Supported self-management for patients with moderate to severe chronic obstructive pulmonary disease (COPD): an evidence synthesis and economic analysis

Definition of self-management used for this review

‘Self-management’ has been defined as the ability of a patient to deal with all that a chronic disease entails, including symptoms, treatment, physical and social consequences and lifestyle changes. 23 SM interventions involve collaboration between the health-care professional and the patient so that the patient acquires and demonstrates the knowledge and skills required to manage their medical regimens, change their health behaviour, improve control of their disease and improve their well-being. 29 This definition of SM was used as a basis to devise a list of SM interventions/components that were considered for this review (Table 2). Because SM interventions are so heterogeneous, we specifically chose to include all possible aspects of SM to ensure completeness. However, we excluded interventions of smoking cessation alone, as there is already good evidence of the benefits of smoking cessation in general, and a large number of systematic reviews addressing the effectiveness of smoking cessation interventions (currently 60 Cochrane systematic reviews alone). Most evidence of the effectiveness of smoking cessation relates to general populations, rather than people with a particular condition. Any study that included smoking cessation as one component of a multicomponent package in people with COPD was included. Similarly, there is already a systematic review of pulmonary rehabilitation (PR) at this time point;55 therefore, it was not considered necessary to repeat it but rather use it for comparison.

Search strategy for effectiveness studies

A comprehensive search strategy was designed and conducted by an experienced information specialist. The searches were kept broad to capture evidence to suit both aims.

Searches for relevant studies were conducted across the following bibliographic databases: MEDLINE, MEDLINE In-Process and Other Non-Indexed Citations and EMBASE (via Ovid), Cochrane Central Register of Controlled Trials (CENTRAL – Wiley) and Science Citation Index (Institute of Scientific Information). Subject-specific databases were also searched: PEDro physiotherapy evidence database, PsycINFO (via Ovid) and the Cochrane Airways Group Register of Trials. Ongoing studies were sourced through the metaRegister of Current Controlled Trials, International Standard Randomised Controlled Trial Number database, World Health Organization International Clinical Trials Registry Platform Portal and ClinicalTrials.gov. Specialist abstract and conference proceedings were sourced through the Institute of Science Information’s Conference Proceedings Citation Index and British Library’s Electronic Table of Contents (Zetoc). Hand-searching through European Respiratory Society, the American Thoracic Society and British Thoracic Society (BTS) conference proceedings from 2010 to 2012 was also undertaken, and selected websites were also examined. No language restrictions or methodological filters were applied to the searches. Electronic database searching was carried out from inception to May 2012, and no updated searches were undertaken beyond this time point. The search strategies used for electronic databases can be found in Appendix 1; terms for COPD were combined with those for SM and, where possible, utilised appropriate medical subject headings.

The citation lists of all included studies and any citations within relevant reviews were scanned for additional relevant studies. Consultations with experts in the field through the investigators identified additional relevant literature.

Reference Manager version 11 (Thomson ResearchSoft, San Francisco, CA, USA) was used to store and manage all search results.

Study selection process

After removal of duplicates, titles and abstracts of the remaining search results were independently reviewed by two reviewers. Full texts were obtained for papers meeting the inclusion criteria or when the abstract was unclear. Full texts were then independently reviewed by two reviewers using detailed and piloted selection criteria concerning study design, populations, interventions, comparators and outcomes for each review. Any discrepancies were resolved by a third reviewer. Any non-English language papers were assessed, based on titles and abstract, but when information was lacking or unclear, translators were used to decide final inclusion. A reviewer worked alongside translators to avoid misinterpretation of the selection criteria. During full-text screening, papers were categorised into their appropriate objectives or were excluded with reasons.

Selection criteria

The selection criteria for this review are summarised in Table 3.

Only primary studies were included. Studies concerning patients with moderate to severe COPD were included, and those with patients with mild or very severe COPD were included only if the majority of the study population was moderate/severe. A COPD study population of approximately 90% was required for inclusion unless data on the subset of patients with COPD were provided separately. Studies were included if the intervention was set within either a hospital or a community. Studies of any SM intervention/package or components of SM interventions were included. For example, medication management, action plans, exercise, inhaler technique and stress management (see Table 2). Comparators consisting of usual care (UC), control/sham or other SM interventions were accepted.

Risk of bias assessment

All RCTs were assessed using the recommended and validated Cochrane Risk of Bias tool. 56 The following six domains were assessed: sequence generation, allocation concealment, blinding of personnel and participants (by outcome), incomplete outcome data (by outcome), selective outcome reporting and other potential threats to validity. Domains were judged as high risk of bias, low risk of bias and unclear risk of bias. For trials with multiple papers, information from all of the studies was used to judge risk of bias. After a piloting process, all studies were assessed by two independent reviewers with a third reviewer overseeing the process. The GRADE57 framework was used to denote overall quality of evidence across studies for each of the primary outcomes and also HRQoL, using a scoring system of 4 (high) to 1 (very low) quality. The findings were summarised in a table, incorporating the results but also aspects that led to the final judgement.

Data extraction and manipulation

Forest plots

Results for each outcome were presented, where relevant, on a forest plot. Interventions were heterogeneous across the studies so results were placed in subgroups most consistent with the intensity and duration of support provided:

1. more-supported SM package – six or more contacts or unspecified contacts but ≥ 6 weeks’ duration

2. less-supported SM package – fewer than six contacts or unspecified contacts and < 6 weeks’ duration

3. exercise-based intervention.

Within each of these subgroups, studies were displayed in order of length of follow-up except for QoL outcomes, which were also grouped by questionnaire [St George’s Respiratory Questionnaire (SGRQ), Chronic Respiratory (Disease) Questionnaire (CRQ), EuroQoL-5 Dimensions (EQ-5D)]. As there were multiple follow-up points, it was decided that for each outcome, only data from the final follow-up period would be displayed in the forest plot and used in any subsequent meta-analysis. The subgroups were specified prior to inspection of the results to allow sensible exploration of the different types of interventions. Meta-regression was not possible owing to the limited number of studies.

Meta-analyses

Search results

Study identification and flow chart

Initial database searches identified 13,355 records, of which 836 remained after scanning titles and abstracts using the inclusion/exclusion criteria (Figure 1). After the same criteria were applied to the full papers, 12 papers reporting 10 trials were finally included in the review. 63–74 Appendix 2 details the reasons for exclusion at each stage. These were largely because patients were not recruited at the appropriate time point during/after discharge. Overall, 5% of all full texts required arbitration by a third reviewer.

FIGURE 1.

The selection process for clinical effectiveness studies.

The inclusion of two trials was particularly difficult to assess. 68,75 Both were comparing ‘hospital at home’ with UC and had substantial SM components. One trial75 was excluded because all patients were seen in the ED then randomised to home compared with hospital (and therefore patients were not admitted at all unless in the control group). In the second study,68 although patients were assessed in the ED, a substantial proportion of patients in both arms were initially admitted and then discharged from hospital. The difference between the two arms was (a) the proportion of patients requiring admissions and (b) the intervention arm had ongoing SM support at home, whereas, once discharged, the control group had usual primary care support. Thus, this trial was included. 68

Conference abstracts meeting the inclusion criteria for this review are listed in Appendix 3. There were a further four trials that were ongoing at the time of the search end date (see Appendix 4).

Characteristics of included randomised controlled trials

Size, setting, recruitment

Randomised controlled trials ranged in size from 3373 to 46463 total participants. One66 was a cluster RCT, based in 45 nursing homes. One paper was the 18-month follow-up of the original study,64,65 and one paper72 referred to the Spanish centre of a European study. 71

Participants were largely recruited in hospital during an exacerbation of COPD or at (or immediately after) discharge. Two papers67,68 also included patients recruited at the ED who may not have been admitted to hospital.

The definition of COPD for inclusion was generally based on a clinical diagnosis (except for Bucknall et al. ,63 which also required patients to meet the spirometric criteria for airflow obstruction). One study70 included a mixed population of patients with chronic lung disease, although 89% had COPD.

Patient exclusion from trials was usually based on inability to provide consent; terminal illness or extreme comorbidities preventing inclusion in rehabilitation/exercise; or social conditions/lack of access to a telephone. All of the studies were set among patients living at home except for the cluster RCT, which was specifically based in nursing homes. 66

Description of included patients

Mean age of participants was similar across the included RCTs (66–74 years), except in the cluster RCT based in nursing homes,66 where the mean age was approximately 80 years. Sex distribution, however, was variable across studies (ranging from 37% to 97% males). Where reported, severity of disease was similar with mean FEV₁ ranging from approximately 31% to 42% of predicted values. Most patients were described as having moderate or severe COPD.

Description of self-management interventions and comparators

Interventions were varied and have been described in full in Table 4. Figure 2 provides a summary diagram of the included RCTs, with interventions grouped into three categories:

FIGURE 2.

Study characteristics of the included RCTs.

1. ‘More supported’ Six or more contacts or ≥ 6 weeks’ duration if contacts not specified. This category included:

   1. – large RCT in the UK of supported SM (based on the Living Well with COPD materials) for 12 months, compared with UC63

   2. – RCT in Spain/The Netherlands of integrated care including supported SM for 12 months,71,72 compared with UC

   3. – RCT in Hong Kong70 of a community nurse-supported discharge programme, including SM support, with weekly visits for 4 weeks and then monthly, with additional telephone hotline and a total follow-up of 6 months, compared with UC

   4. – small RCT in the UK73 of hospital outpatient visit-based SM support over 16 weeks with total 6 months’ follow-up, compared with UC

   5. – cluster RCT in Hong Kong66 of support by community nurses to nursing home staff and patients with a supported SM component, weekly visits for 1 month and thereafter monthly visits for a total of 6 months, compared with UC.

2. ‘Less supported’ Fewer than six contacts or < 6 weeks’ duration if contacts not specified. Including:

   1. – RCT in China of telephone-based SM (based on Bandura’s theories of self-efficacy31), with two telephone calls before week three and total follow-up for 3 months, compared with UC74

   2. – RCT in Australia of SM support provided by two visits after 1 week and 1 month, with total 3 months’ follow-up, compared with UC67

   3. – RCT in Spain of home-based hospitalisation, including SM education and action plans, reinforced during up to five home-visits and telephone contacts over an 8-week period, compared with UC68

   4. – RCT in Australia of case management with SM support with review and two telephone calls and follow-up for 3 months in total, compared with UC. 69

3. ‘Exercise-only intervention’ Home-based exercise-only interventions:

   1. – small RCT of home-based exercise, supervised regularly for 3 months and with 6-month64 and 18-month65 follow-up, compared with general exercise advice.

Primary outcomes

All-cause mortality: no evidence of effect

Six trials63,64,67,68,70–72 contributed mortality results (Figure 3 and Appendix 5). There was a wide range of event rates across the trials. Despite the general heterogeneity of interventions, there was no statistical heterogeneity and the overall results indicated that there was no evidence of effect on mortality (HR 1.07, 95% CI 0.74 to 1.54). Using the GRADE system, we would judge that this is moderate-quality evidence (Table 6).

FIGURE 3.

All-cause mortality. NR, not reported.

Hospital admissions: no evidence of effect

Seven trials63,65,67,68,70,71,73 contributed data to the admissions results (Figure 4 and Appendix 6). The results of three trials66,69,74 could not be included in the combined results because they reported mean number of admissions rather than first admission and one69 also did not provide sufficient information to calculate a SE.

FIGURE 4.

First hospital admission. NR, not reported.

Overall, statistical heterogeneity was high (I² = 70.9%), and subdividing by level of support explained only a fraction of this. Estimates are provided by level of SM support, although there was no evidence of any effect for the non-exercise-based interventions and substantial remaining heterogeneity.

One of the studies that may have contributed to the remaining heterogeneity in the non-exercise-based studies is the small study of 33 participants by Dheda et al. ,73 which was poorly reported, had signs of inadequate randomisation and very high loss to follow-up, especially in the intervention arm. This study73 had the most extreme results in its category.

