Self‐management interventions including action plans for exacerbations versus usual care in patients with chronic obstructive pulmonary disease

Summary of findings for the main comparison. Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD

Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD
Patient or population: patients with chronic obstructive pulmonary disease (COPD) Setting: hospital, outpatient clinic, primary care, home‐based Intervention: self‐management interventions including action plans for COPD exacerbations Comparison: usual care
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
Outcomes	Risk with usual care	Risk with self‐management interventions including action plans for exacerbations	Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
Health‐related quality of life (HRQoL) assessed with: St. George's Respiratory Questionnaire adjusted total score Scale from: 0 to 100 follow up: 12 months	The mean HRQoL ranged from 37.7 to 70.4 points	MD 2.69 points lower (4.49 lower to 0.9 lower)	‐	1582 (10 RCTs)	⊕⊕⊕⊕ HIGH	Lower score indicates better health‐related quality of life.
Respiratory‐related hospital admissions assessed with: number of patients with at least one respiratory‐related hospital admission follow up: range 6 months to 24 months	312 per 1,000	238 per 1,000 (188 to 298)	OR 0.69 (0.51 to 0.94)	3,157 (14 RCTs)	⊕⊕⊕⊝ MODERATE ¹
All‐cause hospital admissions assessed with: number of patients with at least one all‐cause hospital admission follow up: range 6 months to 12 months	427 per 1000	356 per 1,000 (287 to 434)	OR 0.74 (0.54 to 1.03)	2,467 (10 RCTs)	⊕⊕⊕⊝ MODERATE ²
All‐cause mortality assessed with: number of all‐cause deaths follow up: range 3 months to 24 months	102 per 1000	107 per 1,000 (74 to 153)	OR 1.06 (0.71 to 1.59)	3,296 (16 RCTs)	⊕⊕⊕⊝ MODERATE³	Pooled risk difference of 0.0019 (95% CI ‐0.0225 to 0.0263).
Respiratory‐related mortality assessed with: number of respiratory‐related deaths follow up: range 3 months to 24 months	48 per 1000	89 per 1,000 (57 to 136)	OR 1.94 (1.20 to 3.13)	1,219 (7 RCTs)	⊕⊝⊝⊝ VERY LOW ⁴	Pooled risk difference of 0.028 (95% CI 0.0049 to 0.0511).
Dyspnoea assessed with: (modified) Medical Research Council Dyspnoea Scale Scale from: 0 to 4 follow up: 12 months	The mean dyspnoea ranged from 2.4 to 2.6	MD 0.63 lower (1.44 lower to 0.18 higher)	‐	217 (3 RCTs)	⊕⊕⊝⊝ LOW ⁵	Lower score indicates improvement in dyspnoea.
COPD exacerbations assessed with: number of COPD exacerbations per patient follow up: range 3 months to 24 months ⁷	The mean COPD exacerbations ranged from 1.13 to 4.3	MD 0.01 higher (0.28 lower to 0.29 higher)	‐	740 (4 RCTs)	⊕⊕⊕⊝ MODERATE ⁶
Courses of oral steroids assessed with: number of patients who used at least one course of oral steroids follow up: 12 months	497 per 1000	812 per 1000 (352 to 972)	OR 4.38 (0.55 to 34.91)	963 (4 RCTs)	⊕⊕⊝⊝ LOW ⁸
*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; OR: Odds ratio; RCT: randomised controlled trial
GRADE Working Group grades of evidence High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: 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 quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ Heterogeneity was substantial (I² = 57%) (inconsistency ‐1). ² Heterogeneity was substantial (I² = 62%) (inconsistency ‐1). ³ Imprecision of pooled effect size (imprecision ‐1). ⁴ Explorative meta‐analysis. Four studies (Gallefoss 1999; Kheirabadi 2008; Ninot 2011; Tabak 2014) with no events and a high risk of bias for three studies (Bucknall 2012; Tabak 2014; Titova 2015) for incomplete outcome data and selective reporting. Two studies (Bucknall 2012; Fan 2012) dominated the overall effect and heavily influenced the OR (risk of bias ‐1, inconsistency ‐1, imprecision ‐1). ⁵ Heterogeneity was high (I² =86%). Only three studies were included in this meta‐analysis (inconsistency ‐1, imprecision ‐1). ⁶ Only four studies were included in this meta‐analysis (imprecision ‐1). ⁷ COPD exacerbations were defined as worsening of respiratory symptoms beyond normal day‐to‐day variations that required treatment with bronchodilators, oral steroids and/or antibiotics ⁸ Heterogeneity was high (I² = 94%). Only four studies were included in this meta‐analysis (inconsistency ‐1, imprecision ‐1).

Background

Description of the condition

Chronic obstructive pulmonary disease (COPD) is characterised by respiratory symptoms that are caused predominantly by persistent airflow limitation, which is usually progressive. COPD is associated with an enhanced chronic inflammatory response in the lung to noxious particles or gases (GOLD 2017). Many people with COPD experience increasing functional impairment and progressive loss of quality of life over many years (Carrasco Garrido 2006; Celli 2007; Heyworth 2009). Acute exacerbations of COPD (AECOPD), defined as acute deterioration in respiratory health, contribute to functional impairment and risk of mortality in individual people with the disease (Celli 2007; Seemungal 1998).

COPD leads to more than six million deaths annually and has been predicted to be the third leading cause of death worldwide (GOLD 2017; Lozano 2012). Increased mortality is driven mainly by the expanding global epidemic of smoking, reduced mortality from other common causes of death (e.g., ischaemic heart disease, infectious disease) and increasing age of the world population (GOLD 2017). COPD is also a leading cause of morbidity. In 2010, COPD was the fifth largest cause of years of life lived with disability (Vos 2012). Apart from personal distress, COPD confers a substantial and increasing economic and social burden on society (GOLD 2017), with its exacerbations accounting for most direct costs (Toy 2010).

Description of the intervention

Wagner's Chronic Care model (Wagner 1998) suggested improvement of chronic illness care through health systems that: 1) have well‐developed processes and incentives for making change in the care delivery system; 2) assure behaviourally sophisticated self‐management support that gives priority to increasing a person's confidence and skills so they can be the ultimate manager of their illness; 3) re‐organise team function and practice systems (e.g., appointments and follow‐up) to meet the needs of people who are chronically ill; 4) develop and implement evidence‐based guidelines that are supported through provider education, reminders, and increased interaction between generalists and specialists; 5) enhance information systems to facilitate the development of disease registries, tracking systems, and reminders and to give feedback on performance. Patient education, written management plans, access to 24/7 healthcare, and a case manager are required to reduce healthcare utilisation (Wagner 1998).

Self‐management interventions are defined as structured interventions for individuals aimed at improvement in self‐health behaviours and self‐management skills (Lorig 2003). Lorig 2003 indicated that a self‐management programme should ideally include training with feedback to improve the following patient skills: problem solving, decision making, resource utilisation, formating patient‐provider partnerships, action planning and self‐tailoring. Mastery, modelling, interpretation of symptoms and social persuasion skills are believed to contribute to enhanced self‐efficacy (Lorig 2003). People progressively achieve greater confidence in (self) managing their health, and this is a powerful factor in inducing new and sustaining behaviours that provide perceived benefit (Bourbeau 2004; Lorig 2003).

Self‐management has been proposed as an essential part of disease management targeted to helping people develop skills to manage disease more effectively. This is especially important in people with chronic disease (e.g., COPD, for which the individual is responsible for day‐to‐day care for the duration of the illness) (Lorig 2003). COPD self‐management interventions are associated with reduced duration of exacerbations and hospitalisations and decreased healthcare costs, as well as improved health‐related quality of life (HRQoL) (Effing 2009a; Rice 2010; Zwerink 2014). COPD self‐management training aims to help people acquire and improve skills needed to carry out disease‐specific medical regimens (Bourbeau 2009; Effing 2012). Self‐management training also guides changes in health behaviour and provides emotional support for optimal function of people with COPD and control of their disease (Bourbeau 2009; Effing 2012). Self‐management training is considered to be an increasingly important component of treatment and management of COPD. Training should occur as an interactive and iterative process aimed at sustained behavioural change and to instil confidence to recognise when an exacerbation is starting and to self manage it effectively and safely (Bourbeau 2009). Self‐management will not be successful without effective co‐operation between patient and healthcare providers (Bodenheimer 2002). Ongoing case manager support is recognised as an additional component required to achieve effective and safe self‐management (Effing 2012).

Recently, an international expert group reached consensus regarding a conceptual definition for a COPD self‐management intervention (Effing 2016). Self‐management interventions should be structured but personalised and often multi‐component, with goals of motivating, engaging and supporting the patients to positively adapt their behaviour(s) and develop skills to better manage their disease. Our review inclusion criteria were developed in line with this definition.

Action planning is a frequently applied planning technique in generic self‐management programmes and adopted to change behaviour (Hagger 2014; Webb 2010). COPD exacerbation action plans are disease‐specific and considered to be an intrinsic part of COPD self‐management interventions (Effing 2012; Zwerink 2014). People with COPD are trained to use exacerbation action plans if they experience worsening of respiratory symptoms. Appropriate actions can include contacting a healthcare provider for support or initiating self‐treatment (Wood‐Baker 2006). Furthermore, written action plans can include instructions regarding, for example, maintenance treatment.

How the intervention might work

Using action plans for exacerbations of COPD within a self‐management intervention provides training for people with COPD to recognise symptoms earlier, accelerate the initiation of appropriate treatment and lead to better control of deteriorating symptoms. This may lead to improved HRQoL, reduced exacerbation duration and hospitalisations, and decreased healthcare costs for people with COPD.

Why it is important to do this review

A Cochrane Review on COPD self‐management concluded that self‐management is associated with improved HRQoL, reduced respiratory‐related and all‐cause hospitalisations and improved dyspnoea (Zwerink 2014). Subgroup analyses indicate that a standardised exercise component in self‐management interventions did not change the effects of self‐management interventions on HRQoL and respiratory‐related hospital admissions. However, the review could not reveal the effective components within self‐management interventions, not least because of heterogeneity among interventions, study populations, follow‐up time and outcome measures (Zwerink 2014). In recently published individual patient data (IPD) meta‐analyses on the effectiveness of COPD self‐management the included self‐management interventions also differed from each other in terms of dose, mode and content (Jonkman 2016b). Because of the very frequent use of action plans for exacerbations in the included studies, subanalyses on the use of action plans could not be performed by Zwerink 2014. As COPD action plans are currently considered as an intrinsic part of COPD self‐management interventions, in the current new review written action plans for AECOPD were included as part of the self‐management intervention.

Since the publication of Zwerink 2014, several studies have been published and new opinions have been raised regarding the limitations and content of COPD self‐management interventions with exacerbation action plans for people with COPD. So far, the evidence regarding COPD action plans is somewhat contradictory. After two years of follow‐up, a self‐management programme including action plans for the self‐treatment of exacerbations in people with COPD without significant comorbidities resulted in reduced exacerbation duration, exacerbation severity and healthcare utilisation (Zwerink 2016). Furthermore, a review showed that the use of action plans with a single short educational component along with ongoing support, but without a comprehensive self‐management programme, reduces in‐hospital healthcare utilisation and increases treatment of COPD exacerbations (Howcroft 2016). This review showed a small improvement in HRQoL from action plans compared to usual care but it was unlikely to increase or decrease mortality (Howcroft 2016). As a result of using individualised action plans and ongoing support, the impact of exacerbations on health status decreased and the recovery of an exacerbation might be accelerated (Trappenburg 2011). A study evaluating the efficacy of a comprehensive care management programme in reducing the risk for COPD hospitalisations with COPD‐specific action plans was prematurely terminated (Fan 2012) because of significantly higher mortality rates in the intervention group. No definitive explanation for these study outcomes has emerged, and they conflict with the positive study outcomes of another highly comparable self‐management study by Rice 2010. The significantly higher mortality rates in the intervention group reported by Fan 2012 may be partly explained by the use of COPD‐specific action plans for people with COPD and comorbidities. A single‐centre RCT that included nurse support identified only 42% of the intervention group as successful self‐managers. This group of successful self‐managers had a significantly reduced risk of hospital re‐admissions (Bucknall 2012). This study implied that not all people with COPD derive benefit from a COPD self‐management intervention. All COPD self‐management interventions discussed above have included a COPD exacerbation action plan as a key intervention component, underlining that these action plans are currently seen as an intrinsic part of COPD self‐management interventions. Nevertheless, these studies show contradictory results. We assessed the effectiveness of COPD self‐management interventions that include action plans for AECOPD compared with usual care for this review.

Objectives

Methods

Criteria for considering studies for this review

Types of studies

We considered randomised controlled trials (RCTs) reported in full text, those published as abstracts only and unpublished data from RCTs.

Types of participants

We included studies that included participants diagnosed with Chronic Obstructive Pulmonary Disease (COPD) according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) classification criteria (GOLD 2017); people with a post‐bronchodilator forced expiratory volume in one second (FEV₁)‐to‐forced vital capacity (FVC) ratio < 0.70. Participants with primary diagnoses of asthma were excluded.

Types of interventions

We included trials comparing COPD self‐management interventions that included a written action plan for acute exacerbations of COPD (AECOPD) versus usual care. For this review, an action plan refers to specific behaviour to be initiated when respiratory symptoms deteriorate; the plan needed to describe when, where and how one should act. An action plan is an agreed strategy by which people act appropriately when symptoms deteriorate (indicating the start of a COPD exacerbation), for example, by contacting a healthcare provider for support or initiating self‐treatment. It may also include maintenance treatment and advice to avoid situations in which viral infection might be prevalent.

The self‐management intervention needed to include formal training on how and when to use an action plan for AECOPD. To be eligible for inclusion, the formal training programme had to be an iterative process between participants and healthcare provider(s) in which feedback was provided to develop participants’ self‐management skills (e.g., how and when to use an action plan for AECOPD). Training should ideally include techniques directed to achieving behavioural change (Michie 2013). The intervention could also include other components that were directed to achieving behaviour change (e.g., smoking behaviour, exercise or physical activity, diet, use of maintenance medication and correct device use, coping with breathlessness). The intervention content could be delivered to participants verbally, in writing (hard copy or digital) or via audiovisual media.

Disease management programmes classified as pulmonary rehabilitation or exercise classes offered in a hospital, at a rehabilitation centre or in a community‐based setting were excluded to avoid possible overlap with pulmonary rehabilitation as much as possible. The study was considered if the participants were randomised and allocated to self‐management or usual care after pulmonary rehabilitation. The study was excluded if randomisation was performed before pulmonary rehabilitation. Home‐based (unsupervised) exercise programmes that included action plans for AECOPD were included, as these studies asked a more active role of participants and were more clearly aimed at development of self‐management skills compared to supervised exercise programmes.

As the definition, content and focus of COPD self‐management training in particular, and of COPD treatment in general, have dramatically changed over the past 20 years, we excluded studies published before 1995. We included studies that were published in full‐text and excluded abstracts if there was no additional information available from the study authors.

Usual care differs significantly among countries and healthcare systems, and sometimes elements of self‐management interventions were included as part of usual care. We defined usual care as de facto routine clinical care.

Types of outcome measures

Primary outcomes

Health‐related quality of life (HRQoL).
Respiratory‐related hospital admissions.

Secondary outcomes

Number of all‐cause hospital admissions.
Use of (other) healthcare facilities (e.g., number of emergency department (ED) visits, number of all‐cause and respiratory‐related hospitalisation days in total and per patient, general practitioner (GP), number of nurse and specialist visits).
Rescue medication use.
Health status.
Number of COPD exacerbations.
All‐cause mortality.
Self‐efficacy.
Days lost from work.

Reporting one or more of the listed outcomes was not an inclusion criterion for our review. We intended to divide COPD exacerbations into those based on COPD symptom scores (e.g., symptom diary), courses of oral corticosteroids and courses of antibiotics.

Search methods for identification of studies

Electronic searches

We identified studies from the Cochrane Airways Trials Register, which is maintained by the Information Specialist for the Group.

The Cochrane Airways Trials Register contains studies identified from several sources:

Monthly searches of the Cochrane Central Register of Controlled Trials (CENTRAL), through the Cochrane Register of Studies Online (crso.cochrane.org);
Weekly searches of MEDLINE Ovid SP 1946 to date;
Weekly searches of Embase Ovid SP 1974 to date;
Monthly searches of PsycINFO Ovid SP 1967 to date;
Monthly searches of CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature) 1937 to date;
Monthly searches of AMED EBSCO (Allied and Complementary Medicine); and
Handsearches of the proceedings of major respiratory conferences.

Studies contained in the Trials Register are identified through search strategies based on the scope of Cochrane Airways. Details of these strategies, as well as a list of handsearched conference proceedings are in Appendix 1. See Appendix 2 for search terms used to identify studies for this review.

We searched the Cochrane Airways Trials Register from 1995 to May 2016, with no restriction on language of publication.

We contacted the authors of included studies to ask for further information, if needed.

Searching other resources

We checked reference lists of all primary studies and reviewed articles for additional references. We searched for additional trials using ClinicalTrials.gov and the WHO International Clinical Trials Registry Platform (WHO ICTRP, www.who.int/ictrp/en/ databases).

Data collection and analysis

Selection of studies

Two review authors (AL and TE) independently assessed titles and abstracts of all references retrieved. Subsequently, two review authors (AL and TE or MB) independently reviewed full‐text versions of potentially relevant reports, assessed eligibility for inclusion and resolved disagreements by discussion with the third review author (TE or MB).

Data extraction and management

Two review authors (AL and TE or MB) independently assessed trial quality and extracted the following data from included studies: relevant outcome measures; sample size; demographics of included participants; disease severity; setting, duration and content of the intervention and potential effect modifiers. We used standard data extraction forms and spreadsheets. We completed a data extraction form for study characteristics and outcome data that was piloted on two studies in the review.

We noted in Characteristics of included studies tables whether outcome data were reported in a useable way. We resolved disagreements by reaching consensus or by involving a third (TE or MB) or fourth review author (JP or PV). Data were transferred into the Review Manager (RevMan) 5.3 (Review Manager 2014) file (AL) and double‐checked for accuracy by comparing data presented in the systematic review versus data in the study reports (TE).

Assessment of risk of bias in included studies

Two review authors (AL and TE or MB) independently assessed the risk of bias according to recommendations outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) for the following items.

Random sequence generation.
Allocation concealment.
Blinding of participants and personnel.
Blinding of outcome assessment.
Incomplete outcome data.
Selective reporting.
Other potential sources of bias.

For each included study we graded all listed domains to whether high, low or unclear risk of bias was present (AL and TE or MB). An unclear risk indicated that there was insufficient detail of what happened in the study; that what happened in the study was known but the risk of bias was unknown; or that an entry was not relevant to the study at hand. Each judgement of risk of bias is supported by a short description of what was reported to have happened in the specific study. The grade of each potential bias from the included study together with a quote from the study report and justification for our judgement is reported in 'Risk of bias' tables. In the case of cluster‐RCTs, we assessed the risk of recruitment bias, risk of bias for baseline imbalance, risk of bias due to loss of clusters, risk of bias due to incorrect analysis and publication bias. We resolved disagreements by discussion or with involvement of another review author (JP or PV).

