Specialist at the Cochrane Airways editorial base. The Register contains trial reports identified through searches of the following bibliographic databases:

  • monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), through the Cochrane Register of Studies (CRS);

  • weekly searches of MEDLINE Ovid SP; 1946 to December 2020;

  • weekly searches of Embase Ovid SP; 1974 to December 2020;

  • monthly searches of PsycINFO Ovid SP; 1967 to December 2020;

  • monthly searches of CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature); 1937 to December 2020;

  • monthly searches of AMED EBSCO (Allied and Complementary Medicine); all years to December 2020;

  • handsearches of the proceedings of major respiratory conferences.

Full details of the search strategies used to maintain the Cochrane Airways Register of Trials are presented in Appendix 1.

The full search details of the previous versions of this review are presented in Appendix 2. The search terms used for this update can be found in Appendix 3. As some adjustments were made to the inclusion criteria, we conducted the search for this update with no date restrictions. We conducted this most recent search on 21 December 2020 without any language restrictions. 

The current search additionally included the following trial registries; the World Health Organization International Clinical Trials portal (www.who.int/ictrp/en/) and ClinicalTrials.gov (www.clinicaltrials.gov) (see Appendix 3).

Searching other resources

We searched the bibliographies of each RCT for additional papers that may have contained RCTs, and searched the reference lists of relevant clinical guidelines and systematic reviews for additional papers. We contacted authors of identified RCTs for other published and unpublished studies.

Data collection and analysis

Selection of studies

Two review authors independently assessed all identified abstracts; PJW and RSG for the 2002 version, FMS and PJW for the 2013 update, and TR and MD for the current update. When the review authors selected an abstract, we attempted to retrieve the full papers. The same two review authors read these in detail, and resolved any disagreements by consensus.

Data extraction and management

After inclusion of the studies, one author (TR) extracted the study characteristics of all included studies. We then contacted all trial authors by email to request the IPD of the following variables: treatment allocation, age, sex, body mass index, number of exacerbations and hospitalisations, arterial blood gases (PaO2, PaCO2, rate of oxygen flow), lung function (FEV1, FVC, FEV1/FVC), lung volumes (RV%/TLC), maximum static expiratory (PEmax) and inspiratory (PImax) mouth pressures, 6MWD, sleep studies, dyspnoea and HRQL measures (TDI, (modified) MRC dyspnoea scale, SRI, SGRQ), time‐to‐event and event status, NIV settings (inspiratory‐ and expiratory positive airway pressures (IPAP, EPAP)) and NIV compliance (mean hours of usage per night, as read from the ventilator counter). We requested missing data from the included primary studies from the authors. We checked supplied data against study publications, after which we copied raw data from all included studies to one main database.

For the aggregated data, we calculated the mean of the absolute change between the baseline measurement and the value at the follow‐up visits (three and 12 months) of the identified outcome variables (± standard deviations (SD)) from the IPD. If the IPD were not available, two review authors (TR and MD) independently extracted the aggregated data from the studies' primary references.

Assessment of risk of bias in included studies

Two review authors (TR and MD) assessed the risk of bias of each study independently, including the seven studies of the 2013 review. We resolved disagreements by consensus. We used criteria for assessment of risk of bias as provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We considered potential for bias using the following domains:

  • random sequence generation (selection bias);

  • allocation concealment (selection bias);

  • blinding of participants and personnel (performance bias);

  • blinding of outcome assessment (detection bias), separately for:

    • participant‐reported outcomes (HRQL, dyspnoea),

    • lung function, respiratory muscle testing, exercise tests,

    • arterial blood gases, hospitalisations, mortality;

  • incomplete outcome data (attrition bias);

  • selective outcome reporting (reporting bias);

  • other sources of bias.

We judged each domain as 'high', 'low' or 'unclear' risk of bias. If a trial reported insufficient details, the judgment was 'unclear'.

Measures of treatment effect

For continuous data, we calculated and presented an absolute difference in the outcome between the baseline measurement and both follow‐up measurements (three and 12 months). Of these differences, we obtained an overall treatment effect that is presented as the adjusted mean difference (AMD) or standardised mean difference (SMD) for the pooled IPD analysis and as mean difference (MD) or SMD with associated 95% CI for the aggregated data meta‐analysis. We measured the number of COPD exacerbations and hospitalisations as continuous data. We dichotomised these data to participants who experienced none versus at least one exacerbation and obtained the odds ratio (OR). Lastly, we analysed survival data as time‐to‐event. The survival data obtained from the IPD are presented as adjusted hazard ratio (AHR) and as hazard ratio (HR) when obtained from the aggregated data meta‐analysis. 

All outcome measures were presented with associated 95% confidence interval (CI). If available, estimates of the treatment effect were related to the minimal clinical important difference (MCID) for this specific population. Different proportions of participants and studies contributed to the different outcomes.

Unit of analysis issues

We expressed study outcomes in the same natural units across the trials. We analysed arterial blood gases as kPa, and converted data provided as mmHg to kPa by dividing the mmHg value by 7.5. If we obtained data on capillary blood gases, we only used the data for the PaCO2 analysis, as the PaO2 cannot be obtained reliably from capillary blood gases (Magnet 2017). We analysed lung function parameters in absolute values (litres). If we could only collect the percentage of predicted values, we calculated the absolute values using the regression equations of the European Community for Steel and Coal reference values (Quanjer 1993). We standardised the IPD of the SRI and SGRQ. Since the direction of effect differs between the SGRQ and the SRI, we inverted the SGRQ scores after which the data were pooled the data with the SRI. If a study measured both questionnaires, we preferred the data of the SRI over the SGRQ as the SRI was designed and validated specifically for this population (Windisch 2003; Windisch 2008). We could not pool the data on the dyspnoea measurements (TDI and the MRC dyspnoea scale), as the data of the MRC dyspnoea scale were not normally distributed.

In the case of cross‐over studies, we considered only the first study period (prior to the cross‐over). In case of multi‐arm studies, we only considered the control arm (receiving only standard treatment) and the arm receiving NIV in addition to standard treatment. We did not consider cluster‐randomised trials to be relevant to this review.

Dealing with missing data

We contacted the trial authors to obtain the IPD for the included studies. We only used data on continuous variables if both baseline data and data from at least one follow‐up visit were available. For the exacerbation and hospitalisation data, we did not use baseline data. For this outcome, we only used the data of the participants that completed the follow‐up visit. For time‐to‐event data, we used all available data.

If the authors did not respond to our request, we only used the studies for the meta‐analysis of the aggregated trial data and not for the pooled IPD analysis.

Assessment of heterogeneity

We assessed statistical heterogeneity between the studies in each meta‐analysis using the Chi² test and I2 statistic. For the Chi2 test, we considered a P value of < 0.1 significant. The I2 statistic ranges from 0 to 100%, with 0% representing no heterogeneity and 100% representing considerable heterogeneity. We categorised the I2 statistics according to the recommendations in the Cochrane Handbook (Higgins 2011):

  • 0% to 40%: heterogeneity might not be important;

  • 30% to 60%: may represent moderate heterogeneity;

  • 50% to 90%: may represent substantial heterogeneity;

  • 75% to 100%: shows considerable heterogeneity.

Assessment of reporting biases

We created and inspected a funnel plot of the meta‐analysis of the primary outcomes where we were able to meta‐analyse 10 or more studies for an outcome, and if the results of the outcome were homogeneous between studies. 

Data synthesis

For the analysis of the IPD, we used a one‐stage approach. We conducted all our models with a random effect on study level to account for the clustering of participants within the studies. We analysed the continuous IPD using linear mixed‐effect models to obtain the AMD. We analysed age, sex and treatment (NIV versus controls) as fixed factors. Normality of the outcome distributions was required, and we assessed this by visual inspection of normal probability plots and histograms. We used generalised estimating equations to obtain the OR from dichotomous IPD, with age, sex and treatment (NIV versus controls) as fixed factors. We ran separate models for both follow‐up times (3 and 12 months).

We assessed time‐to‐event data using a mixed‐effect Cox regression analysis to obtain the AHR.  For the cohort of people with stable COPD, we entered age, sex, baseline PaCO2 and treatment (NIV versus controls) as fixed factors. For the post‐exacerbation cohort, we entered age, sex, number of hospital admissions (less than three versus three or more admissions) in the year prior to randomisation, baseline PaCO2 and treatment (NIV versus controls). We conducted all analyses in RStudio 2020.

Additionally, conducted a meta‐analysis on the (unadjusted) aggregated data of all included studies using Review Manager Web (RevMan Web 2020). In addition to the aggregated IPD, we included the extracted data from the studies that did not provide any IPD. We calculated the mean differences (MD) and associated 95% CI, and combined them in a random‐effects model to obtain an overall effect estimate.

Subgroup analysis and investigation of heterogeneity

If we identified any substantial heterogeneity between the studies (I2 > 50%), we considered subgroup analyses if sufficient numbers of studies could be included in the analysis (≥ 10 studies). A priori, the following sources of heterogeneity were postulated.

  • Participants with more severe hypercapnia (PaCO2 > 7.3 kPa) might benefit more from NIV; the division of the subgroups was made based on guidelines from a consensus report (Chest 1999).

  • Participants receiving higher ventilatory support (IPAP ≥ 18 cmH2O) might have a greater benefit from NIV; the division of the subgroups was made based on the first trial that showed an effect of NIV on gas exchange (Meecham Jones 1995).

  • The benefits of NIV might be greater among participants who actually adhere to the NIV for at least five hours per night; division of the subgroups was made based on the inclusion criteria for this review that the NIV should be prescribed for at least five hours per night.

Sensitivity analysis

We conducted sensitivity analyses based on methodological quality by repeating the IPD analysis among only those studies that we judged to be of 'high quality.' Studies defined as 'high‐quality' studies were studies that we judged to be at low risk of bias for the following domains: random sequence generation, allocation concealment, incomplete outcome data. We limited the sensitivity analysis to the primary outcomes.

Summary of findings and assessment of the certainty of the evidence

We created the summary of findings tables using GRADEpro GDT software, and presented separate tables for participants with stable COPD and for post‐exacerbation COPD. We included the primary outcomes and the (admission‐free) survival outcomes of the pooled IPD analysis in the tables.

Two review authors (TR and MD) assessed the certainty of the evidence using the GRADE working group criteria (GRADE 2013), considering the following domains: risk of bias, inconsistency, indirectness, imprecision, publication bias. We described our decisions to downgrade the certainty of the evidence.

Results

Description of studies

See: Characteristics of included studies.

Results of the search

The search conducted on 21 December 2020 identified 2291 records. After deduplication 1709 records remained, of which we considered 1653 records ineligible after title and abstract screening. Of the 56 records that remained, we retrieved 22 full‐text papers. The remaining records were conference abstracts (n = 27) or records from trial registries (n = 7). Of the 56 records, we excluded nine records of the current search for the following reasons.

1) Ineligible study design:

  • study not randomised (Ali 2012; Clini 1998; Kamei 1999);

  • duration of NIV was less than four hours per night prescribed use and prescribed at daytime (Diaz 1999);

  • use of NIV was less than three weeks (Lin 1996);

  • participants were randomised to either terminate or continue NIV after an exacerbation (Funk 2011).

2) Ineligible study population:

  • mixed population of participants with obstructive lung diseases (Chiang 2004, two records).

3) Ineligible intervention:

  • intervention group received inspiratory positive pressure breathing during daytime (Nicolini 2014).

Three studies (one full text paper and two conference abstracts) are awaiting classification as the authors did not respond to clarify important issues (Baturova 2013; Guan 2018; Xiang 2007). We identified six studies (seven records) that were ongoing at the time of this review (Ankjaergaard; Beth Israel Deaconess Medical Center; Gonzales; Lamia; Pepin; Wijkstra). Additionally, we identified five records of studies that were included in the previous version of this review.

This update identified 10 studies (18 records) investigating chronic NIV in stable COPD (Bhatt 2013; Cui 2019; Duiverman 2008; Eman Shebl 2015; Garrod 2000; Köhnlein 2014; Ma 2019; Martin‐Marquez 2014; Schneeberger 2017; Zhou 2017). Together with the seven studies from the original review, we included 17 studies on stable COPD in this review update. Additionally, we identified four studies (14 records) on chronic NIV following an acute exacerbation of COPD (Cheung 2010; De Backer 2011; Murphy 2017; Struik 2014). In total, we included 21 studies in this review update. 

Figure 1 presents the PRISMA‐IPD flowchart. 

Figure 1
PRISMA flow diagram

PRISMA flow diagram

Included studies

The Characteristics of included studies table shows full details of the included studies.

Stable COPD

Seventeen studies met the inclusion criteria for the review (Bhatt 2013; Casanova 2000; Clini 2002; Cui 2019; Duiverman 2008; Eman Shebl 2015; Garrod 2000; Gay 1996; Köhnlein 2014; Ma 2019; Martin‐Marquez 2014; McEvoy 2009; Meecham Jones 1995; Schneeberger 2017; Sin 2007; Strumpf 1991; Zhou 2017). Two of these studies used a cross‐over in design (Meecham Jones 1995; Strumpf 1991), the remaining studies adopted a parallel design. Fourteen studies measured outcomes after three months and are classified as 'short‐term', while five studies also measured outcomes after 12 months and are defined as 'long‐term' (Clini 2002; Duiverman 2008; Köhnlein 2014; Ma 2019; McEvoy 2009). We obtained the IPD for the studies from the trial authors, except for four studies (Cui 2019; Eman Shebl 2015; Garrod 2000; Ma 2019). Of these four studies, we could only include the aggregated data of Garrod 2000 for the meta‐analysis. In one study, the treatment arm received either NIV or CPAP, and the trial did not report the number of participants receiving each treatment (Cui 2019). One study only measured the outcomes after six months of NIV, so we did not use the data (Eman Shebl 2015). In the other trial, the study report did not present the data (Ma 2019).

Participants

The 17 included studies were based in different countries: Australia (McEvoy 2009), Canada (Sin 2007), China (Cui 2019; Ma 2019; Zhou 2017), Egypt (Eman Shebl 2015), Germany (Köhnlein 2014; Schneeberger 2017), the Netherlands (Duiverman 2008), Spain (Casanova 2000; Martin‐Marquez 2014), Italy (Clini 2002), the UK (Garrod 2000; Meecham Jones 1995) and the USA (Bhatt 2013; Gay 1996; Strumpf 1991). All studies included people with COPD in a clinically stable condition and included both males and females.