The trial of home-based exercise65 demonstrated a large relative reduction on the rate of first admission (HR 0.17, 95% CI 0.05 to 0.66), although this trial was very small and these results were based only on participants who completed the study (< 60% of those randomised), and thus would also be subject to high risk of bias. Furthermore, although subgrouping by intervention category was an a priori identified analysis, care must be taken in the interpretation of results due to multiple comparisons being performed. The evidence above was judged to be of low quality according to GRADE (see Table 6).

Emergency department visits: no evidence of effect

The effect on ED visits is displayed in Figure 5 (see Appendix 8), for which five trials66–68,70,72 contributed data. Four trials66,68,70,74 reported mean visits per patient and two67,68 reported first visit. The two trials with a longer follow-up of 6 months66,70 failed to demonstrate any evidence of an effect on ED visits. This evidence was also of low quality (see Table 6).

FIGURE 5.

Emergency department visits. NR, not reported.

TABLE 6 - GRADE summary of findings for main outcomes

n/a, not applicable to report as combined values not computed; RR, rate ratio.

Outcome	Control risk	Intervention risk	Results	No. of participants (trials)	Quality rating	Comments
Mortality	56/598	59/581	1.15 (95% CI 0.79 to 1.67)	1179 (6)	+++; moderate	Outcome measurement likely to be unbiased I2 = 0% Generally, low risk of bias except allocation concealment rarely mentioned
Hospital admissions	259/621	211/596	0.78 (95% CI 0.52 to 1.17)	1217 (7)	++; low	Outcome measurement problematic in some trials due to loss to follow-up I2 = 70.9% Wide CIs crossing line of no effect Some study results based on completers only
ED visits	n/a	n/a	Not combined RR ranged from 0.27 to 1.06	932 (5)	++; low	Unclear methods of randomisation and allocation in two trials Follow-up was generally short and results inconsistent
GP consultations	n/a	n/a	Not combined No significant effects	332 (2)	+; very low	Very little information
HRQoL	n/a	n/a	SGRQ 3.84-point improvement (95% CI 1.29 to 6.40 points)	845 (6)	+; very low	Biased follow-up – enormous loss to follow-up Outcome measurement likely to be biased I2 = low

Secondary outcomes

Health-related quality of life: consistent with improvement although potential bias

Six trials63–66,67,68,72,73 contributed data on HRQoL using the SGRQ, the CRQ or the EQ-5D (Figure 6 and Appendix 9). The data from one study69 could not be used as they reported median change only. Five trials63,67,68,72,73 used the SGRQ, two trials63,72 reported the EQ-5D and one trial64 the CRQ.

Overall, SM interventions resulted in an improvement of 3.84 (95% CI 1.29 to 6.40) points on the SGRQ scale compared with control (close to the minimal clinically important difference of four points), although follow-up (where reported) ranged from about 25% to 83% across studies, and, therefore, this result should be treated with caution and contributed to the overall judgement that this evidence was of very low quality (see Table 6). In particular, the study by Dheda et al. ,73 which produced the most extreme results, had many methodological flaws.

FIGURE 6.

Health-related quality of life as measured by SGRQ, CRQ and EQ-5D. NR, not reported.

Lung function: no evidence of effect

Data from four trials64–66,68,72 were plotted (see Appendix 11). Three trials provided results on raw FEV₁ values (Figure 7),64,65,68,72 and two trials the effect on percentage predicted FEV₁ (Figure 8). 65,66 There was no evidence of any effect from any of the individual trial results, and it was not deemed appropriate to pool the individual trials due to the small number of studies and heterogeneity of outcomes, follow-up time and methods of analysis. The findings are in agreement with the fifth trial73 which reported no evidence of effect on FEV₁ but did not provide data. Again, the proportion of patients followed up across the trials was very variable.

FIGURE 7.

Lung function (FEV₁, l). NR, not reported.

FIGURE 8.

Lung function (FEV₁% pred). NR, not reported.

Anxiety and depression: possible improvement in scores

Four trials (see Appendix 12)63,66,69,70 reported on psychological health outcomes; however, only two trials63,66 contributed to the analysis on anxiety (Figure 9) because one trial69 reported only median changes and the other70 did not provide separate results for anxiety. Although there were data on less than half of the sample in the larger study,63 the intervention group had a mean reduction of 1.06 points (95% CI 0.04 to 2.08 points) in the ‘anxiety’ component of the Hospital Anxiety and Depression Scale (HADS) score compared with the control group, and the other trial66 demonstrated a mean reduction of 1.5 points (95% CI 0.62 to 2.38 points) in the ‘anxiety and insomnia’ component of the General Health Questionnaire (GHQ) relative to the control group. 66

FIGURE 9.

Anxiety. NR, not reported.

One of the above trials66 also showed some evidence of a reduction in depression score (MD –1.0, 95% CI –1.97 to –0.03), although follow-up rates were not reported, whereas there was no evidence of effect in the larger trial63 (Figure 10).

FIGURE 10.

Depression. NR, not reported.

Key results

Despite a rigorous search we identified only 10 RCTs63–74 that evaluated the effectiveness of interventions providing SM support to patients shortly after being discharged from hospital with an acute exacerbation of their COPD.

Few of the trials had consistently low risk of bias. Many studies were small and suffered from inadequate reporting and high loss to follow-up, particularly affecting patient-reported outcomes such as HRQoL.

Although the participants seemed relatively homogeneous, interventions were very heterogeneous, with some trials67–69,74 providing low-intensity, short-term support of 2–3 months and others63 a very intensive package lasting for 12 months.

Overall, there was limited evidence of effect on health-related behaviours and outcomes. There was some evidence of improvement in patient knowledge, treatment of exacerbations and inhaler technique,68,69,73 but there was inconsistent evidence of effect on other health-promoting behaviours68,73 or on self-efficacy. 64,75

In terms of health outcomes, the most consistent effects were observed on patients’ QoL, with an overall improvement with data from five trials63,67,68,72,73 of 3.8 points on the SGRQ score (close to the minimally clinically important difference of four points). Notably, though, this estimate should be treated with caution because, although reaching statistical significance, there was substantial and differential loss to follow-up in both arms, which could bias the results in favour of a positive effect. The authors of the largest trial63 indicated, themselves, that the results from their trial could be unreliable. The reduction in anxiety exhibited in two trials,63,66 however, supports some potential effect on patients’ psychological health (although it is not clear whether or not this would be clinically important).

An important outcome for these patients is whether the SM package had any effect on subsequent hospital admissions. We were able to use data that reported time to first all-cause admission. Despite subdividing by intensity of intervention, we were unable to explain the substantial heterogeneity observed, but, overall, there was no clear evidence of effect on hospitalisation (HR 0.78, 95% CI 0.52 to 1.17; I² = 71%). Post hoc inspection of the data suggested a possibility of a greater effect with the exercise-based intervention but would require more data to be explored in depth. It is possible, however, that the effects on admissions would be diluted because we extracted admissions due to any cause (although in our analysis the majority were for respiratory causes).

There was no apparent evidence of effect on mortality and no clear patterns with duration of intervention.

In general, the most positive results across the outcomes were observed in the small trial of home-based exercise64,65 but, given the multiple methodological limitations of the trial in terms of reporting and analysis, the results have to be interpreted with caution.

How this fits with other literature

This is the first systematic review that addresses the effectiveness of SM support provided to patients with COPD soon after hospital discharge. The only other review related to this time point is a Cochrane review of PR,76 which identified nine trials and showed significant reduction in hospital admissions [pooled odds ratio (OR) 0.22, 95% CI 0.08 to 0.58], over 25 weeks and mortality (OR 0.28, 95% CI 0.10 to 0.84) over 107 weeks. Effects of PR on HRQoL were well above the minimal important difference when measured by the CRQ and the SGRQ total score (MD –9.88, 95% CI –14.40 to –5.37). However, in common with our review, trials were small and methodologically inadequate, and, although loss to follow-up was not discussed or assessed in the risk of bias section, a large proportion did not complete the rehabilitation. There was also significant heterogeneity across many of the outcomes. The authors discussed the possibility of publication bias and possible overestimate of effect with small trials but suggested this would not account for the whole effect. The results would fit with our tentative observation that trials with an exercise component might be more effective.

The majority of the studies and reviews of SM support are set among patients who have COPD in a stable state. Our results, although showing few significant effects, are consistent with some of the other systematic reviews. For example, a systematic review of SM education48 showed evidence of a significant reduction in respiratory admissions (OR 0.64, 95% CI 0.47 to 0.89; n = 8 RCTs) and a significant mean improvement of 2.6 points (95% CI 0.2 to 5.0 points) on the SGRQ score (n = 7 trials). A review of PR46 reported an overall mean improvement in SGRQ score of 6 points (95% CI 3 to 9 points; n = 6 trials) and a review of integrated disease management47 found a similar improvement in HRQoL: SGRQ 3.71 points (95% CI 1.6 to 5.8 points; n = 13); CRQ 1.02 points (95% CI 0.67 to 1.36 points; n = 4) and respiratory admissions (OR 0.68, 95% CI 0.47 to 0.99; n = 7) and a similar lack of effect on mortality. Conversely, a review of action plans found little evidence of benefit on HRQoL or health-care utilisation. 50

In the last couple of years, and particularly since the completion of our searches, there have also been a number of individual trials and commentaries that question whether patients are actually able to self-manage. 34,36,77,78 Two of these among patients with COPD35,62 identify a group of successful self-managers in post hoc exploratory analyses. The first of these63 is included in our review but no other studies have explored these subgroups so we were unable to examine this point further.

Conclusions

• Self-management support delivered shortly after an acute exacerbation may have some benefit in terms of HRQoL and possibly admissions but the evidence is thin and unconvincing.

• Exercise may play an important role but there are not enough data to explore this.

• Any future trials should address issues of bias, particularly loss to follow-up, but this may be inherent in the nature of the intervention and the fact that patients are still trying to recover from an exacerbation.

• The evidence is not in support of SM interventions to be put into practice for patients with COPD at, or recently after, hospital discharge.

Definition of self-management

As described for review 1 and tabulated in Table 2.

Search strategy

A comprehensive search strategy was undertaken as described as for review 1. The search was broad and covered many databases and no study design filters were applied. Search terms related to qualitative evidence included ‘patient-centred’ and ‘patient focus’.

Study selection process

As described for review 1.

Selection criteria

Selection criteria were as described for review 1, with the difference of two elements: study design and outcomes. Only qualitative study designs (interviews and focus groups) were sought and outcomes relating to patient satisfaction, acceptance and barriers to SM were assessed.

Study quality, data extraction and synthesis

Study quality was assessed using the Critical Appraisal Skills Programme (CASP) checklist for qualitative research. 80 As well as extracting data related to study and patient characteristics, any quotes, key themes and concepts identified were extracted. As outlined in the protocol, an interpretive analysis (meta-ethnography)81 was planned if findings allowed. This involved looking for similarities (reciprocal translation), differences (refutational translation) or creating a line of argument using concepts proposed in included primary studies. However, as only one study described a small element of qualitative interviewing, this was not undertaken.

Search results

Figure 11 outlines the flow of included studies. One of the included randomised controlled trials (RCTs)69 from review 1 had a limited qualitative element referring to patient satisfaction and was therefore included in this review. There was also one ongoing study (see Appendix 4).

FIGURE 11.

Selection process for qualitative studies.

Characteristics of included study

The included trial was a trial of nursing-based case management among 66 patients in Australia admitted to hospital with COPD. The intervention comprised SM support with review and two telephone calls and follow-up for 3 months in total, compared with usual care. 69 A subgroup of 18 patients and their carers from both arms of the trial were interviewed in more depth using semistructured interviews focusing on issues associated with patients and caregiver satisfaction with care. The interviews were recorded, and then transcribed and coded to identify themes. These interviews revealed that patients were very satisfied with their care in hospital and, for those in the intervention arm, ongoing contact with the community nurse was very helpful in improving their access to resources and communication with the health professionals. For those in the control arm, those with family support or medical contact were satisfied but those without were not. Table 7 describes the patient and caregiver quotes in relation to case managers.