Assessment of bias in conducting the systematic review

We conducted the review according to the published protocol and reported deviations from it in the 'Differences between protocol and review' section of the systematic review.

Measures of treatment effect

We analysed the results of studies using random‐effects modelling in RevMan (Review Manager 2014). We used forest plots to compare results across trials. We expressed the results of each RCT as odds ratios (ORs) with corresponding 95% confidence intervals (95% CIs) for dichotomous outcomes, and as mean differences (MDs) or standardised mean differences (SMDs) for continuous outcomes. For primary analyses, we used the calculator tool in RevMan along with information from adjusted scores (analysis of co‐variance (ANCOVA)), change from baseline scores or final scores to create a single forest plot. We used the calculator tool with the generic inverse variance method for dichotomous or continuous data to allow transformation from data on effect sizes, CIs and standard errors (SE) to data required by RevMan to create forest plots with relative risks (RRs) or mean differences (MDs). We determined the clinical relevance of treatment effects by using the minimal clinically important difference (MCID), when available. If possible, numbers needed to treat for an additional beneficial outcome (NNTB) were calculated for both respiratory‐related and all‐cause hospital admissions using pooled ORs and control group data from individual studies within the meta‐analysis to obtain study‐specific NNTB, with Visual Rx 3 (Cates).

Unit of analysis issues

The participant was the unit of analysis for included RCTs. We intended to include cluster‐RCTs with the cluster as the unit of analysis. We had envisaged that for more recent studies, clusters would have been taken into account in the analyses. However, if this was not the case, we intended to adjust for the clusters.

Dealing with missing data

We contacted the study authors to obtain missing or incomplete outcome data where possible. If study authors did not respond, we made two further attempts to request missing data. If study authors did not respond after a third attempt, we analysed and described the available data and indicated that data were missing.

Assessment of heterogeneity

Variability among studies was explored by performing visual inspection and using the I² statistic (Higgins 2011). If we identified substantial heterogeneity (I² > 50%), we discussed possible explanations and critically reconsidered the appropriateness of a meta‐analysis. We used a random‐effects model, rather than a fixed‐effect model in meta‐analyses to account for heterogeneity.

Assessment of reporting biases

We explored possible reporting bias by assessing asymmetry in funnel plots to determine whether studies were selectively reported (see Assessment of risk of bias in included studies). We constructed a funnel plot when at least ten studies could be included.

Data synthesis

When appropriate, we performed meta‐analysis using RevMan. We considered a meta‐analysis when at least three studies reported sufficient data for the outcome. Because of the nature of the intervention, we expected to see clinical heterogeneity among studies. If pooling was possible, we performed meta‐analyses using the random‐effects model.

Summary of findings Table

Using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), we created a 'Summary of findings' (SoF) table that includes key information concerning the quality of evidence, the magnitude of effect of the self‐management intervention and the sum of available data on the main outcomes. We used the five GRADE (Grades of Recommendation, Assessment, Development and Evaluation) considerations regarding: 1) study limitations; 2) consistency of effect; 3) imprecision; 4) indirectness; and 5) publication bias, to assess the quality of a body of evidence as it relates to studies that contribute data to the meta‐analyses for pre‐specified outcomes. We used methods and recommendations described in Section 8.5 and Chapter 12 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) by using GRADEpro (GRADEpro GDT) software. We justified all decisions to downgrade or upgrade the quality of studies by using footnotes, and we provided comments to aid the reader's understanding of the review when necessary.

Subgroup analysis and investigation of heterogeneity

We considered subgroup analyses when at least three studies could be included in each subgroup. We intended to perform the following subgroup analyses to detect potential explanatory variables and determine whether outcomes differed in terms of the following:

Duration of follow‐up: fewer than 12 months of follow‐up after the start of the study versus 12 or more months of follow‐up after the start of the study. Shorter‐term and longer‐term effects of self‐management interventions including action plans might be different. In addition, we will perform explorative analyses by using different cut‐off points for follow‐up times (e.g., six months, 18 months).
Inclusion of participants in the acute phase: inclusion of participants with COPD in the acute unstable phase (with an acute exacerbation of COPD) versus inclusion of participants in the non‐acute stable phase (at least four weeks post exacerbation and six weeks post hospitalisation). Acute exacerbations may threaten self‐management improvements. Awareness of the clinical sequelae of acute exacerbations of COPD enables approaches such as early post‐exacerbation rehabilitation to mitigate its negative effects (Goldstein 2014).
Use of a standardised exercise programme as part of the intervention: use of an exercise component in self‐management versus no exercise component. Increased exercise capacity may result in better HRQoL and potentially fewer hospital admissions (McCarthy 2015).
Use of a smoking cessation programme in the intervention: smoking cessation component in self‐management versus no smoking cessation component. Smoking cessation may result in improved HRQoL (Cheruvu 2016; van Eerd 2016).
Self‐management as part of usual care: low‐level usual care versus high‐level usual care. Usual care differs significantly among countries and healthcare systems, and sometimes self‐management will already be included as part of usual care. We classified according to whether self‐management was likely to be part of usual care.

We used the formal test for subgroup interactions in RevMan (Review Manager 2014).

In addition, we have assessed the integration of 16 clusters of behavioural change techniques (BCTs) in an explorative subgroup analysis to promote uptake and optimal use of COPD‐specific self‐management behaviour patterns in the intervention:

Goals and planning
Feedback and monitoring
Social support
Shaping of knowledge
Natural consequences
Comparison of behaviours
Associations
Repetition and substitution
Comparison of outcomes
Reward and threat
Regulation
Antecedents
Identity
Scheduled consequences
Self‐belief; and
Covert learning.

The BCT taxonomy is a methodological tool for specifying intervention content (Michie 2013). The BCT taxonomy (version 1) published by Michie et al. (Michie 2013) describes 93 hierarchically clustered techniques in 16 clusters. The BCT must be an observable, replicable and irreducible component of an intervention designed to alter or redirect causal processes that regulate behaviour; that is, a technique that is proposed to be an "active ingredient" (Michie 2011). In this subgroup analysis, we classified interventions by their number of BCT taxonomy clusters ('lower or equal' versus 'higher' than the median of BCT clusters found in all included interventions) (Michie 2013).

In exploratory analyses, we assessed potential effect modifiers by participant and self‐management intervention levels (e.g., case manager support). We also aimed to collect information about the intention of the self‐management intervention and how it was delivered to participants.

Sensitivity analysis

We planned to carry out sensitivity analyses under different assumptions to investigate the robustness of effect sizes found in this review. Sensitivity analyses were performed to identify whether review findings were dependent on study characteristics, using random‐effects versus fixed‐effect modelling.

Results

Description of studies

See Characteristics of included studies.

Results of the search

Searches identified 1,811 titles and abstracts (Figure 1). In total, 255 potentially eligible articles about self‐management interventions including an action plan for acute exacerbations of chronic obstructive pulmonary disease (AECOPD) were identified, of which 22 studies (described in 30 articles) were included. One study (Österlund Efraimsson 2008) could not be included in the quantitative synthesis (meta‐analysis) because insufficient data were provided.

Figure 1

Study flow diagram

This review fully incorporates the results of searches conducted up to May 2016. A further nine reports were identified by a search update conducted in May 2017. However, these have not yet been incorporated into the results and will be addressed in the next update. See Characteristics of studies awaiting classification.

Included studies

All 22 included studies compared a self‐management intervention using an action plan for AECOPD with a usual care control group (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Jennings 2015; Khdour 2009; Kheirabadi 2008; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Rice 2010; Song 2014; Tabak 2014; Titova 2015). Twenty‐one included studies were parallel randomised controlled trials (RCTs) and one was a cluster‐RCT (Rea 2004). Details of participant and intervention characteristics (Table 1 and Table 2, respectively) were tabulated. We structured both tables according to potential effect modifiers on participant and self‐management intervention levels (e.g., lost to follow‐up, duration and delivery of intervention).

Table 1. Participant characteristics, included studies

Study	Included participants (N)		Lost to follow‐up (%)		Age (years; mean (SD))		Gender (% male)		FEV₁ (% predicted unless stated otherwise (SD))
Study	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care
Bischoff 2012	55	55	10.9	20.0	65.5 (11.5)	63.5 (10.3)	67.0	51.0	66.3 (16.5)	67.0 (18.0)
Bösch 2007	38	12	21.1	8.3	63.8 (8.4)	64.6 (6.8)	63.0% of completers		45.9 (17.5)	47.8 (16.9)
Bourbeau 2003	96	95	10.4	16.8	69.4 (6.5)	69.6 (7.4)	52.0	59.0	1.0 L (0.33)	0.98 (0.31)
Bucknall 2012	232	232	9.1	13.8	70.0 (9.3)	68.3 (9.2)	38.0	35.0	41.2 (13.4)	39.8 (13.8)
Casas 2006	65	90	26.2	20.0	70 (9.0)	72 (9.0)	77.0	88.0	43 (20)	41 (15)
Garcia‐Aymerich 2007	44	69	52.3	40.6	72 (10.0)	73 (9.0)	75.0	93.0	1.2 L (IQR 0.8 to 1.4)	1.0 L (IQR 0.8‐1.5)
Fan 2012	209	217	3.8^a; 51.7^b	4.6^a; 50.2^b	66.2 (8.4)	65.8 (8.2)	97.6	96.3	38.2 (14.3)	37.8 (14.5)
Gallefoss 1999	31	31	16.0	13.0	57 (9.0)	58 (10.0)	48.0	52.0	59 (9)	56 (11)
Hernández 2015	71	84	23.9	34.5	73 (8.0)	75 (9.0)	83.0	86.0	41 (19)	44 (20)
Jennings 2015	93	79	0	0	64.9 (10.9)	64.4 (10.5)	43.1	46.8	44.1 (23.1)	48.3 (22.2)
Khdour 2009	86	87	17.4	17.2	65.6 (10.1)	67.3 (9.2)	44.2	43.7	52.0 (15.9)	52 (17.8)
Kheirabadi 2008	21	21	0	0	56.6 (5.7)	56.2 (4.1)	61.9	76.2	N/A	N/A
Martin 2004	44	49	20.5	8.2	71.1 (95% CI 68.7 to 73.5)	69.1 (95% CI 63.5 to 74.7)	34.1	65.3	35.4 (95% CI 31.6 to 39.2)	34.3 (95% CI 31.2 to 37.4)
Mitchell 2014	89	95	26.9	16.8	69 (8.0)	69 (10.1)	60.7	49.5	56.0 (16.8)	59.6 (17.4)
Monninkhof 2003	127	121	3.9	5.8	65 (7.0)	65 (7.0)	85.0	84.0	56.1 (15.4)	58.4 (14.5)
Ninot 2011	23	22	13.0	18.2	65 (range 59 to 74)	61 (range 56 to 65)	90.0	77.8	56 (range 42 to 67)	54 (range 42 to 57)
Österlund Efraimsson 2008	26	26	0	0	66 (9.4)	67 (10.4)	50.0	50.0	N/A	N/A
Rea 2004	83	52	14.5	11.5	68 (range 44 to 84) for the total group		41.5% for the total group		51.8 (18.1)	50.0 (20.3)
Rice 2010	372	371	9.7	12.9	69.1 (9.4)	70.7 (9.7)	97.6	94.8	36.1 (14.5)	38.2 (14.4)
Song 2014	20	20	15.0	15.0	66.6 (7.1)	68.1 (6.5)	55.0	75.0	57.0 (10.0)	60.4 (24.9)
Tabak 2014	15	14	33.3	85.7	64.1 (9.0)	62.8 (7.4)	50.0	50.0	50.0 (IQR 33.3 to 61.5)	36.0 (IQR 26.0 to 53.5)
Titova 2015	91	81	44.0	39.5	74.1 (9.3)	72.6 (9.3)	42.9	43.2	33.6 (9.9)	33.0 (9.7)

^adiscontinued; ^bincomplete baseline and 1‐year study visits; CI: confidence interval; IQR: interquartile range; L: liters; N/A: not applicable.

Table 2. Characteristics of interventions in included studies

Study	Follow‐up (months)	Setting; provision intervention	Duration intervention	Content intervention	Content action plan
Bischoff 2012	24	General practice; trained practice nurse	2 to 4 FTF individual sessions (60 min each) scheduled in 4‐6 consecutive weeks, 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Bösch 2007	12	Outpatient clinic; trained respiratory nurse under supervision of a respiratory specialist	4 FTF group sessions (120 min each) and final session scheduled 6 weeks later	Self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, other: travelling, daily live	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent, contact healthcare providers for support
Bourbeau 2003	24	Hospital (outpatient); trained professionals (nurses, respiratory therapists, a physiotherapist)	7 FTF individual sessions (60 min each) scheduled in 7‐8 consecutive weeks, 18 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: symptom monitoring list linked to appropriate therapeutic actions
Bucknall 2012	12	Hospital (inpatient); trained study nurse	4 FTF individual sessions (40 min each) in 2 months, at least 6 subsequent home visits, 828 phone calls intervention group	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Casas 2006	12	Hospital (inpatient); trained respiratory nurse and GP, physician, nurse, social worker	3 to 13 FTF individual sessions, 1 x group (40 min), 6 phone calls; Barcelona: 1 joint visit at home. Leuven: GP regularly visited patients at home	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: reinforcement of the logistics for treatment of comorbidities and social support	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: reinforcement of the logistics for treatment of comorbidities
Garcia‐Aymerich 2007	12	Hospital (inpatient); trained specialised respiratory nurse and physician, nurse, social worker	3 to 13 FTF individual sessions at the hospital (40 min each) or at home (20 min), 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: reinforcement of the logistics for treatment of comorbidities and social support	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: reinforcement of the logistics for treatment of comorbidities
Fan 2012	12	Outpatient clinic; trained case manager (various health‐related professionals)	4 FTF individual sessions (90 min each) scheduled weekly, 1x group, 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Gallefoss 1999	12	Hospital (outpatient); trained nurse, physiotherapist, pharmacist, medical doctor	1 to 2 FTF individual sessions by a nurse and 1 to 2 by physiotherapist (40 min each), 2 x group (120 min each)	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: compliance, self‐care	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Hernández 2015	12	Hospital (outpatient); trained specialised respiratory nurse, physician, nurse, social worker	Participants with no mobility problems: 1 FTF individual session (40 min) at home by primary care team, 3 x group at outpatient clinic (2 x 90 min, 1x 120 min) Participants with mobility problems: 4 FTF individual sessions (15 min each), 1 x individual (120 min) or 1 x group (40 min), all at home by primary care team	Self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, exercise or physical activity component, other: instructions on non‐pharmacological treatment	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent, contact healthcare providers for support, self‐treatment of comorbidities
Jennings 2015	3	Hospital (inpatient); research team and research nurse	1 FTF individual session (60 min) at the hospital by research team member 24 hours prior to discharge, phone call 48 hours after discharge	Iterative process, education regarding COPD, smoking cessation, other: primary team was notified if patient was identified as having anxiety or depressive symptoms	Contact healthcare providers for support
Khdour 2009	12	Hospital (outpatient); clinical pharmacist, respiratory specialist, respiratory nurse	1 FTF individual session of 45 min (60 min for smokers) and 2 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Kheirabadi 2008	3	Hospital (outpatient); psychologist, trained psychiatric residents	8 FTF group sessions (60 to 90 minutes each) with 1 week interval and follow‐up by phone	Self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Avoid situations in which viral infection might be prevalent
Martin 2004	12	General practice; respiratory physician and nurse, GP, ED consultant, medical staff hospital	4 FTF individual sessions and respiratory nurse visits at 3, 6 and 12 months	Iterative process, self‐recognition of COPD exacerbations	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, self‐treatment of comorbidities, other: when to use oxygen therapy and diuretics
Mitchell 2014	6	General practice; physiotherapist, trainee health psychologist	1 FTF individual session (30‐45 min) by a physiotherapist and 2 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, avoid situations in which viral infection might be prevalent, contact healthcare providers for support, other: self‐administration, requesting rescue medication
Monninkhof 2003	12	Hospital (outpatient); trained respiratory nurse, respiratory physiotherapist	5 FTF group sessions (120 min each) by a respiratory nurse (4 x with a 1‐week interval and 3 months later) and 1 to 2 x groups (30 to 45 min) by a physiotherapist	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Ninot 2011	12	Hospital (outpatient); health professional and qualified exercise trainer	8 FTF group sessions (120 min each) by a health professional for 4 weeks, 8 exercise sessions (30 to 45 min each) by a qualified exercise trainer, 3 phone calls	Self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent
Österlund Efraimsson 2008	3 to 5	Primary healthcare clinic; COPD nurse, physician, if needed: dietician, medical social worker, physical and occupational therapist	2 FTF individual sessions for self‐care education during 3 to 5 months (60 min each) by the nurse	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Rea 2004	12	General practice; respiratory physician, respiratory nurse specialist, GP	At least 17 individual FTF sessions (monthly visits to practice nurse (N = 12), 3‐monthly to GP (N = 4), 1 x home visit by the respiratory nurse specialist, 1 x after admission)	Iterative process, self‐recognition of COPD exacerbations, other: annual influenza vaccination and PR programme attendance	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Rice 2010	12	Hospital (Veterans Affairs medical centres); trained respiratory therapist case manager	1 group session (60 to 90 min) by a respiratory therapist case manager, 12 monthly phone calls (10 to 15 min each)	Iterative process, self‐recognition of COPD exacerbations; education regarding COPD, smoking cessation	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Song 2014	2	Hospital (inpatient); trained nurse interventionists	3 FTF individual sessions (2 x inpatient (90 + 45 min each) on the day before and on the day of discharge, 1 x outpatient (90 min) on the first follow‐up day) by 2 nurse interventionists, 2 phone calls with a 2‐week interval	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations
Tabak 2014	9	Hospital (outpatient); primary care physiotherapy practices; respiratory nurse practitioner, respiratory physiotherapist	2 group sessions (90 min each) by a nurse practitioner, 1 FTF individual session and 1 x intake by the physiotherapist, additional meetings after 1, 3, 6 and 9 months	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Titova 2015	24	Hospital (inpatient); trained specialist nurse	6 FTF individual sessions (1 x at discharge, 5 x home visits at 3 and 14 days, and at 6, 12, 24 months) by the specialist nurse, 1 e‐learning programme (15 min), at least 24 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, avoid situations in which viral infection might be prevalent, contact healthcare providers for support

COPD: Chronic Obstructive Pulmonary Disease; FTF: face‐to‐face; PR: pulmonary rehabilitation

Participants and recruitment

A total of 3,854 participants (self‐management intervention N = 1,931, usual care control N = 1,923) were assessed in the 22 included studies (Table 1). Dropout rates in the studies ranged from 0% to 59%, and in total 3,293 (85%) participants completed the study follow‐up. Seventeen studies recruited participants from hospitals; 12 studies from outpatient clinics (Bösch 2007; Bourbeau 2003; Bucknall 2012; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Kheirabadi 2008; Monninkhof 2003; Ninot 2011; Rice 2010; Tabak 2014) and five from inpatient populations (Casas 2006; Garcia‐Aymerich 2007; Jennings 2015; Song 2014; Titova 2015). Tabak 2014 reported recruitment from both outpatient clinic and primary care physiotherapy practices. Five studies (Bischoff 2012; Martin 2004; Mitchell 2014; Österlund Efraimsson 2008; Rea 2004) recruited participants from general practices or primary healthcare clinics.