In total, we obtained the IPD of 778 participants with stable COPD, representing 68% of the participants included in the studies. The mean age of all participants was 66 ± 8 years. All studies included males and females; 71% were males. Mean FEV1 was 0.72 ± 0.27 L (28.2% predicted) and mean PaCO2 was 7.3 ± 1.1 kPa.

Intervention and comparator

Eleven studies compared chronic NIV with standard treatment (Bhatt 2013; Casanova 2000; Clini 2002; Cui 2019; Eman Shebl 2015; Köhnlein 2014; Ma 2019; McEvoy 2009; Meecham Jones 1995; Strumpf 1991; Zhou 2017), and two studies compared chronic NIV with sham treatment in the form of continuous positive airway pressure (CPAP) at 2 cmH2O and 4 cmH2O (Gay 1996; Sin 2007). In four studies, NIV was applied in addition to a pulmonary rehabilitation or exercise training program and compared to a control group receiving only the rehabilitation or training program (Duiverman 2008; Garrod 2000; Martin‐Marquez 2014; Schneeberger 2017).

Participants allocated to the NIV were initiated in‐hospital (Casanova 2000; Clini 2002; Cui 2019; Duiverman 2008; Eman Shebl 2015; Gay 1996; Köhnlein 2014; McEvoy 2009; Meecham Jones 1995; Schneeberger 2017; Sin 2007), out‐patient based (Strumpf 1991; Zhou 2017) or at home (Bhatt 2013). The remaining three studies did not report the site of initiation (Garrod 2000; Ma 2019; Martin‐Marquez 2014). Bi‐level pressure support ventilation was applied with a back‐up rate in eight studies (Clini 2002; Cui 2019; Duiverman 2008; Gay 1996; Köhnlein 2014; Martin‐Marquez 2014; Strumpf 1991; Zhou 2017), with no back‐up rate (Casanova 2000; Garrod 2000; Meecham Jones 1995), or information on a back‐up rate was not specified (Bhatt 2013; Eman Shebl 2015; McEvoy 2009; Schneeberger 2017; Sin 2007). The ventilation mode was unspecified in one study (Ma 2019). NIV was delivered by nasal masks in six studies (Casanova 2000; Clini 2002; Garrod 2000; Gay 1996; Meecham Jones 1995; Strumpf 1991), and by either nasal or oronasal masks in seven studies (Bhatt 2013; Duiverman 2008; Köhnlein 2014; Martin‐Marquez 2014; McEvoy 2009; Sin 2007; Zhou 2017). In four studies the interface was not specified (Cui 2019; Eman Shebl 2015; Ma 2019; Schneeberger 2017).

At baseline, the mean IPAP and EPAP were 16.5 ± 4.7 cmH2O and 4.7 ± 4.7 cmH2O, respectively. The corresponding values after three months were 17.9 ± 5.2 cmH2O and 4.6 ± 1.4 cmH2O, respectively. After 12 months, mean IPAP was 16.8 ± 5.4 cmH2O and mean EPAP was 4.9 ± 1.5 cmH2O. Chronic NIV was used for a mean of 6.2 ± 2.8 hours after three months and 5.5 ± 3.2 hours after 12 months.

Post‐exacerbation COPD

Four studies met the inclusion criteria for this review (Cheung 2010; De Backer 2011; Murphy 2017; Struik 2014). All studies used a parallel design and measured outcomes after both three and 12 months. The authors of one study did not respond to our request for the IPD (De Backer 2011), so we did not include this study in the pooled IPD analysis. Since De Backer 2011 did not present results in their paper after three or 12 months, we were also unable to use the data from this study in the meta‐analysis. The remaining three studies provided the IPD after both three and 12 months. 

Participants

The four included studies were based in the following countries: Belgium (De Backer 2011), China (Cheung 2010), the Netherlands (Struik 2014), and the UK (Murphy 2017). The studies included people with COPD who were hospitalised due to an exacerbation (De Backer 2011) or who were hospitalised for an episode of ARF due to COPD (Cheung 2010; Murphy 2017; Struik 2014). All studies included both males and females.

We obtained IPD for 364 participants initiated on chronic NIV following an acute exacerbation, representing 96% of the participants included in the studies. The mean age was 65 ± 9 years, and 50% were females. In the year prior to randomisation, the median number of COPD exacerbations was two (interquartile range (IQR) one to three). The mean FEV1 was 0.62 ± 0.23 L and mean (inclusion) PaCO2 was 7.9 ± 1.1 kPa. 

Intervention and comparator

In one study (Cheung 2010), chronic NIV was compared to sham treatment with CPAP at 5 cmH20, the remaining three studies compared NIV to standard treatment. Participants were initiated on chronic NIV if they had persistent hypercapnia on day five to 12 of the admission (De Backer 2011), or after they were weaned from acute NIV > 48 hours (Cheung 2010; Struik 2014) or two to four weeks (Murphy 2017). All participants were initiated in‐hospital, and on bi‐level pressure support ventilation with a backup rate in three studies (Cheung 2010; Murphy 2017; Struik 2014). The other study did not report the mode of ventilation (De Backer 2011). NIV was delivered through oronasal masks (De Backer 2011), through either nasal or oronasal masks (Murphy 2017; Struik 2014), or was not specified (Cheung 2010).

The mean IPAP and EPAP at initiation were 20.0 ± 4.4 cmH2O and 4.8 ± 0.9 cmH2O, respectively. After three months, the mean IPAP was 20.8 ± 4.4 cmH2O and the mean EPAP was 5.0 ± 1.0 cmH2O. Mean IPAP after 12 months was 21.9 ± 4.3 cmH2O, and the corresponding mean EPAP was 5.0 ± 1.1 cmH2O. Chronic NIV was used for a mean of 6.5 ± 2.4 hours after three months and for 6.8 ± 2.7 hours after 12 months.

Funding of trial (all trials)

Thirteen studies were funded by their National Respiratory Society/Foundation (Casanova 2000; Cheung 2010; Clini 2002; Duiverman 2008; Gay 1996; Köhnlein 2014; Martin‐Marquez 2014; McEvoy 2009; Meecham Jones 1995; Murphy 2017; Sin 2007; Struik 2014; Zhou 2017), of which six were also partly funded by an industrial company (Cheung 2010; Clini 2002; Köhnlein 2014; McEvoy 2009; Murphy 2017; Struik 2014). Four studies were funded by an industrial company alone (Bhatt 2013; Garrod 2000; Schneeberger 2017; Strumpf 1991), and four studies did not report any funding (Cui 2019; De Backer 2011; Eman Shebl 2015; Ma 2019).

Excluded studies

The Characteristics of excluded studies table provides full details of the 10 excluded studies.

Risk of bias in included studies

The Characteristics of included studies table provides details of risk of bias in the included studies. Figure 2 shows a summary of our risk of bias judgements across studies. 

Figure 2
Risk of bias summary: review authors' judgements about each methodological quality item for each included studyNote: blank squares indicate that the outcome domain was not measured in the study.

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

Note: blank squares indicate that the outcome domain was not measured in the study.

Allocation

Random sequence generation (selection bias)

All studies were described as randomised. Thirteen studies described the method of the sequence generation adequately, so we judged these to be at low risk of bias. In the other eight studies, insufficient data were available, so we judged these to be at unclear risk of bias (Clini 2002; Cui 2019; De Backer 2011; Garrod 2000; Gay 1996; Ma 2019; Schneeberger 2017; Sin 2007).

Allocation concealment (selection bias)

We also deemed allocation concealment to be adequate in the majority of the studies; 10 studies described a centralised randomisation office or independent person. In six cases, sequentially numbered, opaque, sealed envelopes were used. Five studies provided insufficient data on the allocation concealment, so we judged them to be at unclear risk of bias in this domain (Cui 2019; De Backer 2011; Eman Shebl 2015; Ma 2019; Martin‐Marquez 2014).

Blinding

Blinding of participants and personnel (performance bias)

Given the nature of the intervention it is difficult to blind participants, but three studies attempted to do so and used a sham‐device (Cheung 2010; Gay 1996; Sin 2007). One study also blinded personnel to treatment allocation, making this the only study that we could classify as having a low risk of bias in this domain (Sin 2007). In one other study it was not clear whether personnel were blinded (Gay 1996). We judged all other studies to be at high risk of bias with respect to blinding.

Blinding of outcome assessment (detection bias)
Participant‐reported outcomes

In the majority of the studies, participants were not blinded for treatment (e.g. did not receive sham NIV), so we judged that participant‐reported outcomes (HRQL and dyspnoea measurements) were likely to be affected by detection bias. Therefore, we judged 15 studies to be at high risk of bias. The remaining studies did not measure any participant‐reported outcomes of relevance to this review (Cheung 2010; De Backer 2011; Garrod 2000; Gay 1996; Ma 2019; Sin 2007).

Lung function, respiratory muscle and exercise capacity tests

Most studies did not provide sufficient data on these outcomes, so we judged them to be at unclear risk. Eight studies described the outcome assessors as being unaware of the treatment assignment, so we judged these to be at low risk (Casanova 2000; Clini 2002; Köhnlein 2014; Murphy 2017; Sin 2007; Struik 2014; Strumpf 1991; Zhou 2017). The Cheung 2010 study did not measure these outcomes.

Arterial blood gases, hospitalizations and mortality

Except for one study, we judged all studies to be at low risk of bias as these are either objective measures (e.g. blood gases) or are unlikely to be affected by the outcome assessor (e.g. hospitalisations/mortality). We judged the study of Cheung 2010 as high risk as for the primary outcome, as outcome assessors had to rate whether there was an event in the NIV group. One study did not measure these outcomes (Schneeberger 2017).

Incomplete outcome data

Given the severity of the illness and nature of the intervention, dropouts are to be expected due to the inability of the participants to attend control visits, or because of intolerance to the intervention. Therefore, most studies experienced dropouts, and the majority described the reason for these dropouts adequately. Five studies did not perform intention‐to‐treat (ITT) analyses. We classified these studies as low risk as dropouts due to intolerance were small, baseline characteristics of dropouts did not differ from completers, and the studies presented a clear description of the reason for participants dropping out (Duiverman 2008; Garrod 2000; Martin‐Marquez 2014; Meecham Jones 1995; Sin 2007). Eight studies reported ITT analyses (or stated that “Inclusion of the patients who did not complete the study (intent to treat) did not affect any of the outcomes” Casanova 2000), so we considered these to be at low risk of bias (Casanova 2000; Clini 2002; Cui 2019; Köhnlein 2014; McEvoy 2009; Murphy 2017; Struik 2014; Zhou 2017). In one of the cross‐over studies, only seven out of 19 randomised participants completed both arms (Strumpf 1991). Another study reported that four out of seven participants randomised to NIV completed the study, as opposed to all six in the sham group (Gay 1996). One study performed an ITT analysis on the primary outcome, but participants were withdrawn from the study after the primary outcome was reached (Cheung 2010). We classified these three studies as high risk of attrition bias. We judged five studies to be at unclear risk of bias, as these studies did not provide sufficient data regarding the reasons for participants dropping out (Bhatt 2013; De Backer 2011; Eman Shebl 2015; Ma 2019; Schneeberger 2017).

Selective reporting

Most of the earlier studies were not registered, nor were the original protocols retrieved, so we were unable to check if the prespecified outcomes were all reported in the articles. In this domain the risk of bias was unclear. However, all outcomes listed in the methods sections of the studies were reported in the results sections. One study was registered retrospectively, and we judged this as unclear (McEvoy 2009). One study was not registered and the paper only reported the outcome data of one follow‐up visit, whilst measurements were taken during four follow‐up visits (De Backer 2011). We therefore judged this study to be at high risk. We judged two other studies to be high risk after comparing the paper with the trial registry, due to a change in the primary outcome (Cheung 2010; Zhou 2017). We judged eight studies to be at low risk, as these studies were registered and the data were reported in the paper, or (in case that the data were not reported), the authors provided the IPD (Bhatt 2013; Duiverman 2008; Garrod 2000; Köhnlein 2014; Martin‐Marquez 2014; Murphy 2017; Schneeberger 2017; Struik 2014).

Other potential sources of bias

For 13 of the studies, we did not find any other source of bias. In one study we found a difference in the number of (telephone) contacts with the caregivers between both study groups at the start of the study, but as it was unclear for how long this activity continued (and thus the likelihood of affecting the outcomes after three months), we judged this as an unclear risk of bias (Strumpf 1991). We also observed this in two other studies (Bhatt 2013; Garrod 2000), and the NIV group in one of these studies received NIV during the run‐in period, which could potentially affect the outcomes (Garrod 2000). In two studies, there were concerns regarding the sample size; in one study financial restraint prevented the inclusion of more than 50 participants, and one study did not present a power calculation and the sample sizes of both groups were unequal (Cheung 2010; De Backer 2011). In one study, participants with both hypoxemic and hypercapnic respiratory failure were included and treated with either CPAP or NIV in the treatment arm (Cui 2019). The study did not provide data on the number of participants treated with either treatment, so we judged this study to have a high risk of bias. Lastly, two studies were only available as meeting abstracts (Ma 2019; Schneeberger 2017). Therefore, we judged all seven of these studies to be at high risk of other bias (Bhatt 2013; Cheung 2010; Cui 2019; De Backer 2011; Garrod 2000; Ma 2019; Schneeberger 2017).

Effects of interventions

See: Summary of findings 1 Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD; Summary of findings 2 Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD

There is a difference between the number of participants included in the studies in total and the number included in the analysis, as most of the data were only available from participants who completed that particular study period and dropouts were common in most studies. For the short‐term follow‐up, participants dropped out due to mask intolerance, intercurrent infections or participants no longer meeting the inclusion criteria after a stabilisation period. In the studies also investigating the long‐term effects, dropping out was often due to progression of the disease or death, and reluctance of participants to return to hospital for follow‐up measurements. Finally, not all parameters were measured in all participants and, therefore, there was a difference in number of participants per outcome.

Stable COPD — primary outcomes

In Table 1, the absolute changes in the outcomes are shown separately for the NIV and the control group. In Figure 3, the modelled means obtained from the linear mixed effect models are presented for the primary outcomes. 