Quality of included studies

Table 8 presents an assessment of the quality of this study. The aims of the study were not very clear from the outset, which means that it was difficult to assess which methodology would be appropriate. It was possible to infer that patients’/carers’ views and satisfaction with the intervention would be sought, which would mean that these semistructured qualitative interviews would be appropriate. However, although the authors mentioned that patients were selected to maximise variability, the only factor that they mention is about representing both intervention and control groups.

There is mention of the home setting, although not the justification for it, but they do not mention a topic guide, any modification of the methods, details of any interviewer biases, how the themes were identified or whether or not saturation was reached. Thus, the findings are really not very valuable.

Key results

A comprehensive search of the literature revealed only one RCT69 with a limited qualitative aspect and limited conclusions.

Strengths and weaknesses

The search strategy was broad, conducted across several databases and with no study design filters applied. Additionally, reviewers adhered to a systematic methodology with two independent reviewers assessing studies for inclusion and exclusion against a prespecified selection criteria; therefore, it was unlikely that any relevant qualitative evidence will have been overlooked. However, after a thorough search and identification process, only one study69 was included.

How this fits with the findings of the effectiveness review (review 1)

Unfortunately, the included study69 only gave scant information regarding the qualitative aspects, and it is difficult to be sure of the true purpose of the interviews and how they were actually conducted. The study,69 however, reported that patients who received a case manager were very satisfied with their care, and were made aware of resources available to them and how to use them. Those in the control arm without family support were more concerned about their situation, although those in the control arm with family support seemed reasonably satisfied. This reflects the evidence from the quantitative studies for which the evidence was inconsistent (two studies66–68,72 showed better satisfaction with the intervention and two did not) but does not really gain much further insight.

Conclusions

There is almost no qualitative evidence about patients’ views on SM support programmes that are delivered post discharge, and, given that the quantitative evidence reveals uncertainty about the effectiveness of such interventions in their current form, it is therefore a potential area of research need.

Search strategy

A comprehensive search strategy was undertaken by an experienced information specialist from inception to May 2012. Three electronic databases were searched: MEDLINE and EMBASE via Ovid and Cochrane NHS Economic Evaluation Database (Wiley). Searches were not limited by date nor were any language restrictions applied. The search strategies used for each database can be found in Appendix 17. Relevant literature from the clinical effectiveness searches were also identified and included for review, if they had not already been captured in the searches for cost-effectiveness.

Reference Manager version 11 was used to store and manage all references.

Study selection process

The inclusion and exclusion criteria outlined below were used to select studies. A two-stage review process was applied by two independent reviewers: first, screening titles and abstracts, and then reviewing full papers. Discrepancies were resolved by a third reviewer with expertise/knowledge in the field of health economics.

Selection criteria

Risk of bias assessment

Risk of bias of included studies was assessed using the Drummond checklist, as suggested in the Cochrane Handbook. 56 Risk of bias assessment was undertaken by two reviewers independently, one of whom had expertise in the field of health economics.

Data extraction

Study characteristics and results were extracted independently by two reviewers. Meta-analysis was not considered appropriate for this review because of the paucity of evidence.

Search results

Figure 12 outlines the study identification process. A total of 1611 references were imported into Reference Manager, and 240 duplicates were removed automatically and manually. Overall, 1131 titles and abstracts were screened, with 129 articles being identified for full-text review. Only one study68 met the inclusion criteria for this review, which also formed part of the clinical effectiveness review. A further four studies met initial inclusion criteria but were ongoing and so have been listed in Appendix 4.

FIGURE 12.

Selection process for cost-effectiveness studies.

A number of other studies (n = 27) were identified as potentially useful to inform the independent economic analysis (see Chapter 6). These studies included relevant primary or secondary data on the cost or utilisation of health care associated with SM in patients with COPD; however, the intervention was not delivered to patients in hospital at the point of discharge or within 6 weeks of hospital discharge. Data from some of these studies were used to estimate the costs of SM in the model (see Chapter 6).

Characteristics of included studies

The single included trial by Hernandez et al. 68 was conducted in two tertiary hospitals in Barcelona, Spain, with a total sample size of 222 patients: 121 patients were randomised to the home hospitalisation group and 101 were randomised to conventional care. The hospital-at-home intervention included four phases: assessment by a specialist team during admission to the emergency room; treatment at discharge; home hospitalisation with follow-up; and assessment after 8-week follow-up. Specific SM components implemented as part of the hospital-at-home service included 2 hours of the following delivered at the point of discharge and later reinforced during home visits (along with action plan reinforcement):

• education on disease, adherence to medication and recognition/prevention of triggers of exacerbations

• selecting appropriate equipment at home and training on the correct administration of pharmacological therapy

• smoking cessation

• patient empowerment for daily life activities, including hygiene, dressing, household tasks, leisure activities, breathing exercises and skeletal muscle activity

• nutrition recommendations

• socialisation and changes in lifestyle.

The following outcomes were reported: mortality, readmissions or ED visits, hospitalisation, HRQoL, lung function, patient satisfaction, disease knowledge, inhaler technique, medication prescriptions, and home rehabilitation and costs.

Quality of included studies

The details of a cost analysis economic evaluation68 were described and are summarised in Table 9. The research question was stated with reasons for its importance as well as for the rationale for the intervention and control under comparison. A public insurer perspective was taken but not justified; however, the limitations of taking this viewpoint were highlighted. The source of effectiveness estimates was stated and details of the design and results of the effectiveness study were provided. Outcome measures were clearly outlined and quantities of resource use (per patient) were given separately from unit costs. Details of direct and indirect costs were provided with nurse home visits, prescriptions, telephone calls and transport being calculated directly and inpatient hospital stay, emergency room visits, outpatient visits, primary care consultations and social support visits being calculated indirectly. Costs incurred by patients or carers were not considered. Currency and price data were reported using 2000 price data (euros); however, no details were provided regarding any price adjustments for inflation or currency conversions. The time horizon for the study was 1 year; thus, discount rates were not applied. Tariff prices were applied to resource-use data that were collected to calculate an average annual health-care cost per patient in both arms. Although differences in both costs and outcomes were reported, they did not conduct a full economic evaluation by presenting the relative cost-effectiveness. As the costs for the intervention were lower and outcomes better, this study68 suggests the intervention dominated UC. Sensitivity analyses, and the details thereof, were discussed only briefly.

Cost-effectiveness

As only one study met inclusion criteria for this review, no meta-analysis was undertaken; instead, the cost analysis results from the included study68 are reported.

Costs were reported for the following outcomes (categories) following an 8-week follow-up period: length of hospital stay, emergency room visits (excluding visits that required further hospital admission), outpatient visits, primary care consultations, social support visits, home visits by nurse, prescriptions, telephone calls (both to the nurse from the patient and from the nurse to the patient) and transport costs. Details of the costs reported for each outcome are provided in Table 10.

The cost analysis found the home hospitalisation intervention to be significantly less costly than conventional care (average cost per patient: €1255.12 vs. €2033.51; p = 0.003). Hospital stay, emergency room visits, outpatient and social support visits were at a greater cost per patient for the conventional care group than for the home hospitalisation group, with the difference for hospital stay and emergency room visits reaching significance (p < 0.001 and p = 0.01, respectively). Prescription costs were significantly higher in the intervention group than in the control group (cost per patient: €217.21 vs. €172.06; p = 0.001). Primary care visits were also greater in the intervention group, although the difference was not reported to be statistically significant.

A sensitivity analysis based on resources for home hospitalisation at 50% and 75% of the average cost per patient (to capture intervention costs in the longer term) was undertaken. Cost savings in favour of home hospitalisation were conserved across each assumption.

Summary of results

A comprehensive search strategy identified one study68 that met the inclusion criteria for this review. The overall quality of the study was high, with some issues related to reporting. Meta-analysis was not possible. The study68 revealed that home hospitalisation is less costly than conventional care [£1041.75 vs. £1687.81 (conversion rate €1 = £0.83)].

How this fits with other literature

One relevant study82 published after the search strategy had been completed was subsequently identified. Xin Lie et al. 82 developed a Markov model to evaluate the impact of a hypothetical exacerbation management programme that could detect the risk of exacerbation and divert the risk of hospitalisation. In patients without prior history of exacerbation, they estimated that this would result in savings of US$2900 per patient over 12 years, and in higher-risk patients – with a history of one or two exacerbations per year – this estimate increased to US$16,000 per patient.

Strengths and limitations

This is the first systematic review of the cost-effectiveness literature of SM interventions for patients who have recently been discharged from hospital after an acute exacerbation of COPD. The methods used throughout this cost-effectiveness review were systematic. A comprehensive search strategy was undertaken and the results were reviewed independently by two reviewers, including a health economist. The recommended quality assessment checklist was used.

It should be noted that the patients in this study were recruited from the emergency room rather than after discharge post hospital admission, which may or may not cause variations in the applicability of the results. A scarcity of evidence of the cost-effectiveness of SM interventions was evident. The identified study included the cost of implementing a hospital-at-home programme with components of SM, thus the SM components reflect only a proportion of the costs and cost savings incurred.

Implications for research

There is a need for more economic evaluations, alongside randomised controlled trials, specifically addressing patients who have recently been discharged from an inpatient hospital admission stay after an acute exacerbation of COPD, and who have been treated with SM interventions or components of SM, such as education, action plans, breathing techniques, relaxation and stress management amongst others.

Model description

A Markov decision model, built in TreeAgePro 2014 (TreeAge Software, Inc., Williamstown, MA, USA) was structured to consider short-term increased risks of readmission and mortality, and the long-term natural history of the disease, taking into account exacerbations, increasing COPD severity and mortality (Figure 13). This structure was an adapted version of other COPD Markov models83 with health states linked to GOLD (Global Initiative for Chronic Obstructive Lung Disease) severity. It incorporated additional health states to capture the higher risks reported in patients immediately after discharge in audits of patients admitted to hospital. 84 The model had a time cycle of 1 month and a lifetime time horizon (30 years) was used. All costs and outcomes were considered from a NHS perspective for a price year of 2012.

FIGURE 13.

Markov model structure.

Severity of COPD was defined according to the GOLD classification. GOLD stage 2 was defined as having a forced expiratory volume in 1 second (FEV₁) of ≥ 50%, < 80% predicted; GOLD stage 3, a FEV₁ of ≥ 30%, < 50% predicted; and GOLD stage 4, a FEV₁ of < 30% predicted. GOLD stage 1 (mild COPD) was excluded, as < 16% of patients admitted to UK hospitals with COPD had a FEV₁ of ≥ 80% predicted.

A patient started in the model in one of three health states representing their first month post admission, taking into account their current GOLD severity stage. Those who continued to recover without further exacerbations moved to health states to represent the second and third month of recovery, again related to their GOLD stage. Within this 3-month recovery period, a patient could die from COPD or other causes or have a further exacerbation requiring readmission, which could be fatal. If the patient survived, they were discharged to restart the 3-month recovery period in a ‘first month post-admission’ health state. The patient pathway within each post-admission health state (Figure 14) was similar for each month and severity stage. The post-admission health states allowed the model to consider the immediate increased risk of readmission and COPD-related mortality for 3 months after discharge. Once a patient survived 3 months of recovery with no readmissions, they moved into a stable health state for their GOLD stage.

FIGURE 14.

Pathways within post-admission states.

Once in the stable GOLD stage 2, 3 or 4, a patient could remain in that health state, deteriorate to the next more severe health state, have an exacerbation or die. An exacerbation could be moderate (managed in primary care) or severe (admitted to hospital). Patients who recovered from moderate exacerbations either remained in the same health state or deteriorated. Severe exacerbations could result in death, and surviving patients moved to the relevant ‘first month post-admission’ health state. It was assumed that no patients could improve into a better GOLD stage health state. Figure 15 illustrates the patient pathway in a stable health state.

FIGURE 15.

Pathways within stable health states.

Estimation of model parameters

This section outlines the assumptions applied, and sources used, to populate the base-case, usual-care arm.

