Types of studies

We considered any study meeting the study design criteria from the Cochrane Effective Practice and Organisation of Care Group (EPOC 2017a): randomised trials; non‐randomised trials; controlled before‐and‐after studies with contemporaneous data collection and with two or more control and intervention sites; repeated measures studies and interrupted time series (ITS) studies with a clearly‐defined point in time when the intervention occurred, and at least three data points before and after implementation of the intervention.

We included studies that were available as a full text, regardless of publication status or language. We included abstracts and associated these with full‐text studies. For abstracts of studies that were potentially eligible, we searched for a corresponding publication and also contacted study authors to obtain any unpublished full study report, if available.

Types of participants

We considered study interventions that targeted any type of health and social care team ‐ uniprofessional, multiprofessional or interprofessional ‐ from different professional areas (e.g. chiropodists or podiatrists, complementary therapists, dentists, dieticians, doctors or physicians, hygienists, midwives, nurses, occupational therapists, pharmacists, physiotherapists, psychologists, psychotherapists, radiographers, social workers or speech therapists), with or without the inclusion of patients. We did not exclude studies on the basis of demographic factors or healthcare setting.

Types of interventions

We considered all collective leadership interventions that:

• had an explicit aim to improve professional practice and healthcare outcomes;

• included a focus on sharing decisions in an interactive and interdependent process;

• fostered role clarification and sharing of responsibilities; and

• aimed to enable staff empowerment and motivation.

Interventions could include workshops, coaching (e.g. Caspar 2017), simulation, protocols (e.g. Armstrong 2020), and role‐playing (e.g. Shirazi 2016). 

We were interested in professional‐directed collective leadership interventions, which may also be referred to as collectivistic leadership, distributed leadership, shared leadership, collaborative leadership, participatory leadership, inclusive leadership, democratic leadership, plural leadership, dispersed leadership, empowering leadership, compassionate leadership, informal leadership, peer leadership and team leadership. We were concerned with collective leadership interventions with a focus on patient case management, professional management or organisational strategies.

For this review, we planned the following comparisons in primary, secondary or tertiary healthcare settings:

• collective leadership versus usual approaches to leadership identified as centralised and hierarchical;

• collective leadership educational interventions with or without organisational restructuring;

• collective leadership interventions targeting frontline teams versus management teams.

We excluded interventions that focused exclusively on strengthening professional collaboration without explicit reference to collective leadership; this allowed us to complement and not overlap with the Cochrane Review on interprofessional collaboration by Reeves 2017.

Types of outcome measures

Electronic searches

The review authors developed the database search strategy in consultation with the EPOC Information Specialist. We searched the Cochrane Database of Systematic Reviews (CDSR) and the Database of Abstracts of Reviews of Effects (DARE) for related systematic reviews.

We searched the following databases for primary studies on 5 January 2021:

• Cochrane Central Register of Controlled Trials (CENTRAL; 2021, issue 1), in the Cochrane Library;

• MEDLINE Ovid (1946 to date of search);

• Embase Ovid (1974 to date of search);

• CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature; 1982 to date of search);

• LILACS (Latin American and Caribbean Health Sciences Information database; 1982 to date of search); and

• Web of Science, Conference Proceedings Citation Index ‐ Science (1990 to date of search).

Search strategies were comprised of natural language and controlled vocabulary terms. We applied no limits on language or publication date. In databases where it was possible and appropriate, we used study design filters to limit to the study designs of interest. For randomised trials in MEDLINE, we used a modified version of the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐ and precision‐maximising version (2008 revision) (Lefebvre 2021), with additional terms for other relevant study designs. The MEDLINE strategy was peer reviewed by a second Cochrane Information Specialist. We used limits in CINAHL and Embase to remove MEDLINE records in order to avoid duplication in downloaded results. We de‐duplicated the remaining results in Endnote against each other. Please see all search strategies used in Appendix 1.

Searching other resources

Selection of studies

We downloaded all titles and abstracts retrieved by electronic searching to a reference management database and removed duplicates. Two review author pairs (JAMS with MP, JAMS with VAM) independently screened titles and abstracts for inclusion. We retrieved the full‐text study reports/publications and two review author pairs (JAMS with MP, JAMS with VAM) independently applied the eligibility criteria to the full texts, identified studies for inclusion, and identified and recorded reasons for exclusion of ineligible studies. We resolved any disagreement through discussion or, when required, through consulting a third review author (AX or RH).

We listed studies that initially appeared to meet the inclusion criteria but that we later excluded in the 'Characteristics of excluded studies' table, with the reasons for their exclusion. We collated multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review. We also provided any information we could obtain about ongoing studies. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (Liberati 2009).