Open in table viewer
Table 1. Absolute changes in the outcomes after three and 12 months in stable COPD, separate for the NIV and the control group

 

3 months

12 months

Outcome

n

Change in NIV group

n

Change in control group

n

Change in NIV group

n

Change in control group

PaO2 (kPa)

131

0.17 ± 0.95

140

‐0.07 ± 0.99

88

‐0.03 ± 1.07

83

‐0.06 ± 1.12

PaCO2 (kPa)

238

‐0.81 ± 1.01

237

‐0.17 ± 0.93

123

‐0.46 ± 1.10

109

‐0.05 ± 0.98

6MWD (m)

167

32.2 ± 73.9

163

16.0 ± 75.6

67

15.0 ± 105.9

67

‐10.7 ± 94.8

SRI, summary score

108

7.5 ± 12.5

109

2.2 ± 11.0

42

6.1 ± 14.2

42

0.5 ± 9.3

SGRQ, total score

20

‐3.2 ± 9.3

20

‐0.2 ± 8.9

62

‐0.2 ± 14.0

53

‐0.9 ± 15.3

FEV1 (l)

194

0.01 ± 0.15

189

0.01 ± 0.14

125

‐0.01 ± 0.17

111

‐0.01 ± 0.15

FVC (l)

196

0.05 ± 0.40

188

0.05 ± 0.38

125

‐0.07 ± 0.43

110

‐0.05 ± 0.46

RV/TLC (%)

93

‐0.6 ± 6.8

91

‐0.05 ± 6.8

101

0.6 ± 8.0

94

0.07 ± 9.0

PEmax (cmH2O)

60

8.7 ± 24.6

63

3.0 ± 23.6

16

17.4 ± 46.2

22

13.9 ± 36.1

PImax (cmH2O)

76

9.2 ± 16.7

76

2.2 ± 12.5

45

8.0 ± 16.9

47

2.0 ± 16.5

TDI

84

1.2 ± 2.7

94

0.5 ± 2.7

0

—

0

—

(m)MRC

68

‐0.7 ± 1.0

87

‐0.4 ± 0.9

49

‐0.6 ± 1.2

54

‐0.1 ± 1.0

Sleep efficiency

13

7.0 ± 23.2

11

12.6 ± 16.8

0

—

0

—

COPD exacerbationsa

24

0.5 (0 to 1.8)

32

1.0 (0 to 2.0)

18

3.0 (2.8 to 4.0)

27

3.0 (2.0 to 5.0)

COPD hospitalisationsa

137

0 (0 to 0.5)

131

0 (0 to 1.0)

71

0 (0 to 2.0)

67

0 (0 to 2.0)

Data are based on the studies that provided the IPD, and are presented as mean ± standard deviation or median (interquartile range). 
a absolute number of exacerbations or hospitalisations.

6MWD: 6‐minute walking distance; CI: confidence interval; COPD: Chronic Obstructive Pulmonary Disease; FEV1: forced expiratory volume in 1 second; FVC: forced vital capacity; (m)MRC: (modified) Medical Research Council dyspnoea scale, a decrease indicates an improvement in dyspnoea; NIV: noninvasive ventilation; PaCO2: arterial carbon dioxide tension; PaO2: arterial oxygen tension; PEmax: peak expiratory pressure; PImax: peak inspiratory pressure; RV/TLC: residual volume/total lung capacity; SGRQ: St. George's Respiratory Questionnaire, a decrease indicates an improved health‐related quality of life; SRI: Severe Respiratory Insufficiency questionnaire, an increase indicates an improved health‐related quality of life; TDI: transitional dyspnoea index, a positive value indicates an improvement in dyspnoea.

Figure 3
The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in stable COPDBlack triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in stable COPD

Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

Daytime arterial blood gas tensions
Three months' follow‐up

Eleven studies contributed IPD for this outcome (Bhatt 2013; Casanova 2000; Clini 2002; Duiverman 2008; Gay 1996; Köhnlein 2014; Martin‐Marquez 2014; Meecham Jones 1995; Sin 2007; Strumpf 1991; Zhou 2017). In total, blood gases of 475 participants were analysed, of which 271 were arterial blood gases and contributed to the PaO2 outcome. Participants treated with NIV showed an increase in PaO2 whereas participants allocated to the control group showed a decrease (adjusted mean difference (AMD) 0.27 kPa, 95% CI 0.04 to 0.49; Table 1). A reduction in the PaCO2 was observed in participants in both groups, but the decrease was ‐0.61 kPa larger in participants allocated to NIV (95% CI ‐0.77 to ‐0.45; Table 1).

The meta‐analyses of the blood gas outcomes included data from one additional study (Garrod 2000), and showed a larger treatment effect on the PaO2 (MD 0.31 kPa, 95% CI 0.14 to 0.48; P = 0.44, I2 = 0%; 10 studies, 308 participants; moderate‐certainty evidence; Analysis 1.1). The funnel plot was asymmetrical to the right side and might indicate that a publication bias was present (Figure 4). A smaller treatment effect with substantial statistical heterogeneity between the studies was found for the PaCO2 (MD ‐0.48 kPa, 95% CI ‐0.72 to ‐0.25; P = 0.001, I2 = 65%; 12 studies, 512 participants; high‐certainty evidence; Analysis 1.2). We conducted subgroup analysis to explore the heterogeneity. These subgroup analyses seem to explain the heterogeneity and are presented later.

Figure 4
Funnel plot for the PaO2 outcome after 3 months, in participants with stable COPD

Funnel plot for the PaO2 outcome after 3 months, in participants with stable COPD

12 months' follow‐up

Four studies contributed blood gases of 232 participants (Clini 2002; Duiverman 2008; Köhnlein 2014; McEvoy 2009). Of these blood gases, 171 were arterial and were used for the PaO2 analysis. Both study groups showed a minor decrease in PaO2 (AMD 0.09 kPa, 95% CI ‐0.23 to 0.42; Table 1). In contrast, a reduction in PaCO2 was only present in participants allocated to NIV, whereas participants in the control group did not change (AMD ‐0.42 kPa, 95% CI ‐0.68 to ‐0.16; Table 1).

The meta‐analyses did not include additional data and yielded results comparable to the pooled IPD analysis for the PaO2 (MD 0.08 kPa, 95% CI ‐0.37 to 0.53; P = 0.16, I2 = 46%; 3 studies, 171 participants; low‐certainty evidence; Analysis 1.1) and a slightly smaller effect of NIV on the PaCO2 (MD ‐0.35 kPa, 95% CI ‐0.68 to ‐0.02; P = 0.14, I2 = 45%; 4 studies, 232 participants; high‐certainty evidence; Analysis 1.2). Both outcomes showed moderate statistical heterogeneity between the studies, but as only four studies measured this outcome we could not conduct subgroup analysis.

Exercise capacity
Three months' follow‐up

Exercise capacity was measured using the 6MWD in eight studies (Bhatt 2013; Duiverman 2008; Gay 1996; Köhnlein 2014; Martin‐Marquez 2014; Meecham Jones 1995; Sin 2007; Zhou 2017), and 330 participants contributed to the IPD. The MCID for the 6MWD in people with severe COPD is 26 m (Puhan 2011). Participants in both groups improved on the 6MWD on average, but there was no relevant effect of the NIV (AMD 15.5 m, 95% CI ‐0.8 to 31.7; Table 1). The meta‐analysis did not include additional data and yielded a smaller effect compared to the pooled IPD analysis (MD 7.85 m, 95% CI ‐5.52 to 21.23; P = 0.39, I2 = 6%; 8 studies, 330 participants; low‐certainty evidence; Analysis 1.3).

12 months' follow‐up

The 6MWD was measured in 134 participants in three studies (Clini 2002; Duiverman 2008; Köhnlein 2014). An improvement in the 6MWD was seen in the NIV group, whereas the control group deteriorated (AMD 26.4 m, 95% CI ‐7.6 to 60.5; Table 1). The meta‐analysis of data of the same three studies that provided IPD yielded similar results (MD 23.50 m, 95% CI ‐3.03 to 50.03; P = 0.37, I2 = 0%; 3 studies, 134 participants; very low‐certainty evidence; Analysis 1.3).

Health‐related quality of life
Three months' follow‐up

Three studies measured HRQL using the SRI in 217 participants (Duiverman 2008; Köhnlein 2014; Zhou 2017). The SRI is reported on a 0 to 100 scale, with a higher score indicating better HRQL. An improvement of approximately five units has been established as the MCID (Raveling 2020). Three studies provided SGRQ data for an additional 42 participants (Köhnlein 2014; Meecham Jones 1995; Schneeberger 2017). Scores for the SGRQ are also reported on a 0 to 100 scale, but with a lower score indicating better HRQL. The MCID of the SGRQ in people with severe COPD is approximately seven units (Welling 2015). Participants treated with NIV showed a larger improvement both on the summary score of the SRI and on the total score of the SGRQ, compared to participants allocated to the control group (Table 1). After pooling of the SRI and SGRQ data, there was a benefit of NIV on HRQL (AMD 0.39, 95% CI 0.15 to 0.62). Given a SD of 12.0 and of 10.2 units for the SRI and SGRQ respectively, this would result in an improvement of 4.7 units on the SRI and of 4.0 units on the SGRQ.

Meta‐analyses based on data of the same five studies that provided IPD yielded comparable results to the pooled IPD analysis (SMD 0.42, 95% CI 0.17 to 0.67; P = 0.43, I2 = 0%; 5 studies, 257 participants; very low‐certainty evidence; Analysis 1.4).

12 months' follow‐up

Two studies contributed to the SRI data of 84 participants (Duiverman 2008; Köhnlein 2014), and three studies to the SGRQ data of 116 participants (Köhnlein 2014; Clini 2002; McEvoy 2009). An improvement on the summary score of the SRI was observed in participants treated with NIV, but participants allocated to the control group showed no change. This was not observed in the studies that used the SGRQ, as both groups showed a small improvement (Table 1). After pooling of the SRI and SGRQ data, there was no effect of chronic NIV (AMD 0.15, 95% CI ‐0.13 to 0.43). Given a SD of 12.2 and of 14.6 units for the SRI and SGRQ respectively, this would result in an improvement of 1.8 units on the SRI and of 2.2 units on the SGRQ.

There were no additional data available for the aggregated meta‐analyses, and this analysis yielded comparable results to the pooled IPD analysis (SMD 0.18, 95% CI ‐0.24 to 0.59; P = 0.12, I2 = 49%; 4 studies; 199 participants; very low‐certainty evidence; Analysis 1.4).

Stable COPD — secondary outcomes

Lung function

In one study (Köhnlein 2014), only the percentage of predicted values were recorded. Absolute spirometry values were calculated using regression equations (Quanjer 1993).

Three months' follow‐up

The FEV1 and FVC were measured in 10 studies (Bhatt 2013; Casanova 2000; Duiverman 2008; Gay 1996; Köhnlein 2014; Martin‐Marquez 2014; Meecham Jones 1995; Sin 2007; Strumpf 1991; Zhou 2017). Data were available for 383 participants for the FEV1 and of 384 participants for the FVC. On average, the FEV1 of participants in both groups remained unchanged (AMD 0.01 L, 95% CI ‐0.02 to 0.04; Table 1). Similar results were observed for the FVC as participants in both groups showed a small but equal increase (AMD ‐0.01 L, 95% CI ‐0.09 to 0.07; Table 1). Aggregated data were available for one additional study (Garrod 2000), but this did not result in any new findings for either the FEV1 (MD ‐0.00 L, 95% CI ‐0.04 to 0.03; P = 0.18, I2 = 28%; 11 studies, 420 participants; Analysis 1.5) or the FVC (MD 0.01 L, 95% CI ‐0.06 to 0.08; P = 0.80, I2 = 0%; 11 studies, 421 participants; Analysis 1.6).

Lung hyperinflation (RV/%TLC) was measured in three studies (Duiverman 2008; Köhnlein 2014; Martin‐Marquez 2014), and 184 participants contributed to the IPD. There were only minor changes in RV/%TLC of participants in both study groups (AMD ‐0.6 %, 95% CI ‐2.6 to 1.4; Table 1). The meta‐analysis of the RV/%TLC data yielded similar results to the pooled IPD analysis (MD ‐0.53%, 95% CI ‐2.77 to 1.71; P = 0.30, I2 = 17%; 3 studies, 184 participants; Analysis 1.7), based on data from the same studies.

12 months' follow‐up

Four studies provided the IPD of 236 participants for the FEV1 and of 235 participants for the FVC (Clini 2002; Duiverman 2008; Köhnlein 2014; McEvoy 2009). Participants in both groups showed no change in FEV1 after 12 months (AMD 0.00 L, 95% CI ‐0.04 to 0.04; Table 1). For the FVC a comparable decrease was observed in participants of both groups (AMD ‐0.03 L, 95% CI ‐0.14 to 0.08; Table 1). The meta‐analyses did not include any additional data and yielded similar results for the FEV1 (MD 0.00 L, 95% CI ‐0.04 to 0.04; P = 0.61, I2 = 0%; 4 studies, 236 participants; Analysis 1.5), and for the FVC (MD ‐0.02 L, 95% CI ‐0.13 to 0.09; P = 0.83, I2 = 0%; 4 studies, 235 participants; Analysis 1.6).

The RV/%TLC was measured in 197 participants in four studies (Clini 2002; Duiverman 2008; Köhnlein 2014; McEvoy 2009), and no treatment effect could be observed (AMD 0.4 %, 95% CI ‐2.1 to 2.9; Table 1). The MD obtained from the meta‐analysis of the RV/%TLC data was 0.27 % (95% CI ‐1.98 to 2.53; P = 0.96, I2 = 0%; 4 studies, 195 participants; Analysis 1.7), based on data of the same four studies.

Respiratory muscle function
Three months' follow‐up

Five studies with a total of 123 participants measured the PEmax (Casanova 2000; Duiverman 2008; Gay 1996; Martin‐Marquez 2014; Strumpf 1991). On average, the PEmax improved more in participants treated with NIV compared to participants allocated to the control group (AMD 3.7 cmH2O, 95% CI ‐4.2 to 11.6; Table 1). The PImax was measured in one additional study (Bhatt 2013), so 152 participants contributed IPD to this outcome. Both groups also showed an increase for the PImax, but AMD was 6.9 cmH2O (95% CI 2.3 to 11.5; Table 1) in favour of the NIV group.