Estimation of quality-adjusted life-years

Utility values were required for all health states and exacerbation events, and were combined with information on survival in order to calculate quality-adjusted life-years (QALYs). The model health states were based on COPD severity defined by GOLD stages 2–4. Utility values for these health states were calculated from unpublished data collected from the BLISS cohort. Utility scores for GOLD stages 2–3 were derived from the EQ-5D-5L (a revised version of the EQ-5D questionnaire). The five dimensions (mobility, self-care, usual activities, pain and discomfort, and anxiety and depression) have five levels, compared with the older version, which used three levels. The addition of two more levels may have made the EQ-5D more sensitive to differences in health states and avoid ceiling effects. 87

The EQ-5D-5L was completed by 917 participants enrolled in the BLISS study, with a confirmed diagnosis of COPD, at GOLD stage 2, 3 or 4, and converted to utility scores using the interim crosswalk value set for a UK population reported by EuroQoL. 88 Data from this cohort were deemed suitable for stable health states in the model for two reasons. First, participants were not recovering from an exacerbation at the time of questionnaire completion, therefore the utility scores were expected to reflect QoL in the stable condition. Second, > 80% of patients in each GOLD stage had been admitted to hospital at least once in the past year, therefore presenting a population who suffered exacerbations. EQ-5D-5L responses were converted to utility scores and are reported in Table 16.

The utility scores were compared with values applied in other COPD models that defined health states by GOLD severity stage. The utility values obtained from the BLISS cohort study were similar to those reported in other studies. 83,89–91 Data on utility loss suffered immediately after a moderate or severe exacerbation were extracted from previously published models; however, estimates varied greatly and the evidence underpinning these was poor. 83,89,92–94 It was assumed that there was a loss of utility for 1 month for moderate exacerbations and a utility loss for 3 months for severe exacerbations due to full recovery taking a longer period of time. 95–97 However, in line with other studies, the utility loss for severe exacerbations was assumed to be greatest in the first month, with improvement in QoL in the second and third months post admission.

The utility loss estimate of 15% for moderate exacerbation and 50% for severe exacerbation was obtained from Rutten-van Mölken et al. 89 This was applied to the mean utility score across all three severity stages (as opposed to each individually) to ensure that the utility loss suffered in stage 2 or 3 was not greater than loss experienced by stage 4 patients. The mean utility score found in the BLISS cohort across stages 2–4 was 0.693; therefore, a moderate exacerbation was assumed to result in a loss of 0.104 QALYs for 1 month and a severe exacerbation was assumed to result in a 0.346 loss of QALYs in the first month, reducing to a loss of 0.173 QALYs for months 2 and 3. A summary of all the utilities applied in the base-case analysis is provided in Table 17.

The review presented in Chapter 3 found evidence that SM may have a positive impact on QoL; however, the results were highly uncertain. As the model takes into account reduced rates of readmissions in the SM strategy, which leads to reduced loss of QoL, no further utility gains were applied in the model.

Resource use and costs

The resource use considered within the model was broadly concerned with the SM intervention, primary and secondary health-care professional contacts, and pharmacotherapy. Health-care contacts for each GOLD severity group were estimated with reference to National Institute of Health and Care Excellence (NICE) guidelines and expert opinion. Use of pharmacotherapy was estimated from data provided by the BLISS cohort. Unit costs were primarily obtained from NHS Reference costs98 and Unit Costs of Health and Social Care. 99 When appropriate, unit costs were inflated to 2012 prices using NHS Health Index inflation rates. Annual costs were divided by 12 to derive a monthly cost. Moderate and severe exacerbations were treated as additional one-off costs and assumed to be the same, irrespective of the underlying GOLD stage.

Assessment of cost-effectiveness

The incremental analysis was designed to generate the cost per additional QALY gained for SM delivered within 6 weeks of discharge when compared with UC in a cohort of patients with COPD. In summary, the key assumptions for the base case were as follows:

• The starting cohort was assumed to be aged 72 years, 47.4% male with 39.4% current smokers (see Chapter 6, Base-case cohort).

• The starting distribution of COPD severity was 35% GOLD stage 2, 35% GOLD stage 3 and 30% GOLD stage 4 (see Chapter 6, Base-case cohort).

• Mortality and readmission risks during admission and immediately after discharge were taken from the European Audit84 and applied for 3 months’ post-admission (see Chapter 6, Transition probabilities within post-admission health states).

• Long-term exacerbation, hospitalisation risk and disease progression were taken from large cohort studies of three years or more (see Chapter 6, Transition probabilities within stable health states).

• The estimate for reduction in risk of admission with SM was taken from moderate- to high-intensity programmes (see Chapter 6, Estimate for effectiveness of self-management report).

• Utility values were obtained from the BLISS cohort and an estimate was applied for the utility loss associated with exacerbation (see Chapter 6, Estimation of quality-adjusted life-years).

• The cost of UC was estimated with reference to pharmacotherapy use amongst the BLISS cohort, best practice guidance, expert opinion and NHS reference prices (see Chapter 6, Resource use and costs).

• The cost of SM was estimated to be £597, incurred in the first month and the effect was assumed to last for 2 years (see Chapter 6, Cost of self-management).

Where available, data were entered into the model as distributions so as to fully incorporate the uncertainty around parameter values in order that a probabilistic sensitivity analysis could be undertaken. Beta distributions were applied to the proportion on different treatments and accessing services in primary and secondary care; they were also applied to annual exacerbation rates and the proportion resulting in hospital admissions, as well as the risk reduction expected in the SM arm. Normal distributions were applied to utilities and utility losses. The probabilistic sensitivity analysis was run with 1000 simulations, and cost-effectiveness planes and acceptability curves were produced.

Sensitivity analysis

Additional model runs were undertaken to determine the impact of changing key parameters on the model results. Those parameters for which the incremental cost-effectiveness ratio (ICER) was demonstrated to be particularly sensitive to change were explored in more detail. The following analyses were undertaken:

1. The time horizon was varied, changing from the base-case assumption of 30 years to 6 months, 2 years, 10 years and 20 years.

2. The effect of SM on admissions was varied by substituting the base-case HR of 0.83 (95% CI 0.50 to 1.36) with two alternative HRs reported in Chapter 3. This included a HR derived from a meta-analysis of two low-intensity programmes of 0.96 (95% CI 0.49 to 1.90) and a meta-analysis that included exercise interventions representing a high-intensity programme of 0.78 (95% CI 0.52 to 1.17).

3. The duration of effect was tested for the base-case moderate-high-intensity SM programme, assuming the effect lasted for only (1) 6 months and (2) the lifetime of the cohort (see Chapter 6, Duration of effect).

4. The cost of SM was tested applying a low estimate of £85 and a high estimate of £671 (see Chapter 6, Cost of self-management).

5. An alternative set of utility scores obtained from Borg et al. 83 were applied (higher utility scores for GOLD stages 2 and 3, lower utility scores for GOLD stage 4 and a proportional deduction in utility for each severity stage lasting for 1 month in both moderate or severe exacerbation); see Table 26 (see also Utility values for chronic obstructive pulmonary disease).

6. Subgroup analysis was conducted to test if the decision rules changed if targeted at different subpopulations. This was tested by assuming that (1) all patients were GOLD stage 2; (2) all patients were GOLD stage 3; (3) all patients were GOLD stage 4; (4) there were different start ages; (5) all of the cohort were male; (6) all of the cohort were female; (7) all were smokers; (8) and all were non-smokers (see Chapter 6, Subgroup analysis).

7. Two scenario analyses were conducted: scenario 1 applied the highest estimate of effect 0.78 (95% CI 0.52 to 1.17) and the highest estimate of SM costs, £671; scenario 2 applied the lowest estimate of effect 0.96 (95% CI 0.49 to 1.90) and the lowest estimate of costs of £85 (see Chapter 6, Scenario analysis).

Base-case analysis

The base-case results presented in Table 27 show that, compared with UC, SM (delivered within 6 weeks of discharge) was more costly and resulted in better outcomes, with a £683 cost difference and gain of 0.0831 QALYs. The ICER was £8218 per QALY gained – well below the threshold values of £20,000–30,000 per QALY gained as recommended by NICE. 10

Results from the probabilistic sensitivity analysis are shown in the cost-effectiveness plane in Figure 16, which shows the distribution of 1000 resampled cost–effect difference pairs. The probabilistic sensitivity analysis clearly shows that SM is the more expensive strategy; however, the effectiveness is less certain, with a number of points indicating that SM may give fewer QALYs. The cost-effectiveness acceptability curve in Figure 17 shows that the intervention has a 68% probability of being cost-effective at £20,000 per QALY gained and 71% at a £30,000 threshold.

FIGURE 16.

Base-case cost-effectiveness plane.

FIGURE 17.

Base-case cost-effectiveness acceptability curve.

Sensitivity analysis

Alternative hazard ratios for readmissions in the self-management strategy

Owing to the considerable uncertainty around the effectiveness estimate (HR) for readmissions, this model parameter was expected to be the biggest driver of the cost-effectiveness results. To test this, two alternative HRs reported in the meta-analysis for review 1 were applied in the model and the results are shown in Table 29. The first alternative HR applied was a higher estimate of the effect. This was derived from a meta-analysis that included SM interventions with an exercise component. Applying the higher estimate of effect, the ICER decreased to £6249, and the likelihood of SM being cost-effective at a threshold value of £20,000 per QALY increased to 82%. The second alternative HR applied was a lower estimate of effect derived from a meta-analysis of two low-intensity SM programmes. This increased the ICER to £38,265. This was above the threshold value of £30,000/QALY and hence the probability of SM being cost-effective at £30,000/QALY reduced to 45%.

TABLE 29 - Sensitivity analysis: alternative HRs for admissions in SM support

HR	Cost difference (£)	QALY difference	ICER (£/QALY)	Probability cost-effective at £20,000/QALY (%)	Probability cost-effective at £30,000/QALY (%)
Base case (0.83, 95% CI 0.50 to 1.36)	683	0.0831	8218	68	71
High estimate (0.78, 95% CI 0.52 to 1.17)	676	0.1082	6249	82	84
Low estimate (0.96, 95% CI 0.49 to 1.90)	703	0.0184	38,265	41	45

Figure 18 illustrates the relationship between changing the point estimates of the HR and the mean ICER. At values of < 1 the ICERs are positive, and at all values of < 0.95 the ICERs are below the threshold of £30,000 per QALY. As the HR approaches 0.5, the ICER decreases and the benefits increase. At all values of > 1, SM is a less favourable option. If SM has no effect or increases the risk of hospital admission, it is dominated by UC (negative ICERs in Figure 18) hence why the ICER drops sharply. As the ratio increases to 1.5 the ICER decreases as UC becomes less cost-effective due to lower mortality in the UC arm. The 95% CIs for all three reported estimates crossed 1.

FIGURE 18.

Sensitivity analysis: relationship between HR for readmissions and ICER. a, Illustration of how the ICER changes by varying the HR between 0.5 and 1.5; and (b) magnification of how the ICER changes within the boundaries of a threshold of plus or minus £30,000 per QALY.

Duration of effect

Table 30 presents the results applying different assumptions regarding the duration of effect of SM support. In the base case it was assumed that the effect of SM would last for 2 years. The values were varied in the sensitivity analysis from 6 months to 30 years. When a shorter duration of effect was applied, the ICER increased and the probability of SM being cost-effective decreased. Conversely, applying a higher duration of effect decreased the ICER and increased the likelihood of SM being cost-effective.

TABLE 30 - Sensitivity analysis: alternative durations of effect

Duration of effect	Cost difference (£)	QALY difference	ICER (£/QALY)	Probability cost-effective at £20,000/QALY (%)	Probability cost-effective at £30,000/QALY (%)
Base case (2 years)	683	0.0831	8218	68	71
High estimate: 30 years	383	0.2876	1333	77	77
Low estimate: 6 months	686	0.0414	16,570	55	63

Figure 19 illustrates the relationship between changing the duration of effect and the ICER. At between 6 and 24 months the ICERs drop sharply and the decision rule changes. Most of the studies identified in the effectiveness review were short term in nature, resulting in uncertainty on the duration of this effect.

FIGURE 19.