Data extraction and management

We used the EPOC standard data collection form and adapted it for study characteristics and outcome data (EPOC 2017b); we piloted the form on at least one study in the review. Two review author pairs (JAMS with MP, JAMS with VAM) independently extracted the following study characteristics from the included studies and enter the data into RevMan Web (RevMan Web 2022).

• Methods: study design, number of study centres and location, study setting, withdrawals, date of study, follow‐up.

• Participants: number, mean age, age range, gender, eligibility criteria, other relevant characteristics.

• Interventions: intervention components, comparison, fidelity assessment.

• Outcomes: main and other outcomes specified and collected, time points reported.

• Notes: funding for trial, notable conflicts of interest of trial authors, ethical approval.

Two review author pairs (JAMS with MP, JAMS with VAM) independently extracted outcome data from included studies. We noted in the Characteristics of included studies table if a trial reported outcome data in an unusable way. We resolved disagreements by consensus or by involving a third review author (AX or RH).

Assessment of risk of bias in included studies

Two review author pairs (JAMS with MP, JAMS with VAM, or JAMS with HFA) independently assessed the risk of bias for each study, using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions Section 7.1.2 and Chapter 8 (Higgins 2021a), and the guidance from the EPOC Group (EPOC 2017c). We resolved any disagreements through discussion or by involving a third review author (AX or RH). We assessed the risk of bias in randomised trials 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;

• baseline outcomes measurement;

• baseline characteristics;

• other bias

To analyse the risk of bias in non‐randomised studies, we had planned to use the 'Risk of Bias in Non‐randomized Studies of Interventions' (ROBINS‐I) tool (Sterne 2016). For interrupted time series studies, we had planned to assess the risk of bias according to the following domains (EPOC 2017c):

• independence from other changes;

• specification of the shape of effects before analysis;

• independence from data collection;

• knowledge of intervention allocation;

• incomplete outcome data;

• selective outcome reporting;

• other bias.

We judged each potential source of bias to be high, low or unclear, and where appropriate, provided a quote from the study report 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 assigned an overall risk of bias assessment (high, moderate or low) to each of the included studies, using the approach suggested in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a). We considered studies with a low risk of bias for all key domains, or where it seemed unlikely that bias would have seriously altered the results, to have a low risk of bias. We considered studies where risk of bias in at least one domain was unclear, or where we judged it to have some bias that could plausibly raise doubts about the conclusions, to have an unclear risk of bias. We considered studies to have a high risk of bias if they had a high risk of bias in at least one domain, or if we judged them to have serious bias that decreased the certainty of the conclusions.

We considered blinding separately for different key outcomes where necessary. Where information on risk of bias related to unpublished data or correspondence with a trialist, we noted this in the risk of bias table. We did not exclude studies on the grounds of their risk of bias, but clearly reported the risk of bias when presenting the results of the studies. When considering specific effects, we took into account the risk of bias for the studies that contributed to that outcome.

We conducted the review according to our published protocol and report any deviations from it in the 'Differences between protocol and review' section.

Measures of treatment effect

Where possible, we estimated the effect of the intervention using the risk ratio (RR) and risk difference (RD) for dichotomous data, with the appropriate associated 95% confidence interval (CI); and mean difference (MD) or standardised mean difference (SMD) for continuous data, with their 95% CIs (Higgins 2021a). We ensured that an increase in scores for continuous outcomes was interpreted in the same way for each outcome and explained the direction to the reader. For controlled before‐and‐after, ITS and repeated measures studies, we planned to report measures before and after the intervention, as well as the difference of the periods for specific time points.

Unit of analysis issues

We did not have issues with cluster‐randomised trials that did not account adequately for the effects of the clustering on the effect estimate, so adjustments of the analyses to avoid unit of analysis errors was not required (EPOC 2017d).

Dealing with missing data

We attempted to contact the study authors, asking them to provide missing outcome data. Where this was not possible, we concluded that the missing data introduced serious bias impacting on the overall assessment of results (Sterne 2011).

Assessment of heterogeneity

We planned to assess clinical and methodological heterogeneity by considering study design, participants and how the collective leadership intervention was applied. We were only able to identify one outcome (leadership) for which we judged measures, participants, intervention approach and study designs of the included studies were sufficiently similar to conduct a meta‐analysis and assess statistical heterogeneity (Borenstein 2009). For this outcome, we assessed statistical heterogeneity using the Chi² statistic and related P value, alongside the I² statistic with associated percentage values (Higgins 2021a).

