Published
The SWiM reporting guideline is the checklist for the part of a systematic review where results cannot be pooled into one effect estimate. Campbell and colleagues published it in the BMJ in 2020 as nine items that extend PRISMA, covering how studies are grouped, which common metric is used, which method turns results into a summary, how data are shown, and what limits the conclusions [1]. This guide explains each item in plain language, shows where vote counting goes wrong, and applies the nine items row by row to a real open-access review from 2026 that states it followed SWiM [2]. If your studies can be pooled, start with our guides to how to conduct a meta-analysis and heterogeneity assessment instead.
In Simple Words
SWiM is a short checklist for reviews that cannot calculate one pooled number. You tell the reader how you grouped the studies, which common metric you used (for example direction of effect), which method produced the summary, how you displayed the data, how sure you are, and what limits the answer. The aim is that another person could repeat your synthesis from your Methods and Results. Writing “narrative synthesis” with no detail is the habit SWiM was built to stop.
What the SWiM Reporting Guideline Covers and How It Was Built
SWiM stands for Synthesis Without Meta-analysis. The SWiM reporting guideline is meant for systematic reviews of the quantitative effects of interventions where meta-analysis of effect estimates is not possible, or not appropriate, for at least some outcomes [1]. Common reasons include incompletely reported effect estimates and study designs, interventions or outcomes that differ too much to give a meaningful summary [1]. It is not meant for qualitative evidence, which has its own guidelines such as ENTREQ and eMERGe [1].
The guideline project was registered with the EQUATOR Network. The team assessed current reporting practice, drafted an item list with an advisory group, and ran a three-round Delphi survey. Of 91 methodologists and review authors invited, 44 responded in round one (48%), 37 of 68 in round two (54%) and 32 of 39 in round three (82%) [1]. An expert panel then met, and eight systematic reviewers who had not taken part in the Delphi tried the draft before the items were finalised [1]. The open-access full text is on PubMed Central.
For an author, the practical point is that the SWiM reporting guideline does not replace PRISMA. The guideline authors describe it as a complement and extension, and they note that PRISMA gives limited guidance on how to present and synthesise results when meta-analysis is not used [1][6]. Keep your PRISMA 2020 checklist and add the SWiM rows beside the synthesis items.
When the SWiM Reporting Guideline Applies Instead of a Pooled Estimate
The question your synthesis answers decides the method, and the method decides what you must report. The first three columns paraphrase the guideline authors’ summary of synthesis methods and the minimum data each one needs [1], and the last column is our own mapping to the SWiM items.
| Question the synthesis answers | Method | Minimum data needed | Where to go |
|---|---|---|---|
| What is the average or common intervention effect? | Meta-analysis of effect estimates and its extensions | Effect estimate and variance | Pool the estimates and report them under PRISMA 2020. See our meta-analysis guide above. |
| What is the range and spread of the observed effects? | Summarising effect estimates, such as the median and interquartile range | Effect estimates | SWiM items 2, 3, 7 and 8 apply. |
| Is there evidence of an effect in at least one study? | Combining P values | Direction of effect and a precise P value | SWiM items 2, 3, 7 and 8 apply. |
| Is there any evidence of an effect? | Vote counting based on direction of effect | Direction of effect only | SWiM items 2, 3, 7, 8 and 9 apply, and item 9 must say the question is weaker. |
Vote counting answers the weakest of the four questions, and SWiM item 9 asks authors to say so. A direction-based count tells the reader whether there is any evidence of an effect, not how large an average effect is [1]. If you can pool some outcomes and not others, the SWiM reporting guideline still applies to the outcomes you could not pool. Our meta-analysis worked examples and systematic review versus meta-analysis guides cover the pooled side of that split.
The Nine Items of the SWiM Reporting Guideline Explained
The items of the SWiM reporting guideline sit in three parts of the paper. Items 1 to 7 belong in Methods, item 8 in Results and item 9 in Discussion. Item 1 has two parts, 1a and 1b, so a completed checklist has ten rows [1]. The table gives our plain-language version of each.