Respiratory muscle function data were available from one additional study (Garrod 2000). In contrast to the pooled IPD analysis, the meta‐analyses revealed a larger and statistically significant effect on the PEmax (MD 9.03 cmH2O, 95% CI ‐1.51 to 19.58; P = 0.07, I2 = 50%; 6 studies, 160 participants; Analysis 1.8). This outcome showed substantial statistical heterogeneity between the studies, which could not be explored by subgroup analysis as we included fewer than 10 studies. The meta‐analysis on the PImax showed comparable results to the pooled IPD analysis (MD 6.99 cmH2O, 95% CI 3.29 to 10.62, P = 0.99, I2 = 0%; 7 studies, 192 participants; Analysis 1.9).

12 months' follow‐up

The PEmax was only measured in one study (Duiverman 2008), and 38 participants contributed to the IPD. Data from this study showed a larger increase in PEmax in participants allocated to the NIV group (AMD 3.2 cmH2O, 95% CI ‐24.5 to 31.0; Table 1). Two studies measured the PImax (Clini 2002; Duiverman 2008), and 92 participants contributed to the IPD. The effect after 12 months was comparable to the effect after three months, and was in favour of the NIV group (AMD 6.0 cmH2O, 95% CI ‐0.9 to 12.8; Table 1).

There were no additional data available for the meta‐analyses of these outcomes. The meta‐analyses did not result in any new findings for the PEmax (MD 3.50 cmH2O, 95% CI ‐23.70 to 30.70; 1 study, 38 participants; Analysis 1.8), nor for the PImax (MD 7.04 cmH2O, 95% CI ‐12.29 to 26.37; P = 0.008, I2 = 86%; 2 studies, 92 participants; Analysis 1.9). The results showed considerable statistical heterogeneity between the two studies, but this heterogeneity could not be explored by subgroup analysis due to the low number of studies.

Dyspnoea
Three months' follow‐up

Three studies provided the IPD of 174 participants for the TDI (Bhatt 2013; Duiverman 2008; Zhou 2017). One other study reported the TDI (Strumpf 1991), but we could not extract the aggregated data from the paper. The total score of the TDI ranges from ‐9 points (major deterioration) to +9 points (major improvement), and an improvement of one point has been established as clinically relevant (Witek 2003). On average, participants allocated to both groups showed minor improvements (AMD 0.7, 95% CI ‐0.8 to 1.5; Table 1). Meta‐analysis yielded a comparable effect of NIV (MD 0.77, 95% CI 0.01 to 1.53; P = 0.34, I2 = 8%; 3 studies, 178 participants; Analysis 1.10) but in contrast to the pooled IPD analysis, this effect was statistically significant.

Four studies measured dyspnoea using the (modified)MRC (Casanova 2000; Clini 2002; Duiverman 2008; Martin‐Marquez 2014), and the IPD were available for 155 participants. The mMRC is reported on a 0 to 4 scale, with a score of 4 indicating severe dyspnoea. This outcome was not normally distributed, hence we could not conduct the linear mixed‐effect models. We included summary data of the same four studies in the meta‐analysis, which showed a decrease in the mMRC in participants allocated to NIV (MD ‐0.29, 95% CI ‐0.59 to 0.01; P = 0.21, I2 = 34%; 4 studies, 155 participants; Analysis 1.11). We observed some statistical heterogeneity between the studies, but this was not considered important.

12 months' follow‐up

None of the studies measured the TDI. Two studies measured dyspnoea using a 6‐point MRC (Clini 2002; Duiverman 2008), and the IPD were available for 103 participants. Due to a non‐normal distribution, we did not conduct the linear mixed‐effects models. The meta‐analysis of this data showed a decrease in dyspnoea in participants treated with NIV (MD ‐0.49, 95% CI ‐0.90 to ‐0.08; P = 0.45, I2 = 0%; 2 studies, 103 participants; Analysis 1.11).

Sleep efficiency

We did not find any new data since the original review was performed for this outcome. For consistency, we re‐analysed the data using the current linear mixed‐effect models.

Three months' follow‐up

Three studies with 24 participants provided IPD for sleep efficiency (Gay 1996; Meecham Jones 1995; Strumpf 1991). Participants allocated to both groups showed an improvement in sleep efficiency, but the effect tended to be larger in participants allocated to the control group (AMD ‐7.9, 95% CI ‐21.6 to 5.8; Table 1). The meta‐analysis of data from the same studies showed a MD of ‐2.86% (95% CI ‐8.76 to 3.04; P = 0.37, I2 = 0%; 3 studies, 24 participants; Analysis 1.12).

12 months' follow‐up

None of the included studies measured sleep efficiency as “the percentage time asleep of the total time in bed”, with a follow‐up of 12 months.

COPD exacerbations and hospital admissions

Baseline hospital admission rates were not recorded, or were recorded over different time frames (Bhatt 2013; Clini 2002; Duiverman 2008). We were unable to perform the linear mixed‐effect models as this outcome was not normally distributed.

Three months' follow‐up 

One study provided the IPD of 56 participants (Duiverman 2008). On average, exacerbations did not occur frequently in participants allocated to either group (Table 1). The number of participants that experienced an exacerbation was similar in participants allocated to the NIV group (12 of 24 participants (50%) ≥ 1 exacerbation) and the control group (19 of 32 participants (59%) ≥ 1 exacerbation) (OR 0.69, 95% CI 0.23 to 2.0). Aggregated exacerbation data from Casanova 2000 were available for the meta‐analysis. The OR for an exacerbation was 0.76 (95% CI 0.34 to 1.69; P = 0.77, I2 = 0%; 2 studies, 100 participants; Analysis 1.13), in favour of participants treated with NIV.

Four studies provided the IPD on hospital admissions of 268 participants (Bhatt 2013; Clini 2002; Duiverman 2008; Köhnlein 2014). Hospital admissions for respiratory causes were rare in participants of both groups (Table 1). An equal number of participants had been admitted to the hospital (NIV group: 34 of 137 participants (25%) admitted; control group: 34 of 131 participants (26%) admitted) (OR 0.89, 95% CI 0.51 to 1.57). Aggregated data from one additional study were available for the meta‐analysis (Casanova 2000). The OR for a hospitalisation was 0.64 (95% CI 0.26 to 1.60; P = 0.10, I2 = 49%; 5 studies, 312 participants; Analysis 1.14), in favour of participants treated with NIV. The results showed moderate statistical heterogeneity between the studies.

12 months' follow‐up

One study provided the IPD on COPD exacerbations of 45 participants (Duiverman 2008). The median number of exacerbations was similar in both groups (Table 1). The number of participants that experienced an exacerbation appeared similar in participants allocated to the NIV group (15 of 18 participants (83%) ≥ 1 exacerbation) and the control group (25 of 27 participants (93%) ≥ 1 exacerbation) (OR 0.36, 95% CI 0.05 to 2.47). For the meta‐analysis, the aggregated data of Casanova 2000 could be included. The OR for an exacerbation favoured participants allocated to NIV (OR 0.72, 95% CI 0.25 to 2.05; P = 0.47, I2 = 0%; 2 studies, 89 participants; Analysis 1.13).

Data on hospital admissions were provided for 138 participants (Clini 2002; Duiverman 2008; Köhnlein 2014). In the first study year, admissions for respiratory causes were rare (Table 1). A similar number of participants had been admitted to the hospital (NIV group: 30 of 71 participants (42%) admitted; control group: 26 of 67 participants (39%) admitted) (OR 1.14, 95% CI 0.58 to 2.25). For the meta‐analysis, the aggregated data of Casanova 2000 could be included. There was no effect of NIV (OR 1.06, 95% CI 0.46 to 2.44; P = 0.18, I2 = 39%; 4 studies, 182 participants; Analysis 1.14). There was some statistical heterogeneity between the studies, but we did not consider this to be important.

All‐cause mortality

Three studies provided the mortality data of 405 participants, with a median follow‐up of 30 months (Duiverman 2008; Köhnlein 2014; McEvoy 2009). There was a lower risk for all‐cause mortality in participants treated with NIV (adjusted hazard ratio (AHR) 0.75, 95% CI 0.58 to 0.97). Mortality was reported by one additional study (Casanova 2000), but we could not extract the data or source it from the authors. The meta‐analysis demonstrated a survival benefit for participants treated with NIV, but in contrast to the IPD analysis, this estimate was not statistically significant (HR 0.78, 95% CI 0.58 to 1.05; P = 0.29, I2 = 20%; 3 studies, 405 participants; moderate‐certainty evidence; Analysis 1.15).

Post‐exacerbation COPD — primary outcomes

The absolute changes in all outcomes of participants commenced on NIV following an acute exacerbation is presented in Table 2.  In Figure 5, the modelled means obtained from the linear mixed‐effect models are presented for the primary outcomes. 

Open in table viewer
Table 2. Absolute changes in the outcomes after three and 12 months in post‐exacerbation COPD, separate for the NIV and the control group

 

3 months

12 months

Outcome

n

Change in NIV group

n

Change in control group

n

Change in NIV group

n

Change in control group

PaO2 (kPa)

120

0.08 ± 2.33 

114

0.19 ± 2.12 

86

0.21 ± 1.96 

84

0.48 ± 2.10 

PaCO2 (kPa)

127

‐0.99 ± 1.24 

114

‐0.59 ± 1.19 

90

‐1.12 ± 1.15 

85

‐0.61 ± 1.20 

SRI, summary score

115

4.7 ± 10.6 

101

1.7 ± 11.7 

85

4.8 ± 11.9 

77

1.7 ± 11.8 

FEV1 (l)

108

0.02 ± 0.19 

93

0.04 ± 0.20 

82

0.01 ± 0.21 

75

0.03 ± 0.22 

FVC (l)

108

‐0.01 ± 0.66 

93

0.12 ± 0.63 

83

‐0.01 ± 0.64 

75

0.08 ± 0.68 

COPD exacerbations a

72

0 [0 to 1.0]

67

0 [0 to 1.0] 

95

2.0 [1.0 to 4.0]

86

2.0 [1.0 to 4.0] 

COPD hospitalisations a

0

—

0

—

100

0 (0 to 2.0) 

93

0 (0 to 2.0)

Data are based on the studies that provided the IPD, and are presented as mean ± standard deviation or median (interquartile range). 

a absolute number of exacerbations or hospitalisations.

CI: confidence interval; COPD: Chronic Obstructive Pulmonary Disease; FEV1: forced expiratory volume in 1 second; FVC: forced vital capacity; NIV: noninvasive ventilation; PaCO2: arterial carbon dioxide tension; PaO2: arterial oxygen tension; SRI: Severe Respiratory Insufficiency questionnaire, an increase indicates an improved health‐related quality of life.

Figure 5
The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in post‐exacerbation COPD Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in post‐exacerbation COPD 

Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

Daytime arterial blood gas tensions
Three months' follow‐up 

Three studies provided the IPD for the arterial blood gases of 234 participants for the PaO2 and for 241 participants for the PaCO2 (Cheung 2010; Murphy 2017; Struik 2014). There was no difference on the change in PaO2 between participants allocated to NIV or standard care (AMD ‐0.10 kPa, 95% CI ‐0.65 to 0.45; Table 2). There were improvements in the PaCO2 in both groups, but the effect was larger in participants treated with NIV (AMD ‐0.40 kPa, 95% CI ‐0.70 to ‐0.09; Table 2).

The meta‐analyses of these outcomes were based on data of the same three studies and yielded comparable results to the pooled IPD analyses for both the PaO2 (MD ‐0.08 kPa, 95% CI ‐0.58 to 0.41; P = 0.85, I2 = 0%; 3 studies, 234 participants; low‐certainty evidence; Analysis 2.1) and PaCO2 (MD ‐0.42 kPa, 95% CI ‐0.72 to ‐0.12; P = 0.45, I2 = 0%; 3 studies, 241 participants; low‐certainty evidence; Analysis 2.2).

12 months' follow‐up 

The same three studies provided the IPD of 170 participants for the PaO2 and of 175 participants for the PaCO2 (Cheung 2010; Murphy 2017; Struik 2014) . In the control group, PaO2 tended to improve more compared to participants in the NIV group (AMD ‐0.27 kPa, 95% CI ‐0.86 to 0.32; Table 2). In contrast, the PaCO2 was reduced in both groups, but the effect was larger in the participants treated with NIV (AMD ‐0.52 kPa, 95% CI ‐0.87 to ‐0.18; Table 2).

The meta‐analyses of these outcomes were based on data from the same three studies and yielded comparable results to the pooled IPD analyses for both the PaO2 (MD ‐0.24 kPa, 95% CI ‐0.81 to 0.34; P = 0.64, I2 = 0%; 3 studies, 170 participants; moderate certainty evidence; Analysis 2.1) and PaCO2 (MD ‐0.53 kPa, 95% CI ‐0.88 to ‐0.19; P = 0.94, I2 = 0%; 3 studies, 175 participants; high‐certainty evidence; Analysis 2.2).

Health‐related quality of life
Three months' follow‐up 

Two studies provided the IPD of 216 participants for the SRI (Murphy 2017; Struik 2014). Additional IPD of the SGRQ were available for three participants (Murphy 2017), so we standardised and pooled the data of the SRI and SGRQ. Participants treated with NIV improved more on the summary score of the SRI compared to participants allocated to the control group (Table 2). After pooling of the SRI and SGRQ data, there was no added benefit of the NIV on HRQL (AMD 0.25, 95% CI ‐0.01 to 0.51). Given a SD of 11.2 units for the SRI, this would suggest an improvement of 2.8 units.

The meta‐analysis of aggregated data from the same two studies showed a SMD in favour of the NIV group (SMD 0.27, 95% CI 0.01 to 0.54; P = 0.63, I2 = 0%; 2 studies, 216 participants; very low‐certainty evidence; Analysis 2.3). In contrast to the pooled IPD analysis, this was statistically significant. 