Interventions

Content of the interventions assessed by the 22 included studies were diverse (Table 2). The median follow‐up duration was 12 months (interquartile range (IQR) 5.3 to 12.0). The duration of follow‐up was three months or less in three (14%) studies (Jennings 2015; Kheirabadi 2008; Song 2014), three to five months in one (4%) study (Österlund Efraimsson 2008), six months in one (4%) study (Mitchell 2014), nine months in one (4%) study (Tabak 2014), 12 months in 13 (59%) studies (Bösch 2007; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Martin 2004; Monninkhof 2003; Ninot 2011; Rea 2004; Rice 2010) and 24 months in three (14%) studies (Bischoff 2012; Bourbeau 2003; Titova 2015).

Self‐management interventions were delivered individually in ten (45%) studies (Bischoff 2012; Bucknall 2012; Jennings 2015; Khdour 2009; Martin 2004; Mitchell 2014; Österlund Efraimsson 2008; Rea 2004; Song 2014; Titova 2015) and in small groups in three (14%) studies (Bösch 2007; Bourbeau 2003; Monninkhof 2003), and included both individual and group sessions in nine (41%) studies (Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Kheirabadi 2008; Ninot 2011; Rice 2010; Tabak 2014).

The median duration of the intervention including self‐management reinforcement was nine months (IQR 1.0 to 12.0). The intervention duration was less than one month in two (9%) studies (Gallefoss 1999; Jennings 2015) and one month in four (18%) studies (Casas 2006; Mitchell 2014; Ninot 2011; Song 2014). In four (18%) studies (Bösch 2007; Kheirabadi 2008; Monninkhof 2003; Österlund Efraimsson 2008), the intervention duration was over one month up to six months. The intervention duration was nine months in two (9%) studies (Garcia‐Aymerich 2007; Tabak 2014), 12 months in eight (36%) studies (Bourbeau 2003; Bucknall 2012; Fan 2012; Hernández 2015; Khdour 2009; Martin 2004; Rea 2004; Rice 2010) and 24 months in two (9%) studies (Bischoff 2012; Titova 2015).

In nine (41%) studies (Bourbeau 2003; Hernández 2015; Kheirabadi 2008; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Song 2014; Tabak 2014) a standardised exercise programme was part of the intervention. A smoking cessation programme was part of the intervention in six (27%) studies (Bösch 2007; Hernández 2015; Jennings 2015; Khdour 2009; Österlund Efraimsson 2008; Rice 2010).

Self‐management topics about (maintenance) medication were discussed in all but one study (Jennings 2015), while coping with breathlessness or breathing techniques was discussed in all but two studies (Martin 2004; Rea 2004).

Other major topics addressed were diet or nutrition or both (n = 17; 77%) (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Gallefoss 1999; Hernández 2015; Jennings 2015; Khdour 2009; Kheirabadi 2008; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Tabak 2014; Titova 2015), and correct device use (n = 13; 59%) (Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Hernández 2015; Jennings 2015; Khdour 2009; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Rice 2010; Titova 2015).

The AECOPD action plan components discussed in the interventions were self‐recognition of COPD exacerbations (n = 20) (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Rice 2010; Song 2014; Tabak 2014; Titova 2015), self‐treatment of COPD exacerbations (n = 20) (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Rice 2010; Song 2014; Tabak 2014; Titova 2015), contact healthcare providers for support (n = 18) (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Jennings 2015; Khdour 2009; Mitchell 2014; Monninkhof 2003; Österlund Efraimsson 2008; Rea 2004; Rice 2010; Tabak 2014; Titova 2015), use of maintenance treatment (n = 10) (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Gallefoss 1999; Hernández 2015; Martin 2004; Ninot 2011), avoid situations in which viral infection might be prevalent (n = 6) (Bösch 2007; Hernández 2015; Kheirabadi 2008; Mitchell 2014; Ninot 2011; Titova 2015), and self‐treatment of comorbidities (n = 2) (Hernández 2015; Martin 2004).

A total of 204 behavioural change techniques (BCT) clusters (Michie 2013) were integrated in the interventions with a median of 9.5 (IQR 8.0 to 10.0) clusters per study (minimum 6 BCT clusters (Kheirabadi 2008), maximum 12 BCT clusters (Bucknall 2012)). The behaviour change clusters that were integrated to promote the uptake and optimal use of COPD‐specific self‐management behaviour patterns in the intervention were: goals and planning (n = 22); feedback and monitoring (n = 22); shaping knowledge (n = 22); associations (n = 22); regulation (n = 21; all but one study (Jennings 2015)); antecedents (n = 20; all but two studies (Kheirabadi 2008; Song 2014)); social support (n = 19; all but three studies (Gallefoss 1999; Kheirabadi 2008 Rea 2004)); comparison of behaviour (n = 18; all but four studies (Fan 2012; Kheirabadi 2008; Song 2014; Titova 2015)); repetition and substitution (n = 16; all but six studies (Bösch 2007; Hernández 2015; Kheirabadi 2008; Martin 2004; Österlund Efraimsson 2008; Rea 2004)); natural consequences (n = 15; all but seven studies (Bösch 2007; Hernández 2015; Jennings 2015; Martin 2004; Ninot 2011; Song 2014; Titova 2015)); identity (n = 3) (Mitchell 2014; Österlund Efraimsson 2008; Song 2014); self‐belief (n = 3) (Bucknall 2012; Song 2014; Tabak 2014) and comparison of outcomes (n = 1) (Bucknall 2012). There were no rewards and threats, scheduled consequences or covert learning integrated in any of the self‐management interventions.

Adherence

Half of the studies reported details regarding participants' adherence to the intervention. Of these, six studies reported adherence as the number or percentage of sessions attended by participants. In Bischoff 2012 the number of sessions that were offered depended on the participant's needs, but was at least two sessions. Participants in Bischoff 2012 received a mean of 3.4 (SD 1.5) sessions; 13% did not attend any sessions or received telephone contact. The self‐management education course in Monninkhof 2003 consisted of five group sessions; of these, four were scheduled at one‐week intervals and the final session three months later. Mean attendance frequency was 0.77 (SD 0.22) sessions per week, and five (4%) participants randomised to the intervention group refused to attend the self‐management education course (Monninkhof 2003). Fan 2012 reported that during the entire follow‐up period, eight of 209 participants in the intervention group and 10 of 217 participants in the usual care group either did not attend scheduled visits or formally withdrew from the study. The study authors also reported that in the intervention group 87% completed all four individual educational visits and 57% completed the scheduled group visit (Fan 2012). Early termination after the intervention was enforced by the Data and Safety Monitoring Committee and the apparently low attendance rate of the group visit may well be a consequence (Fan 2012).

Tabak 2014 reported that the self‐management module on the web portal, including the self‐treatment of COPD exacerbations, was used on 86% of treatment days per participant. Ninot 2011 found that one of 23 participants from the intervention group did not fulfil adherence criteria to the four‐week self‐management programme, defined as completing at least seven of the eight sessions. In Gallefoss 1999, the intervention group participants who did not attend the individual or group sessions were withdrawn (N = 5, 16%). Three studies reported adherence according to different definitions. Self‐reported scales in Casas 2006 and Garcia‐Aymerich 2007 showed better adherence to recommended oral treatment in the intervention group than in the control group (90% versus 85%, respectively) and inhaled treatment regimens (71% versus 37%). Khdour 2009 reported that 78% of participants in the intervention group versus 60% of control group participants reported high adherence to maintenance medication after the 12‐month follow‐up, reflecting a lower number of medication omissions in the intervention group compared to the control group.

Comparisons

As per inclusion criteria, self‐management interventions that included an action plan for AECOPD were compared with usual care in 22 studies. Bischoff 2012 reported two intervention groups (one with and one without an action plan for AECOPD) and one usual care group. We used only data from the intervention group that included an action plan for AECOPD and the usual care group for this review.

Outcomes

See Table 3 for details on the number of included studies reporting outcomes of interest.

See: Summary of findings for the main comparison Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD

Table 3. Number of included studies reporting outcomes of interest

Outcome of interest	Number of studies
Primary outcomes
Health‐related quality of life	16
Respiratory‐related hospital admissions	16
Secondary outcomes
All‐cause hospital admissions	11
All‐cause hospitalisation days	8
Respiratory‐related hospitalisation days	5
Emergency department visits	9
General practitioner visits	7
Specialist visits	4
Rescue medication use	2
Health status	3
COPD exacerbations	6
Use of courses of oral corticosteroids or antibiotics	9
All‐cause mortality	16
Respiratory‐related mortality	7
Self‐efficacy	2
Days lost from work	2

COPD: Chronic Obstructive Pulmonary Disease

Missing data

We listed the study authors from whom we received responses to requests for additional data in Acknowledgements. However, not all study authors were able to provide the requested additional information. If the requested data were not provided for meta‐analyses, we described data that were available.

Excluded studies

We excluded 225 studies following assessment of the full‐text (Figure 1). The most frequent reasons for exclusion were: no COPD self‐management intervention (n = 56); no written action plan for AECOPD (n = 48); no usual care control group (n = 30). See Characteristics of excluded studies.

Studies awaiting classification

A total of 12 studies await classification. Koff 2009, Leiva‐Fernández 2014 and Lou 2015 await classification because we could not reach the authors to verify whether the studies met our eligibility criteria. From a search in May 2017, we identified nine studies (Benzo 2016; Chien 2016; Davis 2016; Imanalieva 2016; Licskai 2016; Sánchez‐Nieto 2016; Sano 2016; Silver 2017; Zwar 2016) that could be included in a future update of the review. Thesehave been added to Characteristics of studies awaiting classification and have not been fully incorporated into the review.

Ongoing studies

We identified two ongoing studies (Bourbeau 2016; Lenferink 2013).

Risk of bias in included studies

A summary of our risk of bias assessment is presented in Figure 2. Assessments were performed based on the content of study articles and no extra information was requested from the study authors. Further details and the rationale for judgments can be found in Characteristics of included studies.

Figure 2

Risk of bias summary for each study according to authors' judgements

Allocation

Computer‐generated random number lists or other computerised methods were most frequently used to generate allocation sequences in studies (n = 13) (Bischoff 2012; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Hernández 2015; Jennings 2015; Khdour 2009; Mitchell 2014; Ninot 2011; Rea 2004; Tabak 2014). Two of these studies used stratification or minimisation to balance for potential confounders (Bucknall 2012; Khdour 2009). All these 13 studies had a well‐defined rule for allocating the intervention to participants and were judged as having a low risk of selection bias. Two studies used random number tables or lists in sealed envelopes (Gallefoss 1999; Monninkhof 2003) or an independent person drew lots for allocation (Österlund Efraimsson 2008) and were assessed at low risk of bias. Six studies (Bösch 2007; Kheirabadi 2008; Martin 2004; Rice 2010; Song 2014; Titova 2015) did not report how the allocation sequence was generated and were judged as having an unclear risk of bias.

In most studies (n = 12) (Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Tabak 2014) the investigators or staff were not able to influence the allocation concealment, or the randomisation was performed by an independent person who was not involved in the study; risk of bias was considered to be low. The risk of bias was judged to be unclear in nine studies (Bischoff 2012; Bösch 2007; Jennings 2015; Kheirabadi 2008; Martin 2004; Mitchell 2014; Rice 2010; Song 2014; Titova 2015) which did not report who performed the allocation or methods used for the allocation concealment. One study was cluster‐randomised and no allocation concealment was provided; therefore, the risk of bias was considered to be high (Rea 2004).

Blinding

Because of the nature of the self‐management intervention, blinding of participants and personnel to group assignment is complicated. None of the included studies reported blinding of participants and personnel; performance bias risk was considered to be high in all included studies.

The detection bias was considered to be low in ten studies (Bischoff 2012; Bourbeau 2003; Garcia‐Aymerich 2007; Fan 2012; Hernández 2015; Jennings 2015; Mitchell 2014; Monninkhof 2003; Ninot 2011; Rice 2010), because these studies were investigator blinded, the outcome assessment was performed by an independent assessor, the evaluator was unaware of participant assignment or only objective outcome measures were used. In 11 studies (Bösch 2007; Bucknall 2012; Casas 2006; Gallefoss 1999; Khdour 2009; Kheirabadi 2008; Martin 2004; Rea 2004; Song 2014; Tabak 2014; Titova 2015) the detection bias was judged to be unclear, since the outcome assessment was not reported or the outcome assessment was only partly blinded. In one study the outcome assessments were performed or supervised by the same person who provided the intervention (Österlund Efraimsson 2008) and was considered to have a high risk of detection bias.

Incomplete outcome data

In 12 studies (Bischoff 2012; Bourbeau 2003; Casas 2006; Gallefoss 1999; Khdour 2009; Kheirabadi 2008; Mitchell 2014; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Song 2014), outcome data were complete and there were no systematic differences detected between the intervention and usual care groups in withdrawals. In these 12 studies the risk of attrition bias was considered to be low. There were incomplete data in two studies due to early termination; one as a result of significantly higher mortality rates in the intervention group (Fan 2012), and one because interim analysis at three years did not demonstrate the desired 10% between‐group differences in emergency department visits or rehospitalisations (Jennings 2015). The risk of attrition bias in these two studies was judged to be unclear (Fan 2012; Jennings 2015). The risk of attrition bias was also considered to be unclear in three studies, because there was insufficient information to permit judgment (Hernández 2015), there was no information provided regarding differences in dropout rates (Martin 2004), or only a part of the outcome data were missing (Rice 2010). In five studies (Bösch 2007; Bucknall 2012; Garcia‐Aymerich 2007; Tabak 2014; Titova 2015) the quantities of missing outcome data were high and the risk of attrition bias was considered to be high.

Selective reporting

Five studies (Bischoff 2012; Fan 2012; Gallefoss 1999; Mitchell 2014; Rice 2010) were judged to have low risk for reporting bias; there were no signs of selective outcome reporting when the reported outcomes and study findings were compared with information provided in the study protocols. In 13 studies (Bösch 2007; Bourbeau 2003; Casas 2006; Garcia‐Aymerich 2007; Hernández 2015; Khdour 2009; Kheirabadi 2008; Martin 2004; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Rea 2004; Song 2014) there were no signs of selective reporting. However, for these studies there were no study protocols available and the reporting bias was considered to be unclear. One study reported a slightly different primary outcome in the paper compared to primary outcome as defined in the study protocol; this study was therefore judged as unclear risk of reporting bias (Jennings 2015). Three studies were considered to have a high risk of reporting bias, because not all relevant outcome measures were completely reported (Bucknall 2012; Tabak 2014; Titova 2015).

Other potential sources of bias

We assessed Rea 2004 for biases which are important in cluster‐RCTs. Rea 2004 reported that general practices were randomly assigned before the participants were included. For reasons unknown, the number of participants screened and included was higher in the intervention group than in the usual care group. Rea 2004 reported there were no significantly between‐group differences for baseline characteristics. We considered the risk of recruitment bias to be unclear and the risk of bias for baseline imbalance to be low. The risk of bias due to loss of clusters was judged as low, because no clusters were lost after participant enrolment. Rea 2004 did not correct for clustering in analyses. The risk of bias due to incorrect analysis was considered to be high. No other potential sources of bias were observed in this study.

We judged three studies in which per protocol analyses were performed as having an unclear risk of other bias (Bösch 2007; Tabak 2014; Ninot 2011). In these studies the baseline characteristics were not reported for all randomised participants. However, in Bösch 2007 and Ninot 2011 no differences were reported for baseline characteristics among withdrawals after randomisation and the participants who completed the study. In Tabak 2014 no differences were reported for baseline characteristics between withdrawals after randomisation and participants who completed the questionnaires at inclusion.

In addition, we explored possible reporting bias by assessing asymmetry in funnel plots for health‐related quality of life (HRQoL) (Figure 3) and respiratory‐related hospital admissions (Figure 4). A negative mean difference (MD) of the St. George's Respiratory Questionnaire (SGRQ) total score indicates better HRQoL in the self‐management group compared to usual care. The SGRQ funnel plot, with MD in SGRQ total score plotted against the standard error (SE) of the MD, seems to show a gap on the lower right side of the graph (Figure 3). This could indicate that smaller studies with effects in favour of the usual care group (positive MD in SGRQ scores) are published less frequently. On the contrary, the funnel plot of the odds ratio (OR) per study plotted against the SE (log OR) in respiratory‐related hospital admissions seems to show a gap on the left side of the graph (Figure 4), indicating that smaller studies and studies of moderate size with effects in favour of the self‐management group are published less frequently. We could not rule out the contribution of other study factors to funnel plot asymmetry.

Figure 3

Funnel plot of comparison: Self‐management versus usual care, outcome: 1.1 HRQoL: adjusted SGRQ total score

Figure 4

Funnel plot of comparison: Self‐management versus usual care, outcome: 1.2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission)

Effects of interventions

We included a 'Summary of Findings' table of the 22 included studies that compared self‐management with usual care. summary of findings Table for the main comparison reflects the endpoints related to health‐related quality of life (HRQoL), hospital admissions, mortality, dyspnoea, number of Chronic Obstructive Pulmonary Disease (COPD) exacerbations, and courses of oral steroids.

