Provide a descriptive summary of the included studies

A descriptive summary of the included studies helps to ground the reader in what studies were included and confirm that the studies have similar PICOs to assist in determining whether it is possible to pool their results. We recommend structuring the descriptive summary of included studies around the PICO elements. Additional descriptions such as the dates of studies, locations or health systems in which the studies were conducted, study design(s) and ways in which outcomes were measured and reported should be included as appropriate to the RR question(s). Providing a summary in an evidence table or an alternative format such as a descriptive graphical display can help streamline RR production and presentation.

Evidence tables lay out the important characteristics of a study in a brief tabular form. Some evidence tables include results, or they may be presented in a separate table. Table 2 provides an example of the format of an evidence table that incorporates results, adapted from a recent RR of bariatric surgery.8 An RR by Uhl et al on the use of telehealth for substance use disorder treatment includes a table with details of the treated population, the telehealth intervention, the comparator and the number of studies and participants for each combination of these elements.9 Pisavadia et al conducted an RR of health economic evaluations of care for perinatal anxiety disorders that uses a table to map out how costs of maternal illness, interventions and comparators were evaluated across studies.10 As these three examples illustrate, there is not one correct form for evidence tables, with each constructed to meet the RR’s needs. Appendix K of the National Institute for Health and Care Excellence’s methods document for development of public health guidance provides evidence table templates for various types of studies and reviews.11 The Cochrane Handbook provides guidance related to summarising study characteristics, including using evidence tables to prepare for evidence synthesis.12

Perform a synthesis of the findings

An RR provides a narrative interpretation of the compiled evidence, with or without meta-analysis, to help users fully understand the collected evidence.1 Even when a meta-analysis is feasible and planned, the accompanying narrative synthesis can aid readers in comprehending the outcomes by interpreting and contextualising them rather than solely presenting numerical data. Campbell and colleagues study of SRs of public health interventions employing narrative synthesis of quantitative data found that most fall short in terms of describing the method and theory used, investigation of heterogeneity, and discussion of limitations.13 Their study gives an overview of best practices for narrative synthesis, particularly for complex interventions, based largely on the Economic and Social Research Council guidance by Popay and colleagues.14

Table 3 contains several examples of evidence synthesis statements for several outcomes from the RR of bariatric surgery by Durbin and colleagues, including statements based on meta-analyses as well as narrative summaries of study findings.8 A summary of findings (SOF) table, as described by GRADE (Grading of Recommendations, Assessment, Development and Evaluation), is also an approach to organise the quantitative findings of studies for each selected outcome, along with the finding’s certainty of evidence. The RR of telehealth interventions for substance use disorders by Uhl et al includes such an SOF table.9

The Cochrane Handbook includes a chapter regarding the development of SOF tables.15 While the Cochrane guidance was written with SRs in mind, the principles and formats can easily be adapted for RRs and another Cochrane RRMG article in this series specifically addresses their application of GRADE and certainty of evidence rating for RRs.7

If a meta-analysis is not possible RR authors may also consider using the Synthesis Without Meta-analysis (SWiM) reporting guideline as a guide to help promote transparency and completeness in the narrative presentation of evidence synthesis.16 The SWiM guidelines suggest that authors describe the synthesis of findings for each comparison and outcome, that the synthesis addresses each key question and specifies which included studies contributed to the findings.16

Consider a meta-analysis if appropriate and resources permit

If sufficient data, both in quantity and quality, are available, then a meta-analysis is often the most useful and efficient way of providing data synthesis in an RR. Meta-analysis may be inappropriate not only when there are insufficient studies for a given comparison, but also in situations where, despite having several eligible studies, factors such as heterogeneity in study design, PICO elements or concerns about the quality and consistency of the data, may render it an inappropriate synthesis method.17 The standards for conducting an SR meta-analyses generally apply to RRs, and authors should consult the Cochrane Handbook for the full details regarding meta-analytic techniques.17 When performed, meta-analysis for an RR can be streamlined by conducting only the pooled analyses on comparisons most critical for the decision-making process. Single reviewer data entry with checking by another reviewer is another common shortcut used in RRs.

Consider how to synthesise evidence when including one or more SRs

Some RRs include only primary studies and others may incorporate additional primary studies with one or more existing SRs. Overviews of SRs without additional primary studies are not included in our RR guidance but are addressed in the Cochrane Handbook chapter on overviews of reviews.18 When one or more SRs are available to help address an RR question, the selection criteria for incorporating SRs may include using the most recent, most methodologically robust, or most comprehensive, of the SRs most closely matching the RR research question. We recommend that if multiple SRs are included, the RR also considers the overlap of any primary studies to avoid statistical errors from double counting in meta-analysis. A formal study of overlap does not need to be undertaken, overlapping studies may be presented in a table that identifies which are included in each SR.

When SRs are included in an RR, synthesis may involve adding new primary studies to an existing meta-analysis. Typically, once a primary SR is identified that matches the research question then primary studies that meet the inclusion criteria and were published after the search date of the included SR are added for meta-analyses presented in the RR.

If an updated meta-analysis incorporating new primary studies is planned, it is usually necessary to abstract data from each study in the SR for entry into the software package along with data from the new primary studies. If the included SR does not use meta-analysis, authors should assess whether meta-analysis is now appropriate, combining primary studies from the SR and the additional studies. When it is not possible or feasible to conduct a new meta-analysis then the narrative synthesis should provide an interpretation of the additional studies in the context of the existing SR, exploring the similarities and differences in their PICO elements, settings, study designs and results.