| Item | Section | What to report | Common slip to avoid |
|---|---|---|---|
| 1a | Methods | The groups used in the synthesis (populations, interventions, outcomes, designs) and the reason for them | Groups appear in the Results but are never justified |
| 1b | Methods | Any change to those groups after the protocol, and why | Groups shift after the data are seen, with no note |
| 2 | Methods | The standardised metric for each outcome, why it was chosen, and how study results were converted to it | “Positive” and “negative” are never defined |
| 3 | Methods | The method used to synthesise each outcome, and why meta-analysis was not possible | The method is called “narrative synthesis” and nothing more |
| 4 | Methods | Criteria used to prioritise studies for the main synthesis, such as design, risk of bias or directness | Priority studies are picked after looking at results |
| 5 | Methods | How heterogeneity in reported effects was examined without meta-regression | Subgroup patterns are presented as findings with no note that they were post hoc |
| 6 | Methods | How the certainty of the synthesis findings was judged | Risk of bias is reported but certainty is not |
| 7 | Methods | The tables and graphs used, and the study characteristics used to order studies | Tables are ordered with no stated rule |
| 8 | Results | For each comparison and outcome, the finding, its certainty, and which studies contribute | Counts are given without the list of contributing studies |
| 9 | Discussion | Limits of the metric, method and groupings, and how they affect the conclusions | A generic limitations paragraph that never mentions the synthesis method |
Items 2 and 3 work as a pair
The SWiM reporting guideline treats the standardised metric as the common currency of the synthesis. Examples include risk ratios, mean differences, direction of effect and P values [1]. The Cochrane Handbook adds that every alternative method begins with a standardised metric, that only one outcome per study should contribute to a given synthesis, and that authors should report the specific methods used in place of meta-analysis rather than stating that they did a narrative synthesis [3]. In practice, item 2 is a definition and item 3 is a justification. If you cannot state the metric in one sentence, the synthesis method is not ready.
Item 5 and item 6 need honest wording
Informal heterogeneity checks, such as ordering a table by a suspected modifier, are acceptable when formal subgroup analysis is not possible, but the guideline warns that they are rarely definitive and asks authors to say if they were not pre-specified [1]. For certainty, GRADE is the most widely used framework, although some domains, such as consistency of effects under vote counting, are hard to judge [1]. Our guides to GRADE certainty of evidence and the RoB 2 tool cover the inputs. Whatever you use, name it, because a reader cannot credit a certainty judgement that has no stated method.
Vote Counting and Effect Direction Without the Usual Traps
Vote counting is the method most often done badly, and the SWiM reporting guideline expects the rule to be stated. The Cochrane Handbook says that rules based on statistical significance are problematic. Each effect should first be classed as benefit or harm from its observed direction alone, and in the Handbook’s words, “Neither statistical significance nor the size of the effect are considered in the categorization.” [3]
After the direction of each study is recorded, a sign test asks whether the proportion of studies favouring the intervention differs from 0.5, and the proportion can carry a confidence interval such as the Wilson interval [3]. Boon and Thomson revised the effect direction plot on this basis. They removed statistical significance from the algorithm for combining outcomes within a study, considered a sign test, and warned against over-reading apparent patterns when few studies are included [4]. Ogilvie and colleagues proposed the harvest plot in 2008 as a way to synthesise evidence about the differential effects of interventions when meta-analysis is not always suitable [5].
A short worked calculation shows why the counting rule matters. In the review analysed below, Table 5 gives overall directions of 10 positive, 0 negative and 3 mixed across 13 studies. If the three mixed studies are set aside as ties, a two-sided sign test on 10 of 10 gives P = 0.002, and the Wilson 95% interval for 10 of 10 runs from 72% to 100%. If the three mixed studies are counted as not favouring the intervention, the result is 10 of 13, or 77%, with a Wilson interval from 50% to 92%. The data are the same and both rules are defensible, yet the precision differs widely. This is why item 2 asks authors to write the rule down. These figures are Editverse arithmetic for illustration, and the review’s authors did not run a sign test.
You can build a quick display of your own directions with GraphPro, and the calculators in StatPro cover the sign test inputs.
