Types of studies

We included randomised trials, non‐randomised trials, controlled before‐after studies and interrupted time series studies in any language. Controlled before‐after studies required two intervention and two control sites to be included. Interrupted time series studies needed a clearly defined point in time when the intervention occurred and three measuring points before and after the intervention. We did not expect to identify a broad range of evidence based on randomised trials alone, and therefore chose to include additional study designs that allowed us to explore causal relationships. We excluded reviews from the analysis, but screened their reference lists for possibly relevant articles, because we wanted to include primary studies only in this review.

Types of participants

We included all trials, regardless of the setting, which recruited adult patients (aged 18 years or older), for whom an ethical issue arose and who subsequently received an ethical case intervention. If the study included mixed populations, consisting of adults and children, we only included the study if subgroup or sensitivity analyses (or both) were provided for the adult population.

Types of interventions

We included ethical case interventions of any form (e.g. moral case deliberation, ethics round, ethics committee). At least one person involved must have had expertise in medical ethics. We compared the intervention with ‘usual care’ (e.g. no specific interventions defined) or another active control. We excluded studies which focused solely on the implementation of ethical case interventions, on the topic of research ethics (e.g. research ethics committees) or policy‐making.

Types of outcome measures

Ethical case interventions are complex interventions. Therefore, we took several outcomes into account to measure their effectiveness. We extracted any additional outcomes identified in the studies.

Electronic searches

We searched the following databases on 26 September 2018.

1. Cochrane Central Register of Controlled Trials (CENTRAL; 2018, Issue 9) in the Cochrane Library.
2. MEDLINE Ovid (including Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations and Versions) (1946 onwards).
3. Embase Ovid (1974 onwards).
4. CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature) (1980 onwards).
5. PsycINFO Ovid (1806 onwards).

We conducted scoping searches on the ethics‐specific databases Ethxweb and Ethicsweb. We decided not to conduct full searches on these databases due to the lack of study designs of interest and overlap with MEDLINE. Search strategies were composed of keywords and controlled vocabulary terms. We applied no language limits. See Appendix 1 for the detailed search strategies. We documented the search strategies and processes in a standardised search log (EPOC 2017a).

Searching other resources

We also:

1. searched the reference lists of included studies and any systematic reviews identified through the searches for possibly relevant articles;
2. handsearched the journal Ethik in der Medizin for relevant articles;
3. consulted with experts. We contacted the first author of each included study, as well members of the European Clinical Ethics Network (ECEN) for further retrieval of relevant studies;
4. searched for ongoing studies on ClinicalTrials.gov (www.clinicaltrials.gov) and the International Clinical Trials Registry Platform Search Portal (ICTRP) (www.who.int/trialsearch);
5. conducted cited reference searches for all included studies in February 2018 in ISI WEB of Science (Science Citation Index).

Selection of studies

Two review authors (JH, JS) involved in the selection and data extraction process were trained in systematic reviews by an experienced researcher (SN) and supervised by two experts (CB and MG). We imported the results of the electronic searches to Covidence (Covidence), removing duplicate records. Two review authors (SN, JH) independently screened titles and abstracts of the retrieved records for inclusion, using a study selection form in Covidence. We coded all potentially eligible studies as ‘retrieve’ (eligible or potentially eligible/unclear) or ‘do not retrieve’. We retrieved the full‐text articles and two review authors (SN, JH) independently rated each as ‘irrelevant’, ‘relevant’ or ‘maybe’. We used a ‘maybe’ for studies which provided insufficient information to make a conclusive decision about study eligibility. In these instances, we contacted the study authors for clarification, resolving any disagreement through consultation with a third review author (JS). We listed the excluded studies with reasons for exclusion in the Characteristics of excluded studies table. We collated multiple reports for the same study, so that each study rather than each report was the unit of interest. We reported the selection process in sufficient detail to complete a PRISMA flow diagram (Liberati 2009).

Data extraction and management

We used a modified data collection form for study characteristics and outcome data, which we piloted on two studies in the review (EPOC 2013). Two review authors (SN, JH) independently extracted the following study characteristics and data from each of the included studies.

1. Methods: study design, number of study centres and location, study setting, withdrawals, date of study and follow‐up.
2. Participants: number, mean age, age range, gender, severity of condition, diagnostic criteria, inclusion criteria, exclusion criteria and other relevant characteristics.
3. Interventions: intervention components, comparison, fidelity assessment and conceptual framework/theory underlying the intervention.
4. Outcomes: main and other outcomes specified and collected, time points reported and outcome measures.
5. Notes: funding for trial, notable conflicts of interest of trial authors and ethical approval.
6. Outcome reporting: outcomes reported in the study register (Appendix 2) or trial protocol.