Health‐related quality of life (HRQoL)

COPD‐specific HRQoL was measured by the St. George's Respiratory Questionnaire (SGRQ) in ten studies (Bourbeau 2003; Bucknall 2012; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Monninkhof 2003; Ninot 2011; Rice 2010; N = 1582). We used adjusted mean difference (MD) scores when available. If not available, we included the change from baseline scores and otherwise the mean total scores of these studies on a single forest plot to perform a meta‐analysis on SGRQ total score. Over 12 months of follow‐up, the included studies showed lower mean SGRQ total scores (meaning better HRQoL) in the self‐management intervention compared with the usual care group. The MD of ‐2.69 (95% CI ‐4.49 to ‐0.90), indicating better HRQoL in the intervention group compared to the control group, was statistically significant at the 5% level (Analysis 1.1; Figure 5; I² = 46%). The pooled MD of ‐2.69 did not reach the minimal clinically important difference (MCID) of four points (Jones 2005). However, four studies (Bucknall 2012; Hernández 2015; Ninot 2011; Rice 2010) reached MCID = 4 for SGRQ total score. Only Fan 2012 reported a statistically non‐significant positive MD of 0.31 for the change from baseline SGRQ total score among participants who completed 12 months follow‐up, indicating that the self‐management intervention group decreased by 0.31 points from baseline compared with the usual care group. Three studies (Österlund Efraimsson 2008; Martin 2004; Song 2014) provided insufficient data for inclusion in the meta‐analysis. Österlund Efraimsson 2008 reported significant and clinically relevant lower total SGRQ total scores in the self‐management intervention group (HRQoL was improved by 8.2 points) compared with the usual care group (no change noted). Martin 2004 found no significant difference in SGRQ total score after 12 months of follow‐up. The SGRQ total score in Song 2014 was significantly lower in the intervention group after two months, which meant better HRQoL. Sensitivity analysis using fixed‐effect modelling resulted in a lower effect size of the SGRQ total score (MD ‐2.08, 95% CI ‐3.21 to ‐0.95) compared to the random‐effects model.

Figure 5

Forest plot of comparison: Self‐management versus usual care, outcome: 1.1 HRQoL: adjusted SGRQ total score after 12 months of follow‐up

Three studies (Bischoff 2012; Mitchell 2014; Rea 2004) measured COPD‐specific HRQoL using the Chronic Respiratory Questionnaire (CRQ) for a total of 394 participants. The CRQ consists of four domain scores: dyspnoea, fatigue, emotional function, and mastery (sense of control over the disease) (Guyatt 1987). A higher CRQ domain score indicates better HRQoL and the MCID is reflected by a change in a CRQ domain score of at least 0.5 on a 7‐point scale (Jaeschke 1989; Redelmeier 1996). Rea 2004 reported the CRQ domains on a different scale and did not provide SDs. Rea 2004 could not be included in a meta‐analysis, leaving an insufficient number of two studies to perform a meta‐analysis. In Rea 2004, two of the four CRQ domains, fatigue and mastery, showed statistically significant higher scores, indicating better HRQoL, for the self‐management intervention group (17.7 and 21.4, respectively) compared to usual care (15.7 and 20.7, respectively) after 12‐months follow‐up. Mitchell 2014 reported that both groups improved CRQ dyspnoea over time and only the self‐management group maintained within‐group changes that exceeded the MCID of 0.5. The between‐group differences were non‐significant at six months of follow‐up (Mitchell 2014). A non‐comprehensive approach with a lack of group support, supervised exercise training and healthcare professional‐led education might have limited the effectiveness of the intervention (Mitchell 2014). Bischoff 2012 reported no statistically significant mean treatment difference between the self‐management intervention and usual care group for the CRQ total score at 24 months of follow‐up. Although more participants in the intervention group showed a clinically important improvement compared to the usual care group, this difference was not statistically significant.

Only Rea 2004 used the Short Form‐36 (SF‐36) to measure the generic HRQoL. There were no differences noted between the intervention and usual care group after 12 months of follow‐up for any dimension of the SF‐36.

Bucknall 2012 and Tabak 2014 reported the generic HRQoL using EuroQol‐5 Dimensions (EQ‐5D). Bucknall 2012 reported no significant differences in the EQ‐5D areas under the curve between the groups after 12 months of follow‐up. The study findings reported by Tabak 2014 showed a trend toward a higher EQ‐5D index, indicating better HRQoL in the intervention group compared to the control group after three months follow‐up (mean 0.78 ± SE 0.08 versus mean 0.61 ± SE 0.09). However, these data were reported only descriptively.

In Tabak 2014 the individual participant's HRQoL state was also reported using a vertical Visual Analogue Scale (VAS). There was a trend toward a higher VAS score, indicating better HRQoL reported for self‐management (72.3 ± SE 3.1) compared to usual care (62.4 ± SE 3.5). Again, these data were only reported descriptively. Garcia‐Aymerich 2007 reported slight, non‐significant improvements in quality of life scores in both groups according to the VAS in the follow‐up year (intervention 1.56 ± SD 1.77, control 0.93 ± SD 2.11).

Generic HRQoL and health status were further measured using the short version of the questionnaire validated by the Nottingham Health Profile (NHP) in Ninot 2011. Ninot 2011 reported statistically significant beneficial effects of the self‐management intervention on the energy (between‐group difference ‐19.8, 95% CI ‐38 to ‐1) and emotional reaction (between‐group difference ‐10.4, 95% CI ‐20 to 0) dimensions of the NHP, after adjustment for baseline values.

Respiratory‐related hospital admissions

Respiratory‐related hospital admissions were reported in 14 studies (Bourbeau 2003; Bucknall 2012; Fan 2012; Gallefoss 1999; Hernández 2015; Jennings 2015; Khdour 2009; Mitchell 2014; Monninkhof 2003; Ninot 2011; Rea 2004; Rice 2010; Tabak 2014; Titova 2015; N = 3157). A statistically significant lower probability of at least one respiratory‐related hospital admission was noted among participants who received the self‐management intervention that included an action plan compared with those who received usual care (OR 0.69, 95% CI 0.51 to 0.94, Analysis 1.2; Figure 6). Heterogeneity was high (I² = 57%). Sensitivity analysis using the fixed‐effect model resulted in a similar effect size (OR 0.71, 95% CI 0.60 to 0.85) compared to random‐effects modelling.

Figure 6

Forest plot of comparison: Self‐management versus usual care, outcome: 1.2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission)

Two studies (Bösch 2007; Martin 2004) could not be included in the meta‐analysis due to a lack of required data. In Martin 2004 more respiratory‐related hospitalisations were reported in the intervention group (1.1 per participant per year) compared to usual care (0.7 per participant per year). Due to a lack of SDs, this study could not be included in the meta‐analysis. There were six studies (Bischoff 2012; Casas 2006; Garcia‐Aymerich 2007; Kheirabadi 2008; Österlund Efraimsson 2008; Song 2014) that did not report any data on respiratory‐related hospital admissions and could not be included in the meta‐analysis.

The study‐specific number needed to treat for an additional beneficial outcome (NNTB) for respiratory‐related hospital admissions ranged from 11 (95% CI 7 to 65) to 71 (95% CI 44 to 367). To calculate NNTB, the pooled effect on respiratory‐related hospital admissions (OR 0.69, 95% CI 0.51 to 0.94) was used and applied to the mean control event risks of the studies with the highest and lowest baseline risks. The seven studies (Bourbeau 2003; Bucknall 2012; Jennings 2015; Khdour 2009; Rea 2004; Rice 2010; Tabak 2014) with the highest baseline risks for respiratory‐related hospital admissions had a mean control event risk (mean observed risk of the respiratory‐related hospital admissions in the usual care group) of 38.99 (Figure 7). Over 12 months of follow‐up, 12 participants (95% CI 7 to 69) with high baseline risk of respiratory‐related hospital admissions needed to be treated to prevent one participant with at least one respiratory‐related hospital admission. The seven studies (Fan 2012; Gallefoss 1999; Hernández 2015; Mitchell 2014; Monninkhof 2003; Ninot 2011; Titova 2015) with the lowest baseline risks for respiratory‐related hospital admissions had a mean control event risk of 23.10 (Figure 8). Over 12 months of follow‐up, 17 participants (95% CI 11 to 93) with low baseline risk of respiratory‐related hospital admissions needed to be treated to prevent one participant with at least one respiratory‐related hospital admission.

Figure 7

Cates plot of COPD participants with high baseline risk of respiratory‐related hospital admissions in self‐management interventions including action plans for AECOPD compared to usual care. In the usual care group, 39 of 100 participants had at least one respiratory‐related hospital admission over 52 weeks, compared with 31 (95% CI 25 to 38) of 100 participants in the self‐management intervention group with the highest baseline risks for respiratory‐related hospital admissions

Figure 8

Cates plot of COPD participants with low baseline risk of respiratory‐related hospital admissions in self‐management interventions with action plans for AECOPD compared to usual care. In the usual care group, 23 of 100 participants had at least one respiratory‐related hospital admission over 52 weeks, compared with 17 (95% CI 13 to 22) of 100 participants in the self‐management intervention group

Five studies (Bösch 2007; Bucknall 2012; Jennings 2015; Tabak 2014; Titova 2015) were included in a meta‐analysis on the mean number of respiratory‐related hospital admissions. No difference was found (MD ‐0.15, 95% CI ‐0.36 to 0.05, Analysis 1.3). Using fixed‐effect modelling in the sensitivity analysis produced similar effects.

All‐cause hospital admissions

All‐cause hospital admissions were reported in 10 studies (Bucknall 2012; Casas 2006; Fan 2012; Hernández 2015; Khdour 2009; Mitchell 2014; Ninot 2011; Rea 2004; Rice 2010; Tabak 2014; N = 2467). There was no statistically significant difference in all‐cause hospital admissions (OR 0.74, 95% CI 0.54 to 1.03; Analysis 1.4). Heterogeneity was high (I² = 62%). Sensitivity analysis using fixed‐effect modelling resulted in statistically significant fewer all‐cause hospital admissions in the self‐management group compared to usual care (OR 0.74, 95% CI 0.63 to 0.88). Since the beneficial effect of the self‐management intervention on all‐cause hospital admissions observed was the same when analysing using random‐effects and fixed‐effect models, the presence of small study effects was considered unlikely.

Twelve studies could not be meta‐analysed due to a lack of required information (Bischoff 2012; Bösch 2007; Bourbeau 2003; Garcia‐Aymerich 2007; Gallefoss 1999; Jennings 2015; Kheirabadi 2008; Martin 2004; Monninkhof 2003; Österlund Efraimsson 2008; Song 2014; Titova 2015). It was not possible to calculate the NNTB for all‐cause hospital admissions, because the 95% CI of the pooled OR for at least one all‐cause hospital admission included the possibilities of both benefit and harm.

Four (Bucknall 2012; Casas 2006; Martin 2004; Tabak 2014) of the six studies that reported on the mean number of all‐cause hospital admissions were included in a meta‐analysis. No difference was found (MD ‐0.04, 95% CI ‐0.38 to 0.29, Analysis 1.5). Heterogeneity was non‐significant (I²⁼ 35%). Two studies (Bourbeau 2003; Ninot 2011) could not be included in the meta‐analysis because SDs were not reported. A sensitivity analysis using fixed‐effect modelling resulted in an effect size (MD ‐0.07, 95% CI ‐0.33 to 0.19) similar to the random‐effects model.

Healthcare utilisation

All‐cause hospitalisation days

The total number of all‐cause hospitalisation days was reported in three studies (Bourbeau 2003; Khdour 2009; Rea 2004; N = 469). The data reported in these studies were heavily skewed and unsuitable for meta‐analysis. All three studies (Bourbeau 2003; Khdour 2009; Rea 2004) reported a lower number of all‐cause hospitalisation days in the intervention group (n = 688, n = 164, and n = 263, respectively) compared to the usual care group (n = 1,190, n = 466, n = 352, respectively). This difference was reported to be statistically significant in Khdour 2009, but the other studies did not report significance of the differences.

The number of all‐cause hospitalisation days per patient was assessed in eight studies. Seven studies (Bourbeau 2003; Bucknall 2012; Hernández 2015; Khdour 2009; Monninkhof 2003; Ninot 2011; Rice 2010) with 1982 participants were meta‐analysed and no statistically significant between‐group differences were found (MD ‐0.65, 95% CI ‐2.01 to 0.71; Analysis 1.6). Heterogeneity was high (I² = 60%). Sensitivity analysis using a fixed‐effect model resulted in statistically significantly lower all‐cause hospitalisation days per participant (MD ‐0.69, 95% CI ‐1.36 to ‐0.02). Rea 2004 could not be included in the meta‐analysis because no SD was reported. The mean number of all‐cause bed days in this study was lower in the intervention group than the usual care group (3.2 versus 6.8); however, this difference did not reach statistical significance.

Respiratory‐related hospitalisation days

The total number of respiratory‐related hospitalisation days was reported in three studies (Rea 2004; Tabak 2014; Titova 2015; N = 333). Reported data were unsuitable for meta‐analysis. All three studies (Rea 2004; Tabak 2014; Titova 2015) reported a lower number of respiratory‐related hospitalisation days in the intervention group (n = 90, n = 22, and n = 486, respectively) compared to the usual care group (n = 210, n = 36, n = 954, respectively). The studies did not report on significance of these differences. However, Titova 2015 reported that in the intervention group the number of respiratory‐related hospitalisation days was statistically significantly reduced during the first year of follow‐up and remained low during the second year of follow‐up.

The number of respiratory‐related hospitalisation days per participant was reported in three studies (Gallefoss 1999; Ninot 2011; Rea 2004; N = 226). However, Rea 2004 did not provide SD so the study could not be included in the meta‐analysis. There were an insufficient number of studies to perform a meta‐analysis, and the data provided were heavily skewed. Although Gallefoss 1999 reported a non‐significant lower mean number of respiratory‐related hospitalisation days in the intervention group (0.7 ± SD 2) compared to the usual care group (2.5 ± SD 11), Ninot 2011 reported a non‐significant higher mean number of respiratory‐related hospitalisation days in the intervention group (1.9 ± SD 3.7) compared to usual care (0.3 ± SD 0.7). Rea 2004 reported significantly fewer respiratory‐related hospitalisation days per participant per year in the intervention group (from 2.8 to 1.1) compared to a significant increase for the usual care group (from 3.5 to 4.0 days). Tabak 2014 could not be included in this meta‐analysis because the median length of stay was reported (intervention 5.5 (IQR 4.8 to 6.3), usual care 7.0 (IQR 6.0 to 7.0)).

Emergency department (ED) visits

Nine studies (Bourbeau 2003; Bucknall 2012; Fan 2012; Hernández 2015; Jennings 2015; Khdour 2009; Rea 2004; Rice 2010; Tabak 2014) reported ED visits. Three studies (Bourbeau 2003; Bucknall 2012; Jennings 2015) were included in a meta‐analysis; ED visit data were reported for 827 participants. There was no statistically significant difference between intervention and usual care (MD ‐0.31, 95% CI ‐0.74 to 0.12, Analysis 1.7). Sensitivity analysis using a fixed‐effect model resulted in a statistically significant lower number of ED visits in the intervention group compared to the control group (MD ‐0.35, 95% CI ‐0.43 to ‐0.27). The observed effect sizes in the fixed‐effect (MD ‐0.35) and random‐effects (‐0.31) models were comparable. The presence of small study effects was considered to be unlikely.

Six studies (Fan 2012; Hernández 2015; Khdour 2009; Rea 2004; Rice 2010; Tabak 2014) could not be meta‐analysed because different methods were used to report the outcome. Fan 2012 reported fewer participants who had at least one ED visit in the intervention group (N = 99, 47%) compared to the usual care group (N = 119, 55%) and a lower total number of ED visits in the intervention group (intervention N = 173 versus usual care N = 203) at 12 months follow‐up. It was not reported whether these differences were statistically significant or if numbers were adjusted for incomplete follow‐up. Hernández 2015 reported a lower mean number of respiratory‐related ED visits in the intervention group (10 ± SD 12.11) compared to the usual care group (23 ± SD 27.4). After adjusting for baseline differences, the intervention significantly reduced the risk of ED visits (OR 0.33, 95% CI 0.13 to 0.84). However, these data, could not be meta‐analysed because there was a different process reported for co‐ordination of hospital admissions in both groups; 80% of admissions in the intervention group were co‐ordinated between primary care and the hospital team, thus bypassing the ED (Hernández 2015). By contrast, all admissions in the usual care group were processed as unplanned admissions through the ED (Hernández 2015). The number of ED visits was dependent on group allocation. Khdour 2009 reported a statistically significant lower number of COPD‐related ED visits in the intervention group compared to the usual care group (40 versus 80) after 12 months of follow‐up. Rea 2004 observed five (6%) all‐cause ED visits in the intervention group and seven (13.5%) visits in the usual care group after 12 months of follow‐up. Rice 2010 found significantly fewer all‐cause ED visits in the intervention group than the usual care group (67.0 versus 91.2 per 100 person‐years) after 12 months of follow‐up. Tabak 2014 reported five (42%) participants with at least one COPD‐related ED visit in both groups.

General practitioner (GP) visits

General practitioner (GP) visits were reported in seven studies (Bourbeau 2003; Bucknall 2012; Casas 2006; Gallefoss 1999; Khdour 2009; Martin 2004; Monninkhof 2003). Three studies (Bucknall 2012; Gallefoss 1999; Martin 2004; N = 605) were included in a meta‐analysis. There was no statistically significant difference noted between the intervention and usual care (MD ‐0.36, 95% CI ‐2.64 to 1.93; Analysis 1.8). Sensitivity analysis using a fixed‐effect model resulted in a non‐significant lower effect on GP visits (MD ‐0.09, 95% CI ‐0.24 to 0.06). Four studies (Bourbeau 2003; Casas 2006; Khdour 2009; Monninkhof 2003) could not be included in the meta‐analysis because different methods were used to report the outcome (Casas 2006; Khdour 2009) and because of missing SDs (Bourbeau 2003; Monninkhof 2003). Bourbeau 2003 reported significantly fewer unscheduled GP visits in the intervention group (n = 46) compared to usual care (n = 112) after 12 months of follow‐up. However, the scheduled GP visits were comparable between groups. Monninkhof 2003 showed a reduction in unscheduled doctor and nurse visits per person per year between the intervention and control groups (difference ‐0.4). Casas 2006 reported no statistically significant differences in the number of GP home visits between the intervention (median 10, IQR 7 to 18) and control groups (median 13, IQR 9 to 27). Khdour 2009 reported a similar number of GP visits in both groups; a lower total number of scheduled GP visits in the intervention group (145 versus 183), although the total number of unscheduled visits in this study was somewhat higher in the intervention group (119 versus 75).

Specialist visits

Four studies reported data on specialist visits (Bourbeau 2003; Casas 2006; Jennings 2015; Martin 2004). These studies could not be included in a meta‐analysis, since different methods and definitions were used to report visits. Bourbeau 2003 reported comparable unscheduled (intervention N = 24, control N = 26) and scheduled specialist visits (intervention N = 347, control N = 316) in both groups. Casas 2006 reported a non‐significantly higher number of doctor and nurse visits (defined as unplanned visits to the GP, specialist outside the hospital, chest physician from the hospital, private doctors, domiciliary visits from the primary care team and visits to the day clinic) in the intervention group compared to the usual care group (14 ± SD 24 versus 10 ± SD 23). However, these data were heavily skewed. Martin 2004 reported a non‐significantly higher number of all‐cause doctor and nurse visits in the intervention group compared to the control group (15.6 ± SD 12.68 versus 11.6 ± SD 8.02). Jennings 2015 reported a non‐significantly lower number of primary care provider visits or pulmonary outpatient visits in the intervention group (0.46 ± SD 0.5) compared to the control group (0.53 ± SD 0.5) after three months of follow‐up.