SWiM Reporting Guideline Worked Example on an Open-Access Review
We chose Gao, Qi and Yu, published in Frontiers in Public Health in 2026, a systematic review of neuromuscular training programmes for anterior cruciate ligament injury prevention in female athletes [2]. The review includes 13 studies (n = 2,847), searched PubMed, SPORTDiscus, Scopus and Web of Science for 2014 to 2025, was registered on INPLASY on 14 June 2025, and states that it followed SWiM [2]. The article is licensed CC BY 4.0, which we confirmed from the Crossref licence record for the version of record, the Europe PMC licence field and the JATS permissions in the full text. We read the full text and every table, then pointed each row of the SWiM reporting guideline to the article’s own section names. Each verdict rates the reporting, not the science.
| SWiM item | Where to look | What the article reports | Verdict |
|---|---|---|---|
| 1a | Methods, Data synthesis approach; SWiM framework methodology (item 1) | Studies are grouped by implementation research focus (implementation-focused studies, efficacy trials with secondary implementation data, real-world effectiveness studies), with secondary groupings by strategy type, setting, population and geography. The authors justify not pooling with five sources of heterogeneity, “Narrative synthesis was selected over meta-analysis due to substantial heterogeneity observed in the implementation research evidence base” (Study design and reporting). | Reported |
| 1b | Methods, Protocol registration; Results, Table 7 | The article states, “There were no deviations from the registered protocol.” The Methods list four strategy types as secondary groupings, while Table 7 sorts studies into six strategy categories with different names, for example “Self-directed/minimal supervision”. The text does not say whether the Table 7 labels were in the protocol. | Judgement call No deviation is declared, but a reader cannot trace the six Table 7 groups back to the listed groupings. |
| 2 | Methods, Data synthesis approach; Standardized metrics development (item 2) | Compliance and adherence are put on 0 to 100% scales. Direction is classed as positive, negative or no effect/unclear using pre-specified thresholds, namely p < 0.05, or at least 10% absolute difference in compliance, or at least 20% relative risk reduction in injury. A separate paragraph, Clinical threshold justification, calls these pragmatic benchmarks. | Reported |
| 3 | Methods, Study design and reporting; Synthesis methods protocol (item 3) | Vote counting by direction of effect for implementation outcomes, medians, ranges and interquartile ranges for compliance, and thematic analysis within the CFIR framework for qualitative data. Five reasons for not meta-analysing are listed. | Reported |
| 4 | Methods, Data synthesis approach; Quality assessment | Studies with quantitative implementation measures were prioritised for structured numeric synthesis and process data for thematic analysis. No study was excluded for quality, and a high-quality-only analysis (8 high, 5 moderate MMAT ratings) served as a sensitivity check in Table 9. | Judgement call The criterion is the type of data, not a rule for selecting a main-synthesis subset. All 13 studies feed the main synthesis. |
| 5 | Methods, Heterogeneity investigation protocol; Results, Table 8 | The Methods name subgroup exploration by setting type, implementation strategy, population characteristics and geographic region. Table 8 compares professional support level, setting, session frequency, study quality, outcome type and follow-up duration. The Conclusion calls the frequency finding hypothesis-generating. | Judgement call Session frequency and follow-up duration are not in the Methods list, and the article does not say whether those comparisons were post hoc, which item 5 asks authors to state. |
| 6 | Methods, Quality assessment; Results, Sensitivity analysis | MMAT 2018 ratings feed “evidence certainty discussions in the narrative synthesis”. No GRADE or other certainty framework is named, and no certainty rating is given for any outcome. | Judgement call Whether a discussion counts as a method is a matter of reading. |
| 7 | Methods, Synthesis methods protocol; Results, Tables 5 to 9 | Tables 5 to 7 use arrows for direction, Table 8 sets out heterogeneity comparisons, and Table 9 splits results by quality. The Methods say “Visual synthesis methods were planned including harvest plots” and effect direction plots. The only figure in the full text is the PRISMA flow diagram, and Table 5 does not state the rule used to order studies. | Judgement call The arrow tables work as a display, but the planned plots are not shown and no ordering rule is given. |
| 8 | Results, Effect direction synthesis and heterogeneity assessment (Tables 5 to 8) | Each domain is reported with counts and cited studies, for example “Eight studies (62%) did not report these outcomes”. The counts differ between tables (see the next table), certainty is not stated per outcome, and one sentence concludes that training programmes “are not harmful when conducted”, while the Discussion later says the absence of negative directions is not evidence of safety. | Judgement call Contributing studies are cited, but the numbers cannot be rebuilt consistently. |
| 9 | Results, end of the effect direction section; Discussion, Limitations | The article states that vote counting “assigns equal weight regardless of sample size, cannot detect small effects, and may oversimplify complex patterns”, and the Limitations section adds interpretation bias, publication bias, selective reporting and the inability to establish causal relationships. | Reported |
Four of the ten rows are Reported (1a, 2, 3 and 9) and six are Judgement calls (1b, 4, 5, 6, 7 and 8). We did not mark any row as not reported. The strongest parts of the article are the explicit labels for items 1 to 3 in the Methods, the pre-specified thresholds and the frank statement of what vote counting cannot do.
