Primary outcomes

• Quality of life (e.g. St. George's Respiratory Questionnaire (SGRQ), COPD assessment test (CAT))

• Exacerbations (as defined by trialists, depending on the data available, we extracted either number of participants experiencing one or more exacerbation, or the exacerbation rate, or both)

Secondary outcomes

• Functional status (6‐minute walk distance (6MWD)/incremental shuttle walk test (ISWT))

• All‐cause hospital admissions (also as a proxy for use of services, e.g. reduction of use)

• Respiratory hospital admissions

• Mortality (all causes)

• Pain (as reported in trials)

• Anxiety symptoms (measured by e.g. Hospital Anxiety and Depression Scale (HADS))

• Depression symptoms (measured by e.g. HADS)

For the planned qualitative synthesis, our outcomes were themes arising from the data.

We included studies regardless of whether they report our predefined outcomes.

There was no minimum duration for the interventions, and we used data reported for the last follow‐up point.

Electronic searches

We searched for studies in June 2019, February 2020, and January 2021 in the following databases and trials registries:

• Cochrane Airways Register of Trials through the CRS, from inception onwards;

• Cochrane Central Register of Controlled Trials (CENTRAL, in the Cochrane Library) through the Cochrane Register of Studies (CRS), from inception onwards;

• MEDLINE Ovid SP from 1946 onwards;

• Embase Ovid from 1974 onwards;

• PsycINFO Ovid Sp from 1974 onwards;

• CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature) from 1937 onwards;

• Web of Science Core Collection from 1970 onwards;

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov;

• World Health Organization International Clinical Trials Registry Platform (ICTRP).

Searches for qualitative and quantitative studies were run separately using appropriate study design filters. We combined search terms for the target population with the Cochrane Highly Sensitive Search Strategy to identify reports of RCTs (Lefebvre 2021), and terms from the search strategies developed and tested by DeJean 2016 to find reports of qualitative studies. The search was developed in MEDLINE by ES, with input from the other authors, and was peer‐reviewed by another Information Specialist using the PRESS checklist (McGowan 2016). The MEDLINE search strategy was adapted appropriately for use in the other databases. We did not apply any language limits, and we did not limit the search strategy by population characteristics such as age, gender, or race. Details of the database search strategies, search dates, and the number of results retrieved are presented in Appendix 1.

We initially searched all databases from their inception to June 2019. We updated the searches in February 2020 and January 2021 in a reduced number of databases (Appendix 2) following an analysis of the individual database yield of eligible study references.

Selection of studies

We retrieved many search results, and we therefore used Cochrane's 'Screen4Me' workflow to help assess the results of our search for RCTs. Screen4Me comprises three components: known assessments — a service that matches records in the search results to records that have already been screened in Cochrane Crowd and have been labelled as 'RCT' or as 'Not an RCT'; the RCT classifier — a machine‐learning model that distinguishes RCTs from non‐RCTs (Marshall 2018); and if appropriate, Cochrane Crowd — Cochrane's citizen science platform where 'the crowd' help to identify and describe health evidence (Noel‐Storr 2020).

Following use of the Screen4Me workflow, two of three review authors (SJ, ES and ED) screened each title and abstract of the remaining search results independently and coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We retrieved the full‐text study reports of all potentially eligible studies and two of three review authors (SJ, ES and ED) independently screened each full text for inclusion, and recorded the reasons for exclusion of ineligible studies.

We resolved any disagreement through discussion or, if required, we consulted a third review author (SH). We identified and excluded duplicates and collated multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram and Characteristics of excluded studies table (Moher 2009).

Data extraction and management

Three authors (SJ, ED, ES) screened citations in Covidence. One review author (SJ) piloted the data extraction form on at least one quantitative and one qualitative study before we extracted data from the rest of the included studies. We extracted data into a Microsoft Excel spreadsheet.

Quantitative studies

Three review authors (SJ, ES and ED) assessed risks of bias independently for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). We resolved any disagreements by discussion or by involving another author (SH). We assessed the risk of bias according to the following domains.

• Random sequence generation.

• Allocation concealment.

• Blinding of participants and personnel.

• Blinding of outcome assessment.

• Incomplete outcome data.

• Selective outcome reporting.

• Other potential bias.