Rescue medication use

Two studies included rescue medication use as an outcome (Gallefoss 1999; Rice 2010), but used different definitions. Gallefoss 1999 reported the use of dispensed short‐acting beta₂‐agonists as rescue medication. This was coded as defined daily dosages (DDDs) for comparison of medications within the same chemical therapeutic group. In this study, participants receiving self‐management used statistically significantly less rescue medication (median DDD 125, IQR 100 to 344) than the control group (median DDD 290, IQR 150 to 550) after 12 months of follow‐up. Rice 2010 reported the use of short‐acting beta₂‐agonists as the mean number of metered‐dose inhalers and found no statistically significant differences between intervention and control groups (6.4 ± 8.3 versus 5.6 ± 8.0).

Health status

In only two studies (Kheirabadi 2008; Tabak 2014) the change in severity of COPD was measured by means of the Clinical COPD Questionnaire (CCQ), so meta‐analysis could not be performed. A lower CCQ score indicates better HRQoL and the MCID of the CCQ total score is reflected by a change in score of 0.4 or more on a 6‐point scale. Kheirabadi 2008 reported that the intervention did not have a significant effect on the severity of COPD in the CCQ total score (mean 1.99 for both groups), but it did significantly decrease (meaning better HRQoL) three domain scores of the CCQ (symptoms, functional and mental). This improvement in HRQoL was clinically relevant for the self‐management group as the three domain scores reached the MCID of 0.4 points (Kheirabadi 2008). Tabak 2014 reported the CCQ total score for both groups after one and three months of follow‐up. These data were descriptive only, but showed trends toward a lower CCQ total score for the intervention group after three months of follow‐up (mean 1.8 ± SE 0.24) compared to usual care (mean 2.3 ± SE 0.26).

Dyspnoea symptoms

The effect of a self‐management intervention on dyspnoea as measured by the modified Medical Research Council questionnaire (mMRC) was assessed in three studies (Bösch 2007; Garcia‐Aymerich 2007; Hernández 2015). Garcia‐Aymerich 2007 assessed dyspnoea using the MRC and the other two studies used the mMRC. The outcomes of the three studies were combined in a meta‐analysis representing 217 participants. A non‐significant difference in dyspnoea scores was noted (MD ‐0.63, 95% CI ‐1.44 to 0.18; Analysis 1.9). Sensitivity analysis using a fixed‐effect model resulted in statistically significant lower dyspnoea scores in the intervention group compared with the control group (MD ‐0.59, 95% CI ‐0.89 to ‐0.29). The observed effect sizes in the fixed‐ (MD ‐0.59) and random‐ (‐0.63) effects models were comparable. The presence of small‐study effects was considered to be unlikely.

Bourbeau 2003 reported a non‐significant difference of participant‐recorded dyspnoea deterioration in 90% of acute exacerbations in the intervention group versus 88% in the control group. Monninkhof 2003 used breathlessness extracted from two‐week diary data and reported non‐significant between‐group differences. Song 2014 reported a non‐significant difference in the degree of dyspnoea by using the BORG scale (range 0 to 10) after walking between the intervention (7.4 ± 2.0) and control groups (4.8 ± 2.1) after two months of follow‐up.

Other COPD symptoms

Bourbeau 2003 reported non‐significant increases in sputum volume (intervention 54%; control 57%) and purulent sputum was present in 48% of the intervention group and 53% of the control group. Monninkhof 2003 reported non‐significant differences in sputum production over a two‐week period. Whereas borderline beneficial significant differences in mean cough and sputum colour scores were reported for the self‐management intervention group, the study authors stated that these differences probably were not clinically relevant.

Number of COPD exacerbations

Data from four studies (Bösch 2007; Bischoff 2012; Fan 2012; Jennings 2015) on the mean number of exacerbations per participant were not statistically significant (MD 0.01, 95% CI ‐0.28 to 0.29, N = 740; Analysis 1.10). The same effect was found when a fixed‐effect rather than a random‐effects model was used in a sensitivity analysis. Monninkhof 2003 reported an average of 2.8 exacerbations in the intervention group and 1.5 in the control group. This study could not be included in the meta‐analysis because SDs were not reported.

Similar definitions were used for COPD exacerbations among studies. Bischoff 2012 defined exacerbations as a change for at least two consecutive days in either two or more major symptoms (dyspnoea, sputum purulence, sputum amount) or any one major symptom plus at least one minor symptom (colds, wheeze, sore throat, cough). Fan 2012 defined AECOPD as an increase in or new onset of one or more respiratory symptoms (cough, sputum, wheezing, dyspnoea or chest tightness) persisting for at least two days. Jennings 2015 defined an exacerbation as an acute event characterised by a worsening of the participant’s respiratory symptoms beyond normal day‐to‐day variations, leading to a change in medication. Bösch 2007 did not provide a definition of exacerbations, but indicated that exacerbations were treated with antibiotics. Monninkhof 2003 defined exacerbations as worsening of respiratory symptoms that required treatment with a short course of oral corticosteroids or antibiotics.

The total number of exacerbations were reported in five studies (Bischoff 2012; Bourbeau 2003; Fan 2012; Monninkhof 2003; Tabak 2014). Bischoff 2012 reported 280 exacerbations in the intervention group (N=55) and 235 in the control group (N=55) after 24 months of follow‐up. Bourbeau 2003 reported 299 exacerbations in the intervention group (N = 96) and 362 exacerbations in the control group (N = 95) after 12 months of follow‐up. Fan 2012 reported 600 self‐reported exacerbations in the intervention group (N = 209) and 610 in the control group (N = 217) during the first 12 months of follow‐up. Monninkhof 2003 reported 360 exacerbations in the intervention group (N = 127) and 177 exacerbations in the control group (N = 121) after 12 months of follow‐up.

Use of oral steroids and antibiotics

Thirteen studies (Casas 2006; Bourbeau 2003; Bucknall 2012; Hernández 2015; Jennings 2015; Khdour 2009; Kheirabadi 2008; Monninkhof 2003; Ninot 2011; Österlund Efraimsson 2008; Song 2014; Tabak 2014; Titova 2015) did not report any data on the use of oral steroids or antibiotics or both and could not be included in meta‐analyses. Two studies (Bischoff 2012; Khdour 2009) reported data on combined use of oral steroids and antibiotics. Bischoff 2012 reported a similar number of participants who started prednisolone, antibiotics or both to manage exacerbations in the self‐management group (N = 16, 11%) compared to the usual care group (N = 13, 10%) in the first year of follow‐up. In the second year of follow‐up, a higher number of exacerbations in the self‐management group were managed by starting prednisolone, antibiotics or both (OR 3.98, 95% CI 1.10 to 15.58). Khdour 2009 observed a significant difference with less oral steroids and antibiotic courses used in the intervention group compared with the control group (mean use 3.08, 95% CI 2.57 to 3.59 versus mean use 4.03, 95% CI 3.37 to 4.69).

Courses of oral steroids

The use of oral steroids for respiratory problems was reported by six studies (Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Martin 2004; Rea 2004; Rice 2010). However, the number of participants who used at least one course of steroids was available for four studies (Garcia‐Aymerich 2007; Gallefoss 1999; Rea 2004; Rice 2010); data from these studies were included in a meta‐analysis. A non‐significant higher probability of using at least one course of oral steroids in the self‐management group compared with the control group was observed (OR 4.38, 95% CI 0.55 to 34.91, Analysis 1.11), with high heterogeneity (I² = 94%). In this meta‐analysis the probability of using at least one course of oral steroids was reported to be in favour of the usual care group. However, it could also be argued that the higher probability of using at least one course of oral steroids is in favour of the self‐management group; it might lead to earlier appropriate treatment of AECOPD and may prevent hospital admissions. Rice 2010 was an outlier in our meta‐analysis (Analysis 1.11); it included many more participants than the other three studies. In addition, the proportion of participants who received at least one course of oral steroids in the self‐management group reported by Rice 2010 was relatively high (97.6%) compared with the other studies (Garcia‐Aymerich 2007 = 9.5%, Gallefoss 1999 = 69.2%, Rea 2004 = 47.6%). Rice 2010 reported that the much higher rates of oral steroid use in the intervention group suggested that participants were recognising and self‐(over)treating respiratory events that otherwise might have resulted in ED visits or hospital admissions. The OR in Rice 2010 was 32.7 which is probably an overestimation of the risk ratio due to the fact that the event is common. This meta‐analysis should therefore be interpreted with caution.

Fan 2012 reported a significantly higher mean of 2.5 exacerbations per patient‐year treated with prednisolone in the self‐management group compared with 2.1 in the control group (rate ratio 1.25, 95% CI 1.05 to 1.48). In Martin 2004, the frequency of oral prednisolone courses per 12 months was not statistically significant higher in the intervention group (2.3 courses, 95% CI 1.4 to 3.2) compared to the control group (1.3 courses, 95% CI 0.8 to 1.8).

Courses of antibiotics

The use of antibiotics for respiratory problems was reported by six studies (Bösch 2007; Fan 2012; Martin 2004; Mitchell 2014; Rea 2004; Rice 2010). However, the number of participants who used at least one course of antibiotics was available for only two studies (Rea 2004; Rice 2010). A meta‐analysis was not justified. Rea 2004 reported fewer participants receiving at least one course of antibiotics in the intervention group than in the control group (59% versus 69%), whereas Rice 2010 reported the opposite (92% versus 56%). Again, Rice 2010 reported that the much higher rates of antibiotic use in the intervention group suggested that participants were recognising and self‐(over)treating respiratory events that otherwise might have resulted in ED visits or hospital admissions. Bösch 2007 reported a statistically significant reduction in the mean number of exacerbations (2.0 ± SD 1.4 to 1.4 ± SD 1.6) that were treated with antibiotics in the intervention group, with no changes observed in the control group. Fan 2012 reported a non‐significantly higher mean of 2.7 exacerbations per patient‐year treated with an antibiotic in the self‐management group compared with 2.5 in the control group (rate ratio 1.11, 95% CI 0.97 to 1.27). In Martin 2004, there was no significant difference in the use of antibiotics between the groups (intervention 3.6, 95% CI 2.5 to 4.7 versus control 2.5, 95% CI 1.7 to 3.3) after 12 months of follow‐up. Mitchell 2014 also reported no statistically significant difference between groups in the number of antibiotic courses (intervention N = 82 versus control N = 70, OR 1.20, 95% CI 0.77 to 1.86) six months post‐randomisation.

Mortality

Mortality was reported as an outcome measure in five studies (Bucknall 2012; Casas 2006; Fan 2012; Rice 2010; Titova 2015). We extracted mortality data from sections describing the participant flow and reasons for losses to follow‐up from 11 studies (Bourbeau 2003; Gallefoss 1999; Hernández 2015; Khdour 2009; Kheirabadi 2008; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Rea 2004; Tabak 2014). Mortality data reported by Garcia‐Aymerich 2007 could not be included in the meta‐analysis, since the same data were already incorporated in Casas 2006. Five studies provided no information on mortality (Bischoff 2012, N = 110 participants; Bösch 2007, N = 50 participants; Jennings 2015, N = 172 participants; Österlund Efraimsson 2008, N = 52 participants; Song 2014, N = 40 participants) and could not be included in the meta‐analysis.

All‐cause mortality

We included data from 16 studies (3,296 participants) in a meta‐analysis of all‐cause mortality (Bourbeau 2003; Bucknall 2012; Casas 2006; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Kheirabadi 2008; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Rea 2004; Rice 2010; Tabak 2014; Titova 2015). No statistically significant differences in mortality were found between intervention and control group participants (RD 0.00, 95% CI ‐0.02 to 0.03, I² = 48%, Analysis 1.12; Figure 9). Four studies (Gallefoss 1999; Kheirabadi 2008; Ninot 2011; Tabak 2014) reported no deaths in the self‐management and control groups. Sensitivity analysis using a fixed‐effect model resulted in a similar non‐significant effect for all‐cause mortality (RD 0.01, 95% CI ‐0.01 to 0.03).

Figure 9

Forest plot of comparison: Self‐management versus usual care, outcome: 1.13 All‐cause mortality

We included data from 12 studies (2,620 participants) in a subgroup analysis of one‐year all‐cause mortality (Bourbeau 2003; Bucknall 2012; Casas 2006; Gallefoss 1999; Hernández 2015; Khdour 2009; Martin 2004; Monninkhof 2003; Ninot 2011; Rea 2004; Rice 2010; Titova 2015). No statistically significant differences in mortality were found between intervention and control (RD ‐0.0070, 95% CI ‐0.0326 to 0.0186, I² = 33%, Analysis 1.12; Figure 9). Sensitivity analysis using a fixed‐effect model resulted in a similar non‐significant effect for one‐year all‐cause mortality (RD 0.0078, 95% CI ‐0.0128 to 0.0283).

Only two studies (Bourbeau 2003; Titova 2015) provided data on two‐year all‐cause mortality, so meta‐analysis could not be performed. Bourbeau 2003 reported a non‐significant lower two‐year all‐cause mortality rate in the intervention group compared to the usual care group (MD ‐0.05, 95% CI ‐0.16 to 0.05). Titova 2015 reported a non‐significant higher two‐year all‐cause mortality rate in the intervention group compared to the usual care group (MD 0.13, 95% CI ‐0.01 to 0.26).

Respiratory‐related mortality

We included data from seven studies in a meta‐analysis of respiratory‐related mortality (Bucknall 2012; Fan 2012; Gallefoss 1999; Kheirabadi 2008; Ninot 2011 Tabak 2014; Titova 2015). A small, but statistically significant higher, respiratory‐related mortality rate was found for the intervention group compared to the control group (RD 0.028, 95% CI 0.0049 to 0.0511, 1219 participants, I² = 0%, Analysis 1.13; Figure 10). Four studies (Gallefoss 1999; Kheirabadi 2008; Ninot 2011; Tabak 2014) reported no deaths in the self‐management and control groups after 12, 3, 12 and 9 months of follow‐up, respectively. Two studies (Bucknall 2012; Fan 2012) dominated the overall effect after 12 months of follow‐up. A similar small, but significant higher one‐year respiratory‐related mortality rate was found for self‐management compared to usual care (RD 0.03, 95% CI 0.00 to 0.05, four studies, 981 participants, I² = 0%, Analysis 1.13; Figure 10). Sensitivity analysis using a fixed‐effect model resulted in a similar statistically significantly higher respiratory‐related mortality in the intervention group compared to the control group (RD 0.04, 95% CI 0.01 to 0.07).

Figure 10

Forest plot of comparison: Self‐management versus usual care, outcome: 1.14 Respiratory‐related mortality

Self‐efficacy

Only two studies (Bischoff 2012; Bucknall 2012) reported on self‐efficacy, so it was not possible to perform a meta‐analysis. Both studies measured self‐efficacy using the COPD Self‐Efficacy Scale (CSES). Bischoff 2012 reported no statistically significant changes or difference in participants' self‐efficacy between the intervention and control groups according to the CSES total (MD ‐0.17, 95% CI ‐0.64 to 0.30) and domain scores after 24 months of follow‐up. Bucknall 2012 also reported a non‐significant difference in CSES total scores between the intervention and control groups (MD 2.65, 95% CI ‐5.85 to 11.14).

Days lost from work

Two studies reported days lost from work (Gallefoss 1999; Monninkhof 2003), so it was not possible to perform a meta‐analysis. Gallefoss 1999 reported no significant differences between groups. Almost 50% of participants with COPD in this study were employed. Three of 14 (21%) participants in the intervention group and two of 13 (15%) in the control group reported absences from work. Monninkhof 2003 used the term 'restrictive activity days', defined as days on which work was missed or days when activities were significantly reduced because of health problems. A reduction in the average number of restricted activity days during exacerbation recovery was seen in the intervention compared with the control group (4.1 ± 4.2 versus 5.3 ± 5.3), but no significant between‐group differences were detected.

Subgroup analyses

We performed subgroup analysis for two outcomes; HRQoL and respiratory‐related hospital admissions.

Duration of follow‐up

We performed a subgroup analysis for duration of follow‐up to assess the short‐ and long‐term effects of self‐management compared to usual care. Six studies (Jennings 2015; Kheirabadi 2008; Mitchell 2014; Österlund Efraimsson 2008; Song 2014; Tabak 2014) reported follow‐up periods shorter than 12 months after the start of the study. Sixteen studies (Bischoff 2012; Bösch 2007; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Hernández 2015; Khdour 2009; Martin 2004; Monninkhof 2003; Ninot 2011; Rea 2004; Rice 2010; Titova 2015) reported long‐term follow‐up (12 or more months of follow‐up after the start of the study).

It was not possible to perform follow‐up subgroup analysis for the effects on HRQoL, because follow‐up inf the 10 included studies were all long‐term (≥ 12 months). In addition, a subgroup analysis based on a follow‐up duration with a cut‐off point of 18 months was not possible to perform, since the criterion of at least three studies per subgroup was not met.

There was no statistically significant difference in respiratory‐related hospital admissions among studies with long‐term (n = 11) or short‐term follow‐up (n = 3) (test for subgroup differences: Chi² = 0.02, df = 1 (P = 0.90), I² = 0 %, Analysis 2.1). It was not possible to perform a subgroup analysis with six months as cut‐off point for the effects on respiratory‐related hospitalisations, since this resulted in an insufficient number of studies for the subgroup analysis. A cut‐off point of 18 months for the duration of follow‐up resulted in a subgroup with only two studies (Bourbeau 2003; Titova 2015) and therefore we could not perform a subgroup analysis.

COPD stability at time of inclusion

Five studies (Casas 2006; Garcia‐Aymerich 2007; Jennings 2015; Song 2014; Titova 2015) reported inclusion of participants with COPD who were in the unstable phase; eight studies (Bourbeau 2003; Fan 2012; Hernández 2015; Martin 2004; Mitchell 2014; Monninkhof 2003; Ninot 2011; Tabak 2014) reported inclusion of participants in the stable phase; and nine studies (Bischoff 2012; Bösch 2007; Bucknall 2012; Gallefoss 1999; Khdour 2009; Kheirabadi 2008; Österlund Efraimsson 2008; Rea 2004; Rice 2010) did not report if participants were in stable or unstable phases. It was not possible to perform a subgroup analysis on the inclusion of participants in the unstable phase versus the stable phase for effects on HRQoL or on the number of participants with at least one respiratory‐related hospital admission, because of the relatively small number of studies that reported inclusion of participants in the unstable phase.

Use of a standardised exercise programme

We performed subgroup analyses on the use of a standardised exercise programme as part of the self‐management intervention. No statistically significant difference was observed for the effects on HRQoL observed among studies (n = 4) with an exercise programme and studies (n = 6) without an exercise programme (test for subgroup differences: Chi² = 0.10, df = 1 (P = 0.76), I² = 0%, Analysis 2.2). The difference in effects on respiratory‐related hospital admissions among studies with (n = 6) and without (n = 8) an exercise programme was not statistically significantly different between subgroups (test for subgroup differences: Chi² = 0.79, df = 1 P = 0.37, I² = 0% Analysis 2.3).