Where the counts do not reconcile
SWiM item 8 depends on a reader being able to see which studies produce each count. We recounted every vote from the article’s own Table 5 and compared it with the other places where the same numbers appear. The compliance counts reconcile (12 positive and 1 mixed). The gaps sit in the biomechanical, injury and subgroup rows.
| What is counted | What the article states | What Table 5 gives when recounted |
|---|---|---|
| Studies with a positive biomechanical direction | Table 6 gives 6 of 13 (46%), while the text says “Five studies evaluated biomechanical outcomes” | 5 of 13 (Zebis, Hopper, Ford, Shams, Ferri-Caruana) |
| Studies with a positive injury direction | Table 6 gives 6 of 13, the text says five studies reported injury outcomes, and the abstract says broader injury outcomes were reported in 6 | 5 of 13 (Zebis, Yarsiasat, Foss, Suits, Schoeb) |
| Studies marked no effect for biomechanical and for injury outcomes | Table 6 gives 7 of 13 for each | 8 of 13 for each |
| Moderate-quality studies with a positive injury direction | Table 9 gives 2 of 5, and the text names studies 52 and 54 | 1 of 5 (Schoeb). Sugimoto (54) is marked no effect for injury |
| Studies in the session-frequency comparison | Table 8 gives 6 of 6 positive at two or more sessions a week plus 0 of 3 below, which is 9 studies with an injury direction | Only 5 studies are positive for injury, and the text says five reported injury outcomes |
| Studies in the professional-support comparison | Table 8 gives 12 of 13 positive for high support plus 1 of 2 mixed for minimal support, which is 15 studies | The review has 13 studies in total |
| Positive classifications under stricter thresholds | The Sensitivity analysis says stricter thresholds “reduced” positive classifications to 77% (10 of 13) | The main analysis already gives 10 of 13 (77%) in Table 6 |
A second pattern deserves a Judgement call label. The text says eight studies did not report adoption or reach outcomes, and Table 5 shows eight cells marked “No effect” in that column. The no-effect symbol therefore seems to carry two meanings, a measured null result and an outcome that was not measured. In a vote count these differ, because a study that did not measure an outcome should not enter the denominator. The authors may intend the symbol as shorthand, which is why we do not call it an error.
One more point belongs to method rather than reporting. The Cochrane Handbook advises that vote counting should not depend on statistical significance [3], whereas this review defined a positive direction by p < 0.05 or by size thresholds. The authors state the rule openly and test alternative thresholds in a sensitivity analysis, which is what items 2 and 5 ask for. A direction-only rule with a sign test would be the Handbook route, as in the calculation above.
How to Write Each SWiM Item in Your Own Methods
- Write the synthesis plan into the protocol before extraction and register it, using our PROSPERO registration guide.
- Name each group and give a one-line reason (item 1a). Add a sentence for any later change (item 1b).
- Define the metric in one sentence. If it is direction of effect, define benefit and harm by direction alone (item 2).
- Name the method for each outcome and say why pooling was not possible (item 3).
- State how studies were prioritised, if they were, and fix the rule before looking at results (item 4).
- List the heterogeneity comparisons, and mark any that were added after seeing the data (item 5).
- Choose the certainty framework and say how consistency was judged under your method (item 6).
- Pick the table or plot, state the ordering rule, and show only displays that you planned and built (item 7).
- In the Results, list the contributing studies for each count and keep one set of numbers across abstract, text and tables (item 8).
- In the Discussion, say which question the synthesis answers and how the metric and groupings limit the conclusions (item 9).
Common Mistakes When Reporting a Synthesis Without Meta-Analysis
Most gaps against the SWiM reporting guideline come from a short list of habits.
- Calling it narrative synthesis and stopping there. The Handbook asks for the specific method and its limits [3].
- Counting by statistical significance. Direction alone is the accepted basis for vote counting [3].
- Letting not reported mean no effect. Studies that did not measure an outcome should not sit in the denominator.
- Letting several outcomes from one study enter the same count. One outcome per study is the rule [3].
- Omitting item 6. A risk-of-bias table is not a certainty judgement.
- Announcing plots you never show. If the Methods promise a harvest plot, the Results need one.