We judged each potential source of bias as high, low or unclear and provided a quote from the study report together with a justification for our judgement in the risk of bias table. We summarised the risk of bias judgements across different studies for each of the domains listed. We considered blinding separately for different key outcomes where necessary (e.g. for unblinded outcome assessment, risk of bias for all‐cause mortality may be very different from a patient‐reported quality‐of‐life scale). We planned to note where information on risk of bias was from unpublished data or correspondence with a trialist, but we decided not to contact trial authors for clarification or unpublished data. When considering treatment effects, we took into account the risk of bias for the studies that contribute to that outcome via GRADE assessment of the certainty of the evidence (see below).

We presented a risk of bias table for all studies.

Qualitative studies

Two review authors (SJ, SH) assessed risk of bias independently for each study using the criteria outlined by the Cochrane Quality and Intervention Methods Group (Hannes 2011). We resolved any disagreements by discussion or by involving another review author (AH). We assessed the risk of bias according to the following criteria.

• Quality of reporting (explicitness in reporting of all study aspects).

• Methodological rigour (validity and reliability of study design and process).

• Conceptual depth and breadth (are reported aims, rationale or theory reflected in the study design, process and findings?).

We used the Critical Appraisal Skills Programme (CASP) checklist (Critical Appraisal Skills Programme 2018) to assess risk of bias. We present risk of bias of studies in a table.

We assessed the risk of bias after the identification of relevant data to help us make judgements about the relative strength of messages in the included research.

Mixed‐methods studies

We planned to use the Mixed Methods Appraisal Tool (MMAT) to assess risk of bias (Hong 2018; Pluye 2011). Two review authors (SJ and ED) would have assessed risk of bias independently for each study using the criteria outlined by the MMAT. We planned to resolve any disagreements by discussion or by involving another author (SH). We planned to assess the risk of bias according to the following criteria.

• Is there an adequate rationale for using a mixed‐methods design to address the research question?

• Are the different components of the study effectively integrated to answer the research question?

• Are the outputs of the integration of qualitative and quantitative components adequately addressed?

• Are divergences and inconsistencies between quantitative and qualitative results adequately addressed?

• Do the different components of the study adhere to the quality criteria of each tradition of the methods involved?

Quantitative treatment effects

We analysed dichotomous data as an odds ratios (OR) and continuous data as the mean difference (MD) or standardised mean difference (SMD). Where data from rating scales were combined in a meta‐analysis, we ensured they were entered with a consistent direction of effect (e.g. lower scores always indicate improvement). We undertook meta‐analyses only where this was meaningful: that is, if the treatments, participants and the underlying clinical question were similar enough for pooling to make sense. Where we encountered substantial statistical or clinical heterogeneity, we presented data in graphs, but did not pool them. We described skewed data narratively (for example, as medians and interquartile ranges for each group).

Where multiple trial arms were reported in a single study, we included only the relevant arms for either comparison. Where two comparisons (e.g. intervention A versus intervention B) were combined in the same meta‐analysis, we planned to either combine the active arms or halve the control group to avoid double‐counting, but this was not necessary. If adjusted analyses were available (ANOVA or ANCOVA), we used these as a preference in our meta‐analyses. If both change from baseline and endpoint scores were available for continuous data, we used change from baseline unless there was low correlation between measurements in individuals. Where studies reported outcomes at multiple time points, we used endpoint data. We used intention‐to‐treat (ITT) or 'full analysis set' analyses where they were reported (i.e. those where data have been imputed for participants who were randomly assigned but did not complete the study) instead of completer or per protocol analyses.

Quantitative analysis

For dichotomous outcomes, we used participants, rather than events, as the unit of analysis. Where rate ratios were reported in a study, we planned to analyse them on this basis. We planned to meta‐analyse data from cluster‐RCTs where data had been adjusted (or could be adjusted by us), to account for the clustering.

Quantitative data

We used the five GRADE considerations (risk of bias, consistency of effect, imprecision, indirectness, and publication bias) to assess the quality of the body of evidence as it relates to the following outcomes: Quality of life (SGRQ and CAT), exacerbations, functional status (6MWD), all‐cause hospital admissions, all‐cause mortality, anxiety and depression (unfortunately we did not specify these a priori). We used the methods and recommendations described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021), using GRADEpro GDT software (GRADEpro GDT; Guyatt 2008). We justified all decisions to downgrade the quality of studies in the footnotes of the table, and made comments to aid the reader's understanding of the results where necessary. We presented GRADE findings in a summary of findings table.

Qualitative data

We used the GRADE Confidence in the Evidence from Reviews of Qualitative Research (CERQual) approach to assess our confidence in the evidence of effectiveness arising from studies evaluating interventions (Lewin 2015). One review author (SJ) did this independently and Jane Noyes (from the Cochrane Qualitative and Implementation methods group) checked the completed assessment separately. This assessment allowed us to make judgements on the following four domains.