Use of a smoking cessation programme

Studies included for subgroup analyses on use of a smoking cessation programme reported no statistically significant between‐group baseline differences in smoking status. There were two studies (Khdour 2009; Hernández 2015) with a smoking cessation programme and one study (Titova 2015) without, in which changes in smoking rates over time were observed. Khdour 2009 observed 22.2% self‐reported abstinence in the self‐management group at six‐ and 12‐months follow‐up compared with 5.3% and 10.5% in the usual care group smokers, respectively. However, the differences in stage of change status in relation to smoking did not reach statistical significance (Khdour 2009). After 12 months of follow‐up, Hernández 2015 reported a statistically significantly lower percentage of current smokers (self‐management 3% versus usual care 16%). Titova 2015 reported a non‐significant trend toward reduction in the percentage of current smokers in the self‐management group from 35.3% at baseline to 31.4% after 12 months and 27.5% after 24 months. In the usual care group these percentages were 30.6% at baseline and after 12 months, and 26.5% after 24 months.

Subgroup analyses on the use of a smoking cessation programme as part of the self‐management intervention showed a statistically significantly larger improvement in HRQoL in the three studies (Hernández 2015; Khdour 2009; Rice 2010) with a smoking cessation programme (MD ‐4.98, 95% CI ‐7.17 to ‐2.78, Analysis 2.4) compared to the seven studies (Bourbeau 2003; Bucknall 2012; Garcia‐Aymerich 2007; Fan 2012; Gallefoss 1999; Monninkhof 2003; Ninot 2011) without a smoking cessation programme (MD ‐1.33, 95% CI ‐2.94 to 0.27, test for subgroup differences: Chi² = 6.89, df = 1 (P = 0.009), I² = 85.5%).

No statistically significant effect was observed in a subgroup analysis of four studies with and ten studies without a smoking cessation programme on the probability of respiratory‐related hospital admissions in the self‐management group compared to usual care (test for subgroup differences: Chi² = 0.00, df = 1 (P = 0.98), I² = 0%, Analysis 2.5).

Self‐management as part of usual care

In Song 2014, self‐management was likely to be part of usual care, so it was not possible to perform a subgroup analysis on the level of self‐management as part of usual care. Song 2014 reported that the control group received usual care consisting of education on COPD management, proven benefits of exercise, and maintaining daily activities.

Integration of behavioural change techniques (BCT) clusters

No statistically significant difference was observed for the effects on HRQoL among studies (n = 6) with a high number of BCT clusters (higher than the median number of 9.5) and studies (n = 4) with few BCT clusters (test for subgroup differences: Chi² = 0.01, df = 1 (P = 0.94), I² = 0%, Analysis 2.6).

There were no statistically significant differences observed in respiratory‐related hospital admissions among studies (n = 7) with a high number of BCT clusters versus studies (n = 7) with few BCT clusters (test for subgroup differences: Chi² = 0.82, df = 1 (P = 0.37), I² = 0%, Analysis 2.7). An additional subgroup analysis using a lower cut‐off point for BCT clusters (> 8 BCT clusters (n = 10) versus ≤ 8 BCT clusters (n = 4) integrated) showed no statistically significant differences in respiratory‐related hospital admissions (test for subgroup differences: Chi² = 0.00, df = 1 (P = 0.97), I² = 0%, Analysis 2.8).

Case manager support

In this review, case manager support was defined as unscheduled ongoing support from a case manager based on individual needs and capabilities in which reinforcement is directed to the patient’s self‐management skills, and delivered face‐to‐face, by telephone or by telemedicine. We included ten studies that reported case manager support (Bischoff 2012; Bourbeau 2003; Bucknall 2012; Casas 2006; Garcia‐Aymerich 2007; Fan 2012; Monninkhof 2003; Rice 2010; Tabak 2014; Titova 2015). No statistically significant difference was observed of effects on HRQoL among studies with case manager support (n = 6) and those without case manager support (n = 4) (test for subgroup differences: Chi² = 1.86, df = 1 (P = 0.17), I² = 46.1%, Analysis 2.9).

No statistically significant differences were observed for effects on respiratory‐related hospital admissions among the eight studies with case manager support and the six studies without case manager support (test for subgroup differences: Chi² = 0.13, df = 1 (P = 0.72), I² = 0%, Analysis 2.10).

Duration of intervention

Subgroup analyses on duration of the self‐management intervention showed no statistically significant differences in HRQoL in studies of at least six months of intervention duration (MD ‐2.96, 95% CI ‐5.20 to ‐0.72) compared to studies with less than six months intervention duration (MD ‐2.57, 95% CI ‐6.96 to 1.82, test for subgroup differences: Chi² = 0.02, df = 1 (P = 0.88), I² = 0%, Analysis 2.11).

There was no statistically significant difference in respiratory‐related hospital admissions among studies with longer intervention durations (OR 0.65, 95% CI 0.43 to 0.96) compared to studies of less than six months intervention duration (OR 0.84, 95% CI 0.53 to 1.32, test for subgroup differences: Chi² = 0.68, df = 1 (P = 0.41), I² = 0%, Analysis 2.12).

Action plan components

We performed subgroup analyses on the different components of the action plans for COPD exacerbations. There was no statistically significant difference in HRQoL effect among studies that defined an action for adaptation of maintenance medication (MD ‐3.75, 95% CI ‐6.16 to ‐1.33) and studies that had not defined this action in their action plans for COPD exacerbations (MD ‐2.02, 95% CI ‐4.77 to 0.72, test for subgroup differences: Chi² = 0.85, df = 1 (P = 0.36), I² = 0%, Analysis 2.13). Nor was there a statistically significant difference in effect on respiratory‐related hospital admissions in studies that included an action for adaptation of maintenance medication (OR 1.01, 95% CI 0.54 to 1.88) compared to studies that did not include this action (OR 0.59, 95% CI 0.42 to 0.83, test for subgroup differences: Chi² = 2.16, df = 1 (P = 0.14), I² = 53.7%, Analysis 2.14). Two studies (Hernández 2015; Ninot 2011) defined an action 'when to avoid situations in which viral infections might be prevalent' and reported data on the HRQoL. It was not possible to perform a subgroup analysis on the action plan component of 'avoiding situations in which viral infections might be prevalent'. There was no statistically significant difference observed in respiratory‐related hospital admissions in studies that defined an action 'when to avoid situations in which viral infections might be prevalent' (OR 0.88, 95% CI 0.25 to 3.13) compared to studies that did not include this action (OR 0.68, 95% CI 0.50 to 0.91, test for subgroup differences: Chi² = 0.16, df = 1 (P = 0.69), I² = 0%, Analysis 2.15). Four studies (Kheirabadi 2008; Martin 2004; Ninot 2011; Song 2014) did not define an action 'when to contact healthcare providers for support'. Only one study (Ninot 2011) did not define an action 'when to contact healthcare providers for support' and reported data on HRQoL or respiratory‐related hospital admissions, so we could not perform subgroup analyses. Two studies (Jennings 2015; Kheirabadi 2008) did not include self‐recognition of COPD exacerbations in their action plans and these studies did not define an action of 'when to self‐initiate treatment of a COPD exacerbation'. We were unable to perform subgroup analyses on these action plan components. Two studies (Hernández 2015; Martin 2004) reported an action 'when to initiate self‐treatment of comorbidities'. There were too few studies for subgroup analysis on the self‐initiation of comorbidities as a COPD exacerbation action plan component.

Discussion

Summary of main results

We systematically evaluated 22 randomised controlled trials (RCTs) (described in 30 articles) on the effectiveness of Chronic Obstructive Pulmonary Disease (COPD) self‐management interventions that included an action plan for acute exacerbations of Chronic Obstructive Pulmonary Disease (AECOPD) in comparison with usual care. An action plan was defined as an agreed strategy including actions to be initiated by people with COPD when symptoms deteriorate.

We observed a statistically significant beneficial effect of self‐management on health‐related quality of life (HRQoL) over 12 months, measured by the St. George's Respiratory Questionnaire (SGRQ) adjusted total score (MD ‐2.69, 95% CI ‐4.49 to ‐0.90; 10 studies; N = 1,582). The pooled mean difference (MD) of SGRQ total score did not reach the minimal clinically important difference (MCID) of four points and therefore could not be considered as clinically relevant (Jones 2005).

A beneficial self‐management effect was also observed for respiratory‐related hospital admissions as reported in 14 studies with 3157 participants. Participants in self‐management intervention study arms that included an action plan for AECOPD were at statistically significantly lower risk for at least one respiratory‐related hospital admission compared with participants who received usual care (OR 0.69, 95% CI 0.51 to 0.94). The number needed to treat to prevent one respiratory‐related hospital admission over one year was 12 (95% CI 7 to 69) for participants with high baseline risk and 17 (95% CI 11 to 93) for participants with low baseline risk.

We observed no statistically significant difference in the probability of at least one all‐cause hospital admission in the self‐management intervention group compared to the usual care group (OR 0.74, 95% CI 0.54 to 1.03; 14 studies; N = 2,467). Furthermore, we observed no statistically significant difference in the number of all‐cause hospitalisation days (MD ‐0.65, 95% CI ‐2.01 to 0.71), emergency department (ED) visits (MD ‐0.31, 95% CI ‐0.74 to 0.12), General practitioner (GP) visits (MD ‐0.36, 95% CI ‐2.64 to 1.93) and modified Medical Research Council questionnaire ((m)MRC) dyspnoea scores (MD ‐0.63, 95% CI ‐1.44 to 0.18). There was no statistically significant effect observed for self‐management on the number of COPD exacerbations (MD 0.01, 95% CI ‐0.28 to 0.29) and no excess all‐cause mortality risk was observed (RD 0.0019, 95% CI ‐0.0225 to 0.0263) in 16 studies (N = 3,296). However, a small, but statistically significant higher respiratory‐related mortality rate was observed in the self‐management intervention group compared to the usual care group (RD 0.028, 95% CI 0.0049 to 0.0511) in seven studies (N = 1,219).

Subgroup analyses

Subgroup analyses showed significantly more improvement in health‐related quality of life (HRQoL) in studies that included a smoking cessation programme as part of the self‐management intervention (MD ‐4.98, 95% CI ‐7.17 to ‐2.78) compared to studies with no smoking cessation programme (MD ‐1.33, 95% CI ‐2.94 to 0.27). The number of behavioural change technique (BCT) clusters integrated in the self‐management intervention, intervention duration, and adaptation of maintenance medication as part of an action plan did not affect HRQoL. Subgroup analyses did not detect potential explanatory variables for differences in respiratory‐related hospital admissions among studies.

Overall completeness and applicability of evidence

Our review showed a beneficial effect on HRQoL and respiratory‐related hospital admissions in a group of studies that differed considerably with regard to follow‐up duration, intervention duration, and self‐management and action plan components. The results were based on a total of 3,854 participants with COPD, verified with a post‐bronchodilator FEV₁ to FVC ratio < 0.70. We included studies performed in 14 countries on four continents (14 in Europe, 4 in North America, 2 in Asia and 2 in Oceania).

In our review, self‐management interventions including AECOPD action plans were associated with improvement in HRQoL (measured by the SGRQ) and lower probability of respiratory‐related hospital admissions. Although the improvement in HRQoL did not reach the MCID, self‐management interventions are part of COPD management and should be based on individualised assessment of COPD to reduce: 1) current symptoms to decrease personal burden and improve HRQoL; and 2) future risks of exacerbations, hospitalisations, mortality and costs (GOLD 2017). We observed no statistically significant difference in the probability of all‐cause hospital admissions, the number of all‐cause hospitalisation days, ED visits, GP visits, and dyspnoea scores as measured by the (m)MRC questionnaire for participants in self‐management interventions compared to usual care. No excess all‐cause mortality risk was observed, but exploratory analysis indicated a small significant higher respiratory‐related mortality rate for self‐management compared to usual care. Subgroup analyses indicated significant improvements in HRQoL from self‐management interventions with a smoking cessation programme. The number of BCT clusters integrated in the self‐management intervention, intervention duration, inclusion of a standardised exercise programme, and adaptation of maintenance medication as part of an action plan did not affect HRQoL.

There are some limitations for the generalisability of our results. We had difficulties with information collection from three studies (Koff 2009; Leiva‐Fernández 2014; Lou 2015). We made five attempts to request information from the authors of these studies on whether an action plan for AECOPD was used. No responses were received so we could not verify if these studies met our inclusion criteria. In addition, one included study (Österlund Efraimsson 2008) could not be included in the meta‐analyses because insufficient data were provided.

Three studies (14%) had follow‐up durations of three months or less (Jennings 2015; Kheirabadi 2008; Song 2014). Depending on the time of participant enrolment (e.g., during summer) in these three studies, seasonal variation may have influenced outcomes (e.g., the number of exacerbations) and may have resulted in under‐ or overestimation of the actual effect. It was also difficult to interpret behavioural change effects for studies with short follow‐up durations. Since the study by Fan 2012 was prematurely stopped with a mean follow‐up of 250 days, it is uncertain if a true effect was observed. The results of this study need to be interpreted with caution.

In addition, some hospitalisations may have been triggered by the COPD self‐management intervention because AECOPD action plans encouraged people to seek help when they may not have otherwise and therefore increased healthcare utilisation. However, the reduction in hospitalisations found in this review strengthens our hypothesis that self‐recognition and self‐treatment of symptoms prevent some of the severe exacerbations that otherwise would have needed hospitalisation. The definition of an exacerbation is also a factor that can influence the number of exacerbations found (Effing 2009b). For example, in Monninkhof 2003 an exacerbation was not based on an increase of symptoms, but on the number of courses of prednisolone and an additional course of antibiotics in the case of increased purulent sputum. This number of courses was driven by the self‐management intervention, which was based on symptoms, and the corresponding action plan stated to initiate self‐treatment with prednisolone and antibiotics if needed. For each individual it is important to recognise what constitutes an exacerbation and to identify what the usual symptoms are in a person's stable health state for COPD and comorbidities (Lenferink 2013; Zwerink 2016). Because of heterogeneity in exacerbations and other individual characteristics, tailoring of (standardised) action plans should always be considered.

Furthermore, usual care is diverse among countries, healthcare systems and populations. Although we excluded studies that did not include a usual care group, it was likely that in one study (Song 2014) self‐management was integrated as part of usual care. The study authors indicated that usual care management was directed toward COPD management education, exercise, and maintaining daily activities (Song 2014). Moreover, effects may be the result of optimised COPD management (e.g., medication treatment) during the self‐management intervention or results may reflect better compliance and concordance with medication treatment in the intervention group(Khdour 2009).

Data were skewed for continuous outcomes (number and duration of hospital admissions, number of exacerbations). In the analyses of mean differences these skews may have led to reduced power to detect a treatment difference for these continuous outcomes. The analyses of Incident Rate Ratios using regression models would have been more appropriate to use to reduce the impact of the skew. However, we could not perform these analyses, because we individual study data were not available.

Differences in study design and characteristics of included participants were not taken into account in the analyses of this review. An analysis of individual participant data, such as Jonkman 2016a and Jonkman 2016b, could contribute to the knowledge of factors influencing proper self‐management. The additional results of the recently published studies and the review with individual participant data do not automatically fit with the results reported in the current review. Future review updates should demonstrate how gained knowledge from recent studies influences and fits the results of the current meta‐analyses.

Quality of the evidence

We graded the quality of evidence for HRQoL as high. However, the significant improvement in HRQoL did not reach the MCID, and may therefore only have been clinically relevant for part of the population. We graded the quality of evidence for all‐cause mortality as high; and moderate for respiratory‐related hospital admissions because substantial heterogeneity resulted in inconsistency. We graded the quality of evidence for all other secondary outcomes as moderate to very low; assessments were based on fewer studies or smaller sample sizes, or both. The quality of evidence for respiratory‐related mortality was downgraded to very low because, as well as few studies and small sample sizes, the overall effect was driven by two of the seven studies (Bucknall 2012; Fan 2012); four studies (Gallefoss 1999; Kheirabadi 2008; Ninot 2011; Tabak 2014) had no events, and there was a high risk of bias for incomplete outcome data and selective reporting for three studies (Bucknall 2012; Tabak 2014; Titova 2015).

Potential biases in the review process

Debate about the definition and most effective content of COPD self‐management interventions is ongoing (Effing 2012). Although we included only studies that aligned with the most recent published conceptual definition of COPD self‐management interventions (Effing 2016), the self‐management interventions were diverse in duration (2 to 24 months of follow‐up), self‐management intervention components (one to six self‐management components), and action plan components (one to six actions defined). Furthermore, a large variety of topics were included in the educational sessions. Operationalisation of the conceptual definition of a COPD self‐management intervention would be helpful to refine future eligibility criteria and thus reduce heterogeneity in interventions.

The inclusion of studies in this review was not based on reported outcome measures. Hence, the included studies used a broad spectrum of outcome measures with different methods for assessment (e.g., different questionnaires) and different calculations (e.g., mean number versus the percentage of participants). This added to heterogeneity among studies. Furthermore, there were insufficient data available for some outcome measures, even after contact with study authors. Moreover, some meta‐analyses could not be performed due to insufficient (< 3 studies) reported outcome data.

Because of the nature of the self‐management intervention, we expected a priori to see clinical heterogeneity among studies so we decided to use random‐effects modelling for the meta‐analyses. The random‐effects model weighs by study rather than number of participants when heterogeneity is present. When only a few large studies and many small studies are included, this may result in bias introduced by small‐study effects. We therefore checked the fixed weights in sensitivity analyses. The beneficial effects of the self‐management intervention on all‐cause hospital admissions and all‐cause hospitalisation days became statistically significant when the fixed‐effect model was used instead of random‐effects modelling. Since the observed effect sizes in fixed‐effect and random‐effects modelling were comparable, the bias introduced by small‐study effects was considered to be unlikely.

Agreements and disagreements with other studies or reviews

Action plans for AECOPD

A written action plan for AECOPD was a requisite for inclusion of studies reporting self‐management interventions in this review. Multi‐component self‐management action plans with iterative processes aimed at sustained behavioural change, providing support and instilling confidence for self‐recognition of AECOPD are recognised as important factors to self‐manage symptoms effectively and safely (Bourbeau 2009; Effing 2012). The actions defined for AECOPD differed among studies (e.g., take direct action when symptoms get worse versus start action 48 hours after onset of symptoms if AECOPD symptoms persist or do not improve), and were not always very detailed (e.g., participants could call a team if they think they have an infection and the team would “maybe” advise to take antibiotics (Hernández 2015)). Because AECOPD self‐recognition, self‐treatment, and contacting healthcare providers for support were included in the AECOPD action plans in almost all included studies, we could not perform subgroup analyses. As a result, we were unable to determine the effectiveness of these action plan components and the most effective component of action plans.