- Reading harm into absent negative votes. No negative direction is not evidence of safety.
- Letting counts drift between abstract, text and tables. Recount before submission.
Frequently Asked Questions About the SWiM Reporting Guideline
How many items does the SWiM reporting guideline have?
The SWiM reporting guideline has nine items, and item 1 is split into 1a and 1b, so a completed checklist has ten rows [1].
Does SWiM replace PRISMA?
No. The SWiM reporting guideline is intended to be used alongside PRISMA and other guidelines that cover the rest of the review [1].
Can authors use SWiM if they pooled some outcomes?
Yes. The SWiM reporting guideline applies when meta-analysis is not possible or not appropriate for at least some outcomes, and its authors note that the items are relevant to reporting quantitative synthesis whatever the method [1].
Is SWiM for qualitative evidence synthesis?
No. Qualitative syntheses use other guidelines, such as ENTREQ and eMERGe [1].
Free Research Tools from Editverse
- GraphPro to build publication-ready charts, 25 types, with built-in t-test, ANOVA, correlation, and chi-square p-values. Free, no signup.
- StatPro offers free statistical calculators and a test-selector tool covering group comparisons, categorical outcomes, time-to-event analysis, and diagnostic accuracy.
- Flowchart Pro generates PRISMA 2020, PRISMA-ScR, CONSORT, STROBE, STARD, MOOSE, and ARRIVE flow diagrams directly in your browser.
Professional Support for a Synthesis Without Meta-Analysis
The synthesis plan is easiest to defend when it is fixed in the protocol, and the written synthesis is easiest to check when every row of the SWiM reporting guideline has a pointer. Our work follows COPE, ICMJE and GPP 2022 guidance, and subject-matter experts at Editverse support authors at both stages.
Before you start
$500
PROSPERO-ready systematic review protocol with a PICO framework, eligibility criteria, a search strategy draft, methodology planning and risk of bias tool selection. 2 to 3 weeks.
Full review
$1,500
PRISMA-compliant systematic review for up to 15 included articles, with protocol development, a search across 5+ databases, dual screening, quality assessment and a publication-ready manuscript. 8 to 12 weeks.
Not sure whether your included studies can be pooled? Send us your protocol draft and your review question, and we will suggest a starting point.
Conclusion
The SWiM reporting guideline asks ten short questions that decide whether a synthesis without meta-analysis can be trusted and repeated. State the groups, the metric and the method, say how you ordered and displayed the data, name how you judged certainty, and be plain about what a direction-based count can and cannot show. The worked example shows that even a review that labels every early item well can leave gaps at items 1b and 4 to 8, and that one set of recounted numbers is the cheapest way to close them. Use the SWiM reporting guideline while you write, not after, and recount before you submit.
References
- Campbell M, McKenzie JE, Sowden A, Katikireddi SV, Brennan SE, Ellis S, et al. Synthesis without meta-analysis (SWiM) in systematic reviews: reporting guideline. BMJ. 2020;368:l6890. doi:10.1136/bmj.l6890. PMID 31948937. PMCID PMC7190266.
- Gao Y, Qi L, Yu T. Implementation effectiveness, barriers, and real-world outcomes of neuromuscular training programs for ACL injury prevention in female athletes: systematic review with narrative synthesis using SWiM framework. Front Public Health. 2026;14:1743075. doi:10.3389/fpubh.2026.1743075. PMID 41815969. PMCID PMC12971891. Registered as INPLASY202560057. CC BY 4.0.
- McKenzie JE, Brennan SE. Chapter 12: Synthesizing and presenting findings using other methods. In: Higgins JPT, Thomas J, Chandler J, Cumpston M, Li T, Page MJ, et al., editors. Cochrane Handbook for Systematic Reviews of Interventions. Version 6.5, 2024. Chapter last updated October 2019. cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-12.
- Boon MH, Thomson H. The effect direction plot revisited: application of the 2019 Cochrane Handbook guidance on alternative synthesis methods. Res Synth Methods. 2021;12(1):29-33. doi:10.1002/jrsm.1458. PMID 32979023.
- Ogilvie D, Fayter D, Petticrew M, Sowden A, Thomas S, Whitehead M, et al. The harvest plot: a method for synthesising evidence about the differential effects of interventions. BMC Med Res Methodol. 2008;8:8. doi:10.1186/1471-2288-8-8. PMID 18298827.
- Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71. PMID 33782057.