Sensitivity analysis: relationship between ICER and duration of effect of SM. (a) Illustration of how the ICER changes between 10 months and 30 years; and (b) magnification of how the ICER changes between 3 and 25 months.

Cost of self-management

The impact of changing the cost of SM is presented in Table 31. Applying the high estimate of £671 increases the ICER to £9257 and the probability of SM being cost-effective at a threshold of £20,000 per QALY is similar at 69%. Applying the low estimate of £85 decreases the ICER to £1033 and increases the probability that SM is cost-effective at a threshold of £20,000 per QALY to 76%.

TABLE 31 - Sensitivity analysis: alternative assumptions for cost

Cost of SM (£)	Cost difference (£)	QALY difference	ICER (£/QALY)	Probability cost-effective at £20,000/QALY (%)	Probability cost-effective at £30,000/QALY (%)
Low estimate: 85	86	0.083	1033	76	77
High estimate: 671	768	0.0831	9257	69	69

Figure 20 shows the relationship between SM costs and the ICER. At all values for the cost of SM between £50 and £2200 the ICER is below a willingness-to-pay threshold of £30,000. At costs above £2200 the mean ICER in the base-case scenario is not cost-effective.

FIGURE 20.

Relationship between cost of SM and ICER.

Subgroup analysis

Scenario analysis

Table 37 presents the results of the scenario analysis. The first scenario applied the highest effect on reducing admissions (HR 0.78) and the highest cost estimate of SM (£671). The second scenario applied the lowest effect (HR 0.96) and the lowest estimate of the cost of SM (£85). This suggests that the likelihood of SM being cost-effective is greater in the higher cost, higher-effect scenario, relative to the lower-cost, lowest-effect scenario, but still < £20,000 per QALY.

Key results

This is the first economic model to consider the cost-effectiveness of SM support compared with UC in patients with COPD within 6 weeks of discharge from hospital admission for an exacerbation. Owing to the considerable uncertainty on the impact on readmissions, and the heterogeneity of the trial results, this model-based analysis should be viewed as speculative, and therefore providing only estimates of the potential impact of a SM programme.

The base-case model results suggested that SM support was a cost-effective intervention at the threshold at which NICE is willing to pay at £20,000 per QALY gained, if the assumption that the provision of SM support leads to a reduction in hospital admissions is met. The impact of reduced readmissions in the model led to lower mortality and morbidity from severe exacerbations over the long term. There were fewer costly hospital admissions, and intervention costs were relatively low compared with the cost of a readmission.

The probabilistic sensitivity analysis, which considers parameter uncertainty in the model, suggested that SM had a probability of 68% of being cost-effective at a threshold of £20,000/QALY, demonstrating the uncertainty around the impact of SM on readmissions. The remaining probability, when SM was not cost-effective, was due to the intervention potentially having worse outcomes while being more costly. Furthermore, the one-way sensitivity analyses undertaken were informative in highlighting the key drivers of the model results. As expected, cost-effectiveness was affected by the estimate of effect on readmissions, duration of effect and cost of a SM programme. The base case considered the intervention to have an impact for 2 years; however, the data from trials were collected only over the short term, for example 6 months. The results demonstrated that if the effect lasts for only 6 months then at a threshold of £20,000 per QALY gained the SM support was unlikely to be cost-effective. Currently, the model suggests that as long as the cost of the intervention is < £2200 then it is likely to be cost-effective; however, this threshold value will drop if the intervention is less effective.

Subgroup analysis, changing by considering different cohorts with regards to gender, age or smoking status, found no evidence of effect on the overall result. There was some evidence that SM might be more cost-effective in GOLD stage 4 patients. This is most likely to be due to a higher baseline risk of exacerbation and a higher proportion of exacerbations resulting in hospital admission. Therefore, a risk reduction is likely to have a greater effect.

Strengths and limitations

A key strength of this analysis is that this is the first economic model to consider the cost-effectiveness of SM in this particular patient group and illustrates the key variables that impact on the results. Although no good-quality, long-term data currently exist on the effectiveness of SM, a model structure exists for reanalysis once additional data become available.

The model structure applied was a further strength of this study. It is a modified version of previously published decision models whereby additional post-admission health states were added to incorporate emerging evidence on the higher risks in patients with COPD immediately after discharge. 13 This was thought to be particularly important for measuring costs and outcomes in this model, as the patient population were assumed to receive the intervention within 6 weeks of discharge.

Although there are concerns about the effectiveness estimates, robust data were included in the model to represent the natural history of the condition. The risks applied in the first 3 months were obtained from the UK cohort included in the European Audit84 of patients with COPD admitted to hospital. The model also captures long-term outcomes by disease severity, applying data on long-term exacerbation risks, mortality and disease progression from large longitudinal studies – TORCH85 and ECLIPSE. 13 This study also applies patient-level utility data obtained from a representative sample of UK patients (unpublished data obtained from BLISS cohort). Finally, although there was a great deal of uncertainty around effectiveness data and assumptions applied to this model, distributions were applied to reflect this uncertainty. The probabilistic sensitivity analysis was also supplemented by a one-way sensitivity analysis of all key parameters, thus demonstrating which parameters were mostly likely to influence decisions to implement SM.

There are a number of caveats when considering the results of this economic model. Most importantly, this is a speculative decision model and therefore can be considered as indicative of only the potential cost-effectiveness of SM. As highlighted in the clinical effectiveness and cost-effectiveness review, there is a dearth of high-quality evidence on the long-term costs and outcomes associated with this intervention. This model was based on this weak evidence of effect and thus incorporates considerable uncertainty when assumptions from the literature and estimates from clinical experts were applied in the absence of better-quality data.

In addition to the uncertainty around the effect of SM, there was also some uncertainty around parameters and assumptions applied in the base case for UC. Although the model was able to reflect mortality and readmission risks in the first 3 months after discharge, it was assumed that after those 3 months, those who were not readmitted would move to a stable health state. It is unclear if this really reflects natural history or if the risk of readmissions and mortality remain higher beyond 3 months among those with recent history of exacerbation. Similarly, the data extracted from the TORCH85 and ECLIPSE13 studies represent average exacerbation and hospitalisation rates in stable COPD cohorts over a 3-year period and these data were applied over a 30-year time horizon. In reality this may change over time.

The model highlights a number of areas in which further research is required. Crucially, further evidence is needed on the effectiveness of SM support to confirm if it is indeed cost-effective and with greater certainty. Longer-term evidence beyond 6 months is also required. Follow-up data on cohorts of patients admitted to hospital is needed to provide better estimates of long-term outcomes. A review of costs applied of other SM programmes in patients with COPD suggests that the cost of implementing SM support are likely to range somewhere between £8568 and £671. 63 Although more research is required to develop more accurate cost estimates for implementing SM programmes in this cohort, this is unlikely to change the outcome of this analysis. Finally, better data are required on utility values in COPD populations, particularly the utility loss associated with exacerbation.

Although outside the scope of this report – in light of the uncertainty around the effectiveness and cost-effectiveness of the intervention – it would be beneficial for a value of information analysis to be undertaken in the future. Value of information analysis allows a comparison of the potential benefits of additional research with the costs of further investigation. The value of any further research is based on how much this extra information will reduce the overall decision uncertainty.

Summary

• Currently, there is no published evidence on the cost-effectiveness of SM compared with UC in patients with COPD who have recently been discharged from hospital.

• This is the first economic model to attempt to estimate the cost-effectiveness of SM in this patient group.

• This speculative model indicates that SM is cost-effective if it is assumed that the intervention has a small positive effect on reducing admissions for a minimum of 6 months.

• The model has a number of limitations, the most important related to the large amount of uncertainty around the effectiveness estimate driving the model results.

• The analysis highlights the importance of conducting further research on the effect and duration of effect of the SM intervention delivered post discharge to allow a more robust analysis of cost-effectiveness.

Definition of self-management

As described for review 1 and tabulated in Table 2.

Search strategy

A comprehensive search strategy described as for review 1. Only citation lists of relevant reviews were examined for additional relevant studies.

Study selection process

As described for review 1.

Selection criteria

In contrast with the first review, owing to the likely high volume of relevant studies, the selection criteria included only RCTs and a more limited range of outcomes (Table 38). Although RCTs purely of smoking cessation were excluded, trials described as ‘pulmonary rehabilitation (PR)’ were included, as many PR trials include components of SM and aim to enable participants to self-manage their condition after the PR programme ends. Furthermore, many interventions describe supported SM with a supervised structured exercise programme, which is similar to PR. There is a large overlap of intervention content even with different definitions and we wanted to include as complete a range of evidence as possible.

Risk of bias assessment

As for review 1, all of the RCTs were assessed using the Cochrane Risk of Bias tool. 56 Assessment was limited to primary outcomes. All studies were assessed by one independent reviewer, with a second reviewer independently checking at least 10% of studies, and a third reviewer overseeing the complete process.

Data extraction

Analyses

Owing to a large volume of literature, only those papers that reported any one of the three primary outcomes were taken forward for further analyses:

1. HRQoL scores including subdomain scores

2. numbers of/time to first hospital admissions/readmissions

3. numbers of/time to first exacerbation/s.

Papers with secondary, but no primary, outcomes were tabulated only.

Analyses consisted of mapping and description of the features and elements of the interventions from the included papers; presentation of the results of various combinations and comparisons of components on forest plots; meta-analysis of the data where appropriate; meta-regression; and subgroup analysis to explore heterogeneity.

Included studies

From 13,355 identified titles, 836 full papers were obtained and 283 papers were finally included. Of these, 174 RCTs from 194 papers reported one of the three primary outcomes: HRQoL, hospital admissions/readmissions and exacerbations (Figure 21). A total of 89 papers reported outcomes other than our three primary outcomes and are listed in Appendix 21 alongside the secondary outcomes that they reported. Overall, 553 papers were excluded (see Appendix 2 for full list with reasons for exclusion). Arbitration by a third reviewer was required for 5% of all full texts. In total, 40 ongoing studies were identified as relevant (see Appendix 4).

FIGURE 21.

Summary of the selection process for clinical effectiveness studies.

Within the 174 trials with primary outcomes several studies had multiple arms. Thus there were 229 comparisons of interventions compared with usual care (UC), control or another active intervention.

Characteristics of studies

The study and population characteristics of the 174 included RCTs with relevant primary outcomes are summarised in Appendix 22.

Interventions

The interventions were very heterogeneous. They included structured group-based PR programmes; more limited one-to-one educational SM interventions delivered in an outpatient setting or at a patient’s home, sometimes with telephone follow-up; integrated disease management with multidisciplinary input and often some element of monitoring by health professionals; exercise-only interventions (with some dyspnoea management) and respiratory muscle training (RMT) using threshold devices. Within these various broad categories, there was a range of individual SM components, including some that might be less traditionally part of SM, such as qigong, t’ai chi and singing. Appendix 23 provides detailed descriptions of the intervention and comparator groups with intensity and frequency of interventions delivered.

Description of self-management components in intervention and comparator arms

Within the arms of the 174 trials we categorised 15 types of components (plus other and unspecified). In the intervention groups exercise was the most commonly reported component (76.9%), followed by breathing techniques and management of dyspnoea (64.2%), and general education about COPD and its management (47.2%). Details of the numbers of individual components for the intervention and comparator groups are shown in Table 45. Appendix 24 displays which components were present within the intervention and comparator groups of each study.