Many people with COPD have comorbidities (Annecchino 2007; Vanfleteren 2013), which has an impact on disease severity, hospital admissions and survival (Divo 2012; Vestbo 2013). Tailored approaches with individualised care plans are needed to reduce the treatment burden and optimise care for people with COPD and comorbid conditions (Vanfleteren 2017). Using COPD‐specific action plans for people with COPD and comorbidities may lead to delayed or incorrect treatment due to symptom overlap (e.g., breathlessness may be caused by COPD, but also by heart failure or anxiety). Future COPD self‐management action plans should account for comorbidities. This would not only increase the safety of COPD self‐management interventions by appropriate and timely treatment actions, but would likely also increase benefits for all‐cause hospital admissions. Unfortunately, only two (Hernández 2015; Martin 2004) of the 11 studies that included participants with the added complexity of major comorbidities defined an action for the self‐treatment of comorbidities. Therefore, we were unable to evaluate the effects of tailoring action plans for people with comorbidities in this review.

Health‐related quality of life (HRQoL)

Previously reported COPD self‐management review data on HRQoL showed similar mean differences in SGRQ total scores. In the most recent Cochrane Review evaluating the effects of self‐management interventions in people with COPD, not focusing on action plan use, a MD of ‐3.51 (95% CI ‐5.37 to ‐1.65) was observed for the SGRQ total score and a MD of ‐2.68 (95% CI ‐4.16 to ‐1.20) for the change from baseline SGRQ total score (Zwerink 2014). These results are very comparable to our findings (MD ‐2.69, 95% CI ‐4.49 to ‐0.90). In Zwerink 2014, action plans were part of most study interventions. The review authors could not perform subgroup analyses and were unable to confirm whether action plans were an essential component of self‐management (Zwerink 2014). The main HRQoL effects reported by the current review are also in line with recently published individual participant data (IPD) meta‐analyses on the effectiveness of self‐management (Jonkman 2016a; Jonkman 2016b). Although we were unable to perform a subgroup analysis on follow‐up duration, Jonkman 2016b showed improved HRQoL at 12 months with a standardised mean difference (SMD) of 0.08, but not at six months (SMD 0.05). Subgroup analyses did not show a consistent pattern across health outcomes for participants benefiting most from the self‐management interventions (Jonkman 2016b).

Self‐management interventions aim to change health behaviours (Bourbeau 2004; Lorig 2003), one of which in many people is smoking. Smoking cessation programmes are currently considered by all evidence‐based and society guidelines as an essential component of care to help quit smoking and stay abstinent (GOLD 2017) and should be offered at the earliest possible stage. This implies ensuring that smoking cessation could be routinely offered in primary care. We observed a clinically relevant and significantly better HRQoL resulting from COPD self‐management interventions including smoking cessation programmes (MD ‐4.98) compared to interventions without smoking cessation programmes (MD ‐1.33). Although we could not compare our findings with other reviews, our results indicate that a smoking cessation programme seems to be an essential part of self‐management interventions to achieve a clinically relevantly improved HRQoL. Smoking cessation could also be offered and delivered to people as part of self‐management interventions to achieve optimal improvement in HRQoL.

Hospital admissions

Participants in self‐management interventions that included AECOPD action plans were at a significantly lower risk for at least one respiratory‐related hospital admission compared with those who received usual care (OR 0.69). Earlier reviews show similar beneficial effects of self‐management on respiratory‐related hospital admissions. A lower risk for at least one respiratory‐related hospital admission was observed in the review on self‐management interventions (OR 0.57) (Zwerink 2014). Recent IPD meta‐analysis showed a significant risk reduction at 12 months of follow‐up (RR 0.77) and interventions improved the time to the first respiratory‐related hospital admission (HR 0.79) (Jonkman 2016b). Whether these lower risks are clinically relevant is unclear, because there is no MCID for hospital admissions. However, a lower number of hospital admissions would potentially result in better HRQoL, reduced mortality, and a reduction in healthcare costs (Effing 2009a). Our subgroup analyses did not identify any specific components of self‐management interventions that were linked to the risk reduction of respiratory‐related hospital admissions.

We observed no significant difference in all‐cause hospital admissions. Based on the observed effect size (OR 0.74, 95% CI 0.54 to 1.03), its 95% CI and low power, we could not rule out there is no actual difference. Significant effects on all‐cause hospital admissions were found in previously published COPD self‐management reviews in which self‐management interventions led to a somewhat lower OR for at least one all‐cause hospital admission (OR 0.60) (Zwerink 2014), reduced relative risk for all‐cause hospital admission within 12 months (RR 0.84), and a longer time to the first all‐cause hospital admission (HR 0.80) (Jonkman 2016b). Because all AECOPD action plans were COPD‐specific in our review, it was perhaps unlikely that the interventions would lead to a reduced risk of all‐cause hospital admissions. This could probably only be expected for two studies (Hernández 2015; Martin 2004) that also defined an action for self‐treatment of comorbidities. This was not reflected in the study results; Hernández 2015 showed an unexplained opposite beneficial effect for usual care, and Martin 2004 provided insufficient data to enable meta‐analysis. A trend toward lower probability of respiratory‐related hospital admissions was present when case manager support was included in COPD self‐management interventions.

Mortality

Like this review, the authors of a previous Cochrane Review on COPD self‐management interventions did not observe an effect from self‐management on all‐cause mortality. Zwerink 2014 observed a trend toward lower all‐cause mortality for self‐management compared to usual care (OR 0.79, 95% CI 0.58 to 1.07). However, this current review includes some more recently conducted large studies, including Fan 2012 which was prematurely terminated because of significantly higher mortality rates in the intervention group. No effects were observed on all‐cause mortality (RD 0.00, 95% CI ‐0.02 to 0.03). Nevertheless, we observed a small, but statistically significantly higher respiratory‐related mortality rate in the self‐management intervention group compared to the usual care group (RD 0.04, 95% CI 0.01 to 0.07). However, these respiratory mortality data should be interpreted with caution because: 1) differentiating between ‘mortality with respiratory problems as a contributing factor’ and ‘respiratory‐specific mortality’ is challenging and misclassification is common (Vestbo 2004), future studies should ensure that classification of death is performed in the same way in all study groups to avoid any bias; 2) the overall effect on respiratory‐related mortality was dominated by two studies (Bucknall 2012; Fan 2012); and, most importantly, 3) the robust analyses for all‐cause mortality did not show any effect (nor trend) toward higher mortality due to self‐management. Since none of the seven included studies where respiratory‐related mortality was an a priori defined outcome, there may be a risk that the cause of mortality was defined differently in the study groups (misclassification). Preliminary findings from a recent large home‐based multi‐component COPD self‐management intervention with 319 participants showed unambiguously higher mortality rates in the usual care group (N = 23 (14.2%)) compared to self‐management (N = 3 (1.9%)) that were mainly respiratory‐related (Bourbeau 2016).

Other secondary outcomes

All‐cause hospitalisation days, ED visits, GP visits and (m)MRC dyspnoea scores showed no difference in healthcare utilisation where self‐management with action plans for AECOPD were used. Trappenburg 2011 observed that beneficial effects for self‐management resulted from improved skills for self‐recognition of AECOPD, quicker start of appropriate self‐initiated treatment, and decreased impact of exacerbations on health status and accelerated recovery. A reduction in dyspnoea score was observed in a Cochrane Review on self‐management interventions for COPD (Zwerink 2014). The review authors reasoned that the reduction may be related to components of self‐management interventions directed to learning strategies to cope with breathlessness (Norweg 2013; Zwerink 2014). In this review, coping with breathlessness or breathing techniques was discussed in all but two included studies.

Although we used an established taxonomy (Michie 2013) to assess the integration of behavioural change techniques (BCTs) into self‐management interventions, we observed no differences in HRQoL and respiratory‐related hospital admissions among studies with high and low integration levels of BCT clusters. The additional value of integrating BCT clusters was difficult to determine. Our inclusion criteria required that studies contained at least four BCTs (goals and (action) planning, feedback and monitoring, shaping knowledge, and associations). The lowest number of BCTs that we extracted from the included studies in our review was six. We expect the actual number of applied BCTs to be higher since we only extracted data that what was explicitly reported. To increase the meaningfulness of the BCT subgroup analysis, future studies should provide more detailed information regarding the behavioural techniques that were integrated.

Recently published studies

We searched up to May 2016 and fully incorporated the results of these trials into this review. An update search conducted in 2017 identified several new studies published on the effectiveness of self‐management interventions (Benzo 2016; Chien 2016; Davis 2016; Imanalieva 2016; Koff 2009; Leiva‐Fernández 2014; Licskai 2016; Lou 2015; Sánchez‐Nieto 2016; Sano 2016; Silver 2017; Zwar 2016). These will be fully incorporated in a future update of this review.

Figure 1

Study flow diagram

Figure 2

Risk of bias summary for each study according to authors' judgements

Figure 3

Funnel plot of comparison: Self‐management versus usual care, outcome: 1.1 HRQoL: adjusted SGRQ total score

Figure 4

Funnel plot of comparison: Self‐management versus usual care, outcome: 1.2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission)

Figure 5

Forest plot of comparison: Self‐management versus usual care, outcome: 1.1 HRQoL: adjusted SGRQ total score after 12 months of follow‐up

Figure 6

Forest plot of comparison: Self‐management versus usual care, outcome: 1.2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission)

Figure 7

Figure 8

Figure 9

Forest plot of comparison: Self‐management versus usual care, outcome: 1.13 All‐cause mortality

Figure 10

Forest plot of comparison: Self‐management versus usual care, outcome: 1.14 Respiratory‐related mortality

Analysis 1.1

Comparison 1 Self‐management versus usual care, Outcome 1 HRQoL: adjusted SGRQ total score.

Analysis 1.2

Comparison 1 Self‐management versus usual care, Outcome 2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission).

Analysis 1.3

Comparison 1 Self‐management versus usual care, Outcome 3 Healthcare utilisation: respiratory‐related hospital admissions (mean number per patient).

Analysis 1.4

Comparison 1 Self‐management versus usual care, Outcome 4 Healthcare utilisation: all‐cause hospital admissions (number of patients with at least one admission).

Analysis 1.5

Comparison 1 Self‐management versus usual care, Outcome 5 Healthcare utilisation: all‐cause hospital admissions (mean number per patient).

Analysis 1.6

Comparison 1 Self‐management versus usual care, Outcome 6 Healthcare utilisation: all‐cause hospitalisation days (per patient).

Analysis 1.7

Comparison 1 Self‐management versus usual care, Outcome 7 Healthcare utilisation: emergency department visits (mean number per patient).

Analysis 1.8

Comparison 1 Self‐management versus usual care, Outcome 8 Healthcare utilisation: GP visits (mean number per patient).

Analysis 1.9

Comparison 1 Self‐management versus usual care, Outcome 9 Health status: (modified) Medical Research Council Dyspnoea Scale ((m)MRC).

Analysis 1.10

Comparison 1 Self‐management versus usual care, Outcome 10 COPD exacerbations (mean number per patient).

Analysis 1.11

Comparison 1 Self‐management versus usual care, Outcome 11 Courses of oral steroids (number of patients used at least one course).

Analysis 1.12

Comparison 1 Self‐management versus usual care, Outcome 12 Mortality: all‐cause mortality.

Analysis 1.13

Comparison 1 Self‐management versus usual care, Outcome 13 Mortality: respiratory‐related mortality.

Analysis 2.1

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 1 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by follow‐up duration).

Analysis 2.2

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 2 HRQoL: adjusted SGRQ total score (subgroup by exercise programme).

Analysis 2.3

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 3 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by exercise programme).

Analysis 2.4

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 4 HRQoL: adjusted SGRQ total score (subgroup by smoking cessation programme).

Analysis 2.5

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 5 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by smoking cessation programme).

Analysis 2.6

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 6 HRQoL: adjusted SGRQ total score (subgroup by median number of BCT clusters).

Analysis 2.7

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 7 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by median number of BCT clusters).

Analysis 2.8

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 8 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by number of BCT clusters).

Analysis 2.9

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 9 HRQoL: adjusted SGRQ total score (subgroup by case manager support).

Analysis 2.10

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 10 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by case manager support).

Analysis 2.11

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 11 HRQoL: adjusted SGRQ total score (subgroup by intervention duration).

Analysis 2.12

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 12 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by intervention duration).

Analysis 2.13

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 13 HRQoL: adjusted SGRQ total score (subgroup by action plan component 'adaptation of maintenance medication').

Analysis 2.14

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 14 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by action plan component 'adaptation of maintenance medication'.

Analysis 2.15

Comparison 2 Subgroup analysis self‐management versus usual care, Outcome 15 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by action plan component 'when to avoid situations in which viral infections might be prevalent').

Summary of findings for the main comparison. Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD

Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD
Patient or population: patients with chronic obstructive pulmonary disease (COPD) Setting: hospital, outpatient clinic, primary care, home‐based Intervention: self‐management interventions including action plans for COPD exacerbations Comparison: usual care
Outcomes	*Anticipated absolute effects^ (95% CI)**		Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
Outcomes	Risk with usual care	Risk with self‐management interventions including action plans for exacerbations	Relative effect (95% CI)	№ of participants (studies)	Quality of the evidence (GRADE)	Comments
Health‐related quality of life (HRQoL) assessed with: St. George's Respiratory Questionnaire adjusted total score Scale from: 0 to 100 follow up: 12 months	The mean HRQoL ranged from 37.7 to 70.4 points	MD 2.69 points lower (4.49 lower to 0.9 lower)	‐	1582 (10 RCTs)	⊕⊕⊕⊕ HIGH	Lower score indicates better health‐related quality of life.
Respiratory‐related hospital admissions assessed with: number of patients with at least one respiratory‐related hospital admission follow up: range 6 months to 24 months	312 per 1,000	238 per 1,000 (188 to 298)	OR 0.69 (0.51 to 0.94)	3,157 (14 RCTs)	⊕⊕⊕⊝ MODERATE ¹
All‐cause hospital admissions assessed with: number of patients with at least one all‐cause hospital admission follow up: range 6 months to 12 months	427 per 1000	356 per 1,000 (287 to 434)	OR 0.74 (0.54 to 1.03)	2,467 (10 RCTs)	⊕⊕⊕⊝ MODERATE ²
All‐cause mortality assessed with: number of all‐cause deaths follow up: range 3 months to 24 months	102 per 1000	107 per 1,000 (74 to 153)	OR 1.06 (0.71 to 1.59)	3,296 (16 RCTs)	⊕⊕⊕⊝ MODERATE³	Pooled risk difference of 0.0019 (95% CI ‐0.0225 to 0.0263).
Respiratory‐related mortality assessed with: number of respiratory‐related deaths follow up: range 3 months to 24 months	48 per 1000	89 per 1,000 (57 to 136)	OR 1.94 (1.20 to 3.13)	1,219 (7 RCTs)	⊕⊝⊝⊝ VERY LOW ⁴	Pooled risk difference of 0.028 (95% CI 0.0049 to 0.0511).
Dyspnoea assessed with: (modified) Medical Research Council Dyspnoea Scale Scale from: 0 to 4 follow up: 12 months	The mean dyspnoea ranged from 2.4 to 2.6	MD 0.63 lower (1.44 lower to 0.18 higher)	‐	217 (3 RCTs)	⊕⊕⊝⊝ LOW ⁵	Lower score indicates improvement in dyspnoea.
COPD exacerbations assessed with: number of COPD exacerbations per patient follow up: range 3 months to 24 months ⁷	The mean COPD exacerbations ranged from 1.13 to 4.3	MD 0.01 higher (0.28 lower to 0.29 higher)	‐	740 (4 RCTs)	⊕⊕⊕⊝ MODERATE ⁶
Courses of oral steroids assessed with: number of patients who used at least one course of oral steroids follow up: 12 months	497 per 1000	812 per 1000 (352 to 972)	OR 4.38 (0.55 to 34.91)	963 (4 RCTs)	⊕⊕⊝⊝ LOW ⁸
*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; OR: Odds ratio; RCT: randomised controlled trial
GRADE Working Group grades of evidence High quality: We are very confident that the true effect lies close to that of the estimate of the effect Moderate quality: 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 quality: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low quality: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ Heterogeneity was substantial (I² = 57%) (inconsistency ‐1). ² Heterogeneity was substantial (I² = 62%) (inconsistency ‐1). ³ Imprecision of pooled effect size (imprecision ‐1). ⁴ Explorative meta‐analysis. Four studies (Gallefoss 1999; Kheirabadi 2008; Ninot 2011; Tabak 2014) with no events and a high risk of bias for three studies (Bucknall 2012; Tabak 2014; Titova 2015) for incomplete outcome data and selective reporting. Two studies (Bucknall 2012; Fan 2012) dominated the overall effect and heavily influenced the OR (risk of bias ‐1, inconsistency ‐1, imprecision ‐1). ⁵ Heterogeneity was high (I² =86%). Only three studies were included in this meta‐analysis (inconsistency ‐1, imprecision ‐1). ⁶ Only four studies were included in this meta‐analysis (imprecision ‐1). ⁷ COPD exacerbations were defined as worsening of respiratory symptoms beyond normal day‐to‐day variations that required treatment with bronchodilators, oral steroids and/or antibiotics ⁸ Heterogeneity was high (I² = 94%). Only four studies were included in this meta‐analysis (inconsistency ‐1, imprecision ‐1).