Assessment of reporting biases

We were not able to pool more than 10 trials so did not create and examine a funnel plot to explore possible publication biases (Sterne 2011).

Data synthesis

We could only undertake a meta‐analysis for one outcome, for which pooling made sense. For the remaining outcomes, we provided a narrative description of the results. For the meta‐analysis, we combined continuous data from leadership scales that were sufficiently similar (and where direction of effect was the same) using generic inverse variance and standardised mean difference to account for differences in the scales (Higgins 2021a). We used a random‐effects model for the meta‐analysis to account for possible differences among studies in which conditions of the healthcare setting and approach to collective leadership may have varied. We completed the meta‐analysis using the Review Manager (RevMan Web) calculator.

In order to synthesise the remaining outcomes without meta‐analyses, we considered the nine items suggested by Campbell 2020 to guide our reporting:

• group studies for synthesis;

• describe the standardised metric and transformation method used;

• describe the synthesis methods;

• describe the criteria used to prioritise results for summary and synthesis;

• investigate heterogeneity in reported effects;

• assess the certainty of the evidence;

• describe the data presentation methods;

• report the results; and

• report the limitations of the synthesis.

We noted the scales used to measure continuous outcomes, including the range of possible scores and direction of the effect, alongside the mean difference and 95% confidence interval, to enable more meaningful interpretation of results. In the summary of findings table, we present the standardised mean difference, calculated using the RevMan Web calculator, to allow easier comparison of the effect size of collective leadership on different outcomes. For one dichotomous outcome (death), we reported the relative risk next to the 95% confidence interval.

Subgroup analysis and investigation of heterogeneity

We  included all kinds of health professionals involved across all types of clinical settings, but noted length of the intervention, implementation approach, use of explicit intervention protocols, organisational support and frequency of meetings. We planned for but were unable to carry out any of the following subgroup analyses:

• collective leadership hierarchical interventions subgrouped by setting (primary care, secondary care or tertiary care);

• collective leadership non‐hierarchical interventions subgrouped by setting (primary care, secondary care or tertiary care);

• subgroup of collective leadership hierarchical interventions in high‐income versus low‐income setting;

• subgroup of collective leadership non‐hierarchical interventions in high‐income versus low‐income setting.

To conduct the subgroup analysis, we planned to use all outcomes that were common to at least three studies, performing a separate analysis for each outcome. However, this was not possible with the included studies.

Sensitivity analysis

We planned to perform prespecified sensitivity analyses to assess the robustness of our conclusions and explore their impact on effect sizes. This would have involved the following:

• restricting the analysis to published studies;

• restricting the analysis to studies with a low risk of bias; and

• imputing missing data.

Our included studies and available data did not allow for sensitivity analysis to take place.

Summary of findings and assessment of the certainty of the evidence

Two review authors (JAMS and HFA) independently assessed the certainty of the evidence (high, moderate, low and very low) using the five GRADE considerations (risk of bias, consistency of effect, imprecision, indirectness and publication bias) (Guyatt 2008). We used methods and recommendations described in Chapter 8 and 15 of the Cochrane Handbook for Systematic Reviews of interventions (Higgins 2021a), and the EPOC worksheets (EPOC 2017e), and we used GRADEpro software (GRADEpro GDT). We resolved disagreements on certainty ratings through discussion, provided justification for decisions to downgrade or upgrade the ratings using footnotes in the table, and made comments to aid readers' understanding of the review where necessary. We used plain language statements to report these findings in the review (EPOC 2017e).

We summarised our results in 'summary of findings Table 1' for the main intervention against each comparison described in the Types of interventions. We included the most important outcomes in order to draw conclusions about the certainty of the evidence within the text of the review.

• Professional practice

  • Leadership

  • Team performance

  • Clinical performance

• Healthcare outcomes

• • Health status (inpatient mortality, mortality within 30 days of discharge)

• Staff well‐being

• • Burnout, psychological symptoms (anxiety, depression) and work‐related stress

• Unintended consequences

• • Any unintended safety of care events (including errors, healthcare‐associated complications) or professional, organisational and staff consequences

We synthesised data on prespecified outcomes of collective leadership along with any other unintended consequences found in the included studies. We used the same eligibility criteria to assess intended (beneficial) and unintended (adverse) effects, in terms of types of studies, types of participants and types of interventions (Peryer 2019).

We considered whether there was any additional outcome information that we could not incorporate into the meta‐analysis to note in the 'Comments' section, and stated if it supported or contradicted the information from the meta‐analysis.