12 months' follow‐up

SRI data were available for 162 participants of the same two studies (Murphy 2017; Struik 2014). Participants treated with NIV showed, on average, a larger improvement on the summary score of the SRI (Table 2). Additional IPD of the SGRQ were available for two participants (Murphy 2017), so we standardised and pooled the data.  In line with the three‐month data, there was no added benefit of the NIV on HRQL (AMD 0.25, 95% CI ‐0.06 to 0.55). Given a SD of 11.9 units for the SRI, this would suggest an improvement of 3.0 units.

The meta‐analysis yielded similar results to the pooled IPD analysis (SMD 0.27, 95% CI ‐0.04 to 0.58; P = 0.34, I2 = 0%; 2 studies, 162 participants; very low‐certainty evidence; Analysis 2.3).

Admission‐free survival

This outcome was measured as time from randomisation to readmission (for a pulmonary cause) or death (of any cause) in the first year after randomisation in two studies (Murphy 2017; Struik 2014). The trials provided data from 317 participants. The admission‐free survival was better in participants treated with NIV compared to participants allocated to the control group (AHR 0.71, 95% CI 0.54 to 0.94).

The meta‐analysis of data of the same studies demonstrated an admission‐free survival benefit favouring the NIV group, but in contrast to the pooled IPD analysis this effect was not statistically significant (HR 0.69, 95% CI 0.39 to 1.23; P = 0.05, I2 = 75%; 2 studies, 317 participants; low‐certainty evidence; Analysis 2.4). The results showed considerable statistical heterogeneity between the studies, but this heterogeneity could not be explored by subgroup analysis due to the low number of studies.

Post exacerbation COPD — secondary outcomes

Lung function
Three months' follow‐up 

Two studies provided the IPD of 204 participants for this outcome (Murphy 2017; Struik 2014). For the FEV1, both study groups showed only a minor increase, the increase probably being slightly larger in participants allocated to the control group (AMD ‐0.02 L, 95% CI ‐0.07 to 0.03; Table 2). In contrast, an increase in the FVC was only observed in participants allocated to the control group (AMD ‐0.11 L, 95% CI ‐0.28 to 0.06; Table 2). The meta‐analyses were based on the same two studies that provided the IPD and did not reveal any new findings for both the FEV1 (MD ‐0.03 L, 95% CI ‐0.09 to 0.03; P = 0.28, I2 = 14%; 2 studies, 201 participants; Analysis 2.5) and the FVC (MD ‐0.13 L, 95% CI ‐0.32 to 0.04; P = 0.61, I2 = 0%; 2 studies, 202 participants; Analysis 2.6).

12 months' follow‐up

The results of the 157 participants from the same two studies yielded similar results. Both study groups showed only minor changes in the FEV1 (AMD ‐0.03 L, 95% CI ‐0.09 to 0.04; Table 2). The change in the FVC tended to be in favour of the control group (AMD ‐0.09 L, 95% CI ‐0.29 to 0.12; Table 2), as the FVC in the NIV group seems not affected.

The meta‐analyses yielded comparable results for the FEV1 (MD ‐0.03 L, 95% CI ‐0.09 to 0.03; P = 0.62, I2 = 0%; 2 studies, 157 participants; Analysis 2.5) and the FVC (MD ‐0.11 L, 95% CI ‐0.32 to 0.10; P = 0.29, I2 = 9%; 2 studies, 158 participants; Analysis 2.6).

COPD exacerbations and hospital admissions

Baseline hospital admission rates were recorded in two studies (Murphy 2017; Struik 2014) for the 12 months before inclusion in the study but in one study it was only recorded whether there were < 3 or ≥ 3 admissions. We were unable to perform the linear mixed effect models as this outcome was not normally distributed.

Three months' follow‐up 

One study provided the IPD of 139 participants (Struik 2014). On average, the total numbers of COPD exacerbations appear to be equal between both groups (Table 2). An equal proportion of participants had experienced one or more exacerbation (NIV group: 35 of 76 participants (46%); control group: 31 of 68 participants (46%) ≥ 1 exacerbation). The OR for an exacerbation was 0.93 (95% CI 0.47 to 1.81) in favour of participants in the NIV group. The meta‐analysis of data of the same study yielded also an OR of 0.93 (95% CI 0.48 to 1.81; 1 study, 139 participants; Analysis 2.7).

Hospitalisation data were not available for any of the included studies after three months.

12 months' follow‐up

Three studies contributed COPD exacerbation data of 181 participants (Cheung 2010; Murphy 2017; Struik 2014). There was an equal number of COPD exacerbations in both groups during the study period (Table 2). Also, exacerbation rates appeared similar between participants allocated to NIV (73 of 95 participants (77%) ≥ 1 exacerbation) and to the control group (71 of 86 participants (83%) ≥ 1 exacerbation). The OR was in favour of participants allocated to NIV (OR 0.70, 95% CI 0.33 to 1.46). Meta‐analysis of data of the same three studies showed a comparable result to the IPD analysis (OR 0.67, 95% CI 0.32 to 1.41; P = 0.98, I2 = 0%; 3 studies, 181 participants; Analysis 2.7).

IPD regarding hospital admissions were available for 193 participants of the same three studies. There was no difference in the number of hospital admissions (Table 2). Admission rates appeared similar between participants treated with NIV (54 of 100 participants (54%) admitted) and participants allocated to the control group (53 of 93 participants (57%) admitted). The OR for a hospitalisation was 0.88 (95% CI 0.52 to 1.50) in favour of participants treated with NIV. In the meta‐analysis, we observed a larger OR (OR 0.63, 95% CI 0.19 to 2.09; P = 0.06, I2 = 66%; 3 studies, 193 participants; Analysis 2.8). These results may represent substantial statistical heterogeneity between the studies, but this heterogeneity could not be explored due to the low number of studies.

All‐cause mortality

All‐cause mortality was measured in two studies (Murphy 2017; Struik 2014), which provided the IPD of 317 participants. The median follow‐up of all participants was 21 months. The risk for all‐cause mortality was not affected by NIV (AHR 0.97, 95% CI 0.74 to 1.28). The meta‐analysis of the mortality data of the same two studies yielded similar results to the pooled IPD analysis (HR 0.98, 95% CI 0.74 to 1.29; P = 0.96, I2 = 0%; 2 studies, 317 participants; low‐certainty evidence; Analysis 2.9).

Heterogeneity exploration in PaCO2 response in stable state

In the stable population, the results of the PaCO2 analysis after three months demonstrated substantial statistical heterogeneity (P = 0.001; I2 = 65%;  Analysis 1.2). Subgroup analysis based on the mean baseline PaCO2 of the study revealed a significant subgroup difference (test for subgroup differences: P = 0.001; I2 = 91%; Analysis 1.16; Figure 6). In the PaCO2 < 7.3 kPa subgroup, heterogeneity was reduced to 0%, but remained substantial in the ≥ 7.3 kPa subgroup. As seen in Figure 6, the study by Clini 2002 seems to cause this heterogeneity. This was the only study in this subgroup adopting low ventilator pressures (mean IPAP < 18 cmH2O), and when this study was removed from the subgroup, the heterogeneity was reduced to 0% (P = 0.64).

Additional subgroup analysis based on mean IPAP also found significant subgroup differences (test for subgroup differences: P = 0.005, I2 = 88%; Analysis 1.17; Figure 7). In the IPAP < 18 cmH2O subgroup, the heterogeneity was reduced to 0% (P = 0.55), whilst in the IPAP ≥ 18 cmH2O subgroup, the heterogeneity remained substantial. As seen in Figure 7, the study by Duiverman 2008 seems to cause this heterogeneity. This study mainly included participants with moderate hypercapnia (mean baseline PaCO2 < 7.3 kPa), and removal of this study from the subgroup reduced the heterogeneity to 0% (P = 0.64).

Lastly, subgroup analysis based on compliance also showed significant subgroup differences (test for subgroup differences: P = 0.04, I2 = 77%; Analysis 1.18), but the heterogeneity in the good compliance subgroup remained substantial (P = 0.01, I2 = 59%).

Sensitivity analysis

For the sensitivity analysis we excluded seven studies on stable COPD as we had judged these to be at high or unclear risk of bias for the selection or attrition domain (or both) (Bhatt 2013; Clini 2002; Gay 1996; Martin‐Marquez 2014; Schneeberger 2017; Sin 2007; Strumpf 1991). Removal of these studies from the analysis resulted in a slightly larger treatment effect for PaO2 after three months of NIV (AMD 0.31 kPa, 95% CI 0.02 to 0.62, 103 participants), but did not affect the PaO2 results after 12 months. The treatment effect of NIV on PaCO2 increased both after three months' follow‐up (AMD ‐0.76 kPa, 95% CI ‐0.98 to ‐0.55, 304 participants) and 12 months' follow‐up (AMD ‐0.55 kPa, 95% CI ‐0.84 to ‐0.27, 174 participants). The sensitivity analysis also revealed a larger and statistically significant effect of NIV on the 6MWD, both after three months (AMD 22.3 m, 95% CI 2.1 to 42.5, 245 participants) and after 12 months (AMD 44.0m, 95% CI 0.2 to 87.8, 88 participants). Lastly, a larger effect of NIV was observed on HRQL at both follow‐ups (three months: AMD 0.44, 95% CI 0.18 to 0.68, 229 participants; 12 months: AMD 0.26, 95% CI ‐0.05 to 0.55; 169 participants).

We excluded one post‐exacerbation study due to high or unclear risk of bias (Cheung 2010). When limiting the analysis to the remaining high‐quality studies, the treatment effects for all primary outcomes were in line with the overall treatment effect in this cohort.

Discussion

Summary of main results

In people with severe COPD in a stable condition, the addition of chronic NIV to standard treatment improves diurnal hypercapnia (high‐certainty evidence) for up to 12 months. Additionally, chronic NIV seems to improve short‐term diurnal PaO2 and all‐cause mortality (moderate‐certainty evidence) and might result in a relevant improvement in HRQL at three months (very low‐certainty evidence).  In people who have COPD with persistent hypercapnia following a severe COPD exacerbation with AHRF, the added benefit of chronic NIV seems to be more limited. In this population, chronic NIV seems to improve daytime hypercapnia (moderate to high‐certainty evidence) and might prolong the time to readmission or death (low‐certainty evidence), without a beneficial effect on HRQL. Although the targets of this treatment seem to differ between these populations, the results of this systematic review seem to confirm the application of chronic NIV in people with COPD in a stable clinical state, and might support the initiation of chronic NIV in people who have COPD with persistent hypercapnia after an acute exacerbation of COPD.

Interpretation of the main findings

Gas exchange

The evidence indicates that regardless of the timing of chronic NIV initiation, relevant improvements in hypercapnia may be achieved with chronic NIV. Additionally, the PaO2 of people with stable COPD seems to improve with NIV after three months. Our results again indicate that that in stable COPD, the effect of NIV on diurnal ventilation seems to be larger in people with more severe hypercapnia (PaCO2 ≥ 7.3 kPa) and a better treatment compliance (≥ 5 hours per night) (Struik 2014). Hypercapnia also seems to improve more in people treated with an IPAP ≥ 18 cmH2O. This finding supports the application of NIV adopting high IPAP to target nocturnal PaCO2 reduction, which has become a major NIV strategy during the past decade (Dreher 2010; Windisch 2009). However, it remains uncertain to what extent these improvements in gas exchange contribute to the beneficial effect on individual‐centred outcomes such as HRQL and survival. A better understanding of these mechanisms could improve selection of people with COPD for NIV, and could optimise the titration of the individual's NIV settings. Therefore, we emphasise this should be an aim of future studies.

Exercise capacity

Evidence that chronic NIV improves exercise capacity is currently insufficient. However, if only high‐quality studies are included, it seems that there is a benefit of chronic NIV on the exercise capacity at both three months (AMD 22.3 m, 95% CI 2.1 to 42.5) and 12 months' follow‐up (AMD 44.0 m, 95% CI 0.2 to 87.8). Given the MCID of 26 m identified in this population of people with very severe COPD (Puhan 2011), this improvement only seems to be of relevance after 12 months. We acknowledge that the uncertainty in these estimates is high, as the 95% CI are wide. So although there might be a positive effect, a definite conclusion about the effect of chronic NIV on exercise capacity cannot be drawn at this moment.

Health‐related quality of life

Improving HRQL should be a major target of any treatment in severe or end‐stage COPD. Acknowledging that our confidence in the evidence is very low, the results do indicate that the initiation on chronic NIV might have a beneficial effect on HRQL in people with stable COPD in the short term. This treatment effect after three months seems to reach clinical relevance, at least when measured with the change responsive SRI. The contrasting results at the 3‐ and 12‐month follow‐up periods are interesting. We emphasise that different proportions of studies contributed to either follow‐up, possibly with different treatment and participant characteristics, which might explain these different results. For participants initiated following an exacerbation, there seems to be no benefit of NIV on HRQL and only the upper range of the 95% CI seems to be of clinically relevance for the total score of the SRI. Future studies should therefore continue to investigate the effect of post‐exacerbation chronic NIV on HRQL and identify which individual is most likely to experience a HRQL benefit. Lastly, given the result of the meta‐analysis on the dyspnoea outcomes, there may be a small beneficial effect on dyspnoea, but the data are sparse and the effect estimate does not seem to reach clinical relevance (Witek 2003).

Exacerbations and (re)hospitalisations

We found that chronic NIV might prolong the admission‐free survival in people initiated on NIV following a severe COPD exacerbation. However, continuing NIV after an exacerbation did not affect the total number of COPD admissions in the first year, and did not affect mortality. We hypothesise that, although the first readmission might have been postponed, eventually these people with severe COPD suffer from frequent exacerbations and related mortality despite the chronic NIV. Furthermore, admission‐free survival was investigated in the total population, while exacerbation and admission rates were investigated only in the participants who completed the study period of one year. To conclude, at this moment there is insufficient evidence that chronic NIV following a severe COPD exacerbation reduces the number of COPD exacerbations and admissions, and mortality in these patients remains high.