Summary of findings for the main comparison. Self‐management interventions including action plans for exacerbations compared to usual care for patients with COPD

Table 1. Participant characteristics, included studies

Study	Included participants (N)		Lost to follow‐up (%)		Age (years; mean (SD))		Gender (% male)		FEV₁ (% predicted unless stated otherwise (SD))
Study	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care	Self‐management	Usual care
Bischoff 2012	55	55	10.9	20.0	65.5 (11.5)	63.5 (10.3)	67.0	51.0	66.3 (16.5)	67.0 (18.0)
Bösch 2007	38	12	21.1	8.3	63.8 (8.4)	64.6 (6.8)	63.0% of completers		45.9 (17.5)	47.8 (16.9)
Bourbeau 2003	96	95	10.4	16.8	69.4 (6.5)	69.6 (7.4)	52.0	59.0	1.0 L (0.33)	0.98 (0.31)
Bucknall 2012	232	232	9.1	13.8	70.0 (9.3)	68.3 (9.2)	38.0	35.0	41.2 (13.4)	39.8 (13.8)
Casas 2006	65	90	26.2	20.0	70 (9.0)	72 (9.0)	77.0	88.0	43 (20)	41 (15)
Garcia‐Aymerich 2007	44	69	52.3	40.6	72 (10.0)	73 (9.0)	75.0	93.0	1.2 L (IQR 0.8 to 1.4)	1.0 L (IQR 0.8‐1.5)
Fan 2012	209	217	3.8^a; 51.7^b	4.6^a; 50.2^b	66.2 (8.4)	65.8 (8.2)	97.6	96.3	38.2 (14.3)	37.8 (14.5)
Gallefoss 1999	31	31	16.0	13.0	57 (9.0)	58 (10.0)	48.0	52.0	59 (9)	56 (11)
Hernández 2015	71	84	23.9	34.5	73 (8.0)	75 (9.0)	83.0	86.0	41 (19)	44 (20)
Jennings 2015	93	79	0	0	64.9 (10.9)	64.4 (10.5)	43.1	46.8	44.1 (23.1)	48.3 (22.2)
Khdour 2009	86	87	17.4	17.2	65.6 (10.1)	67.3 (9.2)	44.2	43.7	52.0 (15.9)	52 (17.8)
Kheirabadi 2008	21	21	0	0	56.6 (5.7)	56.2 (4.1)	61.9	76.2	N/A	N/A
Martin 2004	44	49	20.5	8.2	71.1 (95% CI 68.7 to 73.5)	69.1 (95% CI 63.5 to 74.7)	34.1	65.3	35.4 (95% CI 31.6 to 39.2)	34.3 (95% CI 31.2 to 37.4)
Mitchell 2014	89	95	26.9	16.8	69 (8.0)	69 (10.1)	60.7	49.5	56.0 (16.8)	59.6 (17.4)
Monninkhof 2003	127	121	3.9	5.8	65 (7.0)	65 (7.0)	85.0	84.0	56.1 (15.4)	58.4 (14.5)
Ninot 2011	23	22	13.0	18.2	65 (range 59 to 74)	61 (range 56 to 65)	90.0	77.8	56 (range 42 to 67)	54 (range 42 to 57)
Österlund Efraimsson 2008	26	26	0	0	66 (9.4)	67 (10.4)	50.0	50.0	N/A	N/A
Rea 2004	83	52	14.5	11.5	68 (range 44 to 84) for the total group		41.5% for the total group		51.8 (18.1)	50.0 (20.3)
Rice 2010	372	371	9.7	12.9	69.1 (9.4)	70.7 (9.7)	97.6	94.8	36.1 (14.5)	38.2 (14.4)
Song 2014	20	20	15.0	15.0	66.6 (7.1)	68.1 (6.5)	55.0	75.0	57.0 (10.0)	60.4 (24.9)
Tabak 2014	15	14	33.3	85.7	64.1 (9.0)	62.8 (7.4)	50.0	50.0	50.0 (IQR 33.3 to 61.5)	36.0 (IQR 26.0 to 53.5)
Titova 2015	91	81	44.0	39.5	74.1 (9.3)	72.6 (9.3)	42.9	43.2	33.6 (9.9)	33.0 (9.7)
^adiscontinued; ^bincomplete baseline and 1‐year study visits; CI: confidence interval; IQR: interquartile range; L: liters; N/A: not applicable.

Table 1. Participant characteristics, included studies

Table 2. Characteristics of interventions in included studies

Study	Follow‐up (months)	Setting; provision intervention	Duration intervention	Content intervention	Content action plan
Bischoff 2012	24	General practice; trained practice nurse	2 to 4 FTF individual sessions (60 min each) scheduled in 4‐6 consecutive weeks, 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Bösch 2007	12	Outpatient clinic; trained respiratory nurse under supervision of a respiratory specialist	4 FTF group sessions (120 min each) and final session scheduled 6 weeks later	Self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, other: travelling, daily live	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent, contact healthcare providers for support
Bourbeau 2003	24	Hospital (outpatient); trained professionals (nurses, respiratory therapists, a physiotherapist)	7 FTF individual sessions (60 min each) scheduled in 7‐8 consecutive weeks, 18 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: symptom monitoring list linked to appropriate therapeutic actions
Bucknall 2012	12	Hospital (inpatient); trained study nurse	4 FTF individual sessions (40 min each) in 2 months, at least 6 subsequent home visits, 828 phone calls intervention group	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Casas 2006	12	Hospital (inpatient); trained respiratory nurse and GP, physician, nurse, social worker	3 to 13 FTF individual sessions, 1 x group (40 min), 6 phone calls; Barcelona: 1 joint visit at home. Leuven: GP regularly visited patients at home	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: reinforcement of the logistics for treatment of comorbidities and social support	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: reinforcement of the logistics for treatment of comorbidities
Garcia‐Aymerich 2007	12	Hospital (inpatient); trained specialised respiratory nurse and physician, nurse, social worker	3 to 13 FTF individual sessions at the hospital (40 min each) or at home (20 min), 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: reinforcement of the logistics for treatment of comorbidities and social support	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support, other: reinforcement of the logistics for treatment of comorbidities
Fan 2012	12	Outpatient clinic; trained case manager (various health‐related professionals)	4 FTF individual sessions (90 min each) scheduled weekly, 1x group, 6 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Gallefoss 1999	12	Hospital (outpatient); trained nurse, physiotherapist, pharmacist, medical doctor	1 to 2 FTF individual sessions by a nurse and 1 to 2 by physiotherapist (40 min each), 2 x group (120 min each)	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, other: compliance, self‐care	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, contact healthcare providers for support
Hernández 2015	12	Hospital (outpatient); trained specialised respiratory nurse, physician, nurse, social worker	Participants with no mobility problems: 1 FTF individual session (40 min) at home by primary care team, 3 x group at outpatient clinic (2 x 90 min, 1x 120 min) Participants with mobility problems: 4 FTF individual sessions (15 min each), 1 x individual (120 min) or 1 x group (40 min), all at home by primary care team	Self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, exercise or physical activity component, other: instructions on non‐pharmacological treatment	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent, contact healthcare providers for support, self‐treatment of comorbidities
Jennings 2015	3	Hospital (inpatient); research team and research nurse	1 FTF individual session (60 min) at the hospital by research team member 24 hours prior to discharge, phone call 48 hours after discharge	Iterative process, education regarding COPD, smoking cessation, other: primary team was notified if patient was identified as having anxiety or depressive symptoms	Contact healthcare providers for support
Khdour 2009	12	Hospital (outpatient); clinical pharmacist, respiratory specialist, respiratory nurse	1 FTF individual session of 45 min (60 min for smokers) and 2 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Kheirabadi 2008	3	Hospital (outpatient); psychologist, trained psychiatric residents	8 FTF group sessions (60 to 90 minutes each) with 1 week interval and follow‐up by phone	Self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Avoid situations in which viral infection might be prevalent
Martin 2004	12	General practice; respiratory physician and nurse, GP, ED consultant, medical staff hospital	4 FTF individual sessions and respiratory nurse visits at 3, 6 and 12 months	Iterative process, self‐recognition of COPD exacerbations	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, self‐treatment of comorbidities, other: when to use oxygen therapy and diuretics
Mitchell 2014	6	General practice; physiotherapist, trainee health psychologist	1 FTF individual session (30‐45 min) by a physiotherapist and 2 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, avoid situations in which viral infection might be prevalent, contact healthcare providers for support, other: self‐administration, requesting rescue medication
Monninkhof 2003	12	Hospital (outpatient); trained respiratory nurse, respiratory physiotherapist	5 FTF group sessions (120 min each) by a respiratory nurse (4 x with a 1‐week interval and 3 months later) and 1 to 2 x groups (30 to 45 min) by a physiotherapist	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Ninot 2011	12	Hospital (outpatient); health professional and qualified exercise trainer	8 FTF group sessions (120 min each) by a health professional for 4 weeks, 8 exercise sessions (30 to 45 min each) by a qualified exercise trainer, 3 phone calls	Self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, use of maintenance treatment, avoid situations in which viral infection might be prevalent
Österlund Efraimsson 2008	3 to 5	Primary healthcare clinic; COPD nurse, physician, if needed: dietician, medical social worker, physical and occupational therapist	2 FTF individual sessions for self‐care education during 3 to 5 months (60 min each) by the nurse	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, smoking cessation, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Rea 2004	12	General practice; respiratory physician, respiratory nurse specialist, GP	At least 17 individual FTF sessions (monthly visits to practice nurse (N = 12), 3‐monthly to GP (N = 4), 1 x home visit by the respiratory nurse specialist, 1 x after admission)	Iterative process, self‐recognition of COPD exacerbations, other: annual influenza vaccination and PR programme attendance	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Rice 2010	12	Hospital (Veterans Affairs medical centres); trained respiratory therapist case manager	1 group session (60 to 90 min) by a respiratory therapist case manager, 12 monthly phone calls (10 to 15 min each)	Iterative process, self‐recognition of COPD exacerbations; education regarding COPD, smoking cessation	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Song 2014	2	Hospital (inpatient); trained nurse interventionists	3 FTF individual sessions (2 x inpatient (90 + 45 min each) on the day before and on the day of discharge, 1 x outpatient (90 min) on the first follow‐up day) by 2 nurse interventionists, 2 phone calls with a 2‐week interval	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations
Tabak 2014	9	Hospital (outpatient); primary care physiotherapy practices; respiratory nurse practitioner, respiratory physiotherapist	2 group sessions (90 min each) by a nurse practitioner, 1 FTF individual session and 1 x intake by the physiotherapist, additional meetings after 1, 3, 6 and 9 months	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD, exercise or physical activity component	Self‐recognition and self‐treatment of exacerbations, contact healthcare providers for support
Titova 2015	24	Hospital (inpatient); trained specialist nurse	6 FTF individual sessions (1 x at discharge, 5 x home visits at 3 and 14 days, and at 6, 12, 24 months) by the specialist nurse, 1 e‐learning programme (15 min), at least 24 phone calls	Iterative process, self‐recognition of COPD exacerbations, education regarding COPD	Self‐recognition and self‐treatment of exacerbations, avoid situations in which viral infection might be prevalent, contact healthcare providers for support
COPD: Chronic Obstructive Pulmonary Disease; FTF: face‐to‐face; PR: pulmonary rehabilitation

Table 2. Characteristics of interventions in included studies

Table 3. Number of included studies reporting outcomes of interest

Outcome of interest	Number of studies
Primary outcomes
Health‐related quality of life	16
Respiratory‐related hospital admissions	16
Secondary outcomes
All‐cause hospital admissions	11
All‐cause hospitalisation days	8
Respiratory‐related hospitalisation days	5
Emergency department visits	9
General practitioner visits	7
Specialist visits	4
Rescue medication use	2
Health status	3
COPD exacerbations	6
Use of courses of oral corticosteroids or antibiotics	9
All‐cause mortality	16
Respiratory‐related mortality	7
Self‐efficacy	2
Days lost from work	2
COPD: Chronic Obstructive Pulmonary Disease

Table 3. Number of included studies reporting outcomes of interest

Comparison 1. Self‐management versus usual care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 HRQoL: adjusted SGRQ total score Show forest plot	10	1582	Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

2 Healthcare utilisation: respiratory‐related hospital admissions (number of patients with at least one admission) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

3 Healthcare utilisation: respiratory‐related hospital admissions (mean number per patient) Show forest plot	5	873	Mean Difference (IV, Random, 95% CI)	‐0.15 [‐0.36, 0.05]

4 Healthcare utilisation: all‐cause hospital admissions (number of patients with at least one admission) Show forest plot	10	2467	Odds Ratio (M‐H, Random, 95% CI)	0.74 [0.54, 1.03]

5 Healthcare utilisation: all‐cause hospital admissions (mean number per patient) Show forest plot	4	736	Mean Difference (IV, Random, 95% CI)	‐0.04 [‐0.38, 0.29]

6 Healthcare utilisation: all‐cause hospitalisation days (per patient) Show forest plot	7	1982	Mean Difference (IV, Random, 95% CI)	‐0.65 [‐2.01, 0.71]

7 Healthcare utilisation: emergency department visits (mean number per patient) Show forest plot	3	827	Mean Difference (IV, Random, 95% CI)	‐0.31 [‐0.74, 0.12]

8 Healthcare utilisation: GP visits (mean number per patient) Show forest plot	3	605	Mean Difference (IV, Random, 95% CI)	‐0.36 [‐2.64, 1.93]

9 Health status: (modified) Medical Research Council Dyspnoea Scale ((m)MRC) Show forest plot	3	217	Mean Difference (IV, Random, 95% CI)	‐0.63 [‐1.44, 0.18]

10 COPD exacerbations (mean number per patient) Show forest plot	4	740	Mean Difference (IV, Random, 95% CI)	0.01 [‐0.28, 0.29]

11 Courses of oral steroids (number of patients used at least one course) Show forest plot	4	963	Odds Ratio (M‐H, Random, 95% CI)	4.38 [0.55, 34.91]

12 Mortality: all‐cause mortality Show forest plot	16		Risk Difference (M‐H, Random, 95% CI)	Subtotals only

12.1 All‐cause mortality	16	3296	Risk Difference (M‐H, Random, 95% CI)	0.00 [‐0.02, 0.03]
12.2 All‐cause 1‐year mortality	12	2620	Risk Difference (M‐H, Random, 95% CI)	‐0.01 [‐0.03, 0.02]
13 Mortality: respiratory‐related mortality Show forest plot	7		Risk Difference (M‐H, Random, 95% CI)	Subtotals only

13.1 Respiratory‐related mortality	7	1219	Risk Difference (M‐H, Random, 95% CI)	0.03 [0.00, 0.05]
13.2 Respiratory‐related 1‐year mortality	4	981	Risk Difference (M‐H, Random, 95% CI)	0.03 [0.00, 0.05]

Comparison 1. Self‐management versus usual care

Comparison 2. Subgroup analysis self‐management versus usual care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by follow‐up duration) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

1.1 Long‐term follow up (≥ 12 months)	11	2777	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.49, 0.99]
1.2 Short‐term follow‐up (< 12 months)	3	380	Odds Ratio (M‐H, Random, 95% CI)	0.72 [0.39, 1.35]
2 HRQoL: adjusted SGRQ total score (subgroup by exercise programme) Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

2.1 Exercise programme	4		Mean Difference (Random, 95% CI)	‐2.34 [‐5.09, 0.40]
2.2 No exercise programme	6		Mean Difference (Random, 95% CI)	‐2.95 [‐5.63, ‐0.27]
3 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by exercise programme) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

3.1 Exercise programme	6	840	Odds Ratio (M‐H, Random, 95% CI)	0.88 [0.47, 1.65]
3.2 No exercise programme	8	2317	Odds Ratio (M‐H, Random, 95% CI)	0.63 [0.43, 0.91]
4 HRQoL: adjusted SGRQ total score (subgroup by smoking cessation programme) Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

4.1 Smoking cessation programme	3		Mean Difference (Random, 95% CI)	‐4.98 [‐7.17, ‐2.78]
4.2 No smoking cessation programme	7		Mean Difference (Random, 95% CI)	‐1.33 [‐2.94, 0.27]
5 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by smoking cessation programme) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

5.1 Smoking cessation programme	4	1213	Odds Ratio (M‐H, Random, 95% CI)	0.71 [0.34, 1.45]
5.2 No smoking cessation programme	10	1944	Odds Ratio (M‐H, Random, 95% CI)	0.71 [0.51, 1.00]
6 HRQoL: adjusted SGRQ total score (subgroup by median number of BCT clusters) Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

6.1 High number of BCT clusters (> median of 9.5)	6		Mean Difference (Random, 95% CI)	‐2.93 [‐4.85, ‐1.00]
6.2 Low number of BCT clusters (≤ median of 9.5)	4		Mean Difference (Random, 95% CI)	‐3.11 [‐7.65, 1.43]
7 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by median number of BCT clusters) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

7.1 High number of BCT clusters (> median of 9.5)	7	1997	Odds Ratio (M‐H, Random, 95% CI)	0.61 [0.42, 0.89]
7.2 Low number of BCT clusters (≤ median of 9.5)	7	1160	Odds Ratio (M‐H, Random, 95% CI)	0.83 [0.48, 1.43]
8 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by number of BCT clusters) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

8.1 High number of BCT clusters (> 8)	10	2523	Odds Ratio (M‐H, Random, 95% CI)	0.70 [0.51, 0.97]
8.2 Low number of BCT clusters (≤ 8)	4	634	Odds Ratio (M‐H, Random, 95% CI)	0.71 [0.30, 1.68]
9 HRQoL: adjusted SGRQ total score (subgroup by case manager support) Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

9.1 Case manager support	6		Mean Difference (Random, 95% CI)	‐2.15 [‐4.25, ‐0.04]
9.2 No case manager support	4		Mean Difference (Random, 95% CI)	‐5.11 [‐8.81, ‐1.41]
10 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by case manager support) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

10.1 Case manager support	8	2403	Odds Ratio (M‐H, Random, 95% CI)	0.68 [0.49, 0.93]
10.2 No case manager support	6	754	Odds Ratio (M‐H, Random, 95% CI)	0.80 [0.36, 1.77]
11 HRQoL: adjusted SGRQ total score (subgroup by intervention duration) Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

11.1 Intervention duration ≥ 6 months	7		Mean Difference (Random, 95% CI)	‐2.96 [‐5.20, ‐0.72]
11.2 Intervention duration < 6 months	3		Mean Difference (Random, 95% CI)	‐2.57 [‐6.96, 1.82]
12 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by intervention duration) Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

12.1 Intervention duration ≥ 6 months	9	2453	Odds Ratio (M‐H, Random, 95% CI)	0.65 [0.43, 0.96]
12.2 Intervention duration < 6 months	5	704	Odds Ratio (M‐H, Random, 95% CI)	0.84 [0.53, 1.32]
13 HRQoL: adjusted SGRQ total score (subgroup by action plan component 'adaptation of maintenance medication') Show forest plot	10		Mean Difference (Random, 95% CI)	‐2.69 [‐4.49, ‐0.90]

13.1 Action defined for adaptation of maintenance medication	6		Mean Difference (Random, 95% CI)	‐3.75 [‐6.16, ‐1.33]
13.2 No action defined for adaptation of maintenance medication	4		Mean Difference (Random, 95% CI)	‐2.02 [‐4.77, 0.72]
14 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by action plan component 'adaptation of maintenance medication' Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

14.1 Action defined for adaptation of maintenance medication	5	910	Odds Ratio (M‐H, Random, 95% CI)	1.01 [0.54, 1.88]
14.2 No action defined for adaptation of maintenance medication	9	2247	Odds Ratio (M‐H, Random, 95% CI)	0.59 [0.42, 0.83]
15 Healthcare utilisation: respiratory‐related hospital admissions (subgroup by action plan component 'when to avoid situations in which viral infections might be prevalent') Show forest plot	14	3157	Odds Ratio (M‐H, Random, 95% CI)	0.69 [0.51, 0.94]

15.1 Action defined 'when to avoid situations in which viral infections might be prevalent'	4	549	Odds Ratio (M‐H, Random, 95% CI)	0.88 [0.25, 3.13]
15.2 No action defined 'when to avoid situations in which viral infections might be prevalent'	10	2608	Odds Ratio (M‐H, Random, 95% CI)	0.68 [0.50, 0.91]

Comparison 2. Subgroup analysis self‐management versus usual care