TABLE 45 - Self-management components reported in the interventions and comparator groups

Component	Intervention (no. of studies)	Comparator (no. of studies)
	n (%)	n (%)
Exercise	176 (76.9)	96 (41.9)
Breathing techniques/dyspnoea management	147 (64.2)	52 (22.7)
Disease knowledge	108 (47.2)	68 (29.7)
Psychological including relaxation and stress management	77 (33.6)	34 (14.8)
Medication advice	77 (33.6)	43 (18.8)
Nutrition advice	51 (22.3)	28 (12.2)
Enhanced access	50 (21.8)	15 (6.6)
Action planning for self-treating exacerbations	43 (18.8)	9 (3.9)
Smoking cessation advice/support	44 (19.2)	18 (7.9)
Inhaler technique	36 (15.7)	19 (8.3)
Bronchial hygiene/secretion clearance techniques	30 (13.1)	16 (7.0)
Unspecified	24 (10.5)	9 (3.9)
RMT	32 (14.0)	11 (4.8)
Energy conservation	22 (9.6)	7 (3.1)
Other	18 (7.9)	5 (2.2)
Preventative measures to avoid infection	18 (7.9)	11 (4.8)
COPD support groups	7 (3.1)	3 (1.3)

Up to 13 different SM components were included in any one of the intervention arms, and up to 11 in any one of the comparator groups (Table 46). Seventy-three (31.9%) of the intervention arm interventions had six or more components. In the intervention group, 38 (16.6%) were single components with the vast majority of these being exercise-only interventions. In contrast, the majority of the comparators had two or fewer described components [167 (72.9%)], with 34.9% not providing any detail about the SM education or support provided to the comparator group as part of UC (see Table 46).

TABLE 46 - Numbers of components in intervention vs. comparator groups

		No. of components in the comparator groups
		0	1	2	3	4	5	6	7	8	9	10	11	Total, n (%)
No. of components in the intervention groups	1	10	22	2	1	2	0	1	0	0	0	0	0	38 (16.6)
	2	20	18	13	0	0	0	2	0	1	0	0	0	54 (23.6)
	3	7	5	11	6	0	0	0	0	0	0	0	0	29 (12.7)
	4	6	3	2	6	2	0	0	0	0	0	0	0	19 (8.3)
	5	8	2	1	1	3	1	0	0	0	0	0	0	16 (7.0)
	6	7	3	0	0	3	1	6	1	0	0	0	0	21 (9.2)
	7	8	2	0	2	2	0	2	2	0	0	0	0	18 (7.9)
	8	6	0	0	1	2	0	0	2	3	0	0	0	14 (6.1)
	9	2	0	1	0	0	0	0	1	0	1	0	0	5 (2.2)
	10	4	1	0	0	0	0	0	1	0	1	0	0	7 (3.1)
	11	0	1	0	0	0	0	1	1	0	0	1	1	5 (2.2)
	12	1	0	0	0	0	0	0	0	0	0	0	1	2 (0.9)
	13	1	0	0	0	0	0	0	0	0	0	0	0	1 (0.4)
Total no. of studies		80 (34.9)	57 (24.9)	30 (13.1)	17 (7.4)	14 (6.1)	2 (0.9)	12 (5.2)	8 (3.5)	4 (1.7)	2 (0.9)	1 (0.4)	2 (0.9)	229

The content of the components of the intervention are shown according to the total number of components in the intervention in Table 47. Of the single-component interventions, 25 of 38 (65.8%) were exercise only, 9 of 38 (23.7%) were RMT and three (7.9%) were breathing exercises. In the two- and three-component interventions, exercise is frequently combined with breathing/dyspnoea management and disease knowledge. Overall, the most common components were exercise [176 (76.9%) of studies], breathing techniques and dyspnoea management [147 (64.2%)], disease knowledge [108 (47.2%)] and psychological interventions [77 (33.6%)].

In those interventions with six or more components, the most common components were exercise [69 (94.5%) of studies], breathing techniques and dyspnoea management [66 (90.4%)], disease knowledge [65 (89.0%)] and medication advice [60 (82.5%)]. Notably, smoking cessation was mentioned in only 38 (52.1%) of the interventions with six or more components.

TABLE 47 - Content of interventions by the number of components within the SM package

Intervention components	No. of SM components in intervention													Total no. of studies
	1	2	3	4	5	6	7	8	9	10	11	12	13
Action planning for self-treating exacerbations	0	3	2	2	4	2	7	6	4	7	3	3	1	43
Breathing techniques/dyspnoea management	3	37	16	15	10	18	16	12	5	7	5	24	1	147
Bronchial hygiene/secretion clearance techniques	0	0	0	4	2	3	5	4	1	3	5	2	1	30
Disease knowledge	0	8	11	10	14	17	16	13	4	7	5	2	1	108
Energy conservation	0	0	0	2	0	3	1	8	1	2	4	1	0	22
Enhanced access	0	4	6	3	6	4	8	7	3	4	3	1	1	50
Exercise	25	37	21	14	10	18	17	14	5	7	5	2	1	176
Inhaler technique	0	0	0	1	2	4	8	7	4	5	2	2	1	36
Medication advice	0	1	6	3	7	16	13	13	5	6	5	1	1	77
Nutrition advice	0	2	3	1	4	8	10	6	4	5	5	2	1	51
Preventative measures to avoid infection	0	0	3	0	0	2	2	3	2	2	2	1	1	18
Psychological including relaxation and stress management	1	3	8	14	8	14	8	7	2	4	5	2	1	77
RMT	9	7	7	1	0	3	4	0	0	0	1	0	0	32
Smoking cessation advice/support	0	0	0	1	5	6	9	8	4	5	3	2	1	44
COPD support groups	0	0	0	0	1	2	1	1	0	1	0	1	0	7
Unspecified	0	0	1	4	3	6	0	3	0	4	1	1	1	24
Other	0	6	3	1	4	0	1	0	1	1	1	0	0	18

Figure 22 displays the range of different interventions included.

FIGURE 22.

Range of interventions included across 174 trials.

Duration of the intervention

The duration of the intervention was measured to the last behavioural or supportive contact and ranged from 1 day to 2 years (Table 48). A total of 114 trials (58.8%) reported interventions of ≤ 3 months’ duration, with nine trials (4.6%) being longer than 1 year. Five trials71,72,117,142,215,255,280,281 did not report an intervention duration or had a variable duration intervention (see Appendix 23).

TABLE 48 - Duration of intervention

NR, not reported.

Intervention duration (weeks)	No. of studies	%
< 1	2	1.1
4	19	10.9
5–8	59	33.9
9–13	34	19.5
14–26	29	16.7
27–52	22	12.6
53–104	4	2.3
NR/variable duration	5	2.9

Mode of delivery of the intervention

The majority of the interventions were delivered by nurses and respiratory physiotherapists. Half of the interventions had a group-based component; 63 (36.2%) were entirely group based; and an additional 24 (13.8%) had a group component followed by individual support. In 20 studies the mode of delivery was unclear. Details of the mode of delivery are in Table 49 and in the detailed characteristics of intervention table in Appendix 23.

TABLE 49 - Mode of delivery of SM interventions

Mode of delivery	n (studies)	%
Group based	63	36.2
Individual	63	36.2
Mixed: group and one to one	24	13.8
Remote (internet/telemonitoring)	4	2.3
Unclear	20	11.5
Total	174	100.0

Comparator arms

There were 141 comparisons (from 127 trials) of an intervention compared with UC or a control group that was not an active intervention. The UC arm was frequently not described; in other cases it was the standard primary and/or secondary care for people with COPD.

A total of 107 comparisons (from 85 trials) were of two active interventions. Details are provided in Appendix 23.

Risk of bias of included studies

Table 50 summarises the risk of bias. Details of the risk of bias assessment for all of the included studies are tabulated in Appendix 26.

Reporting of the method of generating the randomisation sequence was generally poor, with only 71 (36%) of the trials providing adequate information. Where reported, the randomisation method was adequate to produce a low risk of bias. Similarly, the majority of studies [146 (84%)] did not provide sufficient information about allocation concealment to be able to determine the risk of bias. We considered the risk of bias for self-reported HRQoL to be high unless the participant was blinded to the intervention, either by randomisation to an active intervention in each study arm, or through a sham intervention. This resulted in a high rate of categorisation of high risk of bias for this outcome measure [117 (63%)].

We also assumed that reporting of hospital admission would be unlikely to be influenced by knowledge of allocation; thus, the majority of trials reporting this outcome were categorised as at low risk of bias for this outcome. Loss to follow-up was frequently high, and authors often failed to adequately account for those with missing outcome data or did not describe their characteristics. Relatively few trials reported that they had published a protocol or were registered on a clinical trials database, so only 63 (30%) were categorised as at low risk of bias of selective outcome reporting; however, in most cases all the outcome measures described in the methods were reported in the results section.

A significant other potential cause of bias was due to authors reporting in the abstract the numbers completing the trial rather than randomised and the characteristics of only those who completed the trial. Furthermore, baseline imbalances (e.g. caused by small sample sizes) were not routinely adjusted for in statistical analyses, and thus differences at follow-up (e.g. in QoL) might (partly) be due to baseline differences.

All trials

Figures 23–26 plot the outcomes at all reported time points for all trials for HRQoL measured by the SGRQ and the CRQ, hospital admission and exacerbations. These results have not been combined by meta-analysis due to the heterogeneity of the interventions and comparators.

The trials were ordered by the number of components, and upon visual inspection there does not appear to be any relationship between the size of the effect and the number of components. For HRQoL, many trials reported a large difference between the intervention and comparator group at baseline and, when present, this was rarely adjusted for in the analysis using ANCOVA.

At the last follow-up point, 11 of 56 (19.6%) resulted in a statistically significant reduction in hospital admissions and 4 of 28 (14.3%) a statistically significant reduction in exacerbations. A total of 22 of 87 (25.3%) comparisons showed a statistically significant improvement in total SGRQ score, 16 of 41 (39.0%) in total CRQ, 10 of 24 (41.7%) on the physical components of the Short Form questionnaire-36 items (SF-36) but only 2 of 21 (9.5%) on the mental component of the SF-36. For individual components, the CRQ ‘dyspnoea’ and ‘mastery components had the highest proportions of reported significant improvements (26.4% and 25.8%, respectively).

FIGURE 23.

(a) Health-related quality of life as measured by the SGRQ at all reported follow-up points for all included studies; and (b) key for figure. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. b, Indicates that the number in the control group has been halved, with half of the group used as control for one comparison (nurse-assisted medical management vs. UC) and half for the other comparison (nurse-assisted collaborative management vs. UC).

FIGURE 24.

Health-related quality of life as measured by the CRQ at all reported follow-up points for all included studies. A = rehabilitation (traditional and modern) + qigong + breathing training + limb training vs. UC. B = modern rehabilitation + breathing training + limb training vs. UC. C = traditional rehabilitation + qigong vs. UC. D = rehab (traditional and modern) + qigong + breathing training + limb training vs. modern rehabilitation + breathing training + limb training. E = rehab (traditional and modern) + qigong + breathing training + limb training vs. traditional rehabilitation + qigong. F = rehab (traditional and modern) + qigong + breathing training + limb training vs. traditional rehabilitation + qigong.

FIGURE 25.

Hospital admissions at all reported follow-up points for all included studies. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

FIGURE 26.

Exacerbations at all reported follow-up points for all included studies. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

Multicomponent interventions compared with usual care

Health-related quality of life

Multicomponent interventions were defined as those with three or more components. Forty-one trials reporting 44 interventions compared with UC reported usable total SGRQ or CRQ results; 31 trials reported hospital admissions, of which 18 could be used in meta-analysis; and 12 trials reported exacerbations of which three could be used in the meta-analysis (see Appendix 25).

The meta-analysis findings are reported for three time periods: up to and including 3 months, greater than 3 months but up to 6 months (hereafter described as 3–6 months) and greater than 6 months.

For SGRQ followed up to 3 months, most trials reported a greater improvement from baseline to follow-up in favour of the intervention group. 68,121,142,144,148,149,161,186,194,213,214,221,226,227,237–239,257 The summary meta-analysis result reveals that on average the multicomponent arm had a SGRQ score of 6.50 points (95% CI 3.62 to 9.39 points) higher than the UC arm. However, this is the average of a distribution of trial effects and this distribution is wide due to high heterogeneity (I² = 82.4%). The prediction interval reports the range in which 95% of the distribution of the effects lies. The majority of the interval is > 0 and thus mainly in favour of multicomponent interventions; however, it does overlap zero (95% CI –4.66 to 17.67) indicating that the interventions are not always effective. At the longer follow-up point of 3–6 months, the estimate of the average effect was 4.47 points on the SGRQ (95% CI 1.93 to 7.02 points; I² = 79.6%) and at > 6 months it was 2.40 points (95% CI 0.75 to 4.04 points; I² = 57.9%), both favouring the intervention group. There is some suggestion that the size of the effect was smaller as follow-up was longer, but loss to follow-up was also a more significant problem at the longest follow-up point, which may have biased the effect estimate. The upper boundary of the prediction intervals also lowers with the longer follow-up time.