Two review authors (SN, JH) independently extracted outcome data from included studies. We noted in the Characteristics of included studies table whether outcome data were reported in an unusable way. We sought the relevant missing information on the trial from the corresponding author of the article, if required. We resolved disagreements by discussion in the review team until we reached consensus.

We used the Template for Intervention Description and Replication (TiDieR) to assess whether the intervention was sufficiently described (Hoffmann 2014). We used the ‘modified ORBIT study classification table’ (Kirkham 2010) (Appendix 3) to assess selective reporting.

Assessment of risk of bias in included studies

Two review authors (SN, JH) independently assessed the methodological quality of the included studies, using the ‘Risk of bias’ criteria and qualifiers outlined in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011) and the EPOC guidance (EPOC 2015) to judge whether a study had a low, high or unclear risk of bias for each domain. We provided a quote from the study report together with a justification for our judgement in the ‘Risk of bias’ table. We resolved disagreements in our judgements by discussion in the review team.

We used the following nine standard criteria to assess bias for randomised and non‐randomised trials and controlled before‐after studies (EPOC 2015).

1. Was the allocation sequence adequately generated?
2. Was the allocation adequately concealed?
3. Were baseline outcome measurements similar?
4. Were baseline characteristics similar?
5. Was the study adequately protected against contamination?
6. Were incomplete outcome data adequately addressed?
7. Was knowledge of the allocated interventions adequately prevented during the study?
8. Was the study free from selective outcome reporting?
9. Was the study free from other risks of bias?

We assessed selective reporting through outcome reporting bias (Kirkham 2010) by integrating the data of the ‘Matrix of study endpoints (trials register)’ (Appendix 2), the outcomes reported in the publication and the results of the ‘modified ORBIT study classification table’ (Appendix 3).

We summarised the ‘Risk of bias’ judgements across different studies for each of the domains listed. Where information on risk of bias related to unpublished data or correspondence with a trialist, we noted this in the ‘Risk of bias’ table. When considering treatment effects, we took into account the risks of bias for the studies that contributed to that outcome.

The risk of bias judgement contributed to the GRADE process for recommendations in the data synthesis.

Measures of treatment effect

We estimated the effect of the intervention using risk ratios (RRs) with their 95% confidence intervals (CIs) for dichotomous data, and standardised mean differences (SMDs) with their 95% CIs in case of continuous data. We ensured that an increase in scores for continuous outcomes could be interpreted in the same way for each outcome, explaining the direction to the reader and reporting where the directions were reversed if this proved necessary. We reported the outcomes satisfaction with care and satisfaction with ethics consultation only descriptively using absolute and relative frequencies for the following reasons: for satisfaction with care, there were no reported absolute frequencies for the control and intervention group. For satisfaction of ethics consultation, only data for the intervention group was reported, since the control intervention was usual care.

Unit of analysis issues

All the included studies were parallel randomised trials, where participants were individually allocated to the treatment or control groups.

Dealing with missing data

We contacted investigators by e‐mail to verify key study characteristics and obtained missing outcome data where possible. We sent four reminders. We evaluated important numerical data such as screened, randomised participants and intention‐to‐treat, as‐treated and per‐protocol populations. We also investigated attrition rates. If data were still missing, we calculated standard deviations (Higgins 2003; Wan 2014), and reported them as such. The calculation was based on the underlying assumption that the standard deviations for both groups were identical. We used those data for analysis, if results were not biased by that fact.

Assessment of heterogeneity

We assessed heterogeneity using the I² statistic. I² values of 25% corresponded to low, 50% to medium and 75% to high levels of heterogeneity (Higgins 2003).

Assessment of reporting biases

We contacted study authors for missing outcome data. Where this was not possible, and the missing data were thought to introduce serious bias, we explored the impact of including such studies in the overall assessment of results by a sensitivity analysis.

Data synthesis

We attempted pooling only for treatment costs and mortality, because all other outcomes were heterogeneously reported. We used Review Manager 5 software for meta‐analyses (Review Manager 2014). We expected heterogeneity and variability between the studies due to different outcomes measures, interventions and population of the complex ethical case intervention. Therefore, we assumed that the true effect was related but not necessarily the same in different studies, and used a random‐effects model for analysis. We applied the Mantel‐Haenszel method for dichotomous outcomes and generic inverse variance for continuous outcomes.

We presented a synthesis of findings from the studies included and an assessment of the robustness of the synthesis using a structured analysis. We focused on active components of ethical case interventions, using the TiDieR rationale (Hoffmann 2014), as well as elements of the conceptual frameworks for ethical case interventions, which had been developed by the authors (Schildmann 2019).

Subgroup analysis and investigation of heterogeneity

We did not conduct subgroup analyses due to a lack of studies.

Sensitivity analysis

We performed sensitivity analysis for mortality with large studies only.