Mechanisms

It has been hypothesised that chronic NIV achieves its beneficial effect though improvements in respiratory mechanics, such as stabilisation of FEV1 and a reduction in residual volume. The results of this review do not seem to be supportive of these hypotheses. We did not show an effect of NIV on different lung function outcomes in either of our cohorts. A second hypothesis concerns an improved function of the respiratory muscles that might be obtained by chronic NIV. The results of our analyses indicate an improved PImax. In people with severe COPD, PImax is known to be related to dyspnoea, HRQL and even survival (Ottenheijm 2008), so an improved respiratory muscle function could contribute to the improved outcomes of NIV. It should, however, be noted that PImax measures maximum inspiratory pressure, which is a derivative of respiratory muscle strength and importantly also affected by hyperinflation (Rochester 1985). However, we did not show an improvement in hyperinflation with NIV. In addition, due to its voluntary nature, these measurements are affected not only by strength but also by motivation (Ottenheijm 2008). These caveats prevent strong conclusions regarding improved respiratory function by NIV, but our results do at least not refute this hypothesis. Finally, an effect on sleep quality might be an important aspect in improving HRQL. However, this has unfortunately seldom been investigated, and no additional data could be added to this meta‐analysis to support this hypothesis.

Overall completeness and applicability of evidence

This review included 21 RCTs investigating chronic NIV for people with severe COPD. Fortunately, the trial authors of 16 studies responded to our request to share their trial data, enabling the investigation of the IPD of 1142 participants.  We could not obtain the IPD from five trials. Therefore, we also conducted a meta‐analysis of the summary trial data from all trials. For the majority of the outcomes, the results of the meta‐analysis were in line with the IPD results, except for expiratory muscle strength after three months, which became significant after additional data were added (Garrod 2000). In the post‐exacerbation population, the IPD of the admission‐free survival and mortality outcomes were not fully obtained, as these IPD were not provided for the study by Cheung 2010. This trial might bias the results, as participants were excluded from the trial once the primary outcome (“an episode recurrent AHRF requiring acute NIV or intubation, or death”) was reached. Additionally, it remains unclear if the trial by Xiang 2007 could be included, as we were unable to contact the trial authors.

The inclusion criteria were generally consistent between the studies, and all studies only included participants with severe COPD. The vast majority included participants with chronic hypercapnia; only 61 participants (8%) without hypercapnia were included. We therefore emphasise that the results apply primarily to people with COPD who experience chronic hypercapnia. In the stable population, 72% of the participants were male, which could reflect the higher prevalence of COPD in males (GOLD 2021; Terzikhan 2016). In contrast to the 2013 update, this review also included studies that investigated chronic nocturnal NIV in addition to a pulmonary rehabilitation or exercise training program. This increased the body of evidence as it resulted in the inclusion of four additional studies (Duiverman 2008; Garrod 2000; Martin‐Marquez 2014; Schneeberger 2017), including the IPD of 127 participants. Our results cannot automatically be extended to overlap people who have COPD with concomitant obstructive sleep apnea, as most studies excluded people with concomitant sleep apnea or a very high BMI. Although this might limit the generalisability of our conclusions, in the present analysis effects for people with COPD are reliably displayed without overestimating the effects due to concomitant sleep apnea or serious obesity hypoventilation.  

It should also be noted the majority of the studies, especially the larger studies and studies with positive outcomes, have been conducted in Europe. This could limit the generalisability of the results to other regions as expertise and organisation of care might differ, which could lead to differences in treatment characteristics, adherence and effectiveness.

Quality of the evidence

Overall, the certainty of the evidence for the main comparisons was moderate to very low, according to the  GRADE 2013 recommendations. The PaCO2 outcome was the only outcome that we judged to be of high certainty, and it is believed that the observed treatment effect lies close to the true treatment effect of NIV. Risk of bias was a major reason for downgrading the certainty, as the unblinded design of the included studies is likely to affect outcomes such as the 6MWD and HRQL. We decided not to downgrade the blood gas and survival outcomes on this domain, as we judged that the unblinded design is unlikely to affect these outcomes. Inconsistency was observed in two outcomes, but we did not consider this a reason to downgrade the certainty. First, we observed substantial statistical heterogeneity between the studies for the PaCO2 outcome (P = 0.001, I2 = 65%), but this could be explained by the subgroup analysis. Second, the admission‐free survival demonstrated considerable heterogeneity (P = 0.005, I2 = 75%). As only two studies provided data for this outcome, we were unable to conduct subgroup analysis to explore this heterogeneity. Nevertheless, this heterogeneity may be explained by differences in the timing of NIV initiation and participant selection, as in both studies the intervention targeted nocturnal correction of hypoventilation and was initiated in highly experienced centres. Another major reason to downgrade the certainty was serious imprecision. For most outcomes, the number of studies was low, resulting in wide 95% CIs. Thus, the possibility of harm or no treatment effect could not be excluded for outcomes such as the 6MWD, HRQL and the survival outcomes. Lastly, we did not observe indirectness or publication bias in our results.

Potential biases in the review process

Limitations regarding the setup of the analyses

Our protocol limited the follow‐up duration to three or 12 months of NIV. Two studies measured or reported trial outcomes only after six months of NIV, and we did not use these data (De Backer 2011; Eman Shebl 2015). The results of these two studies seem in line with the results of our review, with the exception that a relevant benefit was found on the 6MWD in stable COPD (Eman Shebl 2015). Although only 45 participants were included in these two studies, we acknowledge that this a limitation as it reduced the body of evidence. Another limitation is that we did not investigate serious or rare adverse effects. However, the (serious) adverse effects of chronic NIV are well known and are usually related to intolerance to the device or mask, leakage of the air delivered or facial irritation due to mask pressures. A recent review on chronic NIV supports this as it found an incidence rate of 0 per person (95% CI 0.00 to 0.01) for severe adverse events and of 0.21 per person (95% CI 0.12 to 0.37) for adverse events, acknowledging that a majority of the trials did not report adverse events (Wilson 2020). Lastly, we could not assess the risk of publication bias as the only outcome with more than 10 included studies (PaCO2) demonstrated heterogeneity between the studies.

Limitations regarding the included studies

We encountered several limitations regarding the included studies. The major limitation concerns the unblinded nature of the vast majority of the trials. Although some trials used a continuous positive airway pressure as sham device (Cheung 2010; Gay 1996; Sin 2007), an optimal sham treatment is not available as these approaches could increase anatomical dead space or alter respiratory mechanisms, possibly also affecting the respiratory failure (Rodway 2010; Saatci 2004). The extent to which the unblinded nature of the trials influences the outcomes differs; e.g. blood gases, hospitalisations and mortality are unlikely to be affected, whilst participant‐reported outcomes are likely affected to a larger extent. Secondly, we have also included trials investigating chronic NIV applied during the night in addition to a training or rehabilitation program. It is, however, difficult to study the additional benefit of these training programs as for the majority of the included studies, it was unclear if participants were allowed to follow such a program, as studies did not record this adequately. Third, data on the actual timing of NIV initiation were missing in the post‐exacerbation trials. This data would be of value to estimate the optimal timing of chronic NIV initiation shortly following an episode of AHRF. Fourth, exacerbations and hospitalisation frequency proved difficult to investigate adequately. Many studies did not report these numbers prior to trial inclusion, pre‐study numbers were not always objectively assured and so could have been affected by recall bias, or studies reported exacerbations and hospitalisations over a different time frame (e.g. one or three years). It was therefore difficult to investigate whether NIV actually reduced the number of exacerbations. Additionally, nine of the included studies investigated HRQL but used multiple tools to assess this. We chose to consider only the disease‐specific SRI and the SGRQ, as these tools have shown good validity and reliability, especially in people with COPD in a stable phase (Struik 2013). We therefore recommend future trials to use disease‐specific HRQL tools such as the SRI. Lastly, one trial only reported the percentage predicted values of the lung function outcomes. Using the regression equations, we were able to calculate the absolute values, but we acknowledge that this could introduce uncertainty and is sensitive to errors.

Agreements and disagreements with other studies or reviews

In recent years, two systematic reviews and meta‐analyses have been conducted. Inclusion criteria appear similar to our review, except that both these reviews included observational or non‐RCTs, and also included trials that compared different NIV strategies. Comparison of mortality and admission outcomes is difficult, as no review has been published that studied these outcomes as time‐to‐event data.

Recently, a systematic review and meta‐analysis was published that included 51,085 people with COPD who experience chronic hypercapnia(33 studies), investigating the effect of chronic home NIV on mortality, all‐cause hospital admissions and quality of life (Wilson 2020). As compared to our systematic review, this review included both RCTs and observational studies, and did not differentiate according to the timing of NIV initiation. However, the review authors also conducted subgroup analysis based on the timing of NIV initiation. These results are, in general, consistent with our finding; a significantly lower mortality risk (OR 0.62, 95% CI 0.42 to 0.92) and no effect on hospital admissions (RR 0.84, 95% CI 0.59 to 1.18) for people initiated in a stable phase, and no significant effect on mortality (OR 0.66, 95% CI 0.41 to 1.06) or HRQL (standardised MD ‐0.03, 95% CI ‐0.25 to 0.20) in people initiated following an acute exacerbation. They did not find a beneficial effect on HRQL in the stable subgroup. This difference might be explained by the tools used in the review by Wilson 2020; the COPD assessment test, the Chronic Respiratory disease Questionnaire, the SGRQ and the Short‐form 36. HRQL tools designed specifically for people with respiratory failure, such as the SRI that was used for this review, might be more sensitive to change (Struik 2013). In addition, we have also used the HRQL data of the RCTs of Clini 2002 and McEvoy 2009, which could explain the observed difference.

In 2016, Dretzke 2016 published their systematic review of both RCTs and non‐RCTs (31 studies), also grouping studies into stable and post‐exacerbation populations. When the  Dretzke 2016 review was published, some large trials included in the present review had not yet been published (Murphy 2017; Zhou 2017), possibly explaining the observed differences. They did not find a survival benefit of NIV in either the stable participants (RR 0.88, 95% CI 0.55 to 1.43) or the post‐exacerbation population (RR 0.89, 95% CI 0.53 to 1.49). Results on the blood gases were not included in the meta‐analysis but support our findings as they suggest improved blood gases (PaCO2 and PaO2) in the stable population, and a potential benefit on PaCO2 in the post‐exacerbation cohort.

Lastly, the results of our review on COPD exacerbation in the post‐exacerbation cohort contradict a major finding of Murphy 2017, who found a reduced exacerbation rate favouring the NIV group. In contrast to Murphy, we conducted our analysis only on participants who completed the 12‐month follow‐up ('completers'). Dropouts could be unequally distributed between both study arms, and the 'time at risk' for an exacerbation could therefore differ between both arms, introducing bias. Although including only completers reduced statistical power, there was not even a trend towards a reduced exacerbation rate in our analysis. We acknowledge that this analysis setup could explain the observed difference.

PRISMA flow diagram

Figures and Tables -
Figure 1

PRISMA flow diagram

Navigate to figure in ReviewOpen in new tab

Risk of bias summary: review authors' judgements about each methodological quality item for each included studyNote: blank squares indicate that the outcome domain was not measured in the study.

Figures and Tables -
Figure 2

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

Note: blank squares indicate that the outcome domain was not measured in the study.

Navigate to figure in ReviewOpen in new tab

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in stable COPDBlack triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.
Figures and Tables -
Figure 3

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in stable COPD

Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

Navigate to figure in ReviewOpen in new tab

Funnel plot for the PaO2 outcome after 3 months, in participants with stable COPD
Figures and Tables -
Figure 4

Funnel plot for the PaO2 outcome after 3 months, in participants with stable COPD

Navigate to figure in ReviewOpen in new tab

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in post‐exacerbation COPD Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

Figures and Tables -
Figure 5

The modelled means of the adjusted mixed‐effect models for the primary outcomes, separate for the NIV and the control group, in post‐exacerbation COPD 

Black triangles: NIV group; red squares: control group. * indicates that the 95% confidence interval of the treatment effect does not cross zero. For the HRQL outcome, a positive value indicates improved HRQL.

Navigate to figure in ReviewOpen in new tab

original image

Figures and Tables -
Figure 6
Navigate to figure in ReviewOpen in new tab

original image

Figures and Tables -
Figure 7
Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 1: Stable COPD: PaO2

Figures and Tables -
Analysis 1.1

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 1: Stable COPD: PaO2

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 2: Stable COPD: PaCO2

Figures and Tables -
Analysis 1.2

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 2: Stable COPD: PaCO2

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 3: Stable COPD: 6‐minute walking distance

Figures and Tables -
Analysis 1.3

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 3: Stable COPD: 6‐minute walking distance

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 4: Stable COPD: HRQL

Figures and Tables -
Analysis 1.4

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 4: Stable COPD: HRQL

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 5: Stable COPD: FEV1

Figures and Tables -
Analysis 1.5

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 5: Stable COPD: FEV1

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 6: Stable COPD: FVC

Figures and Tables -
Analysis 1.6

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 6: Stable COPD: FVC

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 7: Stable COPD: RV/%TLC

Figures and Tables -
Analysis 1.7

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 7: Stable COPD: RV/%TLC

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 8: Stable COPD: PEmax

Figures and Tables -
Analysis 1.8

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 8: Stable COPD: PEmax

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 9: Stable COPD: PImax

Figures and Tables -
Analysis 1.9

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 9: Stable COPD: PImax

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 10: Stable COPD: Transition Dyspnoea Index

Figures and Tables -
Analysis 1.10

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 10: Stable COPD: Transition Dyspnoea Index

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 11: Stable COPD: (modified) Medical Research Council dyspnoea scale

Figures and Tables -
Analysis 1.11

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 11: Stable COPD: (modified) Medical Research Council dyspnoea scale

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 12: Stable COPD: sleep efficiency

Figures and Tables -
Analysis 1.12

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 12: Stable COPD: sleep efficiency

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 13: Stable COPD: exacerbations

Figures and Tables -
Analysis 1.13

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 13: Stable COPD: exacerbations

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 14: Stable COPD: hospitalisations

Figures and Tables -
Analysis 1.14

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 14: Stable COPD: hospitalisations

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 15: Stable COPD: all‐cause mortality

Figures and Tables -
Analysis 1.15

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 15: Stable COPD: all‐cause mortality

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 16: PaCO2 after 3 months ‐ subgroup PaCO2 (stable COPD)

Figures and Tables -
Analysis 1.16

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 16: PaCO2 after 3 months ‐ subgroup PaCO2 (stable COPD)

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 17: PaCO2 after 3 months ‐ subgroup IPAP (stable COPD)

Figures and Tables -
Analysis 1.17

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 17: PaCO2 after 3 months ‐ subgroup IPAP (stable COPD)