There were fewer trials reporting the CRQ. At all three time points the estimate of the average effect was in favour of the intervention group, suggesting that, on average, multicomponent SM interventions were more effective than UC at up to 6 months’ follow-up. However, heterogeneity was high and the prediction intervals at all three time points included zero. The results for HRQoL are summarised in Figures 27 and 28 and Table 51.

FIGURE 27.

Health-related quality-of-life (SGRQ) outcomes for multicomponent SM intervention vs. UC. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. b, Number in the control group has been halved with half of the group used as control for one comparison (nurse-assisted medical management vs. UC) and half for the other comparison (nurse-assisted collaborative management vs. UC). Chan 2010212 B = exercise vs. control; Coultas 2005112 A = nurse-assisted collaborative management vs. UC; Coultas 2005112 B = nurse-assisted medical management vs. UC.

FIGURE 28.

Health-related quality-of-life (CRQ) outcomes for multicomponent SM intervention vs. UC. Xu 2010204 A = rehabilitation (traditional and modern) + qigong + breathing training + limb training vs. UC.

TABLE 51 - Health-related quality-of-life outcomes for multicomponent interventions vs. UC

Outcome	Time frame	No. of studies (comparisons)	Summary MD (95% CI)	I2 (%)	95% prediction interval
SGRQ	≤ 3 months	18	6.50 (3.62 to 9.39)	82.4	–4.66 to 17.67
	> 3 to ≤ 6 months	14 (15)	4.47 (1.93 to 7.02)	79.6	–4.71 to 13.65
	> 6 months	16	2.40 (0.75 to 4.04)	57.9	–2.38 to 7.17
CRQ	≤ 3 months	7	0.40 (0.01 to 0.79)	75.7	–0.84 to 1.64
	> 3 to ≤ 6 months	5	1.02 (0.05 to 1.98)	93.2	–2.66 to 4.70
	> 6 months	3	1.21 (–0.47 to 2.88)	96.2

Hospital admissions

The results of 18 studies63,65,67,68,70,71,73,120,140,142,148,160,168,212,213,227,266,270 of multicomponent interventions compared with UC which reported hospital admissions have been analysed; eight with follow-up at ≤ 3 months,67,68,70,120,148,160,212,227 four with follow-up to 6 months;70,73,160,266 and eight with follow-up at ≥ 1 year. 63,65,71,140,142,168,213,270 Although the summary HRs from meta-analysis favoured the intervention groups at all three follow-up periods, there was much uncertainty leading to only weak statistical evidence of an effect and heterogeneity was high at follow-up times of > 3 months. Details are given in Figure 29 and Table 52.

FIGURE 29.

Hospital admissions for multicomponent SM interventions vs. UC. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. Chan 2010212 B = exercise vs. control.

Exacerbations

Exacerbations were reported in an analysable format by only four studies. 73,182,213,284 The multicomponent interventions had no evident effect – details in Figure 30 and Table 52. At the last follow-up point, only 2 of 12 studies that reported exacerbations reported a statistically significant effect in favour of the multicomponent SM intervention. 140,172

FIGURE 30.

Exacerbations for multicomponent SM interventions vs. UC. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

TABLE 52 - Multicomponent interventions vs. UC: hospital admissions and exacerbations

n/a, not applicable.

Outcome	Time frame	No. of studies	Summary HR (95% CI)	I2 (%)	95% prediction interval
Admissions	≤ 3 months	8	0.94 (0.73 to 1.20)	0.0	0.69 to 1.28
	> 3 to ≤ 6 months	4	0.56 (0.22 to 1.42)	77.8	–
	> 6 months	8	0.79 (0.60 to 1.05)	62.6	0.36 to 1.77
Exacerbations	≤ 3 months	1	3.01 (0.31 to 28.96)	n/a	–
	> 3 to ≤ 6 months	2	1.01 (0.59 to 1.74)	0.0	–
	> 6 months	2	1.09 (0.77 to 1.53)	37.4	–

Summary: multicomponent interventions

Evidence of effectiveness of multicomponent interventions on HRQoL, but considerable uncertainty for hospital admissions and exacerbations.

Exploring specific individual components of self-management interventions

We aimed to explore the effectiveness of specific individual components of SM interventions by examining the trials for which there was a difference in one component between intervention and control arms.

Action plans

Four trials170,188,229,231 reported the addition of an action plan to a SM package. There was no difference in the average effect on HRQoL of the arms including action plans compared with the comparator groups (SGRQ 0.43, 95% CI –1.69 to 2.54; I² = 0%) (Figure 31).

FIGURE 31.

Health-related quality of life (SGRQ) at final follow-up for comparisons assessing the effects of one additional component of SM.

McGeoch et al. 229 further undertook 1-year follow-up of people who were given an action plan and reported no effect on hospital admissions (HR 0.97, 95% CI 0.33 to 2.89) (Figure 32). At 6 months’ follow-up, a large trial by Trappenburg et al. 188 found no additional effect of action plans on exacerbations (HR 1.12, 95% CI 0.77 to 1.62) (Figure 33).

FIGURE 32.

Admissions at final follow-up for comparisons assessing the effects of one additional component of SM.

FIGURE 33.

Exacerbations at final follow-up for comparisons assessing the effects of one additional component of SM.

Breathing techniques

Two trials235,240 reported breathing training or techniques. On average, the breathing training groups had a SGRQ score that was 5.0 points (95% CI 4.06 to 5.94 points) higher than the comparison groups. Although the trials were of a small size, the heterogeneity was low (I² = 0%). Van Gestal et al. 208 reported no difference in the CRQ at 4 weeks’ follow-up (0.17, 95% CI –0.09 to 0.43 points) (Figure 34; see also Figure 31).

FIGURE 34.

Health-related quality of life (CRQ) at final follow-up for comparisons assessing the effects of one additional component of SM. Xu 2010204 D = rehabilitation (traditional and modern) + qigong + breathing training + limb training vs. modern rehabilitation + breathing training + limb training.

Exercise-only interventions compared with usual care or a sham intervention

We further examined the effect of individual components by reviewing trials in which exercise was a single component. The exercise could be supervised or unsupervised, but there were no other SM components.

Eight trials64,183,207,212,240,248,252,282 reported the effects of exercise-only interventions on HRQoL, with no other SM components in a way through which the data could be incorporated in the meta-analyses. The four trials183,212,240,252 with five comparator groups, which reported SGRQ at up to 3 months, reported a significant benefit from the exercise-only intervention (4.87 points, 95% CI 3.96 to 5.79). The prediction interval (3.39–6.36) provides strong evidence that exercise-only interventions are effective in improving the SGRQ score at up to 13 weeks’ follow-up. Only Gohl et al. 207 reported the SGRQ at 1-year follow-up with a large but non-significant effect in favour of exercise, although the sample size was very small with only 19 participants. Three trials64,248,282 reported the CRQ outcome, although all had low rates of follow-up, so results must be interpreted with caution. At 3 months’ follow-up there was a modest effect in favour of the exercise group but not statistically significant, and at 6 months there was a larger, non-significant effect, with a high level of heterogeneity. Details are shown in Figures 35 and 36, and Table 53.

FIGURE 35.

Health-related quality-of-life (SGRQ) outcomes for exercise-only interventions vs. UC/sham intervention. a, Refers to the fact that number in the control group has been halved with half of the group used as control for one comparison (t’ai chi qigong vs. control) and half for the other comparison (exercise vs. control). Chan 2010212 A = t’ai chi qigong vs. control. Chan 2010212 B = exercise vs. control.

FIGURE 36.

Health-related quality-of-life (CRQ) outcomes for exercise-only interventions vs. UC/sham intervention.

Chan et al. 212 compared t’ai chi and exercise to UC in separate comparisons and reported no effects on hospital admission rates (Figure 37). Behnke et al. 64 undertook follow-up at the end of an 18-month exercise-only intervention and showed a significantly lower hospital admission rate in the intervention group (3/14) compared with 9 of 12 in the UC group (HR 0.17, 95% CI 0.05 to 0.64). In addition, Hernandez et al. 282 reported no statistical difference in admission rates between the exercise group and comparator at last follow-up. A small trial by Murphy et al. 252 reported exacerbations at 3 and 6 months, with a suggestion of lower rates, but participant numbers were very small (HR at 3 months 0.02, 95% CI 0.00 to 28.17; HR at 6 months 0.34, 95% CI 0.07 to 1.77) (Figures 37 and 38). Chan et al. 212 reported no difference on exacerbations between groups at 3 months’ follow-up.

FIGURE 37.

Admission outcomes for exercise-only interventions vs. UC/sham intervention. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. b, Refers to the fact that number in the control group has been halved with half of the group used as control for one comparison (t’ai chi qigong vs. control) and half for the other comparison (exercise vs. control). Chan 2010212 A = t’ai chi qigong vs. control; Chan 2010212 B = exercise vs. control.

FIGURE 38.

Exacerbation outcomes for exercise-only interventions vs. UC/sham intervention.

Enhanced care

We wanted to explore the general effects of providing support to patients over just giving simple information approaches, which we termed ‘enhanced care’. It included regular telephone contact to reinforce information or behaviour change techniques, provided encouragement or included scheduled home visits for assessment with reinforcement or encouragement.

Fifteen studies63–65,67,71,72,110,120,140,155,180,188–190,192,193,206,251,270,271,283 provided information in the form of the total SGRQ or CRQ and were included in the meta-analyses. Only three studies67,120,193 reported the SGRQ at 3 months’ follow-up with very heterogeneous results; estimate of average effect was 1.27 points (95% CI –4.28 to 6.82 points, I² = 73.8). At 3–6 months’ and 12 months’ follow-up the enhanced-care arm had a higher SGRQ score than the UC arm – details are given in Figure 39 and Table 54. The estimates of the average CRQ at the three follow-up times all favoured the enhanced-care arm but were not statistically significant and heterogeneity was very high (Figure 40).

FIGURE 39.

Health-related quality-of-life (SGRQ) outcomes for enhanced care (± SM package) vs. UC/SM package. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

FIGURE 40.

Health-related quality-of-life (CRQ) outcomes for enhanced care (± SM package) vs. UC/SM package.

On average, enhanced care had a similar risk of hospital admission at 3 and 6 months (Figure 41 and Table 54), but a lower risk at 1 year or longer (HR 0.78, 95% CI 0.62 to 0.99). However, there was moderate heterogeneity at 1 year and the prediction interval showed that the intervention would frequently not be effective in lowering hospital admissions. Seven studies did not provide data that could be included in the meta-analyses,69,74,136,140,180,188,251 of which three also reported a statistically significant reduction in hospital admissions at 1-year follow-up. 140,180,251

FIGURE 41.

Admissions outcomes for enhanced care (± SM package) vs. UC/SM package. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

Only three trials155,188,206 were included in the meta-analyses for exacerbations with no evidence of any effect of enhanced care on risk of exacerbation (Figure 42). Four other trials120,140,180,251 reported exacerbation rates at last follow-up, one140 of which reported a statistically significant reduction.

FIGURE 42.

Exacerbation outcomes for enhanced care (± SM package) vs. UC/SM package.

The contribution of exercise to multicomponent self-management packages

The following analyses explore the contribution of exercise and its mode of delivery as a contributor to the heterogeneity. Multicomponent interventions:

• with a supervised exercise element compared with UC

• a structured, unsupervised exercise element compared with UC

• exercise counselling only compared with UC

• without an exercise element compared with UC.

Multicomponent interventions with a supervised exercise element compared with usual care

Trials were included in this category if the exercise component of a larger package of care was directly supervised. The majority were group, centre-based interventions (generally referred to as PR).