• Methodological limitations of included studies.

• Relevance of contributing studies to the research question.

• Coherence of study findings.

• Adequacy of data supporting the study findings.

We used the methods and recommendations described in Chapter 21 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021). We summarised findings of the four domains for each outcome in a CERQual Qualitative Evidence Profile (Lewin 2015). We rated the overall assessment of confidence of evidence as high, moderate, low or very low. We presented CERQual findings in a summary of findings table (Glenton 2020). We justified all decisions to downgrade the quality of studies in the footnotes of the table, and we made comments to aid the reader's understanding of the review where necessary.

Quantitative data

We used the I² statistic to measure heterogeneity among the studies in each analysis where possible. We conducted a meta‐analysis using a random‐effects model, as the interventions were varied. We explored possible causes of heterogeneity.

We considered the following I² ranges in the analyses (Deeks 2021).

• 0% to 40%: might not be important

• 30% to 60%: may represent moderate heterogeneity

• 50% to 90%: may represent substantial heterogeneity

• 75% to 100%: considerable heterogeneity

We checked all data where we encountered substantial statistical heterogeneity. We did not pool data where there was substantial heterogeneity.

Quantitative data

If we had been able to pool more than 10 studies, we planned to create and examine a funnel plot to explore possible small‐study and publication biases.

Quantitative data

We used RevMan 5 and RevManWeb to perform quantitative data syntheses (meta‐analyses) (Review Manager 2020); and where data for population or interventions were similar, we pooled such data. Where we felt it was not appropriate to pool data, we present the data in forest plots to show the range of effect sizes where possible. Where we found clinical heterogeneity within the studies we identified, we grouped studies according to interventions and outcomes, and used the random‐effects model in the analyses (Ioannidis 2008). Where it was not possible to perform a meta‐analysis, we considered presenting data graphically and narratively using recommendations in the latest version of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021).

Qualitative data

Where studies were similar in design we planned to synthesise their data using a thematic analysis (Thomas 2008). This method would have allowed us to identify important or recurrent themes that arise from studies and summarise them under thematic headings. We planned to tabulate information, allowing identification of prominent themes and offer structured ways of dealing with the data in each theme. This type of synthesis would have helped us to identify emerging themes that described the experiences of participants, carers and HCPs when receiving or providing care to manage comorbid COPD.

We planned to initially analyse carer and HCP data separately to identify, for example, conflicting views or experiences. If we had found that the views and experiences had been similar, we planned to combine the two subgroups in subsequent syntheses.

Combining quantitative and qualitative data

We planned to use the methods and recommendations described in the Cochrane Handbook for Systematic Reviews of Interventions, and methods outlined by the Cochrane Qualitative and Implementation Methods Group (Higgins 2021 and Harden 2018, respectively). There are two main approaches to integrating qualitative and quantitative data: sequential and convergent. The sequential approach involves synthesising qualitative and quantitative analyses separately, followed by using common frameworks to integrate the findings across syntheses. We planned to use the sequential approach to integrate qualitative data to explain quantitative findings. We planned to analyse quantitative data at the first stage, followed by synthesis of qualitative data in the second stage.

Integrating the qualitative syntheses with the quantitative analyses can be achieved by using a 'matrix' to compare and contrast findings across both types of evidence. The matrix will help to identify gaps in the evidence. This approach can help us to understand why heterogeneity exists that we may find in the quantitative analyses. Other approaches include the development of a 'logic model' by providing a common framework within which both quantitative and qualitative evidence can contribute (Harden 2018). We acknowledge that the methods for integration are dependent on the quantity of studies and extracted evidence available, and quality of description within included studies (e.g. intervention content, context, and study findings). For divergent data (qualitative data that does not match the quantitative data), we aimed to resolve divergence (deviation of the qualitative data from the quantitative data, or vice versa (Erzberger 1997)) where possible by providing a narrative explanation according to research and knowledge of the topic area.

Review author reflexivity

We maintained a reflexive stance throughout the stages of the review process, from study selection to data syntheses. Progress was discussed regularly among the team and decisions made were discussed critically. As a review author team, our expertise has been listed in Contributions of authors. Based on our collective and individual experiences as clinicians, academics and researchers, we anticipated that the findings of our review would identify a combination of organisational, professional and individual factors influencing the implementation of targeted interventions and approaches to care for people with COPD to manage comorbidities. ED has overseen progress of the review, a process that has allowed her to document and reflect on any decisions made.