Navigate to figure in ReviewOpen in new tab

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 18: PaCO2 after 3 months ‐ subgroup compliance (stable COPD)

Figures and Tables -
Analysis 1.18

Comparison 1: Comparison 1: NIV vs standard care in stable COPD, Outcome 18: PaCO2 after 3 months ‐ subgroup compliance (stable COPD)

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 1: Post‐exacerbation COPD: PaO2

Figures and Tables -
Analysis 2.1

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 1: Post‐exacerbation COPD: PaO2

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 2: Post‐exacerbation COPD: PaCO2

Figures and Tables -
Analysis 2.2

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 2: Post‐exacerbation COPD: PaCO2

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 3: Post‐exacerbation COPD: HRQL

Figures and Tables -
Analysis 2.3

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 3: Post‐exacerbation COPD: HRQL

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 4: Post‐exacerbation COPD: admission‐free survival

Figures and Tables -
Analysis 2.4

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 4: Post‐exacerbation COPD: admission‐free survival

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 5: Post‐exacerbation COPD: FEV1

Figures and Tables -
Analysis 2.5

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 5: Post‐exacerbation COPD: FEV1

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 6: Post‐exacerbation COPD: FVC

Figures and Tables -
Analysis 2.6

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 6: Post‐exacerbation COPD: FVC

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 7: Post‐exacerbation COPD: exacerbations

Figures and Tables -
Analysis 2.7

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 7: Post‐exacerbation COPD: exacerbations

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 8: Post‐exacerbation COPD: hospitalisations

Figures and Tables -
Analysis 2.8

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 8: Post‐exacerbation COPD: hospitalisations

Navigate to figure in ReviewOpen in new tab

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 9: Post‐exacerbation COPD: all‐cause mortality

Figures and Tables -
Analysis 2.9

Comparison 2: Comparison 2: NIV vs standard care in post‐exacerbation COPD, Outcome 9: Post‐exacerbation COPD: all‐cause mortality

Navigate to figure in ReviewOpen in new tab
Summary of findings 1. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD

Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD

Patient or population: stable COPD Setting: home treatment Intervention: NIV Comparison: Standard care

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with Standard care

Risk with NIV

Change in arterial partial pressure of oxygen
follow up: 3 months

The mean change in arterial partial pressure of oxygen was ‐0.07 kPa a

MD 0.27 kPa higher
(0.04 higher to 0.49 higher) b

‐

271
(9 RCTs)

⊕⊕⊕⊝
MODERATE c,d

Chronic NIV probably results in a slightly greater increase in arterial partial pressure of oxygen after 3 months.

Change in arterial partial pressure of oxygen
follow up: 12 months

The mean change in arterial partial pressure of oxygen was ‐0.06 kPa

MD 0.09 kPa higher
(0.23 lower to 0.42 higher) b

‐

171
(3 RCTs)

⊕⊕⊝⊝
LOW c,e

Chronic NIV may result in little to no difference in arterial partial pressure of oxygen after 12 months.

Change in arterial partial pressure of carbon dioxide
follow up: 3 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.17 kPa

MD 0.61 kPa lower
(0.77 lower to 0.45 lower) b

‐

475
(11 RCTs)

⊕⊕⊕⊕
HIGH c,f

Chronic NIV results in a reduction in the arterial partial pressure of carbon dioxide after 3 months.

Change in arterial partial pressure of carbon dioxide
follow up: 12 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.05 kPa

MD 0.42 kPa lower
(0.68 lower to 0.16 lower) b

‐

232
(4 RCTs)

⊕⊕⊕⊕
HIGH c

Chronic NIV results in a reduction in arterial partial pressure of carbon dioxide after 12 months.

Change in 6‐minute walking distance
follow up: 3 months

The mean change in 6‐minute walking distance was 16.0 metres

MD 15.5 metres higher
(0.8 lower to 31.7 higher) b

‐

330
(8 RCTs)

⊕⊕⊝⊝
LOW d,g

The evidence suggests that chronic NIV results in little to no difference on the 6‐minute walking distance test. Sensitivity analysis limited to high‐quality studies suggest a significant benefit; however, this treatment estimate does not reach the minimal important difference of 26m.

Change in 6‐minute walking distance
follow up: 12 months

The mean change in 6‐minute walking distance was ‐10.7 metres

MD 26.4 metres higher
(7.6 lower to 60.5 higher) b

‐

134
(3 RCTs)

⊕⊝⊝⊝
VERY LOW g,h

Chronic NIV may increase the 6‐minute walking distance but the evidence is very uncertain. This treatment estimate reaches the clinical relevance of 26m in this specific population.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 3 months i

‐

SMD 0.39 SD higher
(0.15 higher to 0.62 higher) b

‐

259
(5 RCTs)

⊕⊝⊝⊝
VERY LOW d,j

Chronic NIV may increase health‐related quality of life after 3 months but the evidence is very uncertain. Considering the standard deviation of 12 units on the SRI, the observed SMD does seem to reach the minimal important difference of approximately 5 points in this specific population.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 12 months i

‐

SMD 0.15 SD higher
(0.13 lower to 0.43 higher) b

‐

200
(4 RCTs)

⊕⊝⊝⊝
VERY LOW h,j

The evidence is very uncertain about the effect of chronic NIV on health‐related quality of life after 12 months. Considering a standard deviation of 12.2 units on the SRI, the observed SMD does not seem to reach the minimal important difference of approximately 5 points.

All‐cause mortality
follow up: median 30 months

Study population

HR 0.75
(0.58 to 0.97)
[All‐cause mortality] k

405
(3 RCTs)

⊕⊕⊕⊝
MODERATE c,d

Chronic NIV likely reduces all‐cause mortality (number needed to treat for an additional beneficial outcome 14, 95% CI 8 to 120).

655 per 1000

550 per 1000
(461 to 644)

*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; MD: Mean difference; SMD: Standardised mean difference; HR: Hazard 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

See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_417569665332318026.

a. For the risk with standard care, we calculated an absolute difference in the outcome between the baseline measurement and both follow‐up measurements (3 and 12 months)
b. The mean difference was obtained from the pooled individual participant data, and was adjusted for age and sex.
c. Not downgraded; although the participants and investigators were not blinded in the majority of the trials, this is unlikely to affect blood gases and survival.
d. Downgraded by one for serious imprecision; the CI includes the possibility of no (relevant) benefit and of a substantial benefit.
e. Downgraded by one due to serious imprecision: the CI includes the possibility of harm and of a large substantial benefit.
f. Inconsistency not downgraded; although substantial heterogeneity was observed (P = 0.001, I2 = 65%), this could be explained by the subgroup analysis.
g. Downgraded by one due to high risk of bias; the majority of the trials did not blind participants and investigators/personnel to the treatment allocation.
h. Downgraded by two due to serious imprecision: the CI includes the possibility of harm and of a large substantial benefit.
i. SRI: Severe Respiratory Insufficiency questionnaire; SGRQ: St. George's Respiratory Questionnaire
j. Downgraded by one due to high risk of bias; no trials blinded participants and investigators/personnel to the treatment allocation.
k. The hazard ratio was obtained from the pooled individual participant data, and was adjusted for age, sex and the baseline arterial pressure of carbon dioxide.

Figures and Tables -
Summary of findings 1. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment for people with stable COPD
Navigate to table in Review
Summary of findings 2. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD

Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD

Patient or population: post‐exacerbation COPD Setting: home treatment Intervention: NIV Comparison: Standard care

Outcomes

Anticipated absolute effects* (95% CI)

Relative effect
(95% CI)

№ of participants
(studies)

Certainty of the evidence
(GRADE)

Comments

Risk with Standard care

Risk with NIV

Change in arterial partial pressure of oxygen
follow up: 3 months

The mean change in arterial partial pressure of oxygen was 0.19 kPa a

MD 0.1 kPa lower
(0.65 lower to 0.45 higher) b

‐

234
(3 RCTs)

⊕⊕⊝⊝
LOW c,d

The evidence suggests that chronic NIV results in little to no difference in arterial partial pressure of oxygen after 3 months.

Change in arterial partial pressure of oxygen
follow up: 12 months

The mean change in arterial partial pressure of oxygen was 0.48 kPa

MD 0.27 kPa lower
(0.86 lower to 0.32 higher) b

‐

170
(3 RCTs)

⊕⊕⊝⊝
LOW c,d

Chronic NIV may result in a slight reduction in arterial partial pressure of oxygen after 12 months.

Change in arterial partial pressure of carbon dioxide
follow up: 3 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.59 kPa

MD 0.4 kPa lower
(0.7 lower to 0.09 lower) b

‐

241
(3 RCTs)

⊕⊕⊕⊝
MODERATE c,e

Chronic NIV likely reduces the arterial partial pressure of carbon dioxide after 3 months.

Change in arterial partial pressure of carbon dioxide
follow up: 12 months

The mean change in arterial partial pressure of carbon dioxide was ‐0.61 kPa

MD 0.52 kPa lower
(0.87 lower to 0.18 lower) b

‐

175
(3 RCTs)

⊕⊕⊕⊕
HIGH c

Chronic NIV results in a reduction in the arterial partial pressure of carbon dioxide after 12 months.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 3 months f

‐

SMD 0.25 SD higher
(0.01 lower to 0.51 higher) b

‐

219
(2 RCTs)

⊕⊝⊝⊝
VERY LOW e,g

Chronic NIV may increase/have little to no effect on health‐related quality of life after 3 months but the evidence is very uncertain. Considering a standard deviation of 11.2 on the SRI, the SMD does not seem to reach the minimal important difference.

Change in the health‐related quality of life
assessed with: SRI and SGRQ
follow up: 12 months f

‐

SMD 0.25 SD higher
(0.06 lower to 0.55 higher) b

‐

164
(2 RCTs)

⊕⊝⊝⊝
VERY LOW e,g

Chronic NIV may increase/have little to no effect on health‐related quality of life after 12 months but the evidence is very uncertain. Considering a standard deviation of 11.9 on the SRI, the SMD does not seem to reach the minimal important difference.

Admission‐free survival
follow up: 1 year

Study population

HR 0.71
(0.54 to 0.94)
[readmission or death] h

317
(2 RCTs)

⊕⊕⊝⊝
LOW e,i,j

The evidence suggests chronic NIV improves admission‐free survival (number needed to treat for an additional beneficial outcome 12, 95% CI 7 to 61).

333 per 1000

458 per 1000
(356 to 552)

All‐cause mortality
follow up: median 21 months

Study population

HR 0.97
(0.74 to 1.28)
[All‐cause mortality] h

318
(2 RCTs)

⊕⊕⊝⊝
LOW c,d

The evidence suggests that chronic NIV does not improve all‐cause mortality.

642 per 1000

631 per 1000
(532 to 731)

*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; MD: Mean difference; SMD: Standardised mean difference; HR: Hazard 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

See interactive version of this table: https://gdt.gradepro.org/presentations/#/isof/isof_question_revman_web_417570213828006491.

a. For the risk with standard care, we calculated an absolute difference in the outcome between the baseline measurement and both follow‐up measurements (3 and 12 months).
b. The mean difference was obtained from the pooled individual participant data, and was adjusted for age and sex.
c. Not downgraded; although the participants and investigators were not blinded in the majority of the trials, this is unlikely to affect blood gases and survival.
d. Downgraded by two due to serious imprecision: the CI includes the possibility of significant harm and of a substantial benefit.
e. Downgraded by one due to serious imprecision: the CI includes the possibility of no (relevant) benefit and of a substantial benefit.
f. SRI: Severe Respiratory Insufficiency questionnaire; SGRQ: St. George's Respiratory Questionnaire
g. Downgraded by two due to high risk of bias; no trials blinded participants and investigators/personnel to the treatment allocation.
h. The hazard ratio was obtained from the pooled individual participant data, and was adjusted for age sex, baseline arterial partial pressure of carbon dioxide and number of COPD hospitalisations.
i. Downgraded by one; the majority of the participants were unblinded which could affect the likelyhood of attending the emergency department. Other domains low risk of bias
j. Inconsistency not downgraded; although substantial heterogeneity was observed (P = 0.05, I2 = 75%), this heterogeneity might be explained by differences in participant selection and the timing of NIV.

Figures and Tables -
Summary of findings 2. Summary of Findings Table ‐ Chronic non‐invasive ventilation compared to standard treatment following a severe exacerbation for people with severe COPD
Navigate to table in Review
Table 1. Absolute changes in the outcomes after three and 12 months in stable COPD, separate for the NIV and the control group

 

3 months

12 months

Outcome

n

Change in NIV group

n

Change in control group

n

Change in NIV group

n

Change in control group

PaO2 (kPa)

131

0.17 ± 0.95

140

‐0.07 ± 0.99

88

‐0.03 ± 1.07

83

‐0.06 ± 1.12

PaCO2 (kPa)

238

‐0.81 ± 1.01

237

‐0.17 ± 0.93

123

‐0.46 ± 1.10

109

‐0.05 ± 0.98

6MWD (m)

167

32.2 ± 73.9

163

16.0 ± 75.6

67

15.0 ± 105.9

67

‐10.7 ± 94.8

SRI, summary score

108

7.5 ± 12.5

109

2.2 ± 11.0

42

6.1 ± 14.2

42

0.5 ± 9.3

SGRQ, total score

20

‐3.2 ± 9.3

20

‐0.2 ± 8.9

62

‐0.2 ± 14.0

53

‐0.9 ± 15.3

FEV1 (l)

194

0.01 ± 0.15

189

0.01 ± 0.14

125

‐0.01 ± 0.17

111

‐0.01 ± 0.15

FVC (l)

196

0.05 ± 0.40

188

0.05 ± 0.38

125

‐0.07 ± 0.43

110

‐0.05 ± 0.46

RV/TLC (%)

93

‐0.6 ± 6.8

91

‐0.05 ± 6.8

101

0.6 ± 8.0

94

0.07 ± 9.0

PEmax (cmH2O)

60

8.7 ± 24.6

63

3.0 ± 23.6

16

17.4 ± 46.2

22

13.9 ± 36.1

PImax (cmH2O)

76

9.2 ± 16.7

76

2.2 ± 12.5

45

8.0 ± 16.9

47

2.0 ± 16.5

TDI

84

1.2 ± 2.7

94

0.5 ± 2.7

0

—

0

—

(m)MRC

68

‐0.7 ± 1.0

87

‐0.4 ± 0.9

49

‐0.6 ± 1.2

54

‐0.1 ± 1.0

Sleep efficiency

13

7.0 ± 23.2

11

12.6 ± 16.8

0

—

0

—

COPD exacerbationsa

24

0.5 (0 to 1.8)

32

1.0 (0 to 2.0)

18

3.0 (2.8 to 4.0)

27

3.0 (2.0 to 5.0)

COPD hospitalisationsa

137

0 (0 to 0.5)

131

0 (0 to 1.0)

71

0 (0 to 2.0)

67

0 (0 to 2.0)

Data are based on the studies that provided the IPD, and are presented as mean ± standard deviation or median (interquartile range). 
a absolute number of exacerbations or hospitalisations.

6MWD: 6‐minute walking distance; CI: confidence interval; COPD: Chronic Obstructive Pulmonary Disease; FEV1: forced expiratory volume in 1 second; FVC: forced vital capacity; (m)MRC: (modified) Medical Research Council dyspnoea scale, a decrease indicates an improvement in dyspnoea; NIV: noninvasive ventilation; PaCO2: arterial carbon dioxide tension; PaO2: arterial oxygen tension; PEmax: peak expiratory pressure; PImax: peak inspiratory pressure; RV/TLC: residual volume/total lung capacity; SGRQ: St. George's Respiratory Questionnaire, a decrease indicates an improved health‐related quality of life; SRI: Severe Respiratory Insufficiency questionnaire, an increase indicates an improved health‐related quality of life; TDI: transitional dyspnoea index, a positive value indicates an improvement in dyspnoea.

Figures and Tables -
Table 1. Absolute changes in the outcomes after three and 12 months in stable COPD, separate for the NIV and the control group
Navigate to table in Review
Table 2. Absolute changes in the outcomes after three and 12 months in post‐exacerbation COPD, separate for the NIV and the control group

 

3 months

12 months

Outcome

n

Change in NIV group

n

Change in control group

n

Change in NIV group

n

Change in control group

PaO2 (kPa)

120

0.08 ± 2.33 

114

0.19 ± 2.12 

86

0.21 ± 1.96 

84

0.48 ± 2.10 

PaCO2 (kPa)

127

‐0.99 ± 1.24 

114

‐0.59 ± 1.19 

90

‐1.12 ± 1.15 

85

‐0.61 ± 1.20 

SRI, summary score

115

4.7 ± 10.6 

101

1.7 ± 11.7 

85

4.8 ± 11.9 

77

1.7 ± 11.8 

FEV1 (l)

108

0.02 ± 0.19 

93

0.04 ± 0.20 

82

0.01 ± 0.21 

75

0.03 ± 0.22 

FVC (l)

108

‐0.01 ± 0.66 

93

0.12 ± 0.63 

83

‐0.01 ± 0.64 

75

0.08 ± 0.68 

COPD exacerbations a

72

0 [0 to 1.0]

67

0 [0 to 1.0] 

95

2.0 [1.0 to 4.0]

86

2.0 [1.0 to 4.0] 

COPD hospitalisations a

0

—

0

—

100

0 (0 to 2.0) 

93

0 (0 to 2.0)

Data are based on the studies that provided the IPD, and are presented as mean ± standard deviation or median (interquartile range). 

a absolute number of exacerbations or hospitalisations.

CI: confidence interval; COPD: Chronic Obstructive Pulmonary Disease; FEV1: forced expiratory volume in 1 second; FVC: forced vital capacity; NIV: noninvasive ventilation; PaCO2: arterial carbon dioxide tension; PaO2: arterial oxygen tension; SRI: Severe Respiratory Insufficiency questionnaire, an increase indicates an improved health‐related quality of life.

Figures and Tables -
Table 2. Absolute changes in the outcomes after three and 12 months in post‐exacerbation COPD, separate for the NIV and the control group
Navigate to table in Review
Comparison 1. Comparison 1: NIV vs standard care in stable COPD

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

1.1 Stable COPD: PaO2 Show forest plot

11

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.1.1 follow‐up: 3 months

10

308

Mean Difference (IV, Random, 95% CI)

0.31 [0.14, 0.48]

1.1.2 follow‐up: 12 months

3

171

Mean Difference (IV, Random, 95% CI)

0.08 [‐0.37, 0.53]

1.2 Stable COPD: PaCO2 Show forest plot

13

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.2.1 follow‐up: 3 months

12

512

Mean Difference (IV, Random, 95% CI)

‐0.48 [‐0.72, ‐0.25]

1.2.2 follow‐up: 12 months

4

232

Mean Difference (IV, Random, 95% CI)

‐0.35 [‐0.68, ‐0.02]

1.3 Stable COPD: 6‐minute walking distance Show forest plot

9

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.3.1 follow‐up: 3 months

8

330

Mean Difference (IV, Random, 95% CI)

7.85 [‐5.52, 21.23]

1.3.2 follow‐up: 12 months

3

134

Mean Difference (IV, Random, 95% CI)

23.50 [‐3.03, 50.03]

1.4 Stable COPD: HRQL Show forest plot

7

Std. Mean Difference (IV, Random, 95% CI)

Subtotals only

1.4.1 follow‐up: 3 months

5

257

Std. Mean Difference (IV, Random, 95% CI)

0.42 [0.17, 0.67]

1.4.2 follow‐up: 12 months

4

199

Std. Mean Difference (IV, Random, 95% CI)

0.18 [‐0.24, 0.59]

1.5 Stable COPD: FEV1 Show forest plot

13

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.5.1 follow‐up: 3 months

11

420

Mean Difference (IV, Random, 95% CI)

‐0.00 [‐0.04, 0.03]

1.5.2 follow‐up: 12 months

4

236

Mean Difference (IV, Random, 95% CI)

0.00 [‐0.04, 0.04]

1.6 Stable COPD: FVC Show forest plot

13

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.6.1 follow‐up: 3 months

11

421

Mean Difference (IV, Random, 95% CI)

0.01 [‐0.06, 0.08]

1.6.2 follow‐up: 12 months

4

235

Mean Difference (IV, Random, 95% CI)

‐0.02 [‐0.13, 0.09]

1.7 Stable COPD: RV/%TLC Show forest plot

5

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.7.1 follow‐up: 3 months

3

184

Mean Difference (IV, Random, 95% CI)

‐0.53 [‐2.77, 1.71]

1.7.2 follow‐up: 12 months

4

195

Mean Difference (IV, Random, 95% CI)

0.27 [‐1.98, 2.53]

1.8 Stable COPD: PEmax Show forest plot

6

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.8.1 follow‐up: 3 months

6

160

Mean Difference (IV, Random, 95% CI)

9.03 [‐1.51, 19.58]

1.8.2 follow‐up: 12 months

1

38

Mean Difference (IV, Random, 95% CI)

3.50 [‐23.70, 30.70]

1.9 Stable COPD: PImax Show forest plot

8

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.9.1 follow‐up: 3 months

7

192

Mean Difference (IV, Random, 95% CI)

6.96 [3.29, 10.62]

1.9.2 follow‐up: 12 months

2

92

Mean Difference (IV, Random, 95% CI)

7.04 [‐12.29, 26.37]

1.10 Stable COPD: Transition Dyspnoea Index Show forest plot

3

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.10.1 follow‐up: 3 months

3

178

Mean Difference (IV, Random, 95% CI)

0.77 [0.01, 1.53]

1.11 Stable COPD: (modified) Medical Research Council dyspnoea scale Show forest plot

4

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.11.1 follow‐up: 3 months

4

155

Mean Difference (IV, Random, 95% CI)

‐0.29 [‐0.59, 0.01]

1.11.2 follow‐up: 12 months

2

103

Mean Difference (IV, Random, 95% CI)

‐0.49 [‐0.90, ‐0.08]

1.12 Stable COPD: sleep efficiency Show forest plot

3

Mean Difference (IV, Random, 95% CI)

Subtotals only

1.12.1 follow‐up: 3 months

3

24

Mean Difference (IV, Random, 95% CI)

‐2.86 [‐8.76, 3.04]

1.13 Stable COPD: exacerbations Show forest plot

2

Odds Ratio (M‐H, Random, 95% CI)

Subtotals only

1.13.1 follow‐up: 3 months

2

100

Odds Ratio (M‐H, Random, 95% CI)

0.76 [0.34, 1.69]

1.13.2 follow‐up: 12 months

2

89

Odds Ratio (M‐H, Random, 95% CI)

0.72 [0.25, 2.05]

1.14 Stable COPD: hospitalisations Show forest plot

5

Odds Ratio (M‐H, Random, 95% CI)

Subtotals only

1.14.1 follow‐up: 3 months

5

312

Odds Ratio (M‐H, Random, 95% CI)

0.64 [0.26, 1.60]

1.14.2 follow‐up: 12 months

4

182

Odds Ratio (M‐H, Random, 95% CI)

1.06 [0.46, 2.44]

1.15 Stable COPD: all‐cause mortality Show forest plot

3

405

Hazard Ratio (IV, Random, 95% CI)

0.78 [0.58, 1.05]

1.16 PaCO2 after 3 months ‐ subgroup PaCO2 (stable COPD) Show forest plot

12

512

Mean Difference (IV, Random, 95% CI)

‐0.48 [‐0.72, ‐0.25]

1.16.1 Baseline PaCO2 <7.3 kPa

8

217

Mean Difference (IV, Random, 95% CI)

‐0.25 [‐0.43, ‐0.06]

1.16.2 Baseline PaCO2 ≥7.3 kPa

4

295

Mean Difference (IV, Random, 95% CI)

‐0.82 [‐1.10, ‐0.53]

1.17 PaCO2 after 3 months ‐ subgroup IPAP (stable COPD) Show forest plot

12

512

Mean Difference (IV, Random, 95% CI)

‐0.48 [‐0.72, ‐0.25]

1.17.1 mean IPAP <18 cmH2O

8

241

Mean Difference (IV, Random, 95% CI)

‐0.26 [‐0.45, ‐0.07]

1.17.2 mean IPAP ≥18 cmH2O

4

271

Mean Difference (IV, Random, 95% CI)

‐0.80 [‐1.12, ‐0.48]

1.18 PaCO2 after 3 months ‐ subgroup compliance (stable COPD) Show forest plot

12

512

Mean Difference (IV, Random, 95% CI)

‐0.48 [‐0.72, ‐0.25]

1.18.1 Mean compliance <5 hours per night

3

81

Mean Difference (IV, Random, 95% CI)

‐0.15 [‐0.50, 0.19]

1.18.2 Mean compliance ≥5 hours per night

9

431

Mean Difference (IV, Random, 95% CI)

‐0.60 [‐0.84, ‐0.36]
Figures and Tables -
Comparison 1. Comparison 1: NIV vs standard care in stable COPD
Navigate to table in Review
Comparison 2. Comparison 2: NIV vs standard care in post‐exacerbation COPD

Outcome or subgroup title

No. of studies

No. of participants

Statistical method

Effect size

2.1 Post‐exacerbation COPD: PaO2 Show forest plot

4

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.1.1 follow‐up: 3 months

3

234

Mean Difference (IV, Random, 95% CI)

‐0.08 [‐0.58, 0.41]

2.1.2 follow‐up: 12 months

3

170

Mean Difference (IV, Random, 95% CI)

‐0.24 [‐0.81, 0.34]

2.2 Post‐exacerbation COPD: PaCO2 Show forest plot

3

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.2.1 follow‐up: 3 months

3

241

Mean Difference (IV, Random, 95% CI)

‐0.42 [‐0.72, ‐0.12]

2.2.2 follow‐up: 12 months

3

175

Mean Difference (IV, Random, 95% CI)

‐0.53 [‐0.88, ‐0.19]

2.3 Post‐exacerbation COPD: HRQL Show forest plot

2

Std. Mean Difference (IV, Random, 95% CI)

Subtotals only

2.3.1 follow‐up: 3 months

2

216

Std. Mean Difference (IV, Random, 95% CI)

0.27 [0.01, 0.54]

2.3.2 follow‐up: 12 months

2

162

Std. Mean Difference (IV, Random, 95% CI)

0.27 [‐0.04, 0.58]

2.4 Post‐exacerbation COPD: admission‐free survival Show forest plot

2

317

Hazard Ratio (IV, Random, 95% CI)

0.69 [0.39, 1.23]

2.5 Post‐exacerbation COPD: FEV1 Show forest plot

2

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.5.1 follow‐up: 3 months

2

201

Mean Difference (IV, Random, 95% CI)

‐0.03 [‐0.09, 0.03]

2.5.2 follow‐up: 12 months

2

157

Mean Difference (IV, Random, 95% CI)

‐0.03 [‐0.09, 0.03]

2.6 Post‐exacerbation COPD: FVC Show forest plot

2

Mean Difference (IV, Random, 95% CI)

Subtotals only

2.6.1 follow‐up: 3 months

2

202

Mean Difference (IV, Random, 95% CI)

‐0.13 [‐0.30, 0.04]

2.6.2 follow‐up: 12 months

2

158

Mean Difference (IV, Random, 95% CI)

‐0.11 [‐0.32, 0.10]

2.7 Post‐exacerbation COPD: exacerbations Show forest plot

3

Odds Ratio (M‐H, Random, 95% CI)

Subtotals only

2.7.1 follow‐up: 3 months

1

139

Odds Ratio (M‐H, Random, 95% CI)

0.93 [0.48, 1.81]

2.7.2 follow‐up: 12 months

3

181

Odds Ratio (M‐H, Random, 95% CI)

0.67 [0.32, 1.41]

2.8 Post‐exacerbation COPD: hospitalisations Show forest plot

3

Odds Ratio (M‐H, Random, 95% CI)

Subtotals only

2.8.1 follow‐up: 12 months

3

193

Odds Ratio (M‐H, Random, 95% CI)

0.63 [0.19, 2.09]

2.9 Post‐exacerbation COPD: all‐cause mortality Show forest plot

2

317

Hazard Ratio (IV, Random, 95% CI)

0.98 [0.74, 1.29]
Figures and Tables -
Comparison 2. Comparison 2: NIV vs standard care in post‐exacerbation COPD
Navigate to table in Review