Health-related quality of life

Of the 47 trials that reported HRQoL, 26 trials (27 interventions) reported disease-specific HRQoL using the total SGRQ or CRQ (see Appendices 13 and 27). Findings are similar to those for multicomponent interventions overall, with the largest estimate of the average effect in SGRQ at up to 3 months’ follow-up (7.75, 95% CI 3.49 to 12.01) points. There was no difference in average effect between the intervention and UC arms in SGRQ at follow-up of > 6 months. Heterogeneity was very high for most outcomes and follow-up times and loss to follow-up variable. Small but significant improvements in the CRQ were seen at 3 and 6 months’ follow-up favouring the SM group. Details are given in Figures 43 and 44 and Table 55.

FIGURE 43.

Health-related quality-of-life (SGRQ) outcomes for multicomponent SM interventions including supervised exercise vs. UC/control. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. Chan 2010212 B = exercise vs. control.

FIGURE 44.

Health-related quality-of-life (CRQ) outcomes for multicomponent SM interventions including supervised exercise vs. UC/control. Xu 2010204 A = rehabilitation (traditional and modern) + qigong + breathing training + limb training vs. UC.

TABLE 55 - Outcomes for multicomponent interventions with supervised exercise

n/a, not applicable.

Outcome	Time frame	No. of studies (comparisons)	Summary MD (95% CI)	I2 (%)	95% prediction interval
SGRQ	≤ 3 months	14	7.75 (3.49 to 12.01)	80.1	–8.25 to 23.75
	> 3 to ≤ 6 months	6	6.57 (3.24 to 9.90)	77.6	–3.66 to 16.79
	> 6 months	4	1.13 (–2.81 to 5.08)	0	–
CRQ	≤ 3 months	7	0.43 (0.03 to 0.83)	77.8	–0.86 to 1.72
	> 3 to ≤ 6 months	6	1.02 (0.19 to 1.86)	92.0	–1.95 to 3.99
	> 6 months	3	1.21 (–0.47 to 2.88)	95.2	–
			Summary HR (95% CI)
Admissions	≤ 3 months	4	0.78 (0.54 to 1.14)	0.0	–
	> 3 to ≤ 6 months	1	0.55 (0.25 to 1.18)	n/a	–
	> 6 months	2	0.47 (0.08 to 2.60)	83.8	–
Exacerbations	> 3 to ≤ 6 months	2	0.99 (0.56 to 1.75)	0	–
	> 6 months	2	1.09 (0.77 to 1.53)	37.5	–

Hospital admissions

We were able to combine the reports of hospital admission rates in six trials. 65,148,160,192,212,213,227 Although the trend was for the average effect on admission rates to be lower in the intervention arms, this was not a significant effect at any of the follow-up time points (Figure 45). Four other trials172,182,212,213 reported hospital admissions at last follow-up; only one reported a statistically significant reduction in hospital admissions. 172

FIGURE 45.

Admissions outcomes for multicomponent SM interventions including supervised exercise vs. UC/control. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22. Chan 2010212 B = exercise vs. control.

Exacerbations

Three trials182,206,213 reported exacerbation rates in such a way that a HR could be computed; heterogeneity was low or moderate but no evidence of effect was observed (Figure 46 and Table 55). Two additional trials172,212 reported exacerbations at last study follow-up, with Güell et al. 172 reporting a statistically significant reduction in favour of the intervention group.

FIGURE 46.

Exacerbation outcomes for multicomponent SM interventions including supervised exercise vs. UC/control. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

Multicomponent interventions with a structured, unsupervised exercise element compared with usual care

Trials were included in this category if they provided detail about a structured home exercise programme including duration and proposed frequency of exercise within a larger package of care. These were home-based interventions.

Of the eight trials186,251,255,262,264,270,283,284 that reported HRQoL, five used the total SGRQ186,251,255,262,270 and one the CRQ. 283 On average, the multicomponent SM package with structured unsupervised exercise had a larger improvement in SGRQ at 3–6 months’ follow-up than UC [3.59 points (95% CI 1.28 to 5.91 points; I² = 0%)]. However, by 1-year follow-up there was no evidence of effect: SGRQ 0.80 points (95% CI –1.03 to 2.63 points; prediction interval –2.39 to 3.99). Only one trial283 reported the CRQ at 8 weeks’ follow-up (0.61, 95% CI –0.18 to 1.41 points) (Figures 47 and 48, and Table 56).

FIGURE 47.

Health-related quality-of-life (SGRQ) outcomes for multicomponent SM interventions with structured, unsupervised exercise vs. UC/control.

FIGURE 48.

Health-related quality-of-life (CRQ) outcomes for multicomponent SM interventions with structured, unsupervised exercise vs. UC/control.

TABLE 56 - Outcomes for multicomponent interventions with structured, unsupervised exercise

n/a, not applicable.

Outcome	Time frame	No. of studies	Summary MD (95% CI)	I2 (%)
SGRQ	> 3 to ≤ 6 months	4	3.59 (1.28 to 5.91)	0.0
	> 6 months	5	0.80 (–1.03 to 2.63)	2.3
CRQ	≤ 3 months	1	0.61 (–0.18 to 1.41)	n/a
			Summary HR (95% CI)
Admissions	> 6 months	1	0.55 (0.35 to 0.87)	n/a
Exacerbations	≤ 3 months	1	3.01 (0.31 to 28.96)	n/a

Of the four trials182,255,262,270 that reported hospital admissions at last follow-up, two182,270 reported a statistically significant reduction in the intervention group. A HR could be calculated for only one trial,270 which reported a significant reduction in hospital admissions (HR 0.55, 95% CI 0.35 to 0.87) (Figure 49). 270 Moore et al. ,284 who had only 27 participants (and wide CIs), reported exacerbations (HR 3.01, 95% CI 0.31 to 28.96) (see Figure 50). An additional two trials192,251 reported no significant difference in exacerbations at last follow-up point.

FIGURE 49.

Admissions outcomes for multicomponent SM interventions with structured, unsupervised exercise vs. UC/control.

FIGURE 50.

Exacerbation outcomes for multicomponent SM interventions with structured, unsupervised exercise vs. UC/control.

Multicomponent interventions with exercise counselling only compared with usual care

Seven trials67,72,73,140,170,180,266 reported disease-specific HRQoL using total SGRQ or CRQ following multicomponent interventions that included advice about increasing exercise. There were no significant effects on the SGRQ in the combined analyses at any of the three follow-up points and heterogeneity was high (Figure 51 and Table 57). Two trials72,180 had large differences in the SGRQ score at baseline that were not accounted for, and three trials72,73,140 had low or imbalanced follow-up rates.

FIGURE 51.

Health-related quality-of-life (SGRQ) outcomes for multicomponent SM interventions with exercise counselling only vs. UC/control. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

TABLE 57 - Outcomes for multicomponent interventions with exercise counselling only compared with UC

n/a, not applicable.

Outcome	Time frame	No. of studies	Summary MD (95% CI)	I2 (%)
SGRQ	≤ 3 months	1	1.32 (–2.97 to 5.61)	n/a
	> 3 to ≤ 6 months	3	1.87 (–4.43 to 8.18)	71.2
	> 6 months	4	3.88 (–1.39 to 9.14)	74.6
			Summary HR (95% CI)
Admissions	≤ 3 months	2	1.40 (0.93 to 2.11)	0
	> 3 to ≤ 6 months	3	0.52 (0.13 to 2.09)	81.0
	> 6 months	3	0.79 (0.50 to 1.26)	67.9
Exacerbations	> 3 to ≤ 6 months	1	1.0 (0.17 to 5.98)	n/a

Eight trials reported hospital admissions,67,70,71,73,75,140,168,266 of which one168 was not included in the meta-analyses. There were no significant effects on admissions at any of the three follow-up points, and heterogeneity was high at the 6- and 12-month follow-up points. Details are given in Figure 52 and Table 57.

FIGURE 52.

Admission outcomes for multicomponent SM interventions with exercise counselling only vs. UC/control.

Only Dheda et al. 73 reported exacerbation rates at 6 months’ follow-up in a form enabling a HR to be calculated, with no difference between study arms (Figure 53). Three other trials140,180,266 reported exacerbations at 6 months265 or a year,140,180 with Rice et al. 140 reporting a significant reduction in exacerbations at 1 year.

FIGURE 53.

Exacerbation outcomes for multicomponent SM interventions with exercise counselling only vs. UC/control.

Multicomponent interventions without an exercise element compared with usual care

We included five trials63,112,120,185,256 of multicomponent interventions that did not include exercise or even advice about exercise as a component in the meta-analyses.

Although the estimates of average effects of QoL (SGRQ) at all three follow-up points were in the direction favouring the intervention, none was statistically significant and heterogeneity was high (Figure 54 and Table 58).

FIGURE 54.

Health-related quality-of-life (SGRQ) outcomes for multicomponent SM interventions without an exercise element vs. UC/control. a, Indicates that the control group that has been halved in size (split between two comparisons). Coultas 2005112 A = nurse-assisted collaborative management vs. UC; Coultas 2005112 B = nurse-assisted medical management vs. UC.

TABLE 58 - Health-related quality of life and admission outcomes for multicomponent interventions without exercise advice or support

n/a, not applicable.

Outcome	Time frame	No. of studies	Summary MD (95% CI)	I2 (%)
SGRQ	≤ 3 months	3	4.65 (–1.45 to 10.74)	82.0
	> 3 to ≤ 6 months	2	2.75 (–3.24 to 8.74)	0
	> 6 months	3	3.73 (–0.99 to 8.44)	81.1
			Summary HR (95% CI)
Admissions	≤ 3 months	1	0.32 (0.03 to 3.03)	n/a
	> 6 months	2	0.99 (0.76 to 1.30)	0

Results were combined for two trials that reported hospital admission rates at 3–6 months. 63,142 Neither reported a statistically significant reduction in admissions (Figure 55). Three other trials112,136,142 that reported hospital admissions at last follow-up did not find any significant difference between study groups. Koff et al. 120 reported no significant difference in exacerbations at 3 months’ follow-up.

FIGURE 55.

Admissions outcomes for multicomponent SM interventions without an exercise element vs. UC/control.

Summary: role of exercise in multicomponent interventions

• Multicomponent interventions with supervised exercise compared with UC have a positive effect on HRQoL and positive trend for hospital admissions but not reaching statistical significance.

• For multicomponent interventions with structured, unsupervised exercise there is evidence of effectiveness on HRQoL in the medium term, but insufficient evidence for hospital admissions and exacerbations.

• For interventions with advice to increase exercise in an unstructured manner there were few trials, but no evidence of overall effect on HRQoL, hospital admissions or exacerbations.

• There were limited numbers of studies of multicomponent interventions without any exercise counselling. There is some evidence that they may lead to short-term improvements in HRQoL, but there was inconclusive evidence for hospital admissions.

Self-management interventions including an exercise component consisting of aerobic and strength training

Given the possible influence of exercise, we further explored the effect of different types of exercise interventions by first examining the effects of interventions that include both aerobic and strength training.

The summary meta-analysis result indicates that, on average, the combined aerobic/strength exercise arm has a SGRQ score of 7.80 points higher than the UC arm (95% CI 2.82 to 12.79 points). However, this is the average of the distribution of intervention effects and this distribution was wide as a result of high heterogeneity (I² = 81.5%). The prediction interval, in which 95% of the distribution of the effects occur, is –10.60 to 26.21, which is mainly in favour of the intervention, but also indicates that the intervention is not always effective. At the mid follow-up point of 3–6 months, the estimate of the average effect was 3.76 points on the SGRQ (95% CI 2.13 to 5.39 points; I² = 0%) favouring the intervention group. However, at > 6 months there was no evidence of effect (Figure 56 and Table 59). The CRQ results favoured the intervention group up to 6 months’ follow-up (Figure 57).

FIGURE 56.

Health-related quality-of-life (SGRQ) outcomes for combined strength and aerobic exercise interventions with SM components vs. UC. a, Indicates that several papers are represented by this lead publication. Details are given in Appendix 22.

FIGURE 57.

Health-related quality-of-life (CRQ) outcomes for combined strength and aerobic exercise interventions with SM